]>
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 | ||
f0250618 ML |
113 | /* |
114 | * 'strct resync_pages' stores actual pages used for doing the resync | |
115 | * IO, and it is per-bio, so make .bi_private points to it. | |
116 | */ | |
117 | static inline struct resync_pages *get_resync_pages(struct bio *bio) | |
118 | { | |
119 | return bio->bi_private; | |
120 | } | |
121 | ||
122 | /* | |
123 | * for resync bio, r10bio pointer can be retrieved from the per-bio | |
124 | * 'struct resync_pages'. | |
125 | */ | |
126 | static inline struct r10bio *get_resync_r10bio(struct bio *bio) | |
127 | { | |
128 | return get_resync_pages(bio)->raid_bio; | |
129 | } | |
130 | ||
dd0fc66f | 131 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 132 | { |
e879a879 | 133 | struct r10conf *conf = data; |
9f2c9d12 | 134 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 135 | |
69335ef3 N |
136 | /* allocate a r10bio with room for raid_disks entries in the |
137 | * bios array */ | |
7eaceacc | 138 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
139 | } |
140 | ||
141 | static void r10bio_pool_free(void *r10_bio, void *data) | |
142 | { | |
143 | kfree(r10_bio); | |
144 | } | |
145 | ||
0310fa21 N |
146 | /* amount of memory to reserve for resync requests */ |
147 | #define RESYNC_WINDOW (1024*1024) | |
148 | /* maximum number of concurrent requests, memory permitting */ | |
149 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
150 | |
151 | /* | |
152 | * When performing a resync, we need to read and compare, so | |
153 | * we need as many pages are there are copies. | |
154 | * When performing a recovery, we need 2 bios, one for read, | |
155 | * one for write (we recover only one drive per r10buf) | |
156 | * | |
157 | */ | |
dd0fc66f | 158 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 159 | { |
e879a879 | 160 | struct r10conf *conf = data; |
9f2c9d12 | 161 | struct r10bio *r10_bio; |
1da177e4 | 162 | struct bio *bio; |
f0250618 ML |
163 | int j; |
164 | int nalloc, nalloc_rp; | |
165 | struct resync_pages *rps; | |
1da177e4 LT |
166 | |
167 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 168 | if (!r10_bio) |
1da177e4 | 169 | return NULL; |
1da177e4 | 170 | |
3ea7daa5 N |
171 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || |
172 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
1da177e4 LT |
173 | nalloc = conf->copies; /* resync */ |
174 | else | |
175 | nalloc = 2; /* recovery */ | |
176 | ||
f0250618 ML |
177 | /* allocate once for all bios */ |
178 | if (!conf->have_replacement) | |
179 | nalloc_rp = nalloc; | |
180 | else | |
181 | nalloc_rp = nalloc * 2; | |
182 | rps = kmalloc(sizeof(struct resync_pages) * nalloc_rp, gfp_flags); | |
183 | if (!rps) | |
184 | goto out_free_r10bio; | |
185 | ||
1da177e4 LT |
186 | /* |
187 | * Allocate bios. | |
188 | */ | |
189 | for (j = nalloc ; j-- ; ) { | |
6746557f | 190 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
191 | if (!bio) |
192 | goto out_free_bio; | |
193 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
194 | if (!conf->have_replacement) |
195 | continue; | |
196 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
197 | if (!bio) | |
198 | goto out_free_bio; | |
199 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
200 | } |
201 | /* | |
202 | * Allocate RESYNC_PAGES data pages and attach them | |
203 | * where needed. | |
204 | */ | |
f0250618 | 205 | for (j = 0; j < nalloc; j++) { |
69335ef3 | 206 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
f0250618 ML |
207 | struct resync_pages *rp, *rp_repl; |
208 | ||
209 | rp = &rps[j]; | |
210 | if (rbio) | |
211 | rp_repl = &rps[nalloc + j]; | |
212 | ||
1da177e4 | 213 | bio = r10_bio->devs[j].bio; |
f0250618 ML |
214 | |
215 | if (!j || test_bit(MD_RECOVERY_SYNC, | |
216 | &conf->mddev->recovery)) { | |
217 | if (resync_alloc_pages(rp, gfp_flags)) | |
1da177e4 | 218 | goto out_free_pages; |
f0250618 ML |
219 | } else { |
220 | memcpy(rp, &rps[0], sizeof(*rp)); | |
221 | resync_get_all_pages(rp); | |
222 | } | |
1da177e4 | 223 | |
f0250618 ML |
224 | rp->raid_bio = r10_bio; |
225 | bio->bi_private = rp; | |
226 | if (rbio) { | |
227 | memcpy(rp_repl, rp, sizeof(*rp)); | |
228 | rbio->bi_private = rp_repl; | |
1da177e4 LT |
229 | } |
230 | } | |
231 | ||
232 | return r10_bio; | |
233 | ||
234 | out_free_pages: | |
f0250618 ML |
235 | while (--j >= 0) |
236 | resync_free_pages(&rps[j * 2]); | |
237 | ||
5fdd2cf8 | 238 | j = 0; |
1da177e4 | 239 | out_free_bio: |
5fdd2cf8 | 240 | for ( ; j < nalloc; j++) { |
241 | if (r10_bio->devs[j].bio) | |
242 | bio_put(r10_bio->devs[j].bio); | |
69335ef3 N |
243 | if (r10_bio->devs[j].repl_bio) |
244 | bio_put(r10_bio->devs[j].repl_bio); | |
245 | } | |
f0250618 ML |
246 | kfree(rps); |
247 | out_free_r10bio: | |
1da177e4 LT |
248 | r10bio_pool_free(r10_bio, conf); |
249 | return NULL; | |
250 | } | |
251 | ||
252 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
253 | { | |
e879a879 | 254 | struct r10conf *conf = data; |
9f2c9d12 | 255 | struct r10bio *r10bio = __r10_bio; |
1da177e4 | 256 | int j; |
f0250618 | 257 | struct resync_pages *rp = NULL; |
1da177e4 | 258 | |
f0250618 | 259 | for (j = conf->copies; j--; ) { |
1da177e4 | 260 | struct bio *bio = r10bio->devs[j].bio; |
f0250618 ML |
261 | |
262 | rp = get_resync_pages(bio); | |
263 | resync_free_pages(rp); | |
264 | bio_put(bio); | |
265 | ||
69335ef3 N |
266 | bio = r10bio->devs[j].repl_bio; |
267 | if (bio) | |
268 | bio_put(bio); | |
1da177e4 | 269 | } |
f0250618 ML |
270 | |
271 | /* resync pages array stored in the 1st bio's .bi_private */ | |
272 | kfree(rp); | |
273 | ||
1da177e4 LT |
274 | r10bio_pool_free(r10bio, conf); |
275 | } | |
276 | ||
e879a879 | 277 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
278 | { |
279 | int i; | |
280 | ||
281 | for (i = 0; i < conf->copies; i++) { | |
282 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 283 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
284 | bio_put(*bio); |
285 | *bio = NULL; | |
69335ef3 N |
286 | bio = &r10_bio->devs[i].repl_bio; |
287 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
288 | bio_put(*bio); | |
289 | *bio = NULL; | |
1da177e4 LT |
290 | } |
291 | } | |
292 | ||
9f2c9d12 | 293 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 294 | { |
e879a879 | 295 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 296 | |
1da177e4 LT |
297 | put_all_bios(conf, r10_bio); |
298 | mempool_free(r10_bio, conf->r10bio_pool); | |
299 | } | |
300 | ||
9f2c9d12 | 301 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 302 | { |
e879a879 | 303 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
304 | |
305 | mempool_free(r10_bio, conf->r10buf_pool); | |
306 | ||
0a27ec96 | 307 | lower_barrier(conf); |
1da177e4 LT |
308 | } |
309 | ||
9f2c9d12 | 310 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
311 | { |
312 | unsigned long flags; | |
fd01b88c | 313 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 314 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
315 | |
316 | spin_lock_irqsave(&conf->device_lock, flags); | |
317 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 318 | conf->nr_queued ++; |
1da177e4 LT |
319 | spin_unlock_irqrestore(&conf->device_lock, flags); |
320 | ||
388667be AJ |
321 | /* wake up frozen array... */ |
322 | wake_up(&conf->wait_barrier); | |
323 | ||
1da177e4 LT |
324 | md_wakeup_thread(mddev->thread); |
325 | } | |
326 | ||
327 | /* | |
328 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
329 | * operation and are ready to return a success/failure code to the buffer | |
330 | * cache layer. | |
331 | */ | |
9f2c9d12 | 332 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
333 | { |
334 | struct bio *bio = r10_bio->master_bio; | |
e879a879 | 335 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 336 | |
856e08e2 | 337 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) |
4e4cbee9 | 338 | bio->bi_status = BLK_STS_IOERR; |
fd16f2e8 N |
339 | |
340 | bio_endio(bio); | |
341 | /* | |
342 | * Wake up any possible resync thread that waits for the device | |
343 | * to go idle. | |
344 | */ | |
345 | allow_barrier(conf); | |
346 | ||
1da177e4 LT |
347 | free_r10bio(r10_bio); |
348 | } | |
349 | ||
350 | /* | |
351 | * Update disk head position estimator based on IRQ completion info. | |
352 | */ | |
9f2c9d12 | 353 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 354 | { |
e879a879 | 355 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
356 | |
357 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
358 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
359 | } | |
360 | ||
778ca018 NK |
361 | /* |
362 | * Find the disk number which triggered given bio | |
363 | */ | |
e879a879 | 364 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 365 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
366 | { |
367 | int slot; | |
69335ef3 | 368 | int repl = 0; |
778ca018 | 369 | |
69335ef3 | 370 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
371 | if (r10_bio->devs[slot].bio == bio) |
372 | break; | |
69335ef3 N |
373 | if (r10_bio->devs[slot].repl_bio == bio) { |
374 | repl = 1; | |
375 | break; | |
376 | } | |
377 | } | |
778ca018 NK |
378 | |
379 | BUG_ON(slot == conf->copies); | |
380 | update_head_pos(slot, r10_bio); | |
381 | ||
749c55e9 N |
382 | if (slotp) |
383 | *slotp = slot; | |
69335ef3 N |
384 | if (replp) |
385 | *replp = repl; | |
778ca018 NK |
386 | return r10_bio->devs[slot].devnum; |
387 | } | |
388 | ||
4246a0b6 | 389 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 390 | { |
4e4cbee9 | 391 | int uptodate = !bio->bi_status; |
9f2c9d12 | 392 | struct r10bio *r10_bio = bio->bi_private; |
1da177e4 | 393 | int slot, dev; |
abbf098e | 394 | struct md_rdev *rdev; |
e879a879 | 395 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 396 | |
1da177e4 LT |
397 | slot = r10_bio->read_slot; |
398 | dev = r10_bio->devs[slot].devnum; | |
abbf098e | 399 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
400 | /* |
401 | * this branch is our 'one mirror IO has finished' event handler: | |
402 | */ | |
4443ae10 N |
403 | update_head_pos(slot, r10_bio); |
404 | ||
405 | if (uptodate) { | |
1da177e4 LT |
406 | /* |
407 | * Set R10BIO_Uptodate in our master bio, so that | |
408 | * we will return a good error code to the higher | |
409 | * levels even if IO on some other mirrored buffer fails. | |
410 | * | |
411 | * The 'master' represents the composite IO operation to | |
412 | * user-side. So if something waits for IO, then it will | |
413 | * wait for the 'master' bio. | |
414 | */ | |
415 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
416 | } else { |
417 | /* If all other devices that store this block have | |
418 | * failed, we want to return the error upwards rather | |
419 | * than fail the last device. Here we redefine | |
420 | * "uptodate" to mean "Don't want to retry" | |
421 | */ | |
635f6416 N |
422 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
423 | rdev->raid_disk)) | |
fae8cc5e | 424 | uptodate = 1; |
fae8cc5e N |
425 | } |
426 | if (uptodate) { | |
1da177e4 | 427 | raid_end_bio_io(r10_bio); |
abbf098e | 428 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 429 | } else { |
1da177e4 | 430 | /* |
7c4e06ff | 431 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
432 | */ |
433 | char b[BDEVNAME_SIZE]; | |
08464e09 | 434 | pr_err_ratelimited("md/raid10:%s: %s: rescheduling sector %llu\n", |
8bda470e | 435 | mdname(conf->mddev), |
abbf098e | 436 | bdevname(rdev->bdev, b), |
8bda470e | 437 | (unsigned long long)r10_bio->sector); |
856e08e2 | 438 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
439 | reschedule_retry(r10_bio); |
440 | } | |
1da177e4 LT |
441 | } |
442 | ||
9f2c9d12 | 443 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
444 | { |
445 | /* clear the bitmap if all writes complete successfully */ | |
446 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
447 | r10_bio->sectors, | |
448 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
449 | 0); | |
450 | md_write_end(r10_bio->mddev); | |
451 | } | |
452 | ||
9f2c9d12 | 453 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
454 | { |
455 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
456 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
457 | reschedule_retry(r10_bio); | |
458 | else { | |
459 | close_write(r10_bio); | |
460 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
461 | reschedule_retry(r10_bio); | |
462 | else | |
463 | raid_end_bio_io(r10_bio); | |
464 | } | |
465 | } | |
466 | } | |
467 | ||
4246a0b6 | 468 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 469 | { |
9f2c9d12 | 470 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 471 | int dev; |
749c55e9 | 472 | int dec_rdev = 1; |
e879a879 | 473 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 474 | int slot, repl; |
4ca40c2c | 475 | struct md_rdev *rdev = NULL; |
1919cbb2 | 476 | struct bio *to_put = NULL; |
579ed34f SL |
477 | bool discard_error; |
478 | ||
4e4cbee9 | 479 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 480 | |
475b0321 | 481 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 482 | |
475b0321 N |
483 | if (repl) |
484 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
485 | if (!rdev) { |
486 | smp_rmb(); | |
487 | repl = 0; | |
475b0321 | 488 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 489 | } |
1da177e4 LT |
490 | /* |
491 | * this branch is our 'one mirror IO has finished' event handler: | |
492 | */ | |
4e4cbee9 | 493 | if (bio->bi_status && !discard_error) { |
475b0321 N |
494 | if (repl) |
495 | /* Never record new bad blocks to replacement, | |
496 | * just fail it. | |
497 | */ | |
498 | md_error(rdev->mddev, rdev); | |
499 | else { | |
500 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
501 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
502 | set_bit(MD_RECOVERY_NEEDED, | |
503 | &rdev->mddev->recovery); | |
1919cbb2 | 504 | |
475b0321 | 505 | dec_rdev = 0; |
1919cbb2 N |
506 | if (test_bit(FailFast, &rdev->flags) && |
507 | (bio->bi_opf & MD_FAILFAST)) { | |
508 | md_error(rdev->mddev, rdev); | |
509 | if (!test_bit(Faulty, &rdev->flags)) | |
510 | /* This is the only remaining device, | |
511 | * We need to retry the write without | |
512 | * FailFast | |
513 | */ | |
514 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
515 | else { | |
516 | r10_bio->devs[slot].bio = NULL; | |
517 | to_put = bio; | |
518 | dec_rdev = 1; | |
519 | } | |
520 | } else | |
521 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
475b0321 | 522 | } |
749c55e9 | 523 | } else { |
1da177e4 LT |
524 | /* |
525 | * Set R10BIO_Uptodate in our master bio, so that | |
526 | * we will return a good error code for to the higher | |
527 | * levels even if IO on some other mirrored buffer fails. | |
528 | * | |
529 | * The 'master' represents the composite IO operation to | |
530 | * user-side. So if something waits for IO, then it will | |
531 | * wait for the 'master' bio. | |
532 | */ | |
749c55e9 N |
533 | sector_t first_bad; |
534 | int bad_sectors; | |
535 | ||
3056e3ae AL |
536 | /* |
537 | * Do not set R10BIO_Uptodate if the current device is | |
538 | * rebuilding or Faulty. This is because we cannot use | |
539 | * such device for properly reading the data back (we could | |
540 | * potentially use it, if the current write would have felt | |
541 | * before rdev->recovery_offset, but for simplicity we don't | |
542 | * check this here. | |
543 | */ | |
544 | if (test_bit(In_sync, &rdev->flags) && | |
545 | !test_bit(Faulty, &rdev->flags)) | |
546 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 547 | |
749c55e9 | 548 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 549 | if (is_badblock(rdev, |
749c55e9 N |
550 | r10_bio->devs[slot].addr, |
551 | r10_bio->sectors, | |
579ed34f | 552 | &first_bad, &bad_sectors) && !discard_error) { |
749c55e9 | 553 | bio_put(bio); |
475b0321 N |
554 | if (repl) |
555 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
556 | else | |
557 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
558 | dec_rdev = 0; |
559 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
560 | } | |
561 | } | |
562 | ||
1da177e4 LT |
563 | /* |
564 | * | |
565 | * Let's see if all mirrored write operations have finished | |
566 | * already. | |
567 | */ | |
19d5f834 | 568 | one_write_done(r10_bio); |
749c55e9 | 569 | if (dec_rdev) |
884162df | 570 | rdev_dec_pending(rdev, conf->mddev); |
1919cbb2 N |
571 | if (to_put) |
572 | bio_put(to_put); | |
1da177e4 LT |
573 | } |
574 | ||
1da177e4 LT |
575 | /* |
576 | * RAID10 layout manager | |
25985edc | 577 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
578 | * parameters: near_copies and far_copies. |
579 | * near_copies * far_copies must be <= raid_disks. | |
580 | * Normally one of these will be 1. | |
581 | * If both are 1, we get raid0. | |
582 | * If near_copies == raid_disks, we get raid1. | |
583 | * | |
25985edc | 584 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
585 | * first chunk, followed by near_copies copies of the next chunk and |
586 | * so on. | |
587 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
588 | * as described above, we start again with a device offset of near_copies. | |
589 | * So we effectively have another copy of the whole array further down all | |
590 | * the drives, but with blocks on different drives. | |
591 | * With this layout, and block is never stored twice on the one device. | |
592 | * | |
593 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 594 | * on each device that it is on. |
1da177e4 LT |
595 | * |
596 | * raid10_find_virt does the reverse mapping, from a device and a | |
597 | * sector offset to a virtual address | |
598 | */ | |
599 | ||
f8c9e74f | 600 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
601 | { |
602 | int n,f; | |
603 | sector_t sector; | |
604 | sector_t chunk; | |
605 | sector_t stripe; | |
606 | int dev; | |
1da177e4 | 607 | int slot = 0; |
9a3152ab JB |
608 | int last_far_set_start, last_far_set_size; |
609 | ||
610 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
611 | last_far_set_start *= geo->far_set_size; | |
612 | ||
613 | last_far_set_size = geo->far_set_size; | |
614 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
615 | |
616 | /* now calculate first sector/dev */ | |
5cf00fcd N |
617 | chunk = r10bio->sector >> geo->chunk_shift; |
618 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 619 | |
5cf00fcd | 620 | chunk *= geo->near_copies; |
1da177e4 | 621 | stripe = chunk; |
5cf00fcd N |
622 | dev = sector_div(stripe, geo->raid_disks); |
623 | if (geo->far_offset) | |
624 | stripe *= geo->far_copies; | |
1da177e4 | 625 | |
5cf00fcd | 626 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
627 | |
628 | /* and calculate all the others */ | |
5cf00fcd | 629 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 630 | int d = dev; |
475901af | 631 | int set; |
1da177e4 | 632 | sector_t s = sector; |
1da177e4 | 633 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 634 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
635 | slot++; |
636 | ||
5cf00fcd | 637 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 638 | set = d / geo->far_set_size; |
5cf00fcd | 639 | d += geo->near_copies; |
475901af | 640 | |
9a3152ab JB |
641 | if ((geo->raid_disks % geo->far_set_size) && |
642 | (d > last_far_set_start)) { | |
643 | d -= last_far_set_start; | |
644 | d %= last_far_set_size; | |
645 | d += last_far_set_start; | |
646 | } else { | |
647 | d %= geo->far_set_size; | |
648 | d += geo->far_set_size * set; | |
649 | } | |
5cf00fcd | 650 | s += geo->stride; |
1da177e4 LT |
651 | r10bio->devs[slot].devnum = d; |
652 | r10bio->devs[slot].addr = s; | |
653 | slot++; | |
654 | } | |
655 | dev++; | |
5cf00fcd | 656 | if (dev >= geo->raid_disks) { |
1da177e4 | 657 | dev = 0; |
5cf00fcd | 658 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
659 | } |
660 | } | |
f8c9e74f N |
661 | } |
662 | ||
663 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
664 | { | |
665 | struct geom *geo = &conf->geo; | |
666 | ||
667 | if (conf->reshape_progress != MaxSector && | |
668 | ((r10bio->sector >= conf->reshape_progress) != | |
669 | conf->mddev->reshape_backwards)) { | |
670 | set_bit(R10BIO_Previous, &r10bio->state); | |
671 | geo = &conf->prev; | |
672 | } else | |
673 | clear_bit(R10BIO_Previous, &r10bio->state); | |
674 | ||
675 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
676 | } |
677 | ||
e879a879 | 678 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
679 | { |
680 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
681 | /* Never use conf->prev as this is only called during resync |
682 | * or recovery, so reshape isn't happening | |
683 | */ | |
5cf00fcd | 684 | struct geom *geo = &conf->geo; |
475901af JB |
685 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
686 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
687 | int last_far_set_start; |
688 | ||
689 | if (geo->raid_disks % geo->far_set_size) { | |
690 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
691 | last_far_set_start *= geo->far_set_size; | |
692 | ||
693 | if (dev >= last_far_set_start) { | |
694 | far_set_size = geo->far_set_size; | |
695 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
696 | far_set_start = last_far_set_start; | |
697 | } | |
698 | } | |
1da177e4 | 699 | |
5cf00fcd N |
700 | offset = sector & geo->chunk_mask; |
701 | if (geo->far_offset) { | |
c93983bf | 702 | int fc; |
5cf00fcd N |
703 | chunk = sector >> geo->chunk_shift; |
704 | fc = sector_div(chunk, geo->far_copies); | |
705 | dev -= fc * geo->near_copies; | |
475901af JB |
706 | if (dev < far_set_start) |
707 | dev += far_set_size; | |
c93983bf | 708 | } else { |
5cf00fcd N |
709 | while (sector >= geo->stride) { |
710 | sector -= geo->stride; | |
475901af JB |
711 | if (dev < (geo->near_copies + far_set_start)) |
712 | dev += far_set_size - geo->near_copies; | |
c93983bf | 713 | else |
5cf00fcd | 714 | dev -= geo->near_copies; |
c93983bf | 715 | } |
5cf00fcd | 716 | chunk = sector >> geo->chunk_shift; |
c93983bf | 717 | } |
5cf00fcd N |
718 | vchunk = chunk * geo->raid_disks + dev; |
719 | sector_div(vchunk, geo->near_copies); | |
720 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
721 | } |
722 | ||
1da177e4 LT |
723 | /* |
724 | * This routine returns the disk from which the requested read should | |
725 | * be done. There is a per-array 'next expected sequential IO' sector | |
726 | * number - if this matches on the next IO then we use the last disk. | |
727 | * There is also a per-disk 'last know head position' sector that is | |
728 | * maintained from IRQ contexts, both the normal and the resync IO | |
729 | * completion handlers update this position correctly. If there is no | |
730 | * perfect sequential match then we pick the disk whose head is closest. | |
731 | * | |
732 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
733 | * because position is mirror, not device based. | |
734 | * | |
735 | * The rdev for the device selected will have nr_pending incremented. | |
736 | */ | |
737 | ||
738 | /* | |
739 | * FIXME: possibly should rethink readbalancing and do it differently | |
740 | * depending on near_copies / far_copies geometry. | |
741 | */ | |
96c3fd1f N |
742 | static struct md_rdev *read_balance(struct r10conf *conf, |
743 | struct r10bio *r10_bio, | |
744 | int *max_sectors) | |
1da177e4 | 745 | { |
af3a2cd6 | 746 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 747 | int disk, slot; |
856e08e2 N |
748 | int sectors = r10_bio->sectors; |
749 | int best_good_sectors; | |
56d99121 | 750 | sector_t new_distance, best_dist; |
3bbae04b | 751 | struct md_rdev *best_rdev, *rdev = NULL; |
56d99121 N |
752 | int do_balance; |
753 | int best_slot; | |
5cf00fcd | 754 | struct geom *geo = &conf->geo; |
1da177e4 LT |
755 | |
756 | raid10_find_phys(conf, r10_bio); | |
757 | rcu_read_lock(); | |
856e08e2 | 758 | sectors = r10_bio->sectors; |
56d99121 | 759 | best_slot = -1; |
abbf098e | 760 | best_rdev = NULL; |
56d99121 | 761 | best_dist = MaxSector; |
856e08e2 | 762 | best_good_sectors = 0; |
56d99121 | 763 | do_balance = 1; |
8d3ca83d | 764 | clear_bit(R10BIO_FailFast, &r10_bio->state); |
1da177e4 LT |
765 | /* |
766 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
767 | * device if no resync is going on (recovery is ok), or below |
768 | * the resync window. We take the first readable disk when | |
769 | * above the resync window. | |
1da177e4 LT |
770 | */ |
771 | if (conf->mddev->recovery_cp < MaxSector | |
56d99121 N |
772 | && (this_sector + sectors >= conf->next_resync)) |
773 | do_balance = 0; | |
1da177e4 | 774 | |
56d99121 | 775 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
776 | sector_t first_bad; |
777 | int bad_sectors; | |
778 | sector_t dev_sector; | |
779 | ||
56d99121 N |
780 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
781 | continue; | |
1da177e4 | 782 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
783 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
784 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
785 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
786 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 | 787 | if (rdev == NULL || |
8ae12666 | 788 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
789 | continue; |
790 | if (!test_bit(In_sync, &rdev->flags) && | |
791 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
792 | continue; |
793 | ||
856e08e2 N |
794 | dev_sector = r10_bio->devs[slot].addr; |
795 | if (is_badblock(rdev, dev_sector, sectors, | |
796 | &first_bad, &bad_sectors)) { | |
797 | if (best_dist < MaxSector) | |
798 | /* Already have a better slot */ | |
799 | continue; | |
800 | if (first_bad <= dev_sector) { | |
801 | /* Cannot read here. If this is the | |
802 | * 'primary' device, then we must not read | |
803 | * beyond 'bad_sectors' from another device. | |
804 | */ | |
805 | bad_sectors -= (dev_sector - first_bad); | |
806 | if (!do_balance && sectors > bad_sectors) | |
807 | sectors = bad_sectors; | |
808 | if (best_good_sectors > sectors) | |
809 | best_good_sectors = sectors; | |
810 | } else { | |
811 | sector_t good_sectors = | |
812 | first_bad - dev_sector; | |
813 | if (good_sectors > best_good_sectors) { | |
814 | best_good_sectors = good_sectors; | |
815 | best_slot = slot; | |
abbf098e | 816 | best_rdev = rdev; |
856e08e2 N |
817 | } |
818 | if (!do_balance) | |
819 | /* Must read from here */ | |
820 | break; | |
821 | } | |
822 | continue; | |
823 | } else | |
824 | best_good_sectors = sectors; | |
825 | ||
56d99121 N |
826 | if (!do_balance) |
827 | break; | |
1da177e4 | 828 | |
8d3ca83d N |
829 | if (best_slot >= 0) |
830 | /* At least 2 disks to choose from so failfast is OK */ | |
831 | set_bit(R10BIO_FailFast, &r10_bio->state); | |
22dfdf52 N |
832 | /* This optimisation is debatable, and completely destroys |
833 | * sequential read speed for 'far copies' arrays. So only | |
834 | * keep it for 'near' arrays, and review those later. | |
835 | */ | |
5cf00fcd | 836 | if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
8d3ca83d | 837 | new_distance = 0; |
8ed3a195 KS |
838 | |
839 | /* for far > 1 always use the lowest address */ | |
8d3ca83d | 840 | else if (geo->far_copies > 1) |
56d99121 | 841 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 842 | else |
56d99121 N |
843 | new_distance = abs(r10_bio->devs[slot].addr - |
844 | conf->mirrors[disk].head_position); | |
845 | if (new_distance < best_dist) { | |
846 | best_dist = new_distance; | |
847 | best_slot = slot; | |
abbf098e | 848 | best_rdev = rdev; |
1da177e4 LT |
849 | } |
850 | } | |
abbf098e | 851 | if (slot >= conf->copies) { |
56d99121 | 852 | slot = best_slot; |
abbf098e N |
853 | rdev = best_rdev; |
854 | } | |
1da177e4 | 855 | |
56d99121 | 856 | if (slot >= 0) { |
56d99121 | 857 | atomic_inc(&rdev->nr_pending); |
56d99121 N |
858 | r10_bio->read_slot = slot; |
859 | } else | |
96c3fd1f | 860 | rdev = NULL; |
1da177e4 | 861 | rcu_read_unlock(); |
856e08e2 | 862 | *max_sectors = best_good_sectors; |
1da177e4 | 863 | |
96c3fd1f | 864 | return rdev; |
1da177e4 LT |
865 | } |
866 | ||
5c675f83 | 867 | static int raid10_congested(struct mddev *mddev, int bits) |
0d129228 | 868 | { |
e879a879 | 869 | struct r10conf *conf = mddev->private; |
0d129228 N |
870 | int i, ret = 0; |
871 | ||
4452226e | 872 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
873 | conf->pending_count >= max_queued_requests) |
874 | return 1; | |
875 | ||
0d129228 | 876 | rcu_read_lock(); |
f8c9e74f N |
877 | for (i = 0; |
878 | (i < conf->geo.raid_disks || i < conf->prev.raid_disks) | |
879 | && ret == 0; | |
880 | i++) { | |
3cb03002 | 881 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 882 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 883 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 884 | |
dc3b17cc | 885 | ret |= bdi_congested(q->backing_dev_info, bits); |
0d129228 N |
886 | } |
887 | } | |
888 | rcu_read_unlock(); | |
889 | return ret; | |
890 | } | |
891 | ||
e879a879 | 892 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
893 | { |
894 | /* Any writes that have been queued but are awaiting | |
895 | * bitmap updates get flushed here. | |
a35e63ef | 896 | */ |
a35e63ef N |
897 | spin_lock_irq(&conf->device_lock); |
898 | ||
899 | if (conf->pending_bio_list.head) { | |
900 | struct bio *bio; | |
901 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 902 | conf->pending_count = 0; |
a35e63ef N |
903 | spin_unlock_irq(&conf->device_lock); |
904 | /* flush any pending bitmap writes to disk | |
905 | * before proceeding w/ I/O */ | |
906 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 907 | wake_up(&conf->wait_barrier); |
a35e63ef N |
908 | |
909 | while (bio) { /* submit pending writes */ | |
910 | struct bio *next = bio->bi_next; | |
a9ae93c8 | 911 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 912 | bio->bi_next = NULL; |
a9ae93c8 N |
913 | bio->bi_bdev = rdev->bdev; |
914 | if (test_bit(Faulty, &rdev->flags)) { | |
4e4cbee9 | 915 | bio->bi_status = BLK_STS_IOERR; |
a9ae93c8 N |
916 | bio_endio(bio); |
917 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
918 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
532a2a3f | 919 | /* Just ignore it */ |
4246a0b6 | 920 | bio_endio(bio); |
532a2a3f SL |
921 | else |
922 | generic_make_request(bio); | |
a35e63ef N |
923 | bio = next; |
924 | } | |
a35e63ef N |
925 | } else |
926 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 927 | } |
7eaceacc | 928 | |
0a27ec96 N |
929 | /* Barriers.... |
930 | * Sometimes we need to suspend IO while we do something else, | |
931 | * either some resync/recovery, or reconfigure the array. | |
932 | * To do this we raise a 'barrier'. | |
933 | * The 'barrier' is a counter that can be raised multiple times | |
934 | * to count how many activities are happening which preclude | |
935 | * normal IO. | |
936 | * We can only raise the barrier if there is no pending IO. | |
937 | * i.e. if nr_pending == 0. | |
938 | * We choose only to raise the barrier if no-one is waiting for the | |
939 | * barrier to go down. This means that as soon as an IO request | |
940 | * is ready, no other operations which require a barrier will start | |
941 | * until the IO request has had a chance. | |
942 | * | |
943 | * So: regular IO calls 'wait_barrier'. When that returns there | |
944 | * is no backgroup IO happening, It must arrange to call | |
945 | * allow_barrier when it has finished its IO. | |
946 | * backgroup IO calls must call raise_barrier. Once that returns | |
947 | * there is no normal IO happeing. It must arrange to call | |
948 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 949 | */ |
1da177e4 | 950 | |
e879a879 | 951 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 952 | { |
6cce3b23 | 953 | BUG_ON(force && !conf->barrier); |
1da177e4 | 954 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 955 | |
6cce3b23 N |
956 | /* Wait until no block IO is waiting (unless 'force') */ |
957 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
eed8c02e | 958 | conf->resync_lock); |
0a27ec96 N |
959 | |
960 | /* block any new IO from starting */ | |
961 | conf->barrier++; | |
962 | ||
c3b328ac | 963 | /* Now wait for all pending IO to complete */ |
0a27ec96 | 964 | wait_event_lock_irq(conf->wait_barrier, |
0e5313e2 | 965 | !atomic_read(&conf->nr_pending) && conf->barrier < RESYNC_DEPTH, |
eed8c02e | 966 | conf->resync_lock); |
0a27ec96 N |
967 | |
968 | spin_unlock_irq(&conf->resync_lock); | |
969 | } | |
970 | ||
e879a879 | 971 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
972 | { |
973 | unsigned long flags; | |
974 | spin_lock_irqsave(&conf->resync_lock, flags); | |
975 | conf->barrier--; | |
976 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
977 | wake_up(&conf->wait_barrier); | |
978 | } | |
979 | ||
e879a879 | 980 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
981 | { |
982 | spin_lock_irq(&conf->resync_lock); | |
983 | if (conf->barrier) { | |
984 | conf->nr_waiting++; | |
d6b42dcb N |
985 | /* Wait for the barrier to drop. |
986 | * However if there are already pending | |
987 | * requests (preventing the barrier from | |
988 | * rising completely), and the | |
989 | * pre-process bio queue isn't empty, | |
990 | * then don't wait, as we need to empty | |
991 | * that queue to get the nr_pending | |
992 | * count down. | |
993 | */ | |
578b54ad | 994 | raid10_log(conf->mddev, "wait barrier"); |
d6b42dcb N |
995 | wait_event_lock_irq(conf->wait_barrier, |
996 | !conf->barrier || | |
0e5313e2 | 997 | (atomic_read(&conf->nr_pending) && |
d6b42dcb | 998 | current->bio_list && |
f5fe1b51 N |
999 | (!bio_list_empty(¤t->bio_list[0]) || |
1000 | !bio_list_empty(¤t->bio_list[1]))), | |
eed8c02e | 1001 | conf->resync_lock); |
0a27ec96 | 1002 | conf->nr_waiting--; |
0e5313e2 TM |
1003 | if (!conf->nr_waiting) |
1004 | wake_up(&conf->wait_barrier); | |
1da177e4 | 1005 | } |
0e5313e2 | 1006 | atomic_inc(&conf->nr_pending); |
1da177e4 LT |
1007 | spin_unlock_irq(&conf->resync_lock); |
1008 | } | |
1009 | ||
e879a879 | 1010 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 | 1011 | { |
0e5313e2 TM |
1012 | if ((atomic_dec_and_test(&conf->nr_pending)) || |
1013 | (conf->array_freeze_pending)) | |
1014 | wake_up(&conf->wait_barrier); | |
0a27ec96 N |
1015 | } |
1016 | ||
e2d59925 | 1017 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
1018 | { |
1019 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 1020 | * go quiet. |
4443ae10 | 1021 | * We increment barrier and nr_waiting, and then |
e2d59925 | 1022 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
1023 | * This is called in the context of one normal IO request |
1024 | * that has failed. Thus any sync request that might be pending | |
1025 | * will be blocked by nr_pending, and we need to wait for | |
1026 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 1027 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
1028 | * must match the number of pending IOs (nr_pending) before |
1029 | * we continue. | |
4443ae10 N |
1030 | */ |
1031 | spin_lock_irq(&conf->resync_lock); | |
0e5313e2 | 1032 | conf->array_freeze_pending++; |
4443ae10 N |
1033 | conf->barrier++; |
1034 | conf->nr_waiting++; | |
eed8c02e | 1035 | wait_event_lock_irq_cmd(conf->wait_barrier, |
0e5313e2 | 1036 | atomic_read(&conf->nr_pending) == conf->nr_queued+extra, |
eed8c02e LC |
1037 | conf->resync_lock, |
1038 | flush_pending_writes(conf)); | |
c3b328ac | 1039 | |
0e5313e2 | 1040 | conf->array_freeze_pending--; |
4443ae10 N |
1041 | spin_unlock_irq(&conf->resync_lock); |
1042 | } | |
1043 | ||
e879a879 | 1044 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1045 | { |
1046 | /* reverse the effect of the freeze */ | |
1047 | spin_lock_irq(&conf->resync_lock); | |
1048 | conf->barrier--; | |
1049 | conf->nr_waiting--; | |
1050 | wake_up(&conf->wait_barrier); | |
1051 | spin_unlock_irq(&conf->resync_lock); | |
1052 | } | |
1053 | ||
f8c9e74f N |
1054 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1055 | struct md_rdev *rdev) | |
1056 | { | |
1057 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1058 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1059 | return rdev->data_offset; | |
1060 | else | |
1061 | return rdev->new_data_offset; | |
1062 | } | |
1063 | ||
57c67df4 N |
1064 | struct raid10_plug_cb { |
1065 | struct blk_plug_cb cb; | |
1066 | struct bio_list pending; | |
1067 | int pending_cnt; | |
1068 | }; | |
1069 | ||
1070 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1071 | { | |
1072 | struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb, | |
1073 | cb); | |
1074 | struct mddev *mddev = plug->cb.data; | |
1075 | struct r10conf *conf = mddev->private; | |
1076 | struct bio *bio; | |
1077 | ||
874807a8 | 1078 | if (from_schedule || current->bio_list) { |
57c67df4 N |
1079 | spin_lock_irq(&conf->device_lock); |
1080 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1081 | conf->pending_count += plug->pending_cnt; | |
1082 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1083 | wake_up(&conf->wait_barrier); |
57c67df4 N |
1084 | md_wakeup_thread(mddev->thread); |
1085 | kfree(plug); | |
1086 | return; | |
1087 | } | |
1088 | ||
1089 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1090 | bio = bio_list_get(&plug->pending); | |
1091 | bitmap_unplug(mddev->bitmap); | |
1092 | wake_up(&conf->wait_barrier); | |
1093 | ||
1094 | while (bio) { /* submit pending writes */ | |
1095 | struct bio *next = bio->bi_next; | |
a9ae93c8 | 1096 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
57c67df4 | 1097 | bio->bi_next = NULL; |
a9ae93c8 N |
1098 | bio->bi_bdev = rdev->bdev; |
1099 | if (test_bit(Faulty, &rdev->flags)) { | |
4e4cbee9 | 1100 | bio->bi_status = BLK_STS_IOERR; |
a9ae93c8 N |
1101 | bio_endio(bio); |
1102 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
1103 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
32f9f570 | 1104 | /* Just ignore it */ |
4246a0b6 | 1105 | bio_endio(bio); |
32f9f570 SL |
1106 | else |
1107 | generic_make_request(bio); | |
57c67df4 N |
1108 | bio = next; |
1109 | } | |
1110 | kfree(plug); | |
1111 | } | |
1112 | ||
bb5f1ed7 RL |
1113 | static void raid10_read_request(struct mddev *mddev, struct bio *bio, |
1114 | struct r10bio *r10_bio) | |
1da177e4 | 1115 | { |
e879a879 | 1116 | struct r10conf *conf = mddev->private; |
1da177e4 | 1117 | struct bio *read_bio; |
bb5f1ed7 RL |
1118 | const int op = bio_op(bio); |
1119 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
bb5f1ed7 RL |
1120 | int max_sectors; |
1121 | sector_t sectors; | |
1122 | struct md_rdev *rdev; | |
545250f2 N |
1123 | char b[BDEVNAME_SIZE]; |
1124 | int slot = r10_bio->read_slot; | |
1125 | struct md_rdev *err_rdev = NULL; | |
1126 | gfp_t gfp = GFP_NOIO; | |
bb5f1ed7 | 1127 | |
545250f2 N |
1128 | if (r10_bio->devs[slot].rdev) { |
1129 | /* | |
1130 | * This is an error retry, but we cannot | |
1131 | * safely dereference the rdev in the r10_bio, | |
1132 | * we must use the one in conf. | |
1133 | * If it has already been disconnected (unlikely) | |
1134 | * we lose the device name in error messages. | |
1135 | */ | |
1136 | int disk; | |
1137 | /* | |
1138 | * As we are blocking raid10, it is a little safer to | |
1139 | * use __GFP_HIGH. | |
1140 | */ | |
1141 | gfp = GFP_NOIO | __GFP_HIGH; | |
1142 | ||
1143 | rcu_read_lock(); | |
1144 | disk = r10_bio->devs[slot].devnum; | |
1145 | err_rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
1146 | if (err_rdev) | |
1147 | bdevname(err_rdev->bdev, b); | |
1148 | else { | |
1149 | strcpy(b, "???"); | |
1150 | /* This never gets dereferenced */ | |
1151 | err_rdev = r10_bio->devs[slot].rdev; | |
1152 | } | |
1153 | rcu_read_unlock(); | |
1154 | } | |
bb5f1ed7 RL |
1155 | /* |
1156 | * Register the new request and wait if the reconstruction | |
1157 | * thread has put up a bar for new requests. | |
1158 | * Continue immediately if no resync is active currently. | |
1159 | */ | |
1160 | wait_barrier(conf); | |
1161 | ||
fc9977dd | 1162 | sectors = r10_bio->sectors; |
bb5f1ed7 RL |
1163 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
1164 | bio->bi_iter.bi_sector < conf->reshape_progress && | |
1165 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
1166 | /* | |
1167 | * IO spans the reshape position. Need to wait for reshape to | |
1168 | * pass | |
1169 | */ | |
1170 | raid10_log(conf->mddev, "wait reshape"); | |
1171 | allow_barrier(conf); | |
1172 | wait_event(conf->wait_barrier, | |
1173 | conf->reshape_progress <= bio->bi_iter.bi_sector || | |
1174 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1175 | sectors); | |
1176 | wait_barrier(conf); | |
1177 | } | |
1178 | ||
bb5f1ed7 RL |
1179 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1180 | if (!rdev) { | |
545250f2 N |
1181 | if (err_rdev) { |
1182 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1183 | mdname(mddev), b, | |
1184 | (unsigned long long)r10_bio->sector); | |
1185 | } | |
bb5f1ed7 RL |
1186 | raid_end_bio_io(r10_bio); |
1187 | return; | |
1188 | } | |
545250f2 N |
1189 | if (err_rdev) |
1190 | pr_err_ratelimited("md/raid10:%s: %s: redirecting sector %llu to another mirror\n", | |
1191 | mdname(mddev), | |
1192 | bdevname(rdev->bdev, b), | |
1193 | (unsigned long long)r10_bio->sector); | |
fc9977dd N |
1194 | if (max_sectors < bio_sectors(bio)) { |
1195 | struct bio *split = bio_split(bio, max_sectors, | |
545250f2 | 1196 | gfp, conf->bio_split); |
fc9977dd N |
1197 | bio_chain(split, bio); |
1198 | generic_make_request(bio); | |
1199 | bio = split; | |
1200 | r10_bio->master_bio = bio; | |
1201 | r10_bio->sectors = max_sectors; | |
1202 | } | |
bb5f1ed7 RL |
1203 | slot = r10_bio->read_slot; |
1204 | ||
545250f2 | 1205 | read_bio = bio_clone_fast(bio, gfp, mddev->bio_set); |
bb5f1ed7 RL |
1206 | |
1207 | r10_bio->devs[slot].bio = read_bio; | |
1208 | r10_bio->devs[slot].rdev = rdev; | |
1209 | ||
1210 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + | |
1211 | choose_data_offset(r10_bio, rdev); | |
1212 | read_bio->bi_bdev = rdev->bdev; | |
1213 | read_bio->bi_end_io = raid10_end_read_request; | |
1214 | bio_set_op_attrs(read_bio, op, do_sync); | |
1215 | if (test_bit(FailFast, &rdev->flags) && | |
1216 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
1217 | read_bio->bi_opf |= MD_FAILFAST; | |
1218 | read_bio->bi_private = r10_bio; | |
1219 | ||
1220 | if (mddev->gendisk) | |
1221 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1222 | read_bio, disk_devt(mddev->gendisk), | |
1223 | r10_bio->sector); | |
fc9977dd | 1224 | generic_make_request(read_bio); |
bb5f1ed7 RL |
1225 | return; |
1226 | } | |
1227 | ||
27f26a0f GJ |
1228 | static void raid10_write_one_disk(struct mddev *mddev, struct r10bio *r10_bio, |
1229 | struct bio *bio, bool replacement, | |
fc9977dd | 1230 | int n_copy) |
bb5f1ed7 | 1231 | { |
796a5cf0 | 1232 | const int op = bio_op(bio); |
1eff9d32 JA |
1233 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1234 | const unsigned long do_fua = (bio->bi_opf & REQ_FUA); | |
6cce3b23 | 1235 | unsigned long flags; |
57c67df4 N |
1236 | struct blk_plug_cb *cb; |
1237 | struct raid10_plug_cb *plug = NULL; | |
27f26a0f GJ |
1238 | struct r10conf *conf = mddev->private; |
1239 | struct md_rdev *rdev; | |
1240 | int devnum = r10_bio->devs[n_copy].devnum; | |
1241 | struct bio *mbio; | |
1242 | ||
1243 | if (replacement) { | |
1244 | rdev = conf->mirrors[devnum].replacement; | |
1245 | if (rdev == NULL) { | |
1246 | /* Replacement just got moved to main 'rdev' */ | |
1247 | smp_mb(); | |
1248 | rdev = conf->mirrors[devnum].rdev; | |
1249 | } | |
1250 | } else | |
1251 | rdev = conf->mirrors[devnum].rdev; | |
1252 | ||
1253 | mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); | |
27f26a0f GJ |
1254 | if (replacement) |
1255 | r10_bio->devs[n_copy].repl_bio = mbio; | |
1256 | else | |
1257 | r10_bio->devs[n_copy].bio = mbio; | |
1258 | ||
1259 | mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + | |
1260 | choose_data_offset(r10_bio, rdev)); | |
1261 | mbio->bi_bdev = rdev->bdev; | |
1262 | mbio->bi_end_io = raid10_end_write_request; | |
1263 | bio_set_op_attrs(mbio, op, do_sync | do_fua); | |
1264 | if (!replacement && test_bit(FailFast, | |
1265 | &conf->mirrors[devnum].rdev->flags) | |
1266 | && enough(conf, devnum)) | |
1267 | mbio->bi_opf |= MD_FAILFAST; | |
1268 | mbio->bi_private = r10_bio; | |
1269 | ||
1270 | if (conf->mddev->gendisk) | |
1271 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1272 | mbio, disk_devt(conf->mddev->gendisk), | |
1273 | r10_bio->sector); | |
1274 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1275 | mbio->bi_bdev = (void *)rdev; | |
1276 | ||
1277 | atomic_inc(&r10_bio->remaining); | |
1278 | ||
1279 | cb = blk_check_plugged(raid10_unplug, mddev, sizeof(*plug)); | |
1280 | if (cb) | |
1281 | plug = container_of(cb, struct raid10_plug_cb, cb); | |
1282 | else | |
1283 | plug = NULL; | |
27f26a0f GJ |
1284 | if (plug) { |
1285 | bio_list_add(&plug->pending, mbio); | |
1286 | plug->pending_cnt++; | |
1287 | } else { | |
23b245c0 | 1288 | spin_lock_irqsave(&conf->device_lock, flags); |
27f26a0f GJ |
1289 | bio_list_add(&conf->pending_bio_list, mbio); |
1290 | conf->pending_count++; | |
23b245c0 | 1291 | spin_unlock_irqrestore(&conf->device_lock, flags); |
27f26a0f | 1292 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1293 | } |
27f26a0f GJ |
1294 | } |
1295 | ||
1296 | static void raid10_write_request(struct mddev *mddev, struct bio *bio, | |
1297 | struct r10bio *r10_bio) | |
1298 | { | |
1299 | struct r10conf *conf = mddev->private; | |
1300 | int i; | |
1301 | struct md_rdev *blocked_rdev; | |
bb5f1ed7 | 1302 | sector_t sectors; |
d4432c23 | 1303 | int max_sectors; |
1da177e4 | 1304 | |
cc13b1d1 N |
1305 | /* |
1306 | * Register the new request and wait if the reconstruction | |
1307 | * thread has put up a bar for new requests. | |
1308 | * Continue immediately if no resync is active currently. | |
1309 | */ | |
1310 | wait_barrier(conf); | |
1311 | ||
fc9977dd | 1312 | sectors = r10_bio->sectors; |
3ea7daa5 | 1313 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
4f024f37 KO |
1314 | bio->bi_iter.bi_sector < conf->reshape_progress && |
1315 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
bb5f1ed7 RL |
1316 | /* |
1317 | * IO spans the reshape position. Need to wait for reshape to | |
1318 | * pass | |
3ea7daa5 | 1319 | */ |
578b54ad | 1320 | raid10_log(conf->mddev, "wait reshape"); |
3ea7daa5 N |
1321 | allow_barrier(conf); |
1322 | wait_event(conf->wait_barrier, | |
4f024f37 KO |
1323 | conf->reshape_progress <= bio->bi_iter.bi_sector || |
1324 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1325 | sectors); | |
3ea7daa5 N |
1326 | wait_barrier(conf); |
1327 | } | |
bb5f1ed7 | 1328 | |
3ea7daa5 | 1329 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
3ea7daa5 | 1330 | (mddev->reshape_backwards |
4f024f37 KO |
1331 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1332 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1333 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1334 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1335 | /* Need to update reshape_position in metadata */ |
1336 | mddev->reshape_position = conf->reshape_progress; | |
2953079c SL |
1337 | set_mask_bits(&mddev->sb_flags, 0, |
1338 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
3ea7daa5 | 1339 | md_wakeup_thread(mddev->thread); |
578b54ad | 1340 | raid10_log(conf->mddev, "wait reshape metadata"); |
3ea7daa5 | 1341 | wait_event(mddev->sb_wait, |
2953079c | 1342 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
3ea7daa5 N |
1343 | |
1344 | conf->reshape_safe = mddev->reshape_position; | |
1345 | } | |
1346 | ||
34db0cd6 N |
1347 | if (conf->pending_count >= max_queued_requests) { |
1348 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1349 | raid10_log(mddev, "wait queued"); |
34db0cd6 N |
1350 | wait_event(conf->wait_barrier, |
1351 | conf->pending_count < max_queued_requests); | |
1352 | } | |
6bfe0b49 | 1353 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1354 | * inc refcount on their rdev. Record them by setting |
1355 | * bios[x] to bio | |
d4432c23 N |
1356 | * If there are known/acknowledged bad blocks on any device |
1357 | * on which we have seen a write error, we want to avoid | |
1358 | * writing to those blocks. This potentially requires several | |
1359 | * writes to write around the bad blocks. Each set of writes | |
fd16f2e8 | 1360 | * gets its own r10_bio with a set of bios attached. |
1da177e4 | 1361 | */ |
c3b328ac | 1362 | |
69335ef3 | 1363 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1364 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1365 | retry_write: |
cb6969e8 | 1366 | blocked_rdev = NULL; |
1da177e4 | 1367 | rcu_read_lock(); |
d4432c23 N |
1368 | max_sectors = r10_bio->sectors; |
1369 | ||
1da177e4 LT |
1370 | for (i = 0; i < conf->copies; i++) { |
1371 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1372 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1373 | struct md_rdev *rrdev = rcu_dereference( |
1374 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1375 | if (rdev == rrdev) |
1376 | rrdev = NULL; | |
6bfe0b49 DW |
1377 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1378 | atomic_inc(&rdev->nr_pending); | |
1379 | blocked_rdev = rdev; | |
1380 | break; | |
1381 | } | |
475b0321 N |
1382 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { |
1383 | atomic_inc(&rrdev->nr_pending); | |
1384 | blocked_rdev = rrdev; | |
1385 | break; | |
1386 | } | |
8ae12666 | 1387 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1388 | rdev = NULL; |
8ae12666 | 1389 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1390 | rrdev = NULL; |
1391 | ||
d4432c23 | 1392 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1393 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1394 | |
1395 | if (!rdev && !rrdev) { | |
6cce3b23 | 1396 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1397 | continue; |
1398 | } | |
e7c0c3fa | 1399 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1400 | sector_t first_bad; |
1401 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1402 | int bad_sectors; | |
1403 | int is_bad; | |
1404 | ||
bb5f1ed7 | 1405 | is_bad = is_badblock(rdev, dev_sector, max_sectors, |
d4432c23 N |
1406 | &first_bad, &bad_sectors); |
1407 | if (is_bad < 0) { | |
1408 | /* Mustn't write here until the bad block | |
1409 | * is acknowledged | |
1410 | */ | |
1411 | atomic_inc(&rdev->nr_pending); | |
1412 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1413 | blocked_rdev = rdev; | |
1414 | break; | |
1415 | } | |
1416 | if (is_bad && first_bad <= dev_sector) { | |
1417 | /* Cannot write here at all */ | |
1418 | bad_sectors -= (dev_sector - first_bad); | |
1419 | if (bad_sectors < max_sectors) | |
1420 | /* Mustn't write more than bad_sectors | |
1421 | * to other devices yet | |
1422 | */ | |
1423 | max_sectors = bad_sectors; | |
1424 | /* We don't set R10BIO_Degraded as that | |
1425 | * only applies if the disk is missing, | |
1426 | * so it might be re-added, and we want to | |
1427 | * know to recover this chunk. | |
1428 | * In this case the device is here, and the | |
1429 | * fact that this chunk is not in-sync is | |
1430 | * recorded in the bad block log. | |
1431 | */ | |
1432 | continue; | |
1433 | } | |
1434 | if (is_bad) { | |
1435 | int good_sectors = first_bad - dev_sector; | |
1436 | if (good_sectors < max_sectors) | |
1437 | max_sectors = good_sectors; | |
1438 | } | |
6cce3b23 | 1439 | } |
e7c0c3fa N |
1440 | if (rdev) { |
1441 | r10_bio->devs[i].bio = bio; | |
1442 | atomic_inc(&rdev->nr_pending); | |
1443 | } | |
475b0321 N |
1444 | if (rrdev) { |
1445 | r10_bio->devs[i].repl_bio = bio; | |
1446 | atomic_inc(&rrdev->nr_pending); | |
1447 | } | |
1da177e4 LT |
1448 | } |
1449 | rcu_read_unlock(); | |
1450 | ||
6bfe0b49 DW |
1451 | if (unlikely(blocked_rdev)) { |
1452 | /* Have to wait for this device to get unblocked, then retry */ | |
1453 | int j; | |
1454 | int d; | |
1455 | ||
475b0321 | 1456 | for (j = 0; j < i; j++) { |
6bfe0b49 DW |
1457 | if (r10_bio->devs[j].bio) { |
1458 | d = r10_bio->devs[j].devnum; | |
1459 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1460 | } | |
475b0321 | 1461 | if (r10_bio->devs[j].repl_bio) { |
4ca40c2c | 1462 | struct md_rdev *rdev; |
475b0321 | 1463 | d = r10_bio->devs[j].devnum; |
4ca40c2c N |
1464 | rdev = conf->mirrors[d].replacement; |
1465 | if (!rdev) { | |
1466 | /* Race with remove_disk */ | |
1467 | smp_mb(); | |
1468 | rdev = conf->mirrors[d].rdev; | |
1469 | } | |
1470 | rdev_dec_pending(rdev, mddev); | |
475b0321 N |
1471 | } |
1472 | } | |
6bfe0b49 | 1473 | allow_barrier(conf); |
578b54ad | 1474 | raid10_log(conf->mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 DW |
1475 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
1476 | wait_barrier(conf); | |
1477 | goto retry_write; | |
1478 | } | |
1479 | ||
6b6c8110 | 1480 | if (max_sectors < r10_bio->sectors) |
d4432c23 | 1481 | r10_bio->sectors = max_sectors; |
fc9977dd N |
1482 | |
1483 | if (r10_bio->sectors < bio_sectors(bio)) { | |
1484 | struct bio *split = bio_split(bio, r10_bio->sectors, | |
1485 | GFP_NOIO, conf->bio_split); | |
1486 | bio_chain(split, bio); | |
1487 | generic_make_request(bio); | |
1488 | bio = split; | |
1489 | r10_bio->master_bio = bio; | |
d4432c23 | 1490 | } |
d4432c23 | 1491 | |
4e78064f | 1492 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1493 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1494 | |
1da177e4 | 1495 | for (i = 0; i < conf->copies; i++) { |
27f26a0f | 1496 | if (r10_bio->devs[i].bio) |
fc9977dd | 1497 | raid10_write_one_disk(mddev, r10_bio, bio, false, i); |
27f26a0f | 1498 | if (r10_bio->devs[i].repl_bio) |
fc9977dd | 1499 | raid10_write_one_disk(mddev, r10_bio, bio, true, i); |
d4432c23 | 1500 | } |
079fa166 | 1501 | one_write_done(r10_bio); |
20d0189b KO |
1502 | } |
1503 | ||
fc9977dd | 1504 | static void __make_request(struct mddev *mddev, struct bio *bio, int sectors) |
bb5f1ed7 RL |
1505 | { |
1506 | struct r10conf *conf = mddev->private; | |
1507 | struct r10bio *r10_bio; | |
1508 | ||
1509 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1510 | ||
1511 | r10_bio->master_bio = bio; | |
fc9977dd | 1512 | r10_bio->sectors = sectors; |
bb5f1ed7 RL |
1513 | |
1514 | r10_bio->mddev = mddev; | |
1515 | r10_bio->sector = bio->bi_iter.bi_sector; | |
1516 | r10_bio->state = 0; | |
545250f2 | 1517 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * conf->copies); |
bb5f1ed7 RL |
1518 | |
1519 | if (bio_data_dir(bio) == READ) | |
1520 | raid10_read_request(mddev, bio, r10_bio); | |
1521 | else | |
1522 | raid10_write_request(mddev, bio, r10_bio); | |
1523 | } | |
1524 | ||
cc27b0c7 | 1525 | static bool raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1526 | { |
1527 | struct r10conf *conf = mddev->private; | |
1528 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1529 | int chunk_sects = chunk_mask + 1; | |
fc9977dd | 1530 | int sectors = bio_sectors(bio); |
20d0189b | 1531 | |
1eff9d32 | 1532 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
20d0189b | 1533 | md_flush_request(mddev, bio); |
cc27b0c7 | 1534 | return true; |
20d0189b KO |
1535 | } |
1536 | ||
cc27b0c7 N |
1537 | if (!md_write_start(mddev, bio)) |
1538 | return false; | |
1539 | ||
fc9977dd N |
1540 | /* |
1541 | * If this request crosses a chunk boundary, we need to split | |
1542 | * it. | |
1543 | */ | |
1544 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1545 | sectors > chunk_sects | |
1546 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1547 | || conf->prev.near_copies < | |
1548 | conf->prev.raid_disks))) | |
1549 | sectors = chunk_sects - | |
1550 | (bio->bi_iter.bi_sector & | |
1551 | (chunk_sects - 1)); | |
1552 | __make_request(mddev, bio, sectors); | |
079fa166 N |
1553 | |
1554 | /* In case raid10d snuck in to freeze_array */ | |
1555 | wake_up(&conf->wait_barrier); | |
cc27b0c7 | 1556 | return true; |
1da177e4 LT |
1557 | } |
1558 | ||
849674e4 | 1559 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1560 | { |
e879a879 | 1561 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1562 | int i; |
1563 | ||
5cf00fcd | 1564 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1565 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1566 | if (conf->geo.near_copies > 1) |
1567 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1568 | if (conf->geo.far_copies > 1) { | |
1569 | if (conf->geo.far_offset) | |
1570 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1571 | else |
5cf00fcd | 1572 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1573 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1574 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1575 | } |
5cf00fcd N |
1576 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1577 | conf->geo.raid_disks - mddev->degraded); | |
d44b0a92 N |
1578 | rcu_read_lock(); |
1579 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
1580 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1581 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
1582 | } | |
1583 | rcu_read_unlock(); | |
1da177e4 LT |
1584 | seq_printf(seq, "]"); |
1585 | } | |
1586 | ||
700c7213 N |
1587 | /* check if there are enough drives for |
1588 | * every block to appear on atleast one. | |
1589 | * Don't consider the device numbered 'ignore' | |
1590 | * as we might be about to remove it. | |
1591 | */ | |
635f6416 | 1592 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1593 | { |
1594 | int first = 0; | |
725d6e57 | 1595 | int has_enough = 0; |
635f6416 N |
1596 | int disks, ncopies; |
1597 | if (previous) { | |
1598 | disks = conf->prev.raid_disks; | |
1599 | ncopies = conf->prev.near_copies; | |
1600 | } else { | |
1601 | disks = conf->geo.raid_disks; | |
1602 | ncopies = conf->geo.near_copies; | |
1603 | } | |
700c7213 | 1604 | |
725d6e57 | 1605 | rcu_read_lock(); |
700c7213 N |
1606 | do { |
1607 | int n = conf->copies; | |
1608 | int cnt = 0; | |
80b48124 | 1609 | int this = first; |
700c7213 | 1610 | while (n--) { |
725d6e57 N |
1611 | struct md_rdev *rdev; |
1612 | if (this != ignore && | |
1613 | (rdev = rcu_dereference(conf->mirrors[this].rdev)) && | |
1614 | test_bit(In_sync, &rdev->flags)) | |
700c7213 | 1615 | cnt++; |
635f6416 | 1616 | this = (this+1) % disks; |
700c7213 N |
1617 | } |
1618 | if (cnt == 0) | |
725d6e57 | 1619 | goto out; |
635f6416 | 1620 | first = (first + ncopies) % disks; |
700c7213 | 1621 | } while (first != 0); |
725d6e57 N |
1622 | has_enough = 1; |
1623 | out: | |
1624 | rcu_read_unlock(); | |
1625 | return has_enough; | |
700c7213 N |
1626 | } |
1627 | ||
f8c9e74f N |
1628 | static int enough(struct r10conf *conf, int ignore) |
1629 | { | |
635f6416 N |
1630 | /* when calling 'enough', both 'prev' and 'geo' must |
1631 | * be stable. | |
1632 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1633 | * is held. | |
1634 | */ | |
1635 | return _enough(conf, 0, ignore) && | |
1636 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1637 | } |
1638 | ||
849674e4 | 1639 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1640 | { |
1641 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1642 | struct r10conf *conf = mddev->private; |
635f6416 | 1643 | unsigned long flags; |
1da177e4 LT |
1644 | |
1645 | /* | |
1646 | * If it is not operational, then we have already marked it as dead | |
1647 | * else if it is the last working disks, ignore the error, let the | |
1648 | * next level up know. | |
1649 | * else mark the drive as failed | |
1650 | */ | |
635f6416 | 1651 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1652 | if (test_bit(In_sync, &rdev->flags) |
635f6416 | 1653 | && !enough(conf, rdev->raid_disk)) { |
1da177e4 LT |
1654 | /* |
1655 | * Don't fail the drive, just return an IO error. | |
1da177e4 | 1656 | */ |
635f6416 | 1657 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1658 | return; |
635f6416 | 1659 | } |
2446dba0 | 1660 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1661 | mddev->degraded++; |
2446dba0 N |
1662 | /* |
1663 | * If recovery is running, make sure it aborts. | |
1664 | */ | |
1665 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
de393cde | 1666 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1667 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
1668 | set_mask_bits(&mddev->sb_flags, 0, |
1669 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
635f6416 | 1670 | spin_unlock_irqrestore(&conf->device_lock, flags); |
08464e09 N |
1671 | pr_crit("md/raid10:%s: Disk failure on %s, disabling device.\n" |
1672 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
1673 | mdname(mddev), bdevname(rdev->bdev, b), | |
1674 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); | |
1da177e4 LT |
1675 | } |
1676 | ||
e879a879 | 1677 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1678 | { |
1679 | int i; | |
4056ca51 | 1680 | struct md_rdev *rdev; |
1da177e4 | 1681 | |
08464e09 | 1682 | pr_debug("RAID10 conf printout:\n"); |
1da177e4 | 1683 | if (!conf) { |
08464e09 | 1684 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1685 | return; |
1686 | } | |
08464e09 N |
1687 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1688 | conf->geo.raid_disks); | |
1da177e4 | 1689 | |
4056ca51 N |
1690 | /* This is only called with ->reconfix_mutex held, so |
1691 | * rcu protection of rdev is not needed */ | |
5cf00fcd | 1692 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1693 | char b[BDEVNAME_SIZE]; |
4056ca51 N |
1694 | rdev = conf->mirrors[i].rdev; |
1695 | if (rdev) | |
08464e09 N |
1696 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1697 | i, !test_bit(In_sync, &rdev->flags), | |
1698 | !test_bit(Faulty, &rdev->flags), | |
1699 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1700 | } |
1701 | } | |
1702 | ||
e879a879 | 1703 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1704 | { |
0a27ec96 N |
1705 | wait_barrier(conf); |
1706 | allow_barrier(conf); | |
1da177e4 LT |
1707 | |
1708 | mempool_destroy(conf->r10buf_pool); | |
1709 | conf->r10buf_pool = NULL; | |
1710 | } | |
1711 | ||
fd01b88c | 1712 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1713 | { |
1714 | int i; | |
e879a879 | 1715 | struct r10conf *conf = mddev->private; |
dc280d98 | 1716 | struct raid10_info *tmp; |
6b965620 N |
1717 | int count = 0; |
1718 | unsigned long flags; | |
1da177e4 LT |
1719 | |
1720 | /* | |
1721 | * Find all non-in_sync disks within the RAID10 configuration | |
1722 | * and mark them in_sync | |
1723 | */ | |
5cf00fcd | 1724 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1725 | tmp = conf->mirrors + i; |
4ca40c2c N |
1726 | if (tmp->replacement |
1727 | && tmp->replacement->recovery_offset == MaxSector | |
1728 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1729 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1730 | /* Replacement has just become active */ | |
1731 | if (!tmp->rdev | |
1732 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1733 | count++; | |
1734 | if (tmp->rdev) { | |
1735 | /* Replaced device not technically faulty, | |
1736 | * but we need to be sure it gets removed | |
1737 | * and never re-added. | |
1738 | */ | |
1739 | set_bit(Faulty, &tmp->rdev->flags); | |
1740 | sysfs_notify_dirent_safe( | |
1741 | tmp->rdev->sysfs_state); | |
1742 | } | |
1743 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1744 | } else if (tmp->rdev | |
61e4947c | 1745 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
1746 | && !test_bit(Faulty, &tmp->rdev->flags) |
1747 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1748 | count++; |
2863b9eb | 1749 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
1750 | } |
1751 | } | |
6b965620 N |
1752 | spin_lock_irqsave(&conf->device_lock, flags); |
1753 | mddev->degraded -= count; | |
1754 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1755 | |
1756 | print_conf(conf); | |
6b965620 | 1757 | return count; |
1da177e4 LT |
1758 | } |
1759 | ||
fd01b88c | 1760 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1761 | { |
e879a879 | 1762 | struct r10conf *conf = mddev->private; |
199050ea | 1763 | int err = -EEXIST; |
1da177e4 | 1764 | int mirror; |
6c2fce2e | 1765 | int first = 0; |
5cf00fcd | 1766 | int last = conf->geo.raid_disks - 1; |
1da177e4 LT |
1767 | |
1768 | if (mddev->recovery_cp < MaxSector) | |
1769 | /* only hot-add to in-sync arrays, as recovery is | |
1770 | * very different from resync | |
1771 | */ | |
199050ea | 1772 | return -EBUSY; |
635f6416 | 1773 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 1774 | return -EINVAL; |
1da177e4 | 1775 | |
1501efad DW |
1776 | if (md_integrity_add_rdev(rdev, mddev)) |
1777 | return -ENXIO; | |
1778 | ||
a53a6c85 | 1779 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1780 | first = last = rdev->raid_disk; |
1da177e4 | 1781 | |
2c4193df | 1782 | if (rdev->saved_raid_disk >= first && |
6cce3b23 N |
1783 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1784 | mirror = rdev->saved_raid_disk; | |
1785 | else | |
6c2fce2e | 1786 | mirror = first; |
2bb77736 | 1787 | for ( ; mirror <= last ; mirror++) { |
dc280d98 | 1788 | struct raid10_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1789 | if (p->recovery_disabled == mddev->recovery_disabled) |
1790 | continue; | |
b7044d41 N |
1791 | if (p->rdev) { |
1792 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1793 | p->replacement != NULL) | |
1794 | continue; | |
1795 | clear_bit(In_sync, &rdev->flags); | |
1796 | set_bit(Replacement, &rdev->flags); | |
1797 | rdev->raid_disk = mirror; | |
1798 | err = 0; | |
9092c02d JB |
1799 | if (mddev->gendisk) |
1800 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1801 | rdev->data_offset << 9); | |
b7044d41 N |
1802 | conf->fullsync = 1; |
1803 | rcu_assign_pointer(p->replacement, rdev); | |
1804 | break; | |
1805 | } | |
1da177e4 | 1806 | |
9092c02d JB |
1807 | if (mddev->gendisk) |
1808 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1809 | rdev->data_offset << 9); | |
1da177e4 | 1810 | |
2bb77736 | 1811 | p->head_position = 0; |
d890fa2b | 1812 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1813 | rdev->raid_disk = mirror; |
1814 | err = 0; | |
1815 | if (rdev->saved_raid_disk != mirror) | |
1816 | conf->fullsync = 1; | |
1817 | rcu_assign_pointer(p->rdev, rdev); | |
1818 | break; | |
1819 | } | |
ed30be07 | 1820 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
532a2a3f SL |
1821 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1822 | ||
1da177e4 | 1823 | print_conf(conf); |
199050ea | 1824 | return err; |
1da177e4 LT |
1825 | } |
1826 | ||
b8321b68 | 1827 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1828 | { |
e879a879 | 1829 | struct r10conf *conf = mddev->private; |
1da177e4 | 1830 | int err = 0; |
b8321b68 | 1831 | int number = rdev->raid_disk; |
c8ab903e | 1832 | struct md_rdev **rdevp; |
dc280d98 | 1833 | struct raid10_info *p = conf->mirrors + number; |
1da177e4 LT |
1834 | |
1835 | print_conf(conf); | |
c8ab903e N |
1836 | if (rdev == p->rdev) |
1837 | rdevp = &p->rdev; | |
1838 | else if (rdev == p->replacement) | |
1839 | rdevp = &p->replacement; | |
1840 | else | |
1841 | return 0; | |
1842 | ||
1843 | if (test_bit(In_sync, &rdev->flags) || | |
1844 | atomic_read(&rdev->nr_pending)) { | |
1845 | err = -EBUSY; | |
1846 | goto abort; | |
1847 | } | |
d787be40 | 1848 | /* Only remove non-faulty devices if recovery |
c8ab903e N |
1849 | * is not possible. |
1850 | */ | |
1851 | if (!test_bit(Faulty, &rdev->flags) && | |
1852 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1853 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 1854 | number < conf->geo.raid_disks && |
c8ab903e N |
1855 | enough(conf, -1)) { |
1856 | err = -EBUSY; | |
1857 | goto abort; | |
1da177e4 | 1858 | } |
c8ab903e | 1859 | *rdevp = NULL; |
d787be40 N |
1860 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1861 | synchronize_rcu(); | |
1862 | if (atomic_read(&rdev->nr_pending)) { | |
1863 | /* lost the race, try later */ | |
1864 | err = -EBUSY; | |
1865 | *rdevp = rdev; | |
1866 | goto abort; | |
1867 | } | |
1868 | } | |
1869 | if (p->replacement) { | |
4ca40c2c N |
1870 | /* We must have just cleared 'rdev' */ |
1871 | p->rdev = p->replacement; | |
1872 | clear_bit(Replacement, &p->replacement->flags); | |
1873 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1874 | * but will never see neither -- if they are careful. | |
1875 | */ | |
1876 | p->replacement = NULL; | |
e5bc9c3c | 1877 | } |
4ca40c2c | 1878 | |
e5bc9c3c | 1879 | clear_bit(WantReplacement, &rdev->flags); |
c8ab903e N |
1880 | err = md_integrity_register(mddev); |
1881 | ||
1da177e4 LT |
1882 | abort: |
1883 | ||
1884 | print_conf(conf); | |
1885 | return err; | |
1886 | } | |
1887 | ||
81fa1520 | 1888 | static void __end_sync_read(struct r10bio *r10_bio, struct bio *bio, int d) |
1da177e4 | 1889 | { |
e879a879 | 1890 | struct r10conf *conf = r10_bio->mddev->private; |
0eb3ff12 | 1891 | |
4e4cbee9 | 1892 | if (!bio->bi_status) |
0eb3ff12 | 1893 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
1894 | else |
1895 | /* The write handler will notice the lack of | |
1896 | * R10BIO_Uptodate and record any errors etc | |
1897 | */ | |
4dbcdc75 N |
1898 | atomic_add(r10_bio->sectors, |
1899 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1900 | |
1901 | /* for reconstruct, we always reschedule after a read. | |
1902 | * for resync, only after all reads | |
1903 | */ | |
73d5c38a | 1904 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1905 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1906 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1907 | /* we have read all the blocks, | |
1908 | * do the comparison in process context in raid10d | |
1909 | */ | |
1910 | reschedule_retry(r10_bio); | |
1911 | } | |
1da177e4 LT |
1912 | } |
1913 | ||
81fa1520 ML |
1914 | static void end_sync_read(struct bio *bio) |
1915 | { | |
f0250618 | 1916 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
81fa1520 ML |
1917 | struct r10conf *conf = r10_bio->mddev->private; |
1918 | int d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
1919 | ||
1920 | __end_sync_read(r10_bio, bio, d); | |
1921 | } | |
1922 | ||
1923 | static void end_reshape_read(struct bio *bio) | |
1924 | { | |
f0250618 | 1925 | /* reshape read bio isn't allocated from r10buf_pool */ |
81fa1520 ML |
1926 | struct r10bio *r10_bio = bio->bi_private; |
1927 | ||
1928 | __end_sync_read(r10_bio, bio, r10_bio->read_slot); | |
1929 | } | |
1930 | ||
9f2c9d12 | 1931 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1932 | { |
fd01b88c | 1933 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1934 | |
1da177e4 LT |
1935 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1936 | if (r10_bio->master_bio == NULL) { | |
1937 | /* the primary of several recovery bios */ | |
73d5c38a | 1938 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1939 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1940 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1941 | reschedule_retry(r10_bio); |
1942 | else | |
1943 | put_buf(r10_bio); | |
73d5c38a | 1944 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1945 | break; |
1946 | } else { | |
9f2c9d12 | 1947 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1948 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1949 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1950 | reschedule_retry(r10_bio); |
1951 | else | |
1952 | put_buf(r10_bio); | |
1da177e4 LT |
1953 | r10_bio = r10_bio2; |
1954 | } | |
1955 | } | |
1da177e4 LT |
1956 | } |
1957 | ||
4246a0b6 | 1958 | static void end_sync_write(struct bio *bio) |
5e570289 | 1959 | { |
f0250618 | 1960 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
fd01b88c | 1961 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1962 | struct r10conf *conf = mddev->private; |
5e570289 N |
1963 | int d; |
1964 | sector_t first_bad; | |
1965 | int bad_sectors; | |
1966 | int slot; | |
9ad1aefc | 1967 | int repl; |
4ca40c2c | 1968 | struct md_rdev *rdev = NULL; |
5e570289 | 1969 | |
9ad1aefc N |
1970 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1971 | if (repl) | |
1972 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 1973 | else |
9ad1aefc | 1974 | rdev = conf->mirrors[d].rdev; |
5e570289 | 1975 | |
4e4cbee9 | 1976 | if (bio->bi_status) { |
9ad1aefc N |
1977 | if (repl) |
1978 | md_error(mddev, rdev); | |
1979 | else { | |
1980 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
1981 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1982 | set_bit(MD_RECOVERY_NEEDED, | |
1983 | &rdev->mddev->recovery); | |
9ad1aefc N |
1984 | set_bit(R10BIO_WriteError, &r10_bio->state); |
1985 | } | |
1986 | } else if (is_badblock(rdev, | |
5e570289 N |
1987 | r10_bio->devs[slot].addr, |
1988 | r10_bio->sectors, | |
1989 | &first_bad, &bad_sectors)) | |
1990 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
1991 | ||
9ad1aefc | 1992 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
1993 | |
1994 | end_sync_request(r10_bio); | |
1995 | } | |
1996 | ||
1da177e4 LT |
1997 | /* |
1998 | * Note: sync and recover and handled very differently for raid10 | |
1999 | * This code is for resync. | |
2000 | * For resync, we read through virtual addresses and read all blocks. | |
2001 | * If there is any error, we schedule a write. The lowest numbered | |
2002 | * drive is authoritative. | |
2003 | * However requests come for physical address, so we need to map. | |
2004 | * For every physical address there are raid_disks/copies virtual addresses, | |
2005 | * which is always are least one, but is not necessarly an integer. | |
2006 | * This means that a physical address can span multiple chunks, so we may | |
2007 | * have to submit multiple io requests for a single sync request. | |
2008 | */ | |
2009 | /* | |
2010 | * We check if all blocks are in-sync and only write to blocks that | |
2011 | * aren't in sync | |
2012 | */ | |
9f2c9d12 | 2013 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2014 | { |
e879a879 | 2015 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
2016 | int i, first; |
2017 | struct bio *tbio, *fbio; | |
f4380a91 | 2018 | int vcnt; |
cdb76be3 | 2019 | struct page **tpages, **fpages; |
1da177e4 LT |
2020 | |
2021 | atomic_set(&r10_bio->remaining, 1); | |
2022 | ||
2023 | /* find the first device with a block */ | |
2024 | for (i=0; i<conf->copies; i++) | |
4e4cbee9 | 2025 | if (!r10_bio->devs[i].bio->bi_status) |
1da177e4 LT |
2026 | break; |
2027 | ||
2028 | if (i == conf->copies) | |
2029 | goto done; | |
2030 | ||
2031 | first = i; | |
2032 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
2033 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
2034 | fbio->bi_iter.bi_idx = 0; | |
cdb76be3 | 2035 | fpages = get_resync_pages(fbio)->pages; |
1da177e4 | 2036 | |
f4380a91 | 2037 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 2038 | /* now find blocks with errors */ |
0eb3ff12 N |
2039 | for (i=0 ; i < conf->copies ; i++) { |
2040 | int j, d; | |
8d3ca83d | 2041 | struct md_rdev *rdev; |
f0250618 | 2042 | struct resync_pages *rp; |
1da177e4 | 2043 | |
1da177e4 | 2044 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
2045 | |
2046 | if (tbio->bi_end_io != end_sync_read) | |
2047 | continue; | |
2048 | if (i == first) | |
1da177e4 | 2049 | continue; |
cdb76be3 ML |
2050 | |
2051 | tpages = get_resync_pages(tbio)->pages; | |
8d3ca83d N |
2052 | d = r10_bio->devs[i].devnum; |
2053 | rdev = conf->mirrors[d].rdev; | |
4e4cbee9 | 2054 | if (!r10_bio->devs[i].bio->bi_status) { |
0eb3ff12 N |
2055 | /* We know that the bi_io_vec layout is the same for |
2056 | * both 'first' and 'i', so we just compare them. | |
2057 | * All vec entries are PAGE_SIZE; | |
2058 | */ | |
7bb23c49 N |
2059 | int sectors = r10_bio->sectors; |
2060 | for (j = 0; j < vcnt; j++) { | |
2061 | int len = PAGE_SIZE; | |
2062 | if (sectors < (len / 512)) | |
2063 | len = sectors * 512; | |
cdb76be3 ML |
2064 | if (memcmp(page_address(fpages[j]), |
2065 | page_address(tpages[j]), | |
7bb23c49 | 2066 | len)) |
0eb3ff12 | 2067 | break; |
7bb23c49 N |
2068 | sectors -= len/512; |
2069 | } | |
0eb3ff12 N |
2070 | if (j == vcnt) |
2071 | continue; | |
7f7583d4 | 2072 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
2073 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
2074 | /* Don't fix anything. */ | |
2075 | continue; | |
8d3ca83d N |
2076 | } else if (test_bit(FailFast, &rdev->flags)) { |
2077 | /* Just give up on this device */ | |
2078 | md_error(rdev->mddev, rdev); | |
2079 | continue; | |
0eb3ff12 | 2080 | } |
f84ee364 N |
2081 | /* Ok, we need to write this bio, either to correct an |
2082 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
2083 | * First we need to fixup bv_offset, bv_len and |
2084 | * bi_vecs, as the read request might have corrupted these | |
2085 | */ | |
f0250618 | 2086 | rp = get_resync_pages(tbio); |
8be185f2 KO |
2087 | bio_reset(tbio); |
2088 | ||
1da177e4 | 2089 | tbio->bi_vcnt = vcnt; |
cc578588 | 2090 | tbio->bi_iter.bi_size = fbio->bi_iter.bi_size; |
f0250618 ML |
2091 | rp->raid_bio = r10_bio; |
2092 | tbio->bi_private = rp; | |
4f024f37 | 2093 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 | 2094 | tbio->bi_end_io = end_sync_write; |
796a5cf0 | 2095 | bio_set_op_attrs(tbio, REQ_OP_WRITE, 0); |
1da177e4 | 2096 | |
c31df25f KO |
2097 | bio_copy_data(tbio, fbio); |
2098 | ||
1da177e4 LT |
2099 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2100 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2101 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2102 | |
1919cbb2 N |
2103 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
2104 | tbio->bi_opf |= MD_FAILFAST; | |
4f024f37 | 2105 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
1da177e4 LT |
2106 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; |
2107 | generic_make_request(tbio); | |
2108 | } | |
2109 | ||
9ad1aefc N |
2110 | /* Now write out to any replacement devices |
2111 | * that are active | |
2112 | */ | |
2113 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2114 | int d; |
9ad1aefc N |
2115 | |
2116 | tbio = r10_bio->devs[i].repl_bio; | |
2117 | if (!tbio || !tbio->bi_end_io) | |
2118 | continue; | |
2119 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2120 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2121 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2122 | d = r10_bio->devs[i].devnum; |
2123 | atomic_inc(&r10_bio->remaining); | |
2124 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2125 | bio_sectors(tbio)); |
9ad1aefc N |
2126 | generic_make_request(tbio); |
2127 | } | |
2128 | ||
1da177e4 LT |
2129 | done: |
2130 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2131 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2132 | put_buf(r10_bio); | |
2133 | } | |
2134 | } | |
2135 | ||
2136 | /* | |
2137 | * Now for the recovery code. | |
2138 | * Recovery happens across physical sectors. | |
2139 | * We recover all non-is_sync drives by finding the virtual address of | |
2140 | * each, and then choose a working drive that also has that virt address. | |
2141 | * There is a separate r10_bio for each non-in_sync drive. | |
2142 | * Only the first two slots are in use. The first for reading, | |
2143 | * The second for writing. | |
2144 | * | |
2145 | */ | |
9f2c9d12 | 2146 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2147 | { |
2148 | /* We got a read error during recovery. | |
2149 | * We repeat the read in smaller page-sized sections. | |
2150 | * If a read succeeds, write it to the new device or record | |
2151 | * a bad block if we cannot. | |
2152 | * If a read fails, record a bad block on both old and | |
2153 | * new devices. | |
2154 | */ | |
fd01b88c | 2155 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2156 | struct r10conf *conf = mddev->private; |
5e570289 N |
2157 | struct bio *bio = r10_bio->devs[0].bio; |
2158 | sector_t sect = 0; | |
2159 | int sectors = r10_bio->sectors; | |
2160 | int idx = 0; | |
2161 | int dr = r10_bio->devs[0].devnum; | |
2162 | int dw = r10_bio->devs[1].devnum; | |
cdb76be3 | 2163 | struct page **pages = get_resync_pages(bio)->pages; |
5e570289 N |
2164 | |
2165 | while (sectors) { | |
2166 | int s = sectors; | |
3cb03002 | 2167 | struct md_rdev *rdev; |
5e570289 N |
2168 | sector_t addr; |
2169 | int ok; | |
2170 | ||
2171 | if (s > (PAGE_SIZE>>9)) | |
2172 | s = PAGE_SIZE >> 9; | |
2173 | ||
2174 | rdev = conf->mirrors[dr].rdev; | |
2175 | addr = r10_bio->devs[0].addr + sect, | |
2176 | ok = sync_page_io(rdev, | |
2177 | addr, | |
2178 | s << 9, | |
cdb76be3 | 2179 | pages[idx], |
796a5cf0 | 2180 | REQ_OP_READ, 0, false); |
5e570289 N |
2181 | if (ok) { |
2182 | rdev = conf->mirrors[dw].rdev; | |
2183 | addr = r10_bio->devs[1].addr + sect; | |
2184 | ok = sync_page_io(rdev, | |
2185 | addr, | |
2186 | s << 9, | |
cdb76be3 | 2187 | pages[idx], |
796a5cf0 | 2188 | REQ_OP_WRITE, 0, false); |
b7044d41 | 2189 | if (!ok) { |
5e570289 | 2190 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2191 | if (!test_and_set_bit(WantReplacement, |
2192 | &rdev->flags)) | |
2193 | set_bit(MD_RECOVERY_NEEDED, | |
2194 | &rdev->mddev->recovery); | |
2195 | } | |
5e570289 N |
2196 | } |
2197 | if (!ok) { | |
2198 | /* We don't worry if we cannot set a bad block - | |
2199 | * it really is bad so there is no loss in not | |
2200 | * recording it yet | |
2201 | */ | |
2202 | rdev_set_badblocks(rdev, addr, s, 0); | |
2203 | ||
2204 | if (rdev != conf->mirrors[dw].rdev) { | |
2205 | /* need bad block on destination too */ | |
3cb03002 | 2206 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2207 | addr = r10_bio->devs[1].addr + sect; |
2208 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2209 | if (!ok) { | |
2210 | /* just abort the recovery */ | |
08464e09 N |
2211 | pr_notice("md/raid10:%s: recovery aborted due to read error\n", |
2212 | mdname(mddev)); | |
5e570289 N |
2213 | |
2214 | conf->mirrors[dw].recovery_disabled | |
2215 | = mddev->recovery_disabled; | |
2216 | set_bit(MD_RECOVERY_INTR, | |
2217 | &mddev->recovery); | |
2218 | break; | |
2219 | } | |
2220 | } | |
2221 | } | |
2222 | ||
2223 | sectors -= s; | |
2224 | sect += s; | |
2225 | idx++; | |
2226 | } | |
2227 | } | |
1da177e4 | 2228 | |
9f2c9d12 | 2229 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2230 | { |
e879a879 | 2231 | struct r10conf *conf = mddev->private; |
c65060ad | 2232 | int d; |
24afd80d | 2233 | struct bio *wbio, *wbio2; |
1da177e4 | 2234 | |
5e570289 N |
2235 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2236 | fix_recovery_read_error(r10_bio); | |
2237 | end_sync_request(r10_bio); | |
2238 | return; | |
2239 | } | |
2240 | ||
c65060ad NK |
2241 | /* |
2242 | * share the pages with the first bio | |
1da177e4 LT |
2243 | * and submit the write request |
2244 | */ | |
1da177e4 | 2245 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2246 | wbio = r10_bio->devs[1].bio; |
2247 | wbio2 = r10_bio->devs[1].repl_bio; | |
0eb25bb0 N |
2248 | /* Need to test wbio2->bi_end_io before we call |
2249 | * generic_make_request as if the former is NULL, | |
2250 | * the latter is free to free wbio2. | |
2251 | */ | |
2252 | if (wbio2 && !wbio2->bi_end_io) | |
2253 | wbio2 = NULL; | |
24afd80d N |
2254 | if (wbio->bi_end_io) { |
2255 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2256 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
24afd80d N |
2257 | generic_make_request(wbio); |
2258 | } | |
0eb25bb0 | 2259 | if (wbio2) { |
24afd80d N |
2260 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2261 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2262 | bio_sectors(wbio2)); |
24afd80d N |
2263 | generic_make_request(wbio2); |
2264 | } | |
1da177e4 LT |
2265 | } |
2266 | ||
1e50915f RB |
2267 | /* |
2268 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2269 | * We halve the read error count for every hour that has elapsed | |
2270 | * since the last recorded read error. | |
2271 | * | |
2272 | */ | |
fd01b88c | 2273 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f | 2274 | { |
0e3ef49e | 2275 | long cur_time_mon; |
1e50915f RB |
2276 | unsigned long hours_since_last; |
2277 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2278 | ||
0e3ef49e | 2279 | cur_time_mon = ktime_get_seconds(); |
1e50915f | 2280 | |
0e3ef49e | 2281 | if (rdev->last_read_error == 0) { |
1e50915f RB |
2282 | /* first time we've seen a read error */ |
2283 | rdev->last_read_error = cur_time_mon; | |
2284 | return; | |
2285 | } | |
2286 | ||
0e3ef49e AB |
2287 | hours_since_last = (long)(cur_time_mon - |
2288 | rdev->last_read_error) / 3600; | |
1e50915f RB |
2289 | |
2290 | rdev->last_read_error = cur_time_mon; | |
2291 | ||
2292 | /* | |
2293 | * if hours_since_last is > the number of bits in read_errors | |
2294 | * just set read errors to 0. We do this to avoid | |
2295 | * overflowing the shift of read_errors by hours_since_last. | |
2296 | */ | |
2297 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2298 | atomic_set(&rdev->read_errors, 0); | |
2299 | else | |
2300 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2301 | } | |
2302 | ||
3cb03002 | 2303 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2304 | int sectors, struct page *page, int rw) |
2305 | { | |
2306 | sector_t first_bad; | |
2307 | int bad_sectors; | |
2308 | ||
2309 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2310 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2311 | return -1; | |
796a5cf0 | 2312 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
58c54fcc N |
2313 | /* success */ |
2314 | return 1; | |
b7044d41 | 2315 | if (rw == WRITE) { |
58c54fcc | 2316 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2317 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2318 | set_bit(MD_RECOVERY_NEEDED, | |
2319 | &rdev->mddev->recovery); | |
2320 | } | |
58c54fcc N |
2321 | /* need to record an error - either for the block or the device */ |
2322 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2323 | md_error(rdev->mddev, rdev); | |
2324 | return 0; | |
2325 | } | |
2326 | ||
1da177e4 LT |
2327 | /* |
2328 | * This is a kernel thread which: | |
2329 | * | |
2330 | * 1. Retries failed read operations on working mirrors. | |
2331 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2332 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2333 | */ |
2334 | ||
e879a879 | 2335 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2336 | { |
2337 | int sect = 0; /* Offset from r10_bio->sector */ | |
2338 | int sectors = r10_bio->sectors; | |
3cb03002 | 2339 | struct md_rdev*rdev; |
1e50915f | 2340 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2341 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2342 | |
7c4e06ff N |
2343 | /* still own a reference to this rdev, so it cannot |
2344 | * have been cleared recently. | |
2345 | */ | |
2346 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2347 | |
7c4e06ff N |
2348 | if (test_bit(Faulty, &rdev->flags)) |
2349 | /* drive has already been failed, just ignore any | |
2350 | more fix_read_error() attempts */ | |
2351 | return; | |
1e50915f | 2352 | |
7c4e06ff N |
2353 | check_decay_read_errors(mddev, rdev); |
2354 | atomic_inc(&rdev->read_errors); | |
2355 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2356 | char b[BDEVNAME_SIZE]; | |
2357 | bdevname(rdev->bdev, b); | |
1e50915f | 2358 | |
08464e09 N |
2359 | pr_notice("md/raid10:%s: %s: Raid device exceeded read_error threshold [cur %d:max %d]\n", |
2360 | mdname(mddev), b, | |
2361 | atomic_read(&rdev->read_errors), max_read_errors); | |
2362 | pr_notice("md/raid10:%s: %s: Failing raid device\n", | |
2363 | mdname(mddev), b); | |
d683c8e0 | 2364 | md_error(mddev, rdev); |
fae8cc5e | 2365 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2366 | return; |
1e50915f | 2367 | } |
1e50915f | 2368 | |
6814d536 N |
2369 | while(sectors) { |
2370 | int s = sectors; | |
2371 | int sl = r10_bio->read_slot; | |
2372 | int success = 0; | |
2373 | int start; | |
2374 | ||
2375 | if (s > (PAGE_SIZE>>9)) | |
2376 | s = PAGE_SIZE >> 9; | |
2377 | ||
2378 | rcu_read_lock(); | |
2379 | do { | |
8dbed5ce N |
2380 | sector_t first_bad; |
2381 | int bad_sectors; | |
2382 | ||
0544a21d | 2383 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2384 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2385 | if (rdev && | |
8dbed5ce | 2386 | test_bit(In_sync, &rdev->flags) && |
f5b67ae8 | 2387 | !test_bit(Faulty, &rdev->flags) && |
8dbed5ce N |
2388 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, |
2389 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2390 | atomic_inc(&rdev->nr_pending); |
2391 | rcu_read_unlock(); | |
2b193363 | 2392 | success = sync_page_io(rdev, |
6814d536 | 2393 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2394 | sect, |
6814d536 | 2395 | s<<9, |
796a5cf0 MC |
2396 | conf->tmppage, |
2397 | REQ_OP_READ, 0, false); | |
6814d536 N |
2398 | rdev_dec_pending(rdev, mddev); |
2399 | rcu_read_lock(); | |
2400 | if (success) | |
2401 | break; | |
2402 | } | |
2403 | sl++; | |
2404 | if (sl == conf->copies) | |
2405 | sl = 0; | |
2406 | } while (!success && sl != r10_bio->read_slot); | |
2407 | rcu_read_unlock(); | |
2408 | ||
2409 | if (!success) { | |
58c54fcc N |
2410 | /* Cannot read from anywhere, just mark the block |
2411 | * as bad on the first device to discourage future | |
2412 | * reads. | |
2413 | */ | |
6814d536 | 2414 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2415 | rdev = conf->mirrors[dn].rdev; |
2416 | ||
2417 | if (!rdev_set_badblocks( | |
2418 | rdev, | |
2419 | r10_bio->devs[r10_bio->read_slot].addr | |
2420 | + sect, | |
fae8cc5e | 2421 | s, 0)) { |
58c54fcc | 2422 | md_error(mddev, rdev); |
fae8cc5e N |
2423 | r10_bio->devs[r10_bio->read_slot].bio |
2424 | = IO_BLOCKED; | |
2425 | } | |
6814d536 N |
2426 | break; |
2427 | } | |
2428 | ||
2429 | start = sl; | |
2430 | /* write it back and re-read */ | |
2431 | rcu_read_lock(); | |
2432 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2433 | char b[BDEVNAME_SIZE]; |
0544a21d | 2434 | |
6814d536 N |
2435 | if (sl==0) |
2436 | sl = conf->copies; | |
2437 | sl--; | |
2438 | d = r10_bio->devs[sl].devnum; | |
2439 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2440 | if (!rdev || |
f5b67ae8 | 2441 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2442 | !test_bit(In_sync, &rdev->flags)) |
2443 | continue; | |
2444 | ||
2445 | atomic_inc(&rdev->nr_pending); | |
2446 | rcu_read_unlock(); | |
58c54fcc N |
2447 | if (r10_sync_page_io(rdev, |
2448 | r10_bio->devs[sl].addr + | |
2449 | sect, | |
055d3747 | 2450 | s, conf->tmppage, WRITE) |
1294b9c9 N |
2451 | == 0) { |
2452 | /* Well, this device is dead */ | |
08464e09 N |
2453 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %s)\n", |
2454 | mdname(mddev), s, | |
2455 | (unsigned long long)( | |
2456 | sect + | |
2457 | choose_data_offset(r10_bio, | |
2458 | rdev)), | |
2459 | bdevname(rdev->bdev, b)); | |
2460 | pr_notice("md/raid10:%s: %s: failing drive\n", | |
2461 | mdname(mddev), | |
2462 | bdevname(rdev->bdev, b)); | |
6814d536 | 2463 | } |
1294b9c9 N |
2464 | rdev_dec_pending(rdev, mddev); |
2465 | rcu_read_lock(); | |
6814d536 N |
2466 | } |
2467 | sl = start; | |
2468 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2469 | char b[BDEVNAME_SIZE]; |
0544a21d | 2470 | |
6814d536 N |
2471 | if (sl==0) |
2472 | sl = conf->copies; | |
2473 | sl--; | |
2474 | d = r10_bio->devs[sl].devnum; | |
2475 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2476 | if (!rdev || |
f5b67ae8 | 2477 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2478 | !test_bit(In_sync, &rdev->flags)) |
2479 | continue; | |
6814d536 | 2480 | |
1294b9c9 N |
2481 | atomic_inc(&rdev->nr_pending); |
2482 | rcu_read_unlock(); | |
58c54fcc N |
2483 | switch (r10_sync_page_io(rdev, |
2484 | r10_bio->devs[sl].addr + | |
2485 | sect, | |
055d3747 | 2486 | s, conf->tmppage, |
58c54fcc N |
2487 | READ)) { |
2488 | case 0: | |
1294b9c9 | 2489 | /* Well, this device is dead */ |
08464e09 | 2490 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2491 | mdname(mddev), s, |
2492 | (unsigned long long)( | |
f8c9e74f N |
2493 | sect + |
2494 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 | 2495 | bdevname(rdev->bdev, b)); |
08464e09 | 2496 | pr_notice("md/raid10:%s: %s: failing drive\n", |
1294b9c9 N |
2497 | mdname(mddev), |
2498 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2499 | break; |
2500 | case 1: | |
08464e09 | 2501 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2502 | mdname(mddev), s, |
2503 | (unsigned long long)( | |
f8c9e74f N |
2504 | sect + |
2505 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2506 | bdevname(rdev->bdev, b)); |
2507 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2508 | } |
1294b9c9 N |
2509 | |
2510 | rdev_dec_pending(rdev, mddev); | |
2511 | rcu_read_lock(); | |
6814d536 N |
2512 | } |
2513 | rcu_read_unlock(); | |
2514 | ||
2515 | sectors -= s; | |
2516 | sect += s; | |
2517 | } | |
2518 | } | |
2519 | ||
9f2c9d12 | 2520 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2521 | { |
2522 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2523 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2524 | struct r10conf *conf = mddev->private; |
3cb03002 | 2525 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2526 | /* bio has the data to be written to slot 'i' where |
2527 | * we just recently had a write error. | |
2528 | * We repeatedly clone the bio and trim down to one block, | |
2529 | * then try the write. Where the write fails we record | |
2530 | * a bad block. | |
2531 | * It is conceivable that the bio doesn't exactly align with | |
2532 | * blocks. We must handle this. | |
2533 | * | |
2534 | * We currently own a reference to the rdev. | |
2535 | */ | |
2536 | ||
2537 | int block_sectors; | |
2538 | sector_t sector; | |
2539 | int sectors; | |
2540 | int sect_to_write = r10_bio->sectors; | |
2541 | int ok = 1; | |
2542 | ||
2543 | if (rdev->badblocks.shift < 0) | |
2544 | return 0; | |
2545 | ||
f04ebb0b N |
2546 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2547 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2548 | sector = r10_bio->sector; |
2549 | sectors = ((r10_bio->sector + block_sectors) | |
2550 | & ~(sector_t)(block_sectors - 1)) | |
2551 | - sector; | |
2552 | ||
2553 | while (sect_to_write) { | |
2554 | struct bio *wbio; | |
27028626 | 2555 | sector_t wsector; |
bd870a16 N |
2556 | if (sectors > sect_to_write) |
2557 | sectors = sect_to_write; | |
2558 | /* Write at 'sector' for 'sectors' */ | |
d7a10308 | 2559 | wbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
4f024f37 | 2560 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
27028626 TM |
2561 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); |
2562 | wbio->bi_iter.bi_sector = wsector + | |
2563 | choose_data_offset(r10_bio, rdev); | |
bd870a16 | 2564 | wbio->bi_bdev = rdev->bdev; |
796a5cf0 | 2565 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
2566 | |
2567 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 | 2568 | /* Failure! */ |
27028626 | 2569 | ok = rdev_set_badblocks(rdev, wsector, |
bd870a16 N |
2570 | sectors, 0) |
2571 | && ok; | |
2572 | ||
2573 | bio_put(wbio); | |
2574 | sect_to_write -= sectors; | |
2575 | sector += sectors; | |
2576 | sectors = block_sectors; | |
2577 | } | |
2578 | return ok; | |
2579 | } | |
2580 | ||
9f2c9d12 | 2581 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2582 | { |
2583 | int slot = r10_bio->read_slot; | |
560f8e55 | 2584 | struct bio *bio; |
e879a879 | 2585 | struct r10conf *conf = mddev->private; |
abbf098e | 2586 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
109e3765 N |
2587 | dev_t bio_dev; |
2588 | sector_t bio_last_sector; | |
560f8e55 N |
2589 | |
2590 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2591 | * the block and we can fix it. | |
2592 | * We freeze all other IO, and try reading the block from | |
2593 | * other devices. When we find one, we re-write | |
2594 | * and check it that fixes the read error. | |
2595 | * This is all done synchronously while the array is | |
2596 | * frozen. | |
2597 | */ | |
fae8cc5e | 2598 | bio = r10_bio->devs[slot].bio; |
109e3765 N |
2599 | bio_dev = bio->bi_bdev->bd_dev; |
2600 | bio_last_sector = r10_bio->devs[slot].addr + rdev->data_offset + r10_bio->sectors; | |
fae8cc5e N |
2601 | bio_put(bio); |
2602 | r10_bio->devs[slot].bio = NULL; | |
2603 | ||
8d3ca83d N |
2604 | if (mddev->ro) |
2605 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2606 | else if (!test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2607 | freeze_array(conf, 1); |
560f8e55 N |
2608 | fix_read_error(conf, mddev, r10_bio); |
2609 | unfreeze_array(conf); | |
fae8cc5e | 2610 | } else |
8d3ca83d | 2611 | md_error(mddev, rdev); |
fae8cc5e | 2612 | |
abbf098e | 2613 | rdev_dec_pending(rdev, mddev); |
545250f2 N |
2614 | allow_barrier(conf); |
2615 | r10_bio->state = 0; | |
2616 | raid10_read_request(mddev, r10_bio->master_bio, r10_bio); | |
560f8e55 N |
2617 | } |
2618 | ||
e879a879 | 2619 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2620 | { |
2621 | /* Some sort of write request has finished and it | |
2622 | * succeeded in writing where we thought there was a | |
2623 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2624 | * Or possibly if failed and we need to record |
2625 | * a bad block. | |
749c55e9 N |
2626 | */ |
2627 | int m; | |
3cb03002 | 2628 | struct md_rdev *rdev; |
749c55e9 N |
2629 | |
2630 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2631 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2632 | for (m = 0; m < conf->copies; m++) { |
2633 | int dev = r10_bio->devs[m].devnum; | |
2634 | rdev = conf->mirrors[dev].rdev; | |
2635 | if (r10_bio->devs[m].bio == NULL) | |
2636 | continue; | |
4e4cbee9 | 2637 | if (!r10_bio->devs[m].bio->bi_status) { |
749c55e9 N |
2638 | rdev_clear_badblocks( |
2639 | rdev, | |
2640 | r10_bio->devs[m].addr, | |
c6563a8c | 2641 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2642 | } else { |
2643 | if (!rdev_set_badblocks( | |
2644 | rdev, | |
2645 | r10_bio->devs[m].addr, | |
2646 | r10_bio->sectors, 0)) | |
2647 | md_error(conf->mddev, rdev); | |
749c55e9 | 2648 | } |
9ad1aefc N |
2649 | rdev = conf->mirrors[dev].replacement; |
2650 | if (r10_bio->devs[m].repl_bio == NULL) | |
2651 | continue; | |
4246a0b6 | 2652 | |
4e4cbee9 | 2653 | if (!r10_bio->devs[m].repl_bio->bi_status) { |
9ad1aefc N |
2654 | rdev_clear_badblocks( |
2655 | rdev, | |
2656 | r10_bio->devs[m].addr, | |
c6563a8c | 2657 | r10_bio->sectors, 0); |
9ad1aefc N |
2658 | } else { |
2659 | if (!rdev_set_badblocks( | |
2660 | rdev, | |
2661 | r10_bio->devs[m].addr, | |
2662 | r10_bio->sectors, 0)) | |
2663 | md_error(conf->mddev, rdev); | |
2664 | } | |
1a0b7cd8 | 2665 | } |
749c55e9 N |
2666 | put_buf(r10_bio); |
2667 | } else { | |
95af587e | 2668 | bool fail = false; |
bd870a16 N |
2669 | for (m = 0; m < conf->copies; m++) { |
2670 | int dev = r10_bio->devs[m].devnum; | |
2671 | struct bio *bio = r10_bio->devs[m].bio; | |
2672 | rdev = conf->mirrors[dev].rdev; | |
2673 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2674 | rdev_clear_badblocks( |
2675 | rdev, | |
2676 | r10_bio->devs[m].addr, | |
c6563a8c | 2677 | r10_bio->sectors, 0); |
749c55e9 | 2678 | rdev_dec_pending(rdev, conf->mddev); |
4e4cbee9 | 2679 | } else if (bio != NULL && bio->bi_status) { |
95af587e | 2680 | fail = true; |
bd870a16 N |
2681 | if (!narrow_write_error(r10_bio, m)) { |
2682 | md_error(conf->mddev, rdev); | |
2683 | set_bit(R10BIO_Degraded, | |
2684 | &r10_bio->state); | |
2685 | } | |
2686 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2687 | } |
475b0321 N |
2688 | bio = r10_bio->devs[m].repl_bio; |
2689 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2690 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2691 | rdev_clear_badblocks( |
2692 | rdev, | |
2693 | r10_bio->devs[m].addr, | |
c6563a8c | 2694 | r10_bio->sectors, 0); |
475b0321 N |
2695 | rdev_dec_pending(rdev, conf->mddev); |
2696 | } | |
bd870a16 | 2697 | } |
95af587e N |
2698 | if (fail) { |
2699 | spin_lock_irq(&conf->device_lock); | |
2700 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 2701 | conf->nr_queued++; |
95af587e | 2702 | spin_unlock_irq(&conf->device_lock); |
cf25ae78 GJ |
2703 | /* |
2704 | * In case freeze_array() is waiting for condition | |
2705 | * nr_pending == nr_queued + extra to be true. | |
2706 | */ | |
2707 | wake_up(&conf->wait_barrier); | |
95af587e | 2708 | md_wakeup_thread(conf->mddev->thread); |
c340702c N |
2709 | } else { |
2710 | if (test_bit(R10BIO_WriteError, | |
2711 | &r10_bio->state)) | |
2712 | close_write(r10_bio); | |
95af587e | 2713 | raid_end_bio_io(r10_bio); |
c340702c | 2714 | } |
749c55e9 N |
2715 | } |
2716 | } | |
2717 | ||
4ed8731d | 2718 | static void raid10d(struct md_thread *thread) |
1da177e4 | 2719 | { |
4ed8731d | 2720 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2721 | struct r10bio *r10_bio; |
1da177e4 | 2722 | unsigned long flags; |
e879a879 | 2723 | struct r10conf *conf = mddev->private; |
1da177e4 | 2724 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2725 | struct blk_plug plug; |
1da177e4 LT |
2726 | |
2727 | md_check_recovery(mddev); | |
1da177e4 | 2728 | |
95af587e | 2729 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2730 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
95af587e N |
2731 | LIST_HEAD(tmp); |
2732 | spin_lock_irqsave(&conf->device_lock, flags); | |
2953079c | 2733 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
23ddba80 SL |
2734 | while (!list_empty(&conf->bio_end_io_list)) { |
2735 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2736 | conf->nr_queued--; | |
2737 | } | |
95af587e N |
2738 | } |
2739 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2740 | while (!list_empty(&tmp)) { | |
a452744b MP |
2741 | r10_bio = list_first_entry(&tmp, struct r10bio, |
2742 | retry_list); | |
95af587e | 2743 | list_del(&r10_bio->retry_list); |
c340702c N |
2744 | if (mddev->degraded) |
2745 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
2746 | ||
2747 | if (test_bit(R10BIO_WriteError, | |
2748 | &r10_bio->state)) | |
2749 | close_write(r10_bio); | |
95af587e N |
2750 | raid_end_bio_io(r10_bio); |
2751 | } | |
2752 | } | |
2753 | ||
e1dfa0a2 | 2754 | blk_start_plug(&plug); |
1da177e4 | 2755 | for (;;) { |
6cce3b23 | 2756 | |
0021b7bc | 2757 | flush_pending_writes(conf); |
6cce3b23 | 2758 | |
a35e63ef N |
2759 | spin_lock_irqsave(&conf->device_lock, flags); |
2760 | if (list_empty(head)) { | |
2761 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2762 | break; |
a35e63ef | 2763 | } |
9f2c9d12 | 2764 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2765 | list_del(head->prev); |
4443ae10 | 2766 | conf->nr_queued--; |
1da177e4 LT |
2767 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2768 | ||
2769 | mddev = r10_bio->mddev; | |
070ec55d | 2770 | conf = mddev->private; |
bd870a16 N |
2771 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2772 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 2773 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
2774 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
2775 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 2776 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 2777 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2778 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2779 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2780 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2781 | handle_read_error(mddev, r10_bio); |
fc9977dd N |
2782 | else |
2783 | WARN_ON_ONCE(1); | |
560f8e55 | 2784 | |
1d9d5241 | 2785 | cond_resched(); |
2953079c | 2786 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2787 | md_check_recovery(mddev); |
1da177e4 | 2788 | } |
e1dfa0a2 | 2789 | blk_finish_plug(&plug); |
1da177e4 LT |
2790 | } |
2791 | ||
e879a879 | 2792 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2793 | { |
2794 | int buffs; | |
69335ef3 | 2795 | int i; |
1da177e4 LT |
2796 | |
2797 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2798 | BUG_ON(conf->r10buf_pool); |
69335ef3 | 2799 | conf->have_replacement = 0; |
5cf00fcd | 2800 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
2801 | if (conf->mirrors[i].replacement) |
2802 | conf->have_replacement = 1; | |
1da177e4 LT |
2803 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2804 | if (!conf->r10buf_pool) | |
2805 | return -ENOMEM; | |
2806 | conf->next_resync = 0; | |
2807 | return 0; | |
2808 | } | |
2809 | ||
2810 | /* | |
2811 | * perform a "sync" on one "block" | |
2812 | * | |
2813 | * We need to make sure that no normal I/O request - particularly write | |
2814 | * requests - conflict with active sync requests. | |
2815 | * | |
2816 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2817 | * that can be installed to exclude normal IO requests. | |
2818 | * | |
2819 | * Resync and recovery are handled very differently. | |
2820 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2821 | * | |
2822 | * For resync, we iterate over virtual addresses, read all copies, | |
2823 | * and update if there are differences. If only one copy is live, | |
2824 | * skip it. | |
2825 | * For recovery, we iterate over physical addresses, read a good | |
2826 | * value for each non-in_sync drive, and over-write. | |
2827 | * | |
2828 | * So, for recovery we may have several outstanding complex requests for a | |
2829 | * given address, one for each out-of-sync device. We model this by allocating | |
2830 | * a number of r10_bio structures, one for each out-of-sync device. | |
2831 | * As we setup these structures, we collect all bio's together into a list | |
2832 | * which we then process collectively to add pages, and then process again | |
2833 | * to pass to generic_make_request. | |
2834 | * | |
2835 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2836 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2837 | * has its remaining count decremented to 0, the whole complex operation | |
2838 | * is complete. | |
2839 | * | |
2840 | */ | |
2841 | ||
849674e4 | 2842 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 2843 | int *skipped) |
1da177e4 | 2844 | { |
e879a879 | 2845 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2846 | struct r10bio *r10_bio; |
1da177e4 LT |
2847 | struct bio *biolist = NULL, *bio; |
2848 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2849 | int i; |
6cce3b23 | 2850 | int max_sync; |
57dab0bd | 2851 | sector_t sync_blocks; |
1da177e4 LT |
2852 | sector_t sectors_skipped = 0; |
2853 | int chunks_skipped = 0; | |
5cf00fcd | 2854 | sector_t chunk_mask = conf->geo.chunk_mask; |
022e510f | 2855 | int page_idx = 0; |
1da177e4 LT |
2856 | |
2857 | if (!conf->r10buf_pool) | |
2858 | if (init_resync(conf)) | |
57afd89f | 2859 | return 0; |
1da177e4 | 2860 | |
7e83ccbe MW |
2861 | /* |
2862 | * Allow skipping a full rebuild for incremental assembly | |
2863 | * of a clean array, like RAID1 does. | |
2864 | */ | |
2865 | if (mddev->bitmap == NULL && | |
2866 | mddev->recovery_cp == MaxSector && | |
13765120 N |
2867 | mddev->reshape_position == MaxSector && |
2868 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 2869 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 2870 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
2871 | conf->fullsync == 0) { |
2872 | *skipped = 1; | |
13765120 | 2873 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
2874 | } |
2875 | ||
1da177e4 | 2876 | skipped: |
58c0fed4 | 2877 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
2878 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
2879 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
2880 | max_sector = mddev->resync_max_sectors; |
2881 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
2882 | /* If we aborted, we need to abort the |
2883 | * sync on the 'current' bitmap chucks (there can | |
2884 | * be several when recovering multiple devices). | |
2885 | * as we may have started syncing it but not finished. | |
2886 | * We can find the current address in | |
2887 | * mddev->curr_resync, but for recovery, | |
2888 | * we need to convert that to several | |
2889 | * virtual addresses. | |
2890 | */ | |
3ea7daa5 N |
2891 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
2892 | end_reshape(conf); | |
b3968552 | 2893 | close_sync(conf); |
3ea7daa5 N |
2894 | return 0; |
2895 | } | |
2896 | ||
6cce3b23 N |
2897 | if (mddev->curr_resync < max_sector) { /* aborted */ |
2898 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2899 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
2900 | &sync_blocks, 1); | |
5cf00fcd | 2901 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
2902 | sector_t sect = |
2903 | raid10_find_virt(conf, mddev->curr_resync, i); | |
2904 | bitmap_end_sync(mddev->bitmap, sect, | |
2905 | &sync_blocks, 1); | |
2906 | } | |
9ad1aefc N |
2907 | } else { |
2908 | /* completed sync */ | |
2909 | if ((!mddev->bitmap || conf->fullsync) | |
2910 | && conf->have_replacement | |
2911 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
2912 | /* Completed a full sync so the replacements | |
2913 | * are now fully recovered. | |
2914 | */ | |
f90145f3 N |
2915 | rcu_read_lock(); |
2916 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
2917 | struct md_rdev *rdev = | |
2918 | rcu_dereference(conf->mirrors[i].replacement); | |
2919 | if (rdev) | |
2920 | rdev->recovery_offset = MaxSector; | |
2921 | } | |
2922 | rcu_read_unlock(); | |
9ad1aefc | 2923 | } |
6cce3b23 | 2924 | conf->fullsync = 0; |
9ad1aefc | 2925 | } |
6cce3b23 | 2926 | bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2927 | close_sync(conf); |
57afd89f | 2928 | *skipped = 1; |
1da177e4 LT |
2929 | return sectors_skipped; |
2930 | } | |
3ea7daa5 N |
2931 | |
2932 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
2933 | return reshape_request(mddev, sector_nr, skipped); | |
2934 | ||
5cf00fcd | 2935 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
2936 | /* if there has been nothing to do on any drive, |
2937 | * then there is nothing to do at all.. | |
2938 | */ | |
57afd89f N |
2939 | *skipped = 1; |
2940 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
2941 | } |
2942 | ||
c6207277 N |
2943 | if (max_sector > mddev->resync_max) |
2944 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
2945 | ||
1da177e4 LT |
2946 | /* make sure whole request will fit in a chunk - if chunks |
2947 | * are meaningful | |
2948 | */ | |
5cf00fcd N |
2949 | if (conf->geo.near_copies < conf->geo.raid_disks && |
2950 | max_sector > (sector_nr | chunk_mask)) | |
2951 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 | 2952 | |
7ac50447 TM |
2953 | /* |
2954 | * If there is non-resync activity waiting for a turn, then let it | |
2955 | * though before starting on this new sync request. | |
2956 | */ | |
2957 | if (conf->nr_waiting) | |
2958 | schedule_timeout_uninterruptible(1); | |
2959 | ||
1da177e4 LT |
2960 | /* Again, very different code for resync and recovery. |
2961 | * Both must result in an r10bio with a list of bios that | |
2962 | * have bi_end_io, bi_sector, bi_bdev set, | |
2963 | * and bi_private set to the r10bio. | |
2964 | * For recovery, we may actually create several r10bios | |
2965 | * with 2 bios in each, that correspond to the bios in the main one. | |
2966 | * In this case, the subordinate r10bios link back through a | |
2967 | * borrowed master_bio pointer, and the counter in the master | |
2968 | * includes a ref from each subordinate. | |
2969 | */ | |
2970 | /* First, we decide what to do and set ->bi_end_io | |
2971 | * To end_sync_read if we want to read, and | |
2972 | * end_sync_write if we will want to write. | |
2973 | */ | |
2974 | ||
6cce3b23 | 2975 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
2976 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
2977 | /* recovery... the complicated one */ | |
e875ecea | 2978 | int j; |
1da177e4 LT |
2979 | r10_bio = NULL; |
2980 | ||
5cf00fcd | 2981 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 2982 | int still_degraded; |
9f2c9d12 | 2983 | struct r10bio *rb2; |
ab9d47e9 N |
2984 | sector_t sect; |
2985 | int must_sync; | |
e875ecea | 2986 | int any_working; |
dc280d98 | 2987 | struct raid10_info *mirror = &conf->mirrors[i]; |
f90145f3 | 2988 | struct md_rdev *mrdev, *mreplace; |
24afd80d | 2989 | |
f90145f3 N |
2990 | rcu_read_lock(); |
2991 | mrdev = rcu_dereference(mirror->rdev); | |
2992 | mreplace = rcu_dereference(mirror->replacement); | |
2993 | ||
2994 | if ((mrdev == NULL || | |
f5b67ae8 | 2995 | test_bit(Faulty, &mrdev->flags) || |
f90145f3 N |
2996 | test_bit(In_sync, &mrdev->flags)) && |
2997 | (mreplace == NULL || | |
2998 | test_bit(Faulty, &mreplace->flags))) { | |
2999 | rcu_read_unlock(); | |
ab9d47e9 | 3000 | continue; |
f90145f3 | 3001 | } |
1da177e4 | 3002 | |
ab9d47e9 N |
3003 | still_degraded = 0; |
3004 | /* want to reconstruct this device */ | |
3005 | rb2 = r10_bio; | |
3006 | sect = raid10_find_virt(conf, sector_nr, i); | |
fc448a18 N |
3007 | if (sect >= mddev->resync_max_sectors) { |
3008 | /* last stripe is not complete - don't | |
3009 | * try to recover this sector. | |
3010 | */ | |
f90145f3 | 3011 | rcu_read_unlock(); |
fc448a18 N |
3012 | continue; |
3013 | } | |
f5b67ae8 N |
3014 | if (mreplace && test_bit(Faulty, &mreplace->flags)) |
3015 | mreplace = NULL; | |
24afd80d N |
3016 | /* Unless we are doing a full sync, or a replacement |
3017 | * we only need to recover the block if it is set in | |
3018 | * the bitmap | |
ab9d47e9 N |
3019 | */ |
3020 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3021 | &sync_blocks, 1); | |
3022 | if (sync_blocks < max_sync) | |
3023 | max_sync = sync_blocks; | |
3024 | if (!must_sync && | |
f90145f3 | 3025 | mreplace == NULL && |
ab9d47e9 N |
3026 | !conf->fullsync) { |
3027 | /* yep, skip the sync_blocks here, but don't assume | |
3028 | * that there will never be anything to do here | |
3029 | */ | |
3030 | chunks_skipped = -1; | |
f90145f3 | 3031 | rcu_read_unlock(); |
ab9d47e9 N |
3032 | continue; |
3033 | } | |
f90145f3 N |
3034 | atomic_inc(&mrdev->nr_pending); |
3035 | if (mreplace) | |
3036 | atomic_inc(&mreplace->nr_pending); | |
3037 | rcu_read_unlock(); | |
6cce3b23 | 3038 | |
ab9d47e9 | 3039 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 3040 | r10_bio->state = 0; |
ab9d47e9 N |
3041 | raise_barrier(conf, rb2 != NULL); |
3042 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 3043 | |
ab9d47e9 N |
3044 | r10_bio->master_bio = (struct bio*)rb2; |
3045 | if (rb2) | |
3046 | atomic_inc(&rb2->remaining); | |
3047 | r10_bio->mddev = mddev; | |
3048 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
3049 | r10_bio->sector = sect; | |
1da177e4 | 3050 | |
ab9d47e9 N |
3051 | raid10_find_phys(conf, r10_bio); |
3052 | ||
3053 | /* Need to check if the array will still be | |
3054 | * degraded | |
3055 | */ | |
f90145f3 N |
3056 | rcu_read_lock(); |
3057 | for (j = 0; j < conf->geo.raid_disks; j++) { | |
3058 | struct md_rdev *rdev = rcu_dereference( | |
3059 | conf->mirrors[j].rdev); | |
3060 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
ab9d47e9 | 3061 | still_degraded = 1; |
87fc767b | 3062 | break; |
1da177e4 | 3063 | } |
f90145f3 | 3064 | } |
ab9d47e9 N |
3065 | |
3066 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3067 | &sync_blocks, still_degraded); | |
3068 | ||
e875ecea | 3069 | any_working = 0; |
ab9d47e9 | 3070 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3071 | int k; |
ab9d47e9 | 3072 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3073 | sector_t from_addr, to_addr; |
f90145f3 N |
3074 | struct md_rdev *rdev = |
3075 | rcu_dereference(conf->mirrors[d].rdev); | |
40c356ce N |
3076 | sector_t sector, first_bad; |
3077 | int bad_sectors; | |
f90145f3 N |
3078 | if (!rdev || |
3079 | !test_bit(In_sync, &rdev->flags)) | |
ab9d47e9 N |
3080 | continue; |
3081 | /* This is where we read from */ | |
e875ecea | 3082 | any_working = 1; |
40c356ce N |
3083 | sector = r10_bio->devs[j].addr; |
3084 | ||
3085 | if (is_badblock(rdev, sector, max_sync, | |
3086 | &first_bad, &bad_sectors)) { | |
3087 | if (first_bad > sector) | |
3088 | max_sync = first_bad - sector; | |
3089 | else { | |
3090 | bad_sectors -= (sector | |
3091 | - first_bad); | |
3092 | if (max_sync > bad_sectors) | |
3093 | max_sync = bad_sectors; | |
3094 | continue; | |
3095 | } | |
3096 | } | |
ab9d47e9 N |
3097 | bio = r10_bio->devs[0].bio; |
3098 | bio->bi_next = biolist; | |
3099 | biolist = bio; | |
ab9d47e9 | 3100 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3101 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3102 | if (test_bit(FailFast, &rdev->flags)) |
3103 | bio->bi_opf |= MD_FAILFAST; | |
5e570289 | 3104 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3105 | bio->bi_iter.bi_sector = from_addr + |
3106 | rdev->data_offset; | |
24afd80d N |
3107 | bio->bi_bdev = rdev->bdev; |
3108 | atomic_inc(&rdev->nr_pending); | |
3109 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3110 | |
3111 | for (k=0; k<conf->copies; k++) | |
3112 | if (r10_bio->devs[k].devnum == i) | |
3113 | break; | |
3114 | BUG_ON(k == conf->copies); | |
5e570289 | 3115 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3116 | r10_bio->devs[0].devnum = d; |
5e570289 | 3117 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3118 | r10_bio->devs[1].devnum = i; |
5e570289 | 3119 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3120 | |
f90145f3 | 3121 | if (!test_bit(In_sync, &mrdev->flags)) { |
24afd80d N |
3122 | bio = r10_bio->devs[1].bio; |
3123 | bio->bi_next = biolist; | |
3124 | biolist = bio; | |
24afd80d | 3125 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3126 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3127 | bio->bi_iter.bi_sector = to_addr |
f90145f3 N |
3128 | + mrdev->data_offset; |
3129 | bio->bi_bdev = mrdev->bdev; | |
24afd80d N |
3130 | atomic_inc(&r10_bio->remaining); |
3131 | } else | |
3132 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3133 | ||
3134 | /* and maybe write to replacement */ | |
3135 | bio = r10_bio->devs[1].repl_bio; | |
3136 | if (bio) | |
3137 | bio->bi_end_io = NULL; | |
f90145f3 | 3138 | /* Note: if mreplace != NULL, then bio |
24afd80d N |
3139 | * cannot be NULL as r10buf_pool_alloc will |
3140 | * have allocated it. | |
3141 | * So the second test here is pointless. | |
3142 | * But it keeps semantic-checkers happy, and | |
3143 | * this comment keeps human reviewers | |
3144 | * happy. | |
3145 | */ | |
f90145f3 N |
3146 | if (mreplace == NULL || bio == NULL || |
3147 | test_bit(Faulty, &mreplace->flags)) | |
24afd80d N |
3148 | break; |
3149 | bio->bi_next = biolist; | |
3150 | biolist = bio; | |
24afd80d | 3151 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3152 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3153 | bio->bi_iter.bi_sector = to_addr + |
f90145f3 N |
3154 | mreplace->data_offset; |
3155 | bio->bi_bdev = mreplace->bdev; | |
24afd80d | 3156 | atomic_inc(&r10_bio->remaining); |
ab9d47e9 N |
3157 | break; |
3158 | } | |
f90145f3 | 3159 | rcu_read_unlock(); |
ab9d47e9 | 3160 | if (j == conf->copies) { |
e875ecea N |
3161 | /* Cannot recover, so abort the recovery or |
3162 | * record a bad block */ | |
e875ecea N |
3163 | if (any_working) { |
3164 | /* problem is that there are bad blocks | |
3165 | * on other device(s) | |
3166 | */ | |
3167 | int k; | |
3168 | for (k = 0; k < conf->copies; k++) | |
3169 | if (r10_bio->devs[k].devnum == i) | |
3170 | break; | |
24afd80d | 3171 | if (!test_bit(In_sync, |
f90145f3 | 3172 | &mrdev->flags) |
24afd80d | 3173 | && !rdev_set_badblocks( |
f90145f3 | 3174 | mrdev, |
24afd80d N |
3175 | r10_bio->devs[k].addr, |
3176 | max_sync, 0)) | |
3177 | any_working = 0; | |
f90145f3 | 3178 | if (mreplace && |
24afd80d | 3179 | !rdev_set_badblocks( |
f90145f3 | 3180 | mreplace, |
e875ecea N |
3181 | r10_bio->devs[k].addr, |
3182 | max_sync, 0)) | |
3183 | any_working = 0; | |
3184 | } | |
3185 | if (!any_working) { | |
3186 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3187 | &mddev->recovery)) | |
08464e09 | 3188 | pr_warn("md/raid10:%s: insufficient working devices for recovery.\n", |
e875ecea | 3189 | mdname(mddev)); |
24afd80d | 3190 | mirror->recovery_disabled |
e875ecea N |
3191 | = mddev->recovery_disabled; |
3192 | } | |
e8b84915 N |
3193 | put_buf(r10_bio); |
3194 | if (rb2) | |
3195 | atomic_dec(&rb2->remaining); | |
3196 | r10_bio = rb2; | |
f90145f3 N |
3197 | rdev_dec_pending(mrdev, mddev); |
3198 | if (mreplace) | |
3199 | rdev_dec_pending(mreplace, mddev); | |
ab9d47e9 | 3200 | break; |
1da177e4 | 3201 | } |
f90145f3 N |
3202 | rdev_dec_pending(mrdev, mddev); |
3203 | if (mreplace) | |
3204 | rdev_dec_pending(mreplace, mddev); | |
8d3ca83d N |
3205 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { |
3206 | /* Only want this if there is elsewhere to | |
3207 | * read from. 'j' is currently the first | |
3208 | * readable copy. | |
3209 | */ | |
3210 | int targets = 1; | |
3211 | for (; j < conf->copies; j++) { | |
3212 | int d = r10_bio->devs[j].devnum; | |
3213 | if (conf->mirrors[d].rdev && | |
3214 | test_bit(In_sync, | |
3215 | &conf->mirrors[d].rdev->flags)) | |
3216 | targets++; | |
3217 | } | |
3218 | if (targets == 1) | |
3219 | r10_bio->devs[0].bio->bi_opf | |
3220 | &= ~MD_FAILFAST; | |
3221 | } | |
ab9d47e9 | 3222 | } |
1da177e4 LT |
3223 | if (biolist == NULL) { |
3224 | while (r10_bio) { | |
9f2c9d12 N |
3225 | struct r10bio *rb2 = r10_bio; |
3226 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3227 | rb2->master_bio = NULL; |
3228 | put_buf(rb2); | |
3229 | } | |
3230 | goto giveup; | |
3231 | } | |
3232 | } else { | |
3233 | /* resync. Schedule a read for every block at this virt offset */ | |
3234 | int count = 0; | |
6cce3b23 | 3235 | |
c40f341f | 3236 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, 0); |
78200d45 | 3237 | |
6cce3b23 N |
3238 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
3239 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3240 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3241 | &mddev->recovery)) { | |
6cce3b23 N |
3242 | /* We can skip this block */ |
3243 | *skipped = 1; | |
3244 | return sync_blocks + sectors_skipped; | |
3245 | } | |
3246 | if (sync_blocks < max_sync) | |
3247 | max_sync = sync_blocks; | |
1da177e4 | 3248 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 3249 | r10_bio->state = 0; |
1da177e4 | 3250 | |
1da177e4 LT |
3251 | r10_bio->mddev = mddev; |
3252 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3253 | raise_barrier(conf, 0); |
3254 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3255 | |
3256 | r10_bio->master_bio = NULL; | |
3257 | r10_bio->sector = sector_nr; | |
3258 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3259 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3260 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3261 | |
5cf00fcd | 3262 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3263 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3264 | sector_t first_bad, sector; |
3265 | int bad_sectors; | |
f90145f3 | 3266 | struct md_rdev *rdev; |
40c356ce | 3267 | |
9ad1aefc N |
3268 | if (r10_bio->devs[i].repl_bio) |
3269 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3270 | ||
1da177e4 | 3271 | bio = r10_bio->devs[i].bio; |
4e4cbee9 | 3272 | bio->bi_status = BLK_STS_IOERR; |
f90145f3 N |
3273 | rcu_read_lock(); |
3274 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
3275 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3276 | rcu_read_unlock(); | |
1da177e4 | 3277 | continue; |
f90145f3 | 3278 | } |
40c356ce | 3279 | sector = r10_bio->devs[i].addr; |
f90145f3 | 3280 | if (is_badblock(rdev, sector, max_sync, |
40c356ce N |
3281 | &first_bad, &bad_sectors)) { |
3282 | if (first_bad > sector) | |
3283 | max_sync = first_bad - sector; | |
3284 | else { | |
3285 | bad_sectors -= (sector - first_bad); | |
3286 | if (max_sync > bad_sectors) | |
91502f09 | 3287 | max_sync = bad_sectors; |
f90145f3 | 3288 | rcu_read_unlock(); |
40c356ce N |
3289 | continue; |
3290 | } | |
3291 | } | |
f90145f3 | 3292 | atomic_inc(&rdev->nr_pending); |
1da177e4 LT |
3293 | atomic_inc(&r10_bio->remaining); |
3294 | bio->bi_next = biolist; | |
3295 | biolist = bio; | |
1da177e4 | 3296 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3297 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
1cdd1257 | 3298 | if (test_bit(FailFast, &rdev->flags)) |
8d3ca83d | 3299 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 N |
3300 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
3301 | bio->bi_bdev = rdev->bdev; | |
1da177e4 | 3302 | count++; |
9ad1aefc | 3303 | |
f90145f3 N |
3304 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3305 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3306 | rcu_read_unlock(); | |
9ad1aefc | 3307 | continue; |
f90145f3 N |
3308 | } |
3309 | atomic_inc(&rdev->nr_pending); | |
9ad1aefc N |
3310 | |
3311 | /* Need to set up for writing to the replacement */ | |
3312 | bio = r10_bio->devs[i].repl_bio; | |
4e4cbee9 | 3313 | bio->bi_status = BLK_STS_IOERR; |
9ad1aefc N |
3314 | |
3315 | sector = r10_bio->devs[i].addr; | |
9ad1aefc N |
3316 | bio->bi_next = biolist; |
3317 | biolist = bio; | |
9ad1aefc | 3318 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3319 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1cdd1257 | 3320 | if (test_bit(FailFast, &rdev->flags)) |
1919cbb2 | 3321 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 N |
3322 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
3323 | bio->bi_bdev = rdev->bdev; | |
9ad1aefc | 3324 | count++; |
1cdd1257 | 3325 | rcu_read_unlock(); |
1da177e4 LT |
3326 | } |
3327 | ||
3328 | if (count < 2) { | |
3329 | for (i=0; i<conf->copies; i++) { | |
3330 | int d = r10_bio->devs[i].devnum; | |
3331 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3332 | rdev_dec_pending(conf->mirrors[d].rdev, |
3333 | mddev); | |
9ad1aefc N |
3334 | if (r10_bio->devs[i].repl_bio && |
3335 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3336 | rdev_dec_pending( | |
3337 | conf->mirrors[d].replacement, | |
3338 | mddev); | |
1da177e4 LT |
3339 | } |
3340 | put_buf(r10_bio); | |
3341 | biolist = NULL; | |
3342 | goto giveup; | |
3343 | } | |
3344 | } | |
3345 | ||
1da177e4 | 3346 | nr_sectors = 0; |
6cce3b23 N |
3347 | if (sector_nr + max_sync < max_sector) |
3348 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3349 | do { |
3350 | struct page *page; | |
3351 | int len = PAGE_SIZE; | |
1da177e4 LT |
3352 | if (sector_nr + (len>>9) > max_sector) |
3353 | len = (max_sector - sector_nr) << 9; | |
3354 | if (len == 0) | |
3355 | break; | |
3356 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
f0250618 | 3357 | struct resync_pages *rp = get_resync_pages(bio); |
022e510f | 3358 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
3359 | /* |
3360 | * won't fail because the vec table is big enough | |
3361 | * to hold all these pages | |
3362 | */ | |
3363 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
3364 | } |
3365 | nr_sectors += len>>9; | |
3366 | sector_nr += len>>9; | |
022e510f | 3367 | } while (++page_idx < RESYNC_PAGES); |
1da177e4 LT |
3368 | r10_bio->sectors = nr_sectors; |
3369 | ||
3370 | while (biolist) { | |
3371 | bio = biolist; | |
3372 | biolist = biolist->bi_next; | |
3373 | ||
3374 | bio->bi_next = NULL; | |
f0250618 | 3375 | r10_bio = get_resync_r10bio(bio); |
1da177e4 LT |
3376 | r10_bio->sectors = nr_sectors; |
3377 | ||
3378 | if (bio->bi_end_io == end_sync_read) { | |
3379 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
4e4cbee9 | 3380 | bio->bi_status = 0; |
1da177e4 LT |
3381 | generic_make_request(bio); |
3382 | } | |
3383 | } | |
3384 | ||
57afd89f N |
3385 | if (sectors_skipped) |
3386 | /* pretend they weren't skipped, it makes | |
3387 | * no important difference in this case | |
3388 | */ | |
3389 | md_done_sync(mddev, sectors_skipped, 1); | |
3390 | ||
1da177e4 LT |
3391 | return sectors_skipped + nr_sectors; |
3392 | giveup: | |
3393 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3394 | * drives must be failed or in resync, all drives |
3395 | * have a bad block, so try the next chunk... | |
1da177e4 | 3396 | */ |
09b4068a N |
3397 | if (sector_nr + max_sync < max_sector) |
3398 | max_sector = sector_nr + max_sync; | |
3399 | ||
3400 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3401 | chunks_skipped ++; |
3402 | sector_nr = max_sector; | |
1da177e4 | 3403 | goto skipped; |
1da177e4 LT |
3404 | } |
3405 | ||
80c3a6ce | 3406 | static sector_t |
fd01b88c | 3407 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3408 | { |
3409 | sector_t size; | |
e879a879 | 3410 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3411 | |
3412 | if (!raid_disks) | |
3ea7daa5 N |
3413 | raid_disks = min(conf->geo.raid_disks, |
3414 | conf->prev.raid_disks); | |
80c3a6ce | 3415 | if (!sectors) |
dab8b292 | 3416 | sectors = conf->dev_sectors; |
80c3a6ce | 3417 | |
5cf00fcd N |
3418 | size = sectors >> conf->geo.chunk_shift; |
3419 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3420 | size = size * raid_disks; |
5cf00fcd | 3421 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3422 | |
5cf00fcd | 3423 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3424 | } |
3425 | ||
6508fdbf N |
3426 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3427 | { | |
3428 | /* Calculate the number of sectors-per-device that will | |
3429 | * actually be used, and set conf->dev_sectors and | |
3430 | * conf->stride | |
3431 | */ | |
3432 | ||
5cf00fcd N |
3433 | size = size >> conf->geo.chunk_shift; |
3434 | sector_div(size, conf->geo.far_copies); | |
3435 | size = size * conf->geo.raid_disks; | |
3436 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3437 | /* 'size' is now the number of chunks in the array */ |
3438 | /* calculate "used chunks per device" */ | |
3439 | size = size * conf->copies; | |
3440 | ||
3441 | /* We need to round up when dividing by raid_disks to | |
3442 | * get the stride size. | |
3443 | */ | |
5cf00fcd | 3444 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3445 | |
5cf00fcd | 3446 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3447 | |
5cf00fcd N |
3448 | if (conf->geo.far_offset) |
3449 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3450 | else { |
5cf00fcd N |
3451 | sector_div(size, conf->geo.far_copies); |
3452 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3453 | } |
3454 | } | |
dab8b292 | 3455 | |
deb200d0 N |
3456 | enum geo_type {geo_new, geo_old, geo_start}; |
3457 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3458 | { | |
3459 | int nc, fc, fo; | |
3460 | int layout, chunk, disks; | |
3461 | switch (new) { | |
3462 | case geo_old: | |
3463 | layout = mddev->layout; | |
3464 | chunk = mddev->chunk_sectors; | |
3465 | disks = mddev->raid_disks - mddev->delta_disks; | |
3466 | break; | |
3467 | case geo_new: | |
3468 | layout = mddev->new_layout; | |
3469 | chunk = mddev->new_chunk_sectors; | |
3470 | disks = mddev->raid_disks; | |
3471 | break; | |
3472 | default: /* avoid 'may be unused' warnings */ | |
3473 | case geo_start: /* new when starting reshape - raid_disks not | |
3474 | * updated yet. */ | |
3475 | layout = mddev->new_layout; | |
3476 | chunk = mddev->new_chunk_sectors; | |
3477 | disks = mddev->raid_disks + mddev->delta_disks; | |
3478 | break; | |
3479 | } | |
8bce6d35 | 3480 | if (layout >> 19) |
deb200d0 N |
3481 | return -1; |
3482 | if (chunk < (PAGE_SIZE >> 9) || | |
3483 | !is_power_of_2(chunk)) | |
3484 | return -2; | |
3485 | nc = layout & 255; | |
3486 | fc = (layout >> 8) & 255; | |
3487 | fo = layout & (1<<16); | |
3488 | geo->raid_disks = disks; | |
3489 | geo->near_copies = nc; | |
3490 | geo->far_copies = fc; | |
3491 | geo->far_offset = fo; | |
8bce6d35 N |
3492 | switch (layout >> 17) { |
3493 | case 0: /* original layout. simple but not always optimal */ | |
3494 | geo->far_set_size = disks; | |
3495 | break; | |
3496 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3497 | * actually using this, but leave code here just in case.*/ | |
3498 | geo->far_set_size = disks/fc; | |
3499 | WARN(geo->far_set_size < fc, | |
3500 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3501 | break; | |
3502 | case 2: /* "improved" layout fixed to match documentation */ | |
3503 | geo->far_set_size = fc * nc; | |
3504 | break; | |
3505 | default: /* Not a valid layout */ | |
3506 | return -1; | |
3507 | } | |
deb200d0 N |
3508 | geo->chunk_mask = chunk - 1; |
3509 | geo->chunk_shift = ffz(~chunk); | |
3510 | return nc*fc; | |
3511 | } | |
3512 | ||
e879a879 | 3513 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3514 | { |
e879a879 | 3515 | struct r10conf *conf = NULL; |
dab8b292 | 3516 | int err = -EINVAL; |
deb200d0 N |
3517 | struct geom geo; |
3518 | int copies; | |
3519 | ||
3520 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3521 | |
deb200d0 | 3522 | if (copies == -2) { |
08464e09 N |
3523 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n", |
3524 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3525 | goto out; |
1da177e4 | 3526 | } |
2604b703 | 3527 | |
deb200d0 | 3528 | if (copies < 2 || copies > mddev->raid_disks) { |
08464e09 N |
3529 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
3530 | mdname(mddev), mddev->new_layout); | |
1da177e4 LT |
3531 | goto out; |
3532 | } | |
dab8b292 TM |
3533 | |
3534 | err = -ENOMEM; | |
e879a879 | 3535 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3536 | if (!conf) |
1da177e4 | 3537 | goto out; |
dab8b292 | 3538 | |
3ea7daa5 | 3539 | /* FIXME calc properly */ |
dc280d98 | 3540 | conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks + |
78eaa0d4 | 3541 | max(0,-mddev->delta_disks)), |
dab8b292 TM |
3542 | GFP_KERNEL); |
3543 | if (!conf->mirrors) | |
3544 | goto out; | |
4443ae10 N |
3545 | |
3546 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3547 | if (!conf->tmppage) | |
dab8b292 TM |
3548 | goto out; |
3549 | ||
deb200d0 N |
3550 | conf->geo = geo; |
3551 | conf->copies = copies; | |
dab8b292 TM |
3552 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, |
3553 | r10bio_pool_free, conf); | |
3554 | if (!conf->r10bio_pool) | |
3555 | goto out; | |
3556 | ||
011067b0 | 3557 | conf->bio_split = bioset_create(BIO_POOL_SIZE, 0, 0); |
fc9977dd N |
3558 | if (!conf->bio_split) |
3559 | goto out; | |
3560 | ||
6508fdbf | 3561 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3562 | if (mddev->reshape_position == MaxSector) { |
3563 | conf->prev = conf->geo; | |
3564 | conf->reshape_progress = MaxSector; | |
3565 | } else { | |
3566 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3567 | err = -EINVAL; | |
3568 | goto out; | |
3569 | } | |
3570 | conf->reshape_progress = mddev->reshape_position; | |
3571 | if (conf->prev.far_offset) | |
3572 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
3573 | else | |
3574 | /* far_copies must be 1 */ | |
3575 | conf->prev.stride = conf->dev_sectors; | |
3576 | } | |
299b0685 | 3577 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 3578 | spin_lock_init(&conf->device_lock); |
dab8b292 | 3579 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 3580 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 TM |
3581 | |
3582 | spin_lock_init(&conf->resync_lock); | |
3583 | init_waitqueue_head(&conf->wait_barrier); | |
0e5313e2 | 3584 | atomic_set(&conf->nr_pending, 0); |
dab8b292 | 3585 | |
0232605d | 3586 | conf->thread = md_register_thread(raid10d, mddev, "raid10"); |
dab8b292 TM |
3587 | if (!conf->thread) |
3588 | goto out; | |
3589 | ||
dab8b292 TM |
3590 | conf->mddev = mddev; |
3591 | return conf; | |
3592 | ||
3593 | out: | |
dab8b292 | 3594 | if (conf) { |
644df1a8 | 3595 | mempool_destroy(conf->r10bio_pool); |
dab8b292 TM |
3596 | kfree(conf->mirrors); |
3597 | safe_put_page(conf->tmppage); | |
fc9977dd N |
3598 | if (conf->bio_split) |
3599 | bioset_free(conf->bio_split); | |
dab8b292 TM |
3600 | kfree(conf); |
3601 | } | |
3602 | return ERR_PTR(err); | |
3603 | } | |
3604 | ||
849674e4 | 3605 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 3606 | { |
e879a879 | 3607 | struct r10conf *conf; |
dab8b292 | 3608 | int i, disk_idx, chunk_size; |
dc280d98 | 3609 | struct raid10_info *disk; |
3cb03002 | 3610 | struct md_rdev *rdev; |
dab8b292 | 3611 | sector_t size; |
3ea7daa5 N |
3612 | sector_t min_offset_diff = 0; |
3613 | int first = 1; | |
532a2a3f | 3614 | bool discard_supported = false; |
dab8b292 | 3615 | |
a415c0f1 N |
3616 | if (mddev_init_writes_pending(mddev) < 0) |
3617 | return -ENOMEM; | |
3618 | ||
dab8b292 TM |
3619 | if (mddev->private == NULL) { |
3620 | conf = setup_conf(mddev); | |
3621 | if (IS_ERR(conf)) | |
3622 | return PTR_ERR(conf); | |
3623 | mddev->private = conf; | |
3624 | } | |
3625 | conf = mddev->private; | |
3626 | if (!conf) | |
3627 | goto out; | |
3628 | ||
dab8b292 TM |
3629 | mddev->thread = conf->thread; |
3630 | conf->thread = NULL; | |
3631 | ||
8f6c2e4b | 3632 | chunk_size = mddev->chunk_sectors << 9; |
cc4d1efd | 3633 | if (mddev->queue) { |
532a2a3f SL |
3634 | blk_queue_max_discard_sectors(mddev->queue, |
3635 | mddev->chunk_sectors); | |
5026d7a9 | 3636 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 3637 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
cc4d1efd JB |
3638 | blk_queue_io_min(mddev->queue, chunk_size); |
3639 | if (conf->geo.raid_disks % conf->geo.near_copies) | |
3640 | blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks); | |
3641 | else | |
3642 | blk_queue_io_opt(mddev->queue, chunk_size * | |
3643 | (conf->geo.raid_disks / conf->geo.near_copies)); | |
3644 | } | |
8f6c2e4b | 3645 | |
dafb20fa | 3646 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 3647 | long long diff; |
34b343cf | 3648 | |
1da177e4 | 3649 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
3650 | if (disk_idx < 0) |
3651 | continue; | |
3652 | if (disk_idx >= conf->geo.raid_disks && | |
3653 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
3654 | continue; |
3655 | disk = conf->mirrors + disk_idx; | |
3656 | ||
56a2559b N |
3657 | if (test_bit(Replacement, &rdev->flags)) { |
3658 | if (disk->replacement) | |
3659 | goto out_free_conf; | |
3660 | disk->replacement = rdev; | |
3661 | } else { | |
3662 | if (disk->rdev) | |
3663 | goto out_free_conf; | |
3664 | disk->rdev = rdev; | |
3665 | } | |
3ea7daa5 N |
3666 | diff = (rdev->new_data_offset - rdev->data_offset); |
3667 | if (!mddev->reshape_backwards) | |
3668 | diff = -diff; | |
3669 | if (diff < 0) | |
3670 | diff = 0; | |
3671 | if (first || diff < min_offset_diff) | |
3672 | min_offset_diff = diff; | |
56a2559b | 3673 | |
cc4d1efd JB |
3674 | if (mddev->gendisk) |
3675 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
3676 | rdev->data_offset << 9); | |
1da177e4 LT |
3677 | |
3678 | disk->head_position = 0; | |
532a2a3f SL |
3679 | |
3680 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) | |
3681 | discard_supported = true; | |
6f287ca6 | 3682 | first = 0; |
1da177e4 | 3683 | } |
3ea7daa5 | 3684 | |
ed30be07 JB |
3685 | if (mddev->queue) { |
3686 | if (discard_supported) | |
3687 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3688 | mddev->queue); | |
3689 | else | |
3690 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3691 | mddev->queue); | |
3692 | } | |
6d508242 | 3693 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3694 | if (!enough(conf, -1)) { |
08464e09 | 3695 | pr_err("md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3696 | mdname(mddev)); |
1da177e4 LT |
3697 | goto out_free_conf; |
3698 | } | |
3699 | ||
3ea7daa5 N |
3700 | if (conf->reshape_progress != MaxSector) { |
3701 | /* must ensure that shape change is supported */ | |
3702 | if (conf->geo.far_copies != 1 && | |
3703 | conf->geo.far_offset == 0) | |
3704 | goto out_free_conf; | |
3705 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 3706 | conf->prev.far_offset == 0) |
3ea7daa5 N |
3707 | goto out_free_conf; |
3708 | } | |
3709 | ||
1da177e4 | 3710 | mddev->degraded = 0; |
f8c9e74f N |
3711 | for (i = 0; |
3712 | i < conf->geo.raid_disks | |
3713 | || i < conf->prev.raid_disks; | |
3714 | i++) { | |
1da177e4 LT |
3715 | |
3716 | disk = conf->mirrors + i; | |
3717 | ||
56a2559b N |
3718 | if (!disk->rdev && disk->replacement) { |
3719 | /* The replacement is all we have - use it */ | |
3720 | disk->rdev = disk->replacement; | |
3721 | disk->replacement = NULL; | |
3722 | clear_bit(Replacement, &disk->rdev->flags); | |
3723 | } | |
3724 | ||
5fd6c1dc | 3725 | if (!disk->rdev || |
2e333e89 | 3726 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3727 | disk->head_position = 0; |
3728 | mddev->degraded++; | |
0b59bb64 N |
3729 | if (disk->rdev && |
3730 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 3731 | conf->fullsync = 1; |
1da177e4 | 3732 | } |
d890fa2b | 3733 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3734 | } |
3735 | ||
8c6ac868 | 3736 | if (mddev->recovery_cp != MaxSector) |
08464e09 N |
3737 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", |
3738 | mdname(mddev)); | |
3739 | pr_info("md/raid10:%s: active with %d out of %d devices\n", | |
5cf00fcd N |
3740 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
3741 | conf->geo.raid_disks); | |
1da177e4 LT |
3742 | /* |
3743 | * Ok, everything is just fine now | |
3744 | */ | |
dab8b292 TM |
3745 | mddev->dev_sectors = conf->dev_sectors; |
3746 | size = raid10_size(mddev, 0, 0); | |
3747 | md_set_array_sectors(mddev, size); | |
3748 | mddev->resync_max_sectors = size; | |
46533ff7 | 3749 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
1da177e4 | 3750 | |
cc4d1efd | 3751 | if (mddev->queue) { |
5cf00fcd | 3752 | int stripe = conf->geo.raid_disks * |
9d8f0363 | 3753 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
cc4d1efd JB |
3754 | |
3755 | /* Calculate max read-ahead size. | |
3756 | * We need to readahead at least twice a whole stripe.... | |
3757 | * maybe... | |
3758 | */ | |
5cf00fcd | 3759 | stripe /= conf->geo.near_copies; |
dc3b17cc JK |
3760 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
3761 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
1da177e4 LT |
3762 | } |
3763 | ||
a91a2785 MP |
3764 | if (md_integrity_register(mddev)) |
3765 | goto out_free_conf; | |
3766 | ||
3ea7daa5 N |
3767 | if (conf->reshape_progress != MaxSector) { |
3768 | unsigned long before_length, after_length; | |
3769 | ||
3770 | before_length = ((1 << conf->prev.chunk_shift) * | |
3771 | conf->prev.far_copies); | |
3772 | after_length = ((1 << conf->geo.chunk_shift) * | |
3773 | conf->geo.far_copies); | |
3774 | ||
3775 | if (max(before_length, after_length) > min_offset_diff) { | |
3776 | /* This cannot work */ | |
08464e09 | 3777 | pr_warn("md/raid10: offset difference not enough to continue reshape\n"); |
3ea7daa5 N |
3778 | goto out_free_conf; |
3779 | } | |
3780 | conf->offset_diff = min_offset_diff; | |
3781 | ||
3ea7daa5 N |
3782 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
3783 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3784 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3785 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3786 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3787 | "reshape"); | |
3788 | } | |
3789 | ||
1da177e4 LT |
3790 | return 0; |
3791 | ||
3792 | out_free_conf: | |
01f96c0a | 3793 | md_unregister_thread(&mddev->thread); |
644df1a8 | 3794 | mempool_destroy(conf->r10bio_pool); |
1345b1d8 | 3795 | safe_put_page(conf->tmppage); |
990a8baf | 3796 | kfree(conf->mirrors); |
1da177e4 LT |
3797 | kfree(conf); |
3798 | mddev->private = NULL; | |
3799 | out: | |
3800 | return -EIO; | |
3801 | } | |
3802 | ||
afa0f557 | 3803 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 3804 | { |
afa0f557 | 3805 | struct r10conf *conf = priv; |
1da177e4 | 3806 | |
644df1a8 | 3807 | mempool_destroy(conf->r10bio_pool); |
0fea7ed8 | 3808 | safe_put_page(conf->tmppage); |
990a8baf | 3809 | kfree(conf->mirrors); |
c4796e21 N |
3810 | kfree(conf->mirrors_old); |
3811 | kfree(conf->mirrors_new); | |
fc9977dd N |
3812 | if (conf->bio_split) |
3813 | bioset_free(conf->bio_split); | |
1da177e4 | 3814 | kfree(conf); |
1da177e4 LT |
3815 | } |
3816 | ||
fd01b88c | 3817 | static void raid10_quiesce(struct mddev *mddev, int state) |
6cce3b23 | 3818 | { |
e879a879 | 3819 | struct r10conf *conf = mddev->private; |
6cce3b23 N |
3820 | |
3821 | switch(state) { | |
3822 | case 1: | |
3823 | raise_barrier(conf, 0); | |
3824 | break; | |
3825 | case 0: | |
3826 | lower_barrier(conf); | |
3827 | break; | |
3828 | } | |
6cce3b23 | 3829 | } |
1da177e4 | 3830 | |
006a09a0 N |
3831 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3832 | { | |
3833 | /* Resize of 'far' arrays is not supported. | |
3834 | * For 'near' and 'offset' arrays we can set the | |
3835 | * number of sectors used to be an appropriate multiple | |
3836 | * of the chunk size. | |
3837 | * For 'offset', this is far_copies*chunksize. | |
3838 | * For 'near' the multiplier is the LCM of | |
3839 | * near_copies and raid_disks. | |
3840 | * So if far_copies > 1 && !far_offset, fail. | |
3841 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
3842 | * multiply by chunk_size. Then round to this number. | |
3843 | * This is mostly done by raid10_size() | |
3844 | */ | |
3845 | struct r10conf *conf = mddev->private; | |
3846 | sector_t oldsize, size; | |
3847 | ||
f8c9e74f N |
3848 | if (mddev->reshape_position != MaxSector) |
3849 | return -EBUSY; | |
3850 | ||
5cf00fcd | 3851 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
3852 | return -EINVAL; |
3853 | ||
3854 | oldsize = raid10_size(mddev, 0, 0); | |
3855 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
3856 | if (mddev->external_size && |
3857 | mddev->array_sectors > size) | |
006a09a0 | 3858 | return -EINVAL; |
a4a6125a N |
3859 | if (mddev->bitmap) { |
3860 | int ret = bitmap_resize(mddev->bitmap, size, 0, 0); | |
3861 | if (ret) | |
3862 | return ret; | |
3863 | } | |
3864 | md_set_array_sectors(mddev, size); | |
006a09a0 N |
3865 | if (sectors > mddev->dev_sectors && |
3866 | mddev->recovery_cp > oldsize) { | |
3867 | mddev->recovery_cp = oldsize; | |
3868 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3869 | } | |
6508fdbf N |
3870 | calc_sectors(conf, sectors); |
3871 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
3872 | mddev->resync_max_sectors = size; |
3873 | return 0; | |
3874 | } | |
3875 | ||
53a6ab4d | 3876 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 3877 | { |
3cb03002 | 3878 | struct md_rdev *rdev; |
e879a879 | 3879 | struct r10conf *conf; |
dab8b292 TM |
3880 | |
3881 | if (mddev->degraded > 0) { | |
08464e09 N |
3882 | pr_warn("md/raid10:%s: Error: degraded raid0!\n", |
3883 | mdname(mddev)); | |
dab8b292 TM |
3884 | return ERR_PTR(-EINVAL); |
3885 | } | |
53a6ab4d | 3886 | sector_div(size, devs); |
dab8b292 | 3887 | |
dab8b292 TM |
3888 | /* Set new parameters */ |
3889 | mddev->new_level = 10; | |
3890 | /* new layout: far_copies = 1, near_copies = 2 */ | |
3891 | mddev->new_layout = (1<<8) + 2; | |
3892 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
3893 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
3894 | mddev->raid_disks *= 2; |
3895 | /* make sure it will be not marked as dirty */ | |
3896 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 3897 | mddev->dev_sectors = size; |
dab8b292 TM |
3898 | |
3899 | conf = setup_conf(mddev); | |
02214dc5 | 3900 | if (!IS_ERR(conf)) { |
dafb20fa | 3901 | rdev_for_each(rdev, mddev) |
53a6ab4d | 3902 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 3903 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
3904 | rdev->sectors = size; |
3905 | } | |
02214dc5 KW |
3906 | conf->barrier = 1; |
3907 | } | |
3908 | ||
dab8b292 TM |
3909 | return conf; |
3910 | } | |
3911 | ||
fd01b88c | 3912 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 3913 | { |
e373ab10 | 3914 | struct r0conf *raid0_conf; |
dab8b292 TM |
3915 | |
3916 | /* raid10 can take over: | |
3917 | * raid0 - providing it has only two drives | |
3918 | */ | |
3919 | if (mddev->level == 0) { | |
3920 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
3921 | raid0_conf = mddev->private; |
3922 | if (raid0_conf->nr_strip_zones > 1) { | |
08464e09 N |
3923 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n", |
3924 | mdname(mddev)); | |
dab8b292 TM |
3925 | return ERR_PTR(-EINVAL); |
3926 | } | |
53a6ab4d N |
3927 | return raid10_takeover_raid0(mddev, |
3928 | raid0_conf->strip_zone->zone_end, | |
3929 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
3930 | } |
3931 | return ERR_PTR(-EINVAL); | |
3932 | } | |
3933 | ||
3ea7daa5 N |
3934 | static int raid10_check_reshape(struct mddev *mddev) |
3935 | { | |
3936 | /* Called when there is a request to change | |
3937 | * - layout (to ->new_layout) | |
3938 | * - chunk size (to ->new_chunk_sectors) | |
3939 | * - raid_disks (by delta_disks) | |
3940 | * or when trying to restart a reshape that was ongoing. | |
3941 | * | |
3942 | * We need to validate the request and possibly allocate | |
3943 | * space if that might be an issue later. | |
3944 | * | |
3945 | * Currently we reject any reshape of a 'far' mode array, | |
3946 | * allow chunk size to change if new is generally acceptable, | |
3947 | * allow raid_disks to increase, and allow | |
3948 | * a switch between 'near' mode and 'offset' mode. | |
3949 | */ | |
3950 | struct r10conf *conf = mddev->private; | |
3951 | struct geom geo; | |
3952 | ||
3953 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
3954 | return -EINVAL; | |
3955 | ||
3956 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
3957 | /* mustn't change number of copies */ | |
3958 | return -EINVAL; | |
3959 | if (geo.far_copies > 1 && !geo.far_offset) | |
3960 | /* Cannot switch to 'far' mode */ | |
3961 | return -EINVAL; | |
3962 | ||
3963 | if (mddev->array_sectors & geo.chunk_mask) | |
3964 | /* not factor of array size */ | |
3965 | return -EINVAL; | |
3966 | ||
3ea7daa5 N |
3967 | if (!enough(conf, -1)) |
3968 | return -EINVAL; | |
3969 | ||
3970 | kfree(conf->mirrors_new); | |
3971 | conf->mirrors_new = NULL; | |
3972 | if (mddev->delta_disks > 0) { | |
3973 | /* allocate new 'mirrors' list */ | |
3974 | conf->mirrors_new = kzalloc( | |
dc280d98 | 3975 | sizeof(struct raid10_info) |
3ea7daa5 N |
3976 | *(mddev->raid_disks + |
3977 | mddev->delta_disks), | |
3978 | GFP_KERNEL); | |
3979 | if (!conf->mirrors_new) | |
3980 | return -ENOMEM; | |
3981 | } | |
3982 | return 0; | |
3983 | } | |
3984 | ||
3985 | /* | |
3986 | * Need to check if array has failed when deciding whether to: | |
3987 | * - start an array | |
3988 | * - remove non-faulty devices | |
3989 | * - add a spare | |
3990 | * - allow a reshape | |
3991 | * This determination is simple when no reshape is happening. | |
3992 | * However if there is a reshape, we need to carefully check | |
3993 | * both the before and after sections. | |
3994 | * This is because some failed devices may only affect one | |
3995 | * of the two sections, and some non-in_sync devices may | |
3996 | * be insync in the section most affected by failed devices. | |
3997 | */ | |
3998 | static int calc_degraded(struct r10conf *conf) | |
3999 | { | |
4000 | int degraded, degraded2; | |
4001 | int i; | |
4002 | ||
4003 | rcu_read_lock(); | |
4004 | degraded = 0; | |
4005 | /* 'prev' section first */ | |
4006 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
4007 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4008 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4009 | degraded++; | |
4010 | else if (!test_bit(In_sync, &rdev->flags)) | |
4011 | /* When we can reduce the number of devices in | |
4012 | * an array, this might not contribute to | |
4013 | * 'degraded'. It does now. | |
4014 | */ | |
4015 | degraded++; | |
4016 | } | |
4017 | rcu_read_unlock(); | |
4018 | if (conf->geo.raid_disks == conf->prev.raid_disks) | |
4019 | return degraded; | |
4020 | rcu_read_lock(); | |
4021 | degraded2 = 0; | |
4022 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
4023 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4024 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4025 | degraded2++; | |
4026 | else if (!test_bit(In_sync, &rdev->flags)) { | |
4027 | /* If reshape is increasing the number of devices, | |
4028 | * this section has already been recovered, so | |
4029 | * it doesn't contribute to degraded. | |
4030 | * else it does. | |
4031 | */ | |
4032 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
4033 | degraded2++; | |
4034 | } | |
4035 | } | |
4036 | rcu_read_unlock(); | |
4037 | if (degraded2 > degraded) | |
4038 | return degraded2; | |
4039 | return degraded; | |
4040 | } | |
4041 | ||
4042 | static int raid10_start_reshape(struct mddev *mddev) | |
4043 | { | |
4044 | /* A 'reshape' has been requested. This commits | |
4045 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
4046 | * This also checks if there are enough spares and adds them | |
4047 | * to the array. | |
4048 | * We currently require enough spares to make the final | |
4049 | * array non-degraded. We also require that the difference | |
4050 | * between old and new data_offset - on each device - is | |
4051 | * enough that we never risk over-writing. | |
4052 | */ | |
4053 | ||
4054 | unsigned long before_length, after_length; | |
4055 | sector_t min_offset_diff = 0; | |
4056 | int first = 1; | |
4057 | struct geom new; | |
4058 | struct r10conf *conf = mddev->private; | |
4059 | struct md_rdev *rdev; | |
4060 | int spares = 0; | |
bb63a701 | 4061 | int ret; |
3ea7daa5 N |
4062 | |
4063 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4064 | return -EBUSY; | |
4065 | ||
4066 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
4067 | return -EINVAL; | |
4068 | ||
4069 | before_length = ((1 << conf->prev.chunk_shift) * | |
4070 | conf->prev.far_copies); | |
4071 | after_length = ((1 << conf->geo.chunk_shift) * | |
4072 | conf->geo.far_copies); | |
4073 | ||
4074 | rdev_for_each(rdev, mddev) { | |
4075 | if (!test_bit(In_sync, &rdev->flags) | |
4076 | && !test_bit(Faulty, &rdev->flags)) | |
4077 | spares++; | |
4078 | if (rdev->raid_disk >= 0) { | |
4079 | long long diff = (rdev->new_data_offset | |
4080 | - rdev->data_offset); | |
4081 | if (!mddev->reshape_backwards) | |
4082 | diff = -diff; | |
4083 | if (diff < 0) | |
4084 | diff = 0; | |
4085 | if (first || diff < min_offset_diff) | |
4086 | min_offset_diff = diff; | |
b506335e | 4087 | first = 0; |
3ea7daa5 N |
4088 | } |
4089 | } | |
4090 | ||
4091 | if (max(before_length, after_length) > min_offset_diff) | |
4092 | return -EINVAL; | |
4093 | ||
4094 | if (spares < mddev->delta_disks) | |
4095 | return -EINVAL; | |
4096 | ||
4097 | conf->offset_diff = min_offset_diff; | |
4098 | spin_lock_irq(&conf->device_lock); | |
4099 | if (conf->mirrors_new) { | |
4100 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4101 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4102 | smp_mb(); |
c4796e21 | 4103 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4104 | conf->mirrors_old = conf->mirrors; |
4105 | conf->mirrors = conf->mirrors_new; | |
4106 | conf->mirrors_new = NULL; | |
4107 | } | |
4108 | setup_geo(&conf->geo, mddev, geo_start); | |
4109 | smp_mb(); | |
4110 | if (mddev->reshape_backwards) { | |
4111 | sector_t size = raid10_size(mddev, 0, 0); | |
4112 | if (size < mddev->array_sectors) { | |
4113 | spin_unlock_irq(&conf->device_lock); | |
08464e09 N |
4114 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n", |
4115 | mdname(mddev)); | |
3ea7daa5 N |
4116 | return -EINVAL; |
4117 | } | |
4118 | mddev->resync_max_sectors = size; | |
4119 | conf->reshape_progress = size; | |
4120 | } else | |
4121 | conf->reshape_progress = 0; | |
299b0685 | 4122 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4123 | spin_unlock_irq(&conf->device_lock); |
4124 | ||
bb63a701 N |
4125 | if (mddev->delta_disks && mddev->bitmap) { |
4126 | ret = bitmap_resize(mddev->bitmap, | |
4127 | raid10_size(mddev, 0, | |
4128 | conf->geo.raid_disks), | |
4129 | 0, 0); | |
4130 | if (ret) | |
4131 | goto abort; | |
4132 | } | |
3ea7daa5 N |
4133 | if (mddev->delta_disks > 0) { |
4134 | rdev_for_each(rdev, mddev) | |
4135 | if (rdev->raid_disk < 0 && | |
4136 | !test_bit(Faulty, &rdev->flags)) { | |
4137 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4138 | if (rdev->raid_disk >= | |
4139 | conf->prev.raid_disks) | |
4140 | set_bit(In_sync, &rdev->flags); | |
4141 | else | |
4142 | rdev->recovery_offset = 0; | |
4143 | ||
4144 | if (sysfs_link_rdev(mddev, rdev)) | |
4145 | /* Failure here is OK */; | |
4146 | } | |
4147 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4148 | && !test_bit(Faulty, &rdev->flags)) { | |
4149 | /* This is a spare that was manually added */ | |
4150 | set_bit(In_sync, &rdev->flags); | |
4151 | } | |
4152 | } | |
4153 | /* When a reshape changes the number of devices, | |
4154 | * ->degraded is measured against the larger of the | |
4155 | * pre and post numbers. | |
4156 | */ | |
4157 | spin_lock_irq(&conf->device_lock); | |
4158 | mddev->degraded = calc_degraded(conf); | |
4159 | spin_unlock_irq(&conf->device_lock); | |
4160 | mddev->raid_disks = conf->geo.raid_disks; | |
4161 | mddev->reshape_position = conf->reshape_progress; | |
2953079c | 4162 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 N |
4163 | |
4164 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4165 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4166 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 N |
4167 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
4168 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4169 | ||
4170 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4171 | "reshape"); | |
4172 | if (!mddev->sync_thread) { | |
bb63a701 N |
4173 | ret = -EAGAIN; |
4174 | goto abort; | |
3ea7daa5 N |
4175 | } |
4176 | conf->reshape_checkpoint = jiffies; | |
4177 | md_wakeup_thread(mddev->sync_thread); | |
4178 | md_new_event(mddev); | |
4179 | return 0; | |
bb63a701 N |
4180 | |
4181 | abort: | |
4182 | mddev->recovery = 0; | |
4183 | spin_lock_irq(&conf->device_lock); | |
4184 | conf->geo = conf->prev; | |
4185 | mddev->raid_disks = conf->geo.raid_disks; | |
4186 | rdev_for_each(rdev, mddev) | |
4187 | rdev->new_data_offset = rdev->data_offset; | |
4188 | smp_wmb(); | |
4189 | conf->reshape_progress = MaxSector; | |
299b0685 | 4190 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4191 | mddev->reshape_position = MaxSector; |
4192 | spin_unlock_irq(&conf->device_lock); | |
4193 | return ret; | |
3ea7daa5 N |
4194 | } |
4195 | ||
4196 | /* Calculate the last device-address that could contain | |
4197 | * any block from the chunk that includes the array-address 's' | |
4198 | * and report the next address. | |
4199 | * i.e. the address returned will be chunk-aligned and after | |
4200 | * any data that is in the chunk containing 's'. | |
4201 | */ | |
4202 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4203 | { | |
4204 | s = (s | geo->chunk_mask) + 1; | |
4205 | s >>= geo->chunk_shift; | |
4206 | s *= geo->near_copies; | |
4207 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4208 | s *= geo->far_copies; | |
4209 | s <<= geo->chunk_shift; | |
4210 | return s; | |
4211 | } | |
4212 | ||
4213 | /* Calculate the first device-address that could contain | |
4214 | * any block from the chunk that includes the array-address 's'. | |
4215 | * This too will be the start of a chunk | |
4216 | */ | |
4217 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4218 | { | |
4219 | s >>= geo->chunk_shift; | |
4220 | s *= geo->near_copies; | |
4221 | sector_div(s, geo->raid_disks); | |
4222 | s *= geo->far_copies; | |
4223 | s <<= geo->chunk_shift; | |
4224 | return s; | |
4225 | } | |
4226 | ||
4227 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4228 | int *skipped) | |
4229 | { | |
4230 | /* We simply copy at most one chunk (smallest of old and new) | |
4231 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4232 | * or we hit a bad block or something. | |
4233 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4234 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4235 | * can record any location. |
4236 | * | |
4237 | * If we will want to write to a location that isn't | |
4238 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4239 | * we need to flush all reshape requests and update the metadata. | |
4240 | * | |
4241 | * When reshaping forwards (e.g. to more devices), we interpret | |
4242 | * 'safe' as the earliest block which might not have been copied | |
4243 | * down yet. We divide this by previous stripe size and multiply | |
4244 | * by previous stripe length to get lowest device offset that we | |
4245 | * cannot write to yet. | |
4246 | * We interpret 'sector_nr' as an address that we want to write to. | |
4247 | * From this we use last_device_address() to find where we might | |
4248 | * write to, and first_device_address on the 'safe' position. | |
4249 | * If this 'next' write position is after the 'safe' position, | |
4250 | * we must update the metadata to increase the 'safe' position. | |
4251 | * | |
4252 | * When reshaping backwards, we round in the opposite direction | |
4253 | * and perform the reverse test: next write position must not be | |
4254 | * less than current safe position. | |
4255 | * | |
4256 | * In all this the minimum difference in data offsets | |
4257 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4258 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4259 | * |
4260 | * We need to prepare all the bios here before we start any IO | |
4261 | * to ensure the size we choose is acceptable to all devices. | |
4262 | * The means one for each copy for write-out and an extra one for | |
4263 | * read-in. | |
4264 | * We store the read-in bio in ->master_bio and the others in | |
4265 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4266 | */ | |
4267 | struct r10conf *conf = mddev->private; | |
4268 | struct r10bio *r10_bio; | |
4269 | sector_t next, safe, last; | |
4270 | int max_sectors; | |
4271 | int nr_sectors; | |
4272 | int s; | |
4273 | struct md_rdev *rdev; | |
4274 | int need_flush = 0; | |
4275 | struct bio *blist; | |
4276 | struct bio *bio, *read_bio; | |
4277 | int sectors_done = 0; | |
f0250618 | 4278 | struct page **pages; |
3ea7daa5 N |
4279 | |
4280 | if (sector_nr == 0) { | |
4281 | /* If restarting in the middle, skip the initial sectors */ | |
4282 | if (mddev->reshape_backwards && | |
4283 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4284 | sector_nr = (raid10_size(mddev, 0, 0) | |
4285 | - conf->reshape_progress); | |
4286 | } else if (!mddev->reshape_backwards && | |
4287 | conf->reshape_progress > 0) | |
4288 | sector_nr = conf->reshape_progress; | |
4289 | if (sector_nr) { | |
4290 | mddev->curr_resync_completed = sector_nr; | |
4291 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
4292 | *skipped = 1; | |
4293 | return sector_nr; | |
4294 | } | |
4295 | } | |
4296 | ||
4297 | /* We don't use sector_nr to track where we are up to | |
4298 | * as that doesn't work well for ->reshape_backwards. | |
4299 | * So just use ->reshape_progress. | |
4300 | */ | |
4301 | if (mddev->reshape_backwards) { | |
4302 | /* 'next' is the earliest device address that we might | |
4303 | * write to for this chunk in the new layout | |
4304 | */ | |
4305 | next = first_dev_address(conf->reshape_progress - 1, | |
4306 | &conf->geo); | |
4307 | ||
4308 | /* 'safe' is the last device address that we might read from | |
4309 | * in the old layout after a restart | |
4310 | */ | |
4311 | safe = last_dev_address(conf->reshape_safe - 1, | |
4312 | &conf->prev); | |
4313 | ||
4314 | if (next + conf->offset_diff < safe) | |
4315 | need_flush = 1; | |
4316 | ||
4317 | last = conf->reshape_progress - 1; | |
4318 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4319 | & conf->prev.chunk_mask); | |
4320 | if (sector_nr + RESYNC_BLOCK_SIZE/512 < last) | |
4321 | sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512; | |
4322 | } else { | |
4323 | /* 'next' is after the last device address that we | |
4324 | * might write to for this chunk in the new layout | |
4325 | */ | |
4326 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4327 | ||
4328 | /* 'safe' is the earliest device address that we might | |
4329 | * read from in the old layout after a restart | |
4330 | */ | |
4331 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4332 | ||
4333 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4334 | * as that would possibly corrupt data | |
4335 | */ | |
4336 | if (next > safe + conf->offset_diff) | |
4337 | need_flush = 1; | |
4338 | ||
4339 | sector_nr = conf->reshape_progress; | |
4340 | last = sector_nr | (conf->geo.chunk_mask | |
4341 | & conf->prev.chunk_mask); | |
4342 | ||
4343 | if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last) | |
4344 | last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1; | |
4345 | } | |
4346 | ||
4347 | if (need_flush || | |
4348 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4349 | /* Need to update reshape_position in metadata */ | |
4350 | wait_barrier(conf); | |
4351 | mddev->reshape_position = conf->reshape_progress; | |
4352 | if (mddev->reshape_backwards) | |
4353 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4354 | - conf->reshape_progress; | |
4355 | else | |
4356 | mddev->curr_resync_completed = conf->reshape_progress; | |
4357 | conf->reshape_checkpoint = jiffies; | |
2953079c | 4358 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 | 4359 | md_wakeup_thread(mddev->thread); |
2953079c | 4360 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
4361 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4362 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4363 | allow_barrier(conf); | |
4364 | return sectors_done; | |
4365 | } | |
3ea7daa5 N |
4366 | conf->reshape_safe = mddev->reshape_position; |
4367 | allow_barrier(conf); | |
4368 | } | |
4369 | ||
4370 | read_more: | |
4371 | /* Now schedule reads for blocks from sector_nr to last */ | |
4372 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); | |
cb8b12b5 | 4373 | r10_bio->state = 0; |
3ea7daa5 N |
4374 | raise_barrier(conf, sectors_done != 0); |
4375 | atomic_set(&r10_bio->remaining, 0); | |
4376 | r10_bio->mddev = mddev; | |
4377 | r10_bio->sector = sector_nr; | |
4378 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4379 | r10_bio->sectors = last - sector_nr + 1; | |
4380 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4381 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4382 | ||
4383 | if (!rdev) { | |
4384 | /* Cannot read from here, so need to record bad blocks | |
4385 | * on all the target devices. | |
4386 | */ | |
4387 | // FIXME | |
e337aead | 4388 | mempool_free(r10_bio, conf->r10buf_pool); |
3ea7daa5 N |
4389 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4390 | return sectors_done; | |
4391 | } | |
4392 | ||
4393 | read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev); | |
4394 | ||
4395 | read_bio->bi_bdev = rdev->bdev; | |
4f024f37 | 4396 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4397 | + rdev->data_offset); |
4398 | read_bio->bi_private = r10_bio; | |
81fa1520 | 4399 | read_bio->bi_end_io = end_reshape_read; |
796a5cf0 | 4400 | bio_set_op_attrs(read_bio, REQ_OP_READ, 0); |
ce0b0a46 | 4401 | read_bio->bi_flags &= (~0UL << BIO_RESET_BITS); |
4e4cbee9 | 4402 | read_bio->bi_status = 0; |
3ea7daa5 | 4403 | read_bio->bi_vcnt = 0; |
4f024f37 | 4404 | read_bio->bi_iter.bi_size = 0; |
3ea7daa5 N |
4405 | r10_bio->master_bio = read_bio; |
4406 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4407 | ||
4408 | /* Now find the locations in the new layout */ | |
4409 | __raid10_find_phys(&conf->geo, r10_bio); | |
4410 | ||
4411 | blist = read_bio; | |
4412 | read_bio->bi_next = NULL; | |
4413 | ||
d094d686 | 4414 | rcu_read_lock(); |
3ea7daa5 N |
4415 | for (s = 0; s < conf->copies*2; s++) { |
4416 | struct bio *b; | |
4417 | int d = r10_bio->devs[s/2].devnum; | |
4418 | struct md_rdev *rdev2; | |
4419 | if (s&1) { | |
d094d686 | 4420 | rdev2 = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4421 | b = r10_bio->devs[s/2].repl_bio; |
4422 | } else { | |
d094d686 | 4423 | rdev2 = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4424 | b = r10_bio->devs[s/2].bio; |
4425 | } | |
4426 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4427 | continue; | |
8be185f2 | 4428 | |
3ea7daa5 | 4429 | b->bi_bdev = rdev2->bdev; |
4f024f37 KO |
4430 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4431 | rdev2->new_data_offset; | |
3ea7daa5 | 4432 | b->bi_end_io = end_reshape_write; |
796a5cf0 | 4433 | bio_set_op_attrs(b, REQ_OP_WRITE, 0); |
3ea7daa5 | 4434 | b->bi_next = blist; |
3ea7daa5 N |
4435 | blist = b; |
4436 | } | |
4437 | ||
4438 | /* Now add as many pages as possible to all of these bios. */ | |
4439 | ||
4440 | nr_sectors = 0; | |
f0250618 | 4441 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; |
3ea7daa5 | 4442 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { |
f0250618 | 4443 | struct page *page = pages[s / (PAGE_SIZE >> 9)]; |
3ea7daa5 N |
4444 | int len = (max_sectors - s) << 9; |
4445 | if (len > PAGE_SIZE) | |
4446 | len = PAGE_SIZE; | |
4447 | for (bio = blist; bio ; bio = bio->bi_next) { | |
c85ba149 ML |
4448 | /* |
4449 | * won't fail because the vec table is big enough | |
4450 | * to hold all these pages | |
4451 | */ | |
4452 | bio_add_page(bio, page, len, 0); | |
3ea7daa5 N |
4453 | } |
4454 | sector_nr += len >> 9; | |
4455 | nr_sectors += len >> 9; | |
4456 | } | |
d094d686 | 4457 | rcu_read_unlock(); |
3ea7daa5 N |
4458 | r10_bio->sectors = nr_sectors; |
4459 | ||
4460 | /* Now submit the read */ | |
4461 | md_sync_acct(read_bio->bi_bdev, r10_bio->sectors); | |
4462 | atomic_inc(&r10_bio->remaining); | |
4463 | read_bio->bi_next = NULL; | |
4464 | generic_make_request(read_bio); | |
4465 | sector_nr += nr_sectors; | |
4466 | sectors_done += nr_sectors; | |
4467 | if (sector_nr <= last) | |
4468 | goto read_more; | |
4469 | ||
4470 | /* Now that we have done the whole section we can | |
4471 | * update reshape_progress | |
4472 | */ | |
4473 | if (mddev->reshape_backwards) | |
4474 | conf->reshape_progress -= sectors_done; | |
4475 | else | |
4476 | conf->reshape_progress += sectors_done; | |
4477 | ||
4478 | return sectors_done; | |
4479 | } | |
4480 | ||
4481 | static void end_reshape_request(struct r10bio *r10_bio); | |
4482 | static int handle_reshape_read_error(struct mddev *mddev, | |
4483 | struct r10bio *r10_bio); | |
4484 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4485 | { | |
4486 | /* Reshape read completed. Hopefully we have a block | |
4487 | * to write out. | |
4488 | * If we got a read error then we do sync 1-page reads from | |
4489 | * elsewhere until we find the data - or give up. | |
4490 | */ | |
4491 | struct r10conf *conf = mddev->private; | |
4492 | int s; | |
4493 | ||
4494 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4495 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4496 | /* Reshape has been aborted */ | |
4497 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4498 | return; | |
4499 | } | |
4500 | ||
4501 | /* We definitely have the data in the pages, schedule the | |
4502 | * writes. | |
4503 | */ | |
4504 | atomic_set(&r10_bio->remaining, 1); | |
4505 | for (s = 0; s < conf->copies*2; s++) { | |
4506 | struct bio *b; | |
4507 | int d = r10_bio->devs[s/2].devnum; | |
4508 | struct md_rdev *rdev; | |
d094d686 | 4509 | rcu_read_lock(); |
3ea7daa5 | 4510 | if (s&1) { |
d094d686 | 4511 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4512 | b = r10_bio->devs[s/2].repl_bio; |
4513 | } else { | |
d094d686 | 4514 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4515 | b = r10_bio->devs[s/2].bio; |
4516 | } | |
d094d686 N |
4517 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
4518 | rcu_read_unlock(); | |
3ea7daa5 | 4519 | continue; |
d094d686 | 4520 | } |
3ea7daa5 | 4521 | atomic_inc(&rdev->nr_pending); |
d094d686 | 4522 | rcu_read_unlock(); |
3ea7daa5 N |
4523 | md_sync_acct(b->bi_bdev, r10_bio->sectors); |
4524 | atomic_inc(&r10_bio->remaining); | |
4525 | b->bi_next = NULL; | |
4526 | generic_make_request(b); | |
4527 | } | |
4528 | end_reshape_request(r10_bio); | |
4529 | } | |
4530 | ||
4531 | static void end_reshape(struct r10conf *conf) | |
4532 | { | |
4533 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
4534 | return; | |
4535 | ||
4536 | spin_lock_irq(&conf->device_lock); | |
4537 | conf->prev = conf->geo; | |
4538 | md_finish_reshape(conf->mddev); | |
4539 | smp_wmb(); | |
4540 | conf->reshape_progress = MaxSector; | |
299b0685 | 4541 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
4542 | spin_unlock_irq(&conf->device_lock); |
4543 | ||
4544 | /* read-ahead size must cover two whole stripes, which is | |
4545 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4546 | */ | |
4547 | if (conf->mddev->queue) { | |
4548 | int stripe = conf->geo.raid_disks * | |
4549 | ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE); | |
4550 | stripe /= conf->geo.near_copies; | |
dc3b17cc JK |
4551 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
4552 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
3ea7daa5 N |
4553 | } |
4554 | conf->fullsync = 0; | |
4555 | } | |
4556 | ||
3ea7daa5 N |
4557 | static int handle_reshape_read_error(struct mddev *mddev, |
4558 | struct r10bio *r10_bio) | |
4559 | { | |
4560 | /* Use sync reads to get the blocks from somewhere else */ | |
4561 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 4562 | struct r10conf *conf = mddev->private; |
e0ee7785 N |
4563 | struct { |
4564 | struct r10bio r10_bio; | |
4565 | struct r10dev devs[conf->copies]; | |
4566 | } on_stack; | |
4567 | struct r10bio *r10b = &on_stack.r10_bio; | |
3ea7daa5 N |
4568 | int slot = 0; |
4569 | int idx = 0; | |
2d06e3b7 ML |
4570 | struct page **pages; |
4571 | ||
4572 | /* reshape IOs share pages from .devs[0].bio */ | |
4573 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; | |
3ea7daa5 | 4574 | |
e0ee7785 N |
4575 | r10b->sector = r10_bio->sector; |
4576 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
4577 | |
4578 | while (sectors) { | |
4579 | int s = sectors; | |
4580 | int success = 0; | |
4581 | int first_slot = slot; | |
4582 | ||
4583 | if (s > (PAGE_SIZE >> 9)) | |
4584 | s = PAGE_SIZE >> 9; | |
4585 | ||
d094d686 | 4586 | rcu_read_lock(); |
3ea7daa5 | 4587 | while (!success) { |
e0ee7785 | 4588 | int d = r10b->devs[slot].devnum; |
d094d686 | 4589 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4590 | sector_t addr; |
4591 | if (rdev == NULL || | |
4592 | test_bit(Faulty, &rdev->flags) || | |
4593 | !test_bit(In_sync, &rdev->flags)) | |
4594 | goto failed; | |
4595 | ||
e0ee7785 | 4596 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
d094d686 N |
4597 | atomic_inc(&rdev->nr_pending); |
4598 | rcu_read_unlock(); | |
3ea7daa5 N |
4599 | success = sync_page_io(rdev, |
4600 | addr, | |
4601 | s << 9, | |
2d06e3b7 | 4602 | pages[idx], |
796a5cf0 | 4603 | REQ_OP_READ, 0, false); |
d094d686 N |
4604 | rdev_dec_pending(rdev, mddev); |
4605 | rcu_read_lock(); | |
3ea7daa5 N |
4606 | if (success) |
4607 | break; | |
4608 | failed: | |
4609 | slot++; | |
4610 | if (slot >= conf->copies) | |
4611 | slot = 0; | |
4612 | if (slot == first_slot) | |
4613 | break; | |
4614 | } | |
d094d686 | 4615 | rcu_read_unlock(); |
3ea7daa5 N |
4616 | if (!success) { |
4617 | /* couldn't read this block, must give up */ | |
4618 | set_bit(MD_RECOVERY_INTR, | |
4619 | &mddev->recovery); | |
4620 | return -EIO; | |
4621 | } | |
4622 | sectors -= s; | |
4623 | idx++; | |
4624 | } | |
4625 | return 0; | |
4626 | } | |
4627 | ||
4246a0b6 | 4628 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 4629 | { |
f0250618 | 4630 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
3ea7daa5 N |
4631 | struct mddev *mddev = r10_bio->mddev; |
4632 | struct r10conf *conf = mddev->private; | |
4633 | int d; | |
4634 | int slot; | |
4635 | int repl; | |
4636 | struct md_rdev *rdev = NULL; | |
4637 | ||
4638 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
4639 | if (repl) | |
4640 | rdev = conf->mirrors[d].replacement; | |
4641 | if (!rdev) { | |
4642 | smp_mb(); | |
4643 | rdev = conf->mirrors[d].rdev; | |
4644 | } | |
4645 | ||
4e4cbee9 | 4646 | if (bio->bi_status) { |
3ea7daa5 N |
4647 | /* FIXME should record badblock */ |
4648 | md_error(mddev, rdev); | |
4649 | } | |
4650 | ||
4651 | rdev_dec_pending(rdev, mddev); | |
4652 | end_reshape_request(r10_bio); | |
4653 | } | |
4654 | ||
4655 | static void end_reshape_request(struct r10bio *r10_bio) | |
4656 | { | |
4657 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
4658 | return; | |
4659 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
4660 | bio_put(r10_bio->master_bio); | |
4661 | put_buf(r10_bio); | |
4662 | } | |
4663 | ||
4664 | static void raid10_finish_reshape(struct mddev *mddev) | |
4665 | { | |
4666 | struct r10conf *conf = mddev->private; | |
4667 | ||
4668 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
4669 | return; | |
4670 | ||
4671 | if (mddev->delta_disks > 0) { | |
4672 | sector_t size = raid10_size(mddev, 0, 0); | |
4673 | md_set_array_sectors(mddev, size); | |
4674 | if (mddev->recovery_cp > mddev->resync_max_sectors) { | |
4675 | mddev->recovery_cp = mddev->resync_max_sectors; | |
4676 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4677 | } | |
4678 | mddev->resync_max_sectors = size; | |
859644f0 HM |
4679 | if (mddev->queue) { |
4680 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4681 | revalidate_disk(mddev->gendisk); | |
4682 | } | |
63aced61 N |
4683 | } else { |
4684 | int d; | |
d094d686 | 4685 | rcu_read_lock(); |
63aced61 N |
4686 | for (d = conf->geo.raid_disks ; |
4687 | d < conf->geo.raid_disks - mddev->delta_disks; | |
4688 | d++) { | |
d094d686 | 4689 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
63aced61 N |
4690 | if (rdev) |
4691 | clear_bit(In_sync, &rdev->flags); | |
d094d686 | 4692 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
63aced61 N |
4693 | if (rdev) |
4694 | clear_bit(In_sync, &rdev->flags); | |
4695 | } | |
d094d686 | 4696 | rcu_read_unlock(); |
3ea7daa5 N |
4697 | } |
4698 | mddev->layout = mddev->new_layout; | |
4699 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
4700 | mddev->reshape_position = MaxSector; | |
4701 | mddev->delta_disks = 0; | |
4702 | mddev->reshape_backwards = 0; | |
4703 | } | |
4704 | ||
84fc4b56 | 4705 | static struct md_personality raid10_personality = |
1da177e4 LT |
4706 | { |
4707 | .name = "raid10", | |
2604b703 | 4708 | .level = 10, |
1da177e4 | 4709 | .owner = THIS_MODULE, |
849674e4 SL |
4710 | .make_request = raid10_make_request, |
4711 | .run = raid10_run, | |
afa0f557 | 4712 | .free = raid10_free, |
849674e4 SL |
4713 | .status = raid10_status, |
4714 | .error_handler = raid10_error, | |
1da177e4 LT |
4715 | .hot_add_disk = raid10_add_disk, |
4716 | .hot_remove_disk= raid10_remove_disk, | |
4717 | .spare_active = raid10_spare_active, | |
849674e4 | 4718 | .sync_request = raid10_sync_request, |
6cce3b23 | 4719 | .quiesce = raid10_quiesce, |
80c3a6ce | 4720 | .size = raid10_size, |
006a09a0 | 4721 | .resize = raid10_resize, |
dab8b292 | 4722 | .takeover = raid10_takeover, |
3ea7daa5 N |
4723 | .check_reshape = raid10_check_reshape, |
4724 | .start_reshape = raid10_start_reshape, | |
4725 | .finish_reshape = raid10_finish_reshape, | |
5c675f83 | 4726 | .congested = raid10_congested, |
1da177e4 LT |
4727 | }; |
4728 | ||
4729 | static int __init raid_init(void) | |
4730 | { | |
2604b703 | 4731 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
4732 | } |
4733 | ||
4734 | static void raid_exit(void) | |
4735 | { | |
2604b703 | 4736 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
4737 | } |
4738 | ||
4739 | module_init(raid_init); | |
4740 | module_exit(raid_exit); | |
4741 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 4742 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 4743 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 4744 | MODULE_ALIAS("md-raid10"); |
2604b703 | 4745 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
4746 | |
4747 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |