]>
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> |
bff61975 | 24 | #include <linux/seq_file.h> |
43b2e5d8 | 25 | #include "md.h" |
ef740c37 | 26 | #include "raid10.h" |
dab8b292 | 27 | #include "raid0.h" |
ef740c37 | 28 | #include "bitmap.h" |
1da177e4 LT |
29 | |
30 | /* | |
31 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
32 | * The layout of data is defined by | |
33 | * chunk_size | |
34 | * raid_disks | |
35 | * near_copies (stored in low byte of layout) | |
36 | * far_copies (stored in second byte of layout) | |
c93983bf | 37 | * far_offset (stored in bit 16 of layout ) |
1da177e4 LT |
38 | * |
39 | * The data to be stored is divided into chunks using chunksize. | |
40 | * Each device is divided into far_copies sections. | |
41 | * In each section, chunks are laid out in a style similar to raid0, but | |
42 | * near_copies copies of each chunk is stored (each on a different drive). | |
43 | * The starting device for each section is offset near_copies from the starting | |
44 | * device of the previous section. | |
c93983bf | 45 | * Thus they are (near_copies*far_copies) of each chunk, and each is on a different |
1da177e4 LT |
46 | * drive. |
47 | * near_copies and far_copies must be at least one, and their product is at most | |
48 | * raid_disks. | |
c93983bf N |
49 | * |
50 | * If far_offset is true, then the far_copies are handled a bit differently. | |
51 | * The copies are still in different stripes, but instead of be very far apart | |
52 | * on disk, there are adjacent stripes. | |
1da177e4 LT |
53 | */ |
54 | ||
55 | /* | |
56 | * Number of guaranteed r10bios in case of extreme VM load: | |
57 | */ | |
58 | #define NR_RAID10_BIOS 256 | |
59 | ||
0a27ec96 N |
60 | static void allow_barrier(conf_t *conf); |
61 | static void lower_barrier(conf_t *conf); | |
62 | ||
dd0fc66f | 63 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
64 | { |
65 | conf_t *conf = data; | |
1da177e4 LT |
66 | int size = offsetof(struct r10bio_s, devs[conf->copies]); |
67 | ||
68 | /* allocate a r10bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 69 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
70 | } |
71 | ||
72 | static void r10bio_pool_free(void *r10_bio, void *data) | |
73 | { | |
74 | kfree(r10_bio); | |
75 | } | |
76 | ||
0310fa21 | 77 | /* Maximum size of each resync request */ |
1da177e4 | 78 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 79 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
80 | /* amount of memory to reserve for resync requests */ |
81 | #define RESYNC_WINDOW (1024*1024) | |
82 | /* maximum number of concurrent requests, memory permitting */ | |
83 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
84 | |
85 | /* | |
86 | * When performing a resync, we need to read and compare, so | |
87 | * we need as many pages are there are copies. | |
88 | * When performing a recovery, we need 2 bios, one for read, | |
89 | * one for write (we recover only one drive per r10buf) | |
90 | * | |
91 | */ | |
dd0fc66f | 92 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
93 | { |
94 | conf_t *conf = data; | |
95 | struct page *page; | |
96 | r10bio_t *r10_bio; | |
97 | struct bio *bio; | |
98 | int i, j; | |
99 | int nalloc; | |
100 | ||
101 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 102 | if (!r10_bio) |
1da177e4 | 103 | return NULL; |
1da177e4 LT |
104 | |
105 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | |
106 | nalloc = conf->copies; /* resync */ | |
107 | else | |
108 | nalloc = 2; /* recovery */ | |
109 | ||
110 | /* | |
111 | * Allocate bios. | |
112 | */ | |
113 | for (j = nalloc ; j-- ; ) { | |
6746557f | 114 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
115 | if (!bio) |
116 | goto out_free_bio; | |
117 | r10_bio->devs[j].bio = bio; | |
118 | } | |
119 | /* | |
120 | * Allocate RESYNC_PAGES data pages and attach them | |
121 | * where needed. | |
122 | */ | |
123 | for (j = 0 ; j < nalloc; j++) { | |
124 | bio = r10_bio->devs[j].bio; | |
125 | for (i = 0; i < RESYNC_PAGES; i++) { | |
126 | page = alloc_page(gfp_flags); | |
127 | if (unlikely(!page)) | |
128 | goto out_free_pages; | |
129 | ||
130 | bio->bi_io_vec[i].bv_page = page; | |
131 | } | |
132 | } | |
133 | ||
134 | return r10_bio; | |
135 | ||
136 | out_free_pages: | |
137 | for ( ; i > 0 ; i--) | |
1345b1d8 | 138 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
139 | while (j--) |
140 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 141 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
142 | j = -1; |
143 | out_free_bio: | |
144 | while ( ++j < nalloc ) | |
145 | bio_put(r10_bio->devs[j].bio); | |
146 | r10bio_pool_free(r10_bio, conf); | |
147 | return NULL; | |
148 | } | |
149 | ||
150 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
151 | { | |
152 | int i; | |
153 | conf_t *conf = data; | |
154 | r10bio_t *r10bio = __r10_bio; | |
155 | int j; | |
156 | ||
157 | for (j=0; j < conf->copies; j++) { | |
158 | struct bio *bio = r10bio->devs[j].bio; | |
159 | if (bio) { | |
160 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 161 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
162 | bio->bi_io_vec[i].bv_page = NULL; |
163 | } | |
164 | bio_put(bio); | |
165 | } | |
166 | } | |
167 | r10bio_pool_free(r10bio, conf); | |
168 | } | |
169 | ||
170 | static void put_all_bios(conf_t *conf, r10bio_t *r10_bio) | |
171 | { | |
172 | int i; | |
173 | ||
174 | for (i = 0; i < conf->copies; i++) { | |
175 | struct bio **bio = & r10_bio->devs[i].bio; | |
0eb3ff12 | 176 | if (*bio && *bio != IO_BLOCKED) |
1da177e4 LT |
177 | bio_put(*bio); |
178 | *bio = NULL; | |
179 | } | |
180 | } | |
181 | ||
858119e1 | 182 | static void free_r10bio(r10bio_t *r10_bio) |
1da177e4 | 183 | { |
070ec55d | 184 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
185 | |
186 | /* | |
187 | * Wake up any possible resync thread that waits for the device | |
188 | * to go idle. | |
189 | */ | |
0a27ec96 | 190 | allow_barrier(conf); |
1da177e4 LT |
191 | |
192 | put_all_bios(conf, r10_bio); | |
193 | mempool_free(r10_bio, conf->r10bio_pool); | |
194 | } | |
195 | ||
858119e1 | 196 | static void put_buf(r10bio_t *r10_bio) |
1da177e4 | 197 | { |
070ec55d | 198 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
199 | |
200 | mempool_free(r10_bio, conf->r10buf_pool); | |
201 | ||
0a27ec96 | 202 | lower_barrier(conf); |
1da177e4 LT |
203 | } |
204 | ||
205 | static void reschedule_retry(r10bio_t *r10_bio) | |
206 | { | |
207 | unsigned long flags; | |
208 | mddev_t *mddev = r10_bio->mddev; | |
070ec55d | 209 | conf_t *conf = mddev->private; |
1da177e4 LT |
210 | |
211 | spin_lock_irqsave(&conf->device_lock, flags); | |
212 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 213 | conf->nr_queued ++; |
1da177e4 LT |
214 | spin_unlock_irqrestore(&conf->device_lock, flags); |
215 | ||
388667be AJ |
216 | /* wake up frozen array... */ |
217 | wake_up(&conf->wait_barrier); | |
218 | ||
1da177e4 LT |
219 | md_wakeup_thread(mddev->thread); |
220 | } | |
221 | ||
222 | /* | |
223 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
224 | * operation and are ready to return a success/failure code to the buffer | |
225 | * cache layer. | |
226 | */ | |
227 | static void raid_end_bio_io(r10bio_t *r10_bio) | |
228 | { | |
229 | struct bio *bio = r10_bio->master_bio; | |
230 | ||
6712ecf8 | 231 | bio_endio(bio, |
1da177e4 LT |
232 | test_bit(R10BIO_Uptodate, &r10_bio->state) ? 0 : -EIO); |
233 | free_r10bio(r10_bio); | |
234 | } | |
235 | ||
236 | /* | |
237 | * Update disk head position estimator based on IRQ completion info. | |
238 | */ | |
239 | static inline void update_head_pos(int slot, r10bio_t *r10_bio) | |
240 | { | |
070ec55d | 241 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
242 | |
243 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
244 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
245 | } | |
246 | ||
6712ecf8 | 247 | static void raid10_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
248 | { |
249 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 250 | r10bio_t *r10_bio = bio->bi_private; |
1da177e4 | 251 | int slot, dev; |
070ec55d | 252 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 | 253 | |
1da177e4 LT |
254 | |
255 | slot = r10_bio->read_slot; | |
256 | dev = r10_bio->devs[slot].devnum; | |
257 | /* | |
258 | * this branch is our 'one mirror IO has finished' event handler: | |
259 | */ | |
4443ae10 N |
260 | update_head_pos(slot, r10_bio); |
261 | ||
262 | if (uptodate) { | |
1da177e4 LT |
263 | /* |
264 | * Set R10BIO_Uptodate in our master bio, so that | |
265 | * we will return a good error code to the higher | |
266 | * levels even if IO on some other mirrored buffer fails. | |
267 | * | |
268 | * The 'master' represents the composite IO operation to | |
269 | * user-side. So if something waits for IO, then it will | |
270 | * wait for the 'master' bio. | |
271 | */ | |
272 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 273 | raid_end_bio_io(r10_bio); |
4443ae10 | 274 | } else { |
1da177e4 LT |
275 | /* |
276 | * oops, read error: | |
277 | */ | |
278 | char b[BDEVNAME_SIZE]; | |
279 | if (printk_ratelimit()) | |
128595ed N |
280 | printk(KERN_ERR "md/raid10:%s: %s: rescheduling sector %llu\n", |
281 | mdname(conf->mddev), | |
1da177e4 LT |
282 | bdevname(conf->mirrors[dev].rdev->bdev,b), (unsigned long long)r10_bio->sector); |
283 | reschedule_retry(r10_bio); | |
284 | } | |
285 | ||
286 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); | |
1da177e4 LT |
287 | } |
288 | ||
6712ecf8 | 289 | static void raid10_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
290 | { |
291 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 292 | r10bio_t *r10_bio = bio->bi_private; |
1da177e4 | 293 | int slot, dev; |
070ec55d | 294 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 | 295 | |
1da177e4 LT |
296 | for (slot = 0; slot < conf->copies; slot++) |
297 | if (r10_bio->devs[slot].bio == bio) | |
298 | break; | |
299 | dev = r10_bio->devs[slot].devnum; | |
300 | ||
301 | /* | |
302 | * this branch is our 'one mirror IO has finished' event handler: | |
303 | */ | |
6cce3b23 | 304 | if (!uptodate) { |
1da177e4 | 305 | md_error(r10_bio->mddev, conf->mirrors[dev].rdev); |
6cce3b23 N |
306 | /* an I/O failed, we can't clear the bitmap */ |
307 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
308 | } else | |
1da177e4 LT |
309 | /* |
310 | * Set R10BIO_Uptodate in our master bio, so that | |
311 | * we will return a good error code for to the higher | |
312 | * levels even if IO on some other mirrored buffer fails. | |
313 | * | |
314 | * The 'master' represents the composite IO operation to | |
315 | * user-side. So if something waits for IO, then it will | |
316 | * wait for the 'master' bio. | |
317 | */ | |
318 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
319 | ||
320 | update_head_pos(slot, r10_bio); | |
321 | ||
322 | /* | |
323 | * | |
324 | * Let's see if all mirrored write operations have finished | |
325 | * already. | |
326 | */ | |
327 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
6cce3b23 N |
328 | /* clear the bitmap if all writes complete successfully */ |
329 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
330 | r10_bio->sectors, | |
331 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
332 | 0); | |
1da177e4 LT |
333 | md_write_end(r10_bio->mddev); |
334 | raid_end_bio_io(r10_bio); | |
335 | } | |
336 | ||
337 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); | |
1da177e4 LT |
338 | } |
339 | ||
340 | ||
341 | /* | |
342 | * RAID10 layout manager | |
25985edc | 343 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
344 | * parameters: near_copies and far_copies. |
345 | * near_copies * far_copies must be <= raid_disks. | |
346 | * Normally one of these will be 1. | |
347 | * If both are 1, we get raid0. | |
348 | * If near_copies == raid_disks, we get raid1. | |
349 | * | |
25985edc | 350 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
351 | * first chunk, followed by near_copies copies of the next chunk and |
352 | * so on. | |
353 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
354 | * as described above, we start again with a device offset of near_copies. | |
355 | * So we effectively have another copy of the whole array further down all | |
356 | * the drives, but with blocks on different drives. | |
357 | * With this layout, and block is never stored twice on the one device. | |
358 | * | |
359 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 360 | * on each device that it is on. |
1da177e4 LT |
361 | * |
362 | * raid10_find_virt does the reverse mapping, from a device and a | |
363 | * sector offset to a virtual address | |
364 | */ | |
365 | ||
366 | static void raid10_find_phys(conf_t *conf, r10bio_t *r10bio) | |
367 | { | |
368 | int n,f; | |
369 | sector_t sector; | |
370 | sector_t chunk; | |
371 | sector_t stripe; | |
372 | int dev; | |
373 | ||
374 | int slot = 0; | |
375 | ||
376 | /* now calculate first sector/dev */ | |
377 | chunk = r10bio->sector >> conf->chunk_shift; | |
378 | sector = r10bio->sector & conf->chunk_mask; | |
379 | ||
380 | chunk *= conf->near_copies; | |
381 | stripe = chunk; | |
382 | dev = sector_div(stripe, conf->raid_disks); | |
c93983bf N |
383 | if (conf->far_offset) |
384 | stripe *= conf->far_copies; | |
1da177e4 LT |
385 | |
386 | sector += stripe << conf->chunk_shift; | |
387 | ||
388 | /* and calculate all the others */ | |
389 | for (n=0; n < conf->near_copies; n++) { | |
390 | int d = dev; | |
391 | sector_t s = sector; | |
392 | r10bio->devs[slot].addr = sector; | |
393 | r10bio->devs[slot].devnum = d; | |
394 | slot++; | |
395 | ||
396 | for (f = 1; f < conf->far_copies; f++) { | |
397 | d += conf->near_copies; | |
398 | if (d >= conf->raid_disks) | |
399 | d -= conf->raid_disks; | |
400 | s += conf->stride; | |
401 | r10bio->devs[slot].devnum = d; | |
402 | r10bio->devs[slot].addr = s; | |
403 | slot++; | |
404 | } | |
405 | dev++; | |
406 | if (dev >= conf->raid_disks) { | |
407 | dev = 0; | |
408 | sector += (conf->chunk_mask + 1); | |
409 | } | |
410 | } | |
411 | BUG_ON(slot != conf->copies); | |
412 | } | |
413 | ||
414 | static sector_t raid10_find_virt(conf_t *conf, sector_t sector, int dev) | |
415 | { | |
416 | sector_t offset, chunk, vchunk; | |
417 | ||
1da177e4 | 418 | offset = sector & conf->chunk_mask; |
c93983bf N |
419 | if (conf->far_offset) { |
420 | int fc; | |
421 | chunk = sector >> conf->chunk_shift; | |
422 | fc = sector_div(chunk, conf->far_copies); | |
423 | dev -= fc * conf->near_copies; | |
424 | if (dev < 0) | |
425 | dev += conf->raid_disks; | |
426 | } else { | |
64a742bc | 427 | while (sector >= conf->stride) { |
c93983bf N |
428 | sector -= conf->stride; |
429 | if (dev < conf->near_copies) | |
430 | dev += conf->raid_disks - conf->near_copies; | |
431 | else | |
432 | dev -= conf->near_copies; | |
433 | } | |
434 | chunk = sector >> conf->chunk_shift; | |
435 | } | |
1da177e4 LT |
436 | vchunk = chunk * conf->raid_disks + dev; |
437 | sector_div(vchunk, conf->near_copies); | |
438 | return (vchunk << conf->chunk_shift) + offset; | |
439 | } | |
440 | ||
441 | /** | |
442 | * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged | |
443 | * @q: request queue | |
cc371e66 | 444 | * @bvm: properties of new bio |
1da177e4 LT |
445 | * @biovec: the request that could be merged to it. |
446 | * | |
447 | * Return amount of bytes we can accept at this offset | |
448 | * If near_copies == raid_disk, there are no striping issues, | |
449 | * but in that case, the function isn't called at all. | |
450 | */ | |
cc371e66 AK |
451 | static int raid10_mergeable_bvec(struct request_queue *q, |
452 | struct bvec_merge_data *bvm, | |
453 | struct bio_vec *biovec) | |
1da177e4 LT |
454 | { |
455 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 456 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
1da177e4 | 457 | int max; |
9d8f0363 | 458 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 459 | unsigned int bio_sectors = bvm->bi_size >> 9; |
1da177e4 LT |
460 | |
461 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
462 | if (max < 0) max = 0; /* bio_add cannot handle a negative return */ | |
cc371e66 AK |
463 | if (max <= biovec->bv_len && bio_sectors == 0) |
464 | return biovec->bv_len; | |
1da177e4 LT |
465 | else |
466 | return max; | |
467 | } | |
468 | ||
469 | /* | |
470 | * This routine returns the disk from which the requested read should | |
471 | * be done. There is a per-array 'next expected sequential IO' sector | |
472 | * number - if this matches on the next IO then we use the last disk. | |
473 | * There is also a per-disk 'last know head position' sector that is | |
474 | * maintained from IRQ contexts, both the normal and the resync IO | |
475 | * completion handlers update this position correctly. If there is no | |
476 | * perfect sequential match then we pick the disk whose head is closest. | |
477 | * | |
478 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
479 | * because position is mirror, not device based. | |
480 | * | |
481 | * The rdev for the device selected will have nr_pending incremented. | |
482 | */ | |
483 | ||
484 | /* | |
485 | * FIXME: possibly should rethink readbalancing and do it differently | |
486 | * depending on near_copies / far_copies geometry. | |
487 | */ | |
488 | static int read_balance(conf_t *conf, r10bio_t *r10_bio) | |
489 | { | |
af3a2cd6 | 490 | const sector_t this_sector = r10_bio->sector; |
1da177e4 LT |
491 | int disk, slot, nslot; |
492 | const int sectors = r10_bio->sectors; | |
493 | sector_t new_distance, current_distance; | |
d6065f7b | 494 | mdk_rdev_t *rdev; |
1da177e4 LT |
495 | |
496 | raid10_find_phys(conf, r10_bio); | |
497 | rcu_read_lock(); | |
498 | /* | |
499 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
500 | * device if no resync is going on (recovery is ok), or below |
501 | * the resync window. We take the first readable disk when | |
502 | * above the resync window. | |
1da177e4 LT |
503 | */ |
504 | if (conf->mddev->recovery_cp < MaxSector | |
505 | && (this_sector + sectors >= conf->next_resync)) { | |
506 | /* make sure that disk is operational */ | |
507 | slot = 0; | |
508 | disk = r10_bio->devs[slot].devnum; | |
509 | ||
d6065f7b | 510 | while ((rdev = rcu_dereference(conf->mirrors[disk].rdev)) == NULL || |
0eb3ff12 | 511 | r10_bio->devs[slot].bio == IO_BLOCKED || |
b2d444d7 | 512 | !test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
513 | slot++; |
514 | if (slot == conf->copies) { | |
515 | slot = 0; | |
516 | disk = -1; | |
517 | break; | |
518 | } | |
519 | disk = r10_bio->devs[slot].devnum; | |
520 | } | |
521 | goto rb_out; | |
522 | } | |
523 | ||
524 | ||
525 | /* make sure the disk is operational */ | |
526 | slot = 0; | |
527 | disk = r10_bio->devs[slot].devnum; | |
d6065f7b | 528 | while ((rdev=rcu_dereference(conf->mirrors[disk].rdev)) == NULL || |
0eb3ff12 | 529 | r10_bio->devs[slot].bio == IO_BLOCKED || |
b2d444d7 | 530 | !test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
531 | slot ++; |
532 | if (slot == conf->copies) { | |
533 | disk = -1; | |
534 | goto rb_out; | |
535 | } | |
536 | disk = r10_bio->devs[slot].devnum; | |
537 | } | |
538 | ||
539 | ||
3ec67ac1 N |
540 | current_distance = abs(r10_bio->devs[slot].addr - |
541 | conf->mirrors[disk].head_position); | |
1da177e4 | 542 | |
8ed3a195 KS |
543 | /* Find the disk whose head is closest, |
544 | * or - for far > 1 - find the closest to partition beginning */ | |
1da177e4 LT |
545 | |
546 | for (nslot = slot; nslot < conf->copies; nslot++) { | |
547 | int ndisk = r10_bio->devs[nslot].devnum; | |
548 | ||
549 | ||
d6065f7b | 550 | if ((rdev=rcu_dereference(conf->mirrors[ndisk].rdev)) == NULL || |
0eb3ff12 | 551 | r10_bio->devs[nslot].bio == IO_BLOCKED || |
b2d444d7 | 552 | !test_bit(In_sync, &rdev->flags)) |
1da177e4 LT |
553 | continue; |
554 | ||
22dfdf52 N |
555 | /* This optimisation is debatable, and completely destroys |
556 | * sequential read speed for 'far copies' arrays. So only | |
557 | * keep it for 'near' arrays, and review those later. | |
558 | */ | |
559 | if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending)) { | |
1da177e4 LT |
560 | disk = ndisk; |
561 | slot = nslot; | |
562 | break; | |
563 | } | |
8ed3a195 KS |
564 | |
565 | /* for far > 1 always use the lowest address */ | |
566 | if (conf->far_copies > 1) | |
567 | new_distance = r10_bio->devs[nslot].addr; | |
568 | else | |
569 | new_distance = abs(r10_bio->devs[nslot].addr - | |
570 | conf->mirrors[ndisk].head_position); | |
1da177e4 LT |
571 | if (new_distance < current_distance) { |
572 | current_distance = new_distance; | |
573 | disk = ndisk; | |
574 | slot = nslot; | |
575 | } | |
576 | } | |
577 | ||
578 | rb_out: | |
579 | r10_bio->read_slot = slot; | |
580 | /* conf->next_seq_sect = this_sector + sectors;*/ | |
581 | ||
d6065f7b | 582 | if (disk >= 0 && (rdev=rcu_dereference(conf->mirrors[disk].rdev))!= NULL) |
1da177e4 | 583 | atomic_inc(&conf->mirrors[disk].rdev->nr_pending); |
29fc7e3e N |
584 | else |
585 | disk = -1; | |
1da177e4 LT |
586 | rcu_read_unlock(); |
587 | ||
588 | return disk; | |
589 | } | |
590 | ||
0d129228 N |
591 | static int raid10_congested(void *data, int bits) |
592 | { | |
593 | mddev_t *mddev = data; | |
070ec55d | 594 | conf_t *conf = mddev->private; |
0d129228 N |
595 | int i, ret = 0; |
596 | ||
3fa841d7 N |
597 | if (mddev_congested(mddev, bits)) |
598 | return 1; | |
0d129228 | 599 | rcu_read_lock(); |
84707f38 | 600 | for (i = 0; i < conf->raid_disks && ret == 0; i++) { |
0d129228 N |
601 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
602 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
165125e1 | 603 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
604 | |
605 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
606 | } | |
607 | } | |
608 | rcu_read_unlock(); | |
609 | return ret; | |
610 | } | |
611 | ||
7eaceacc | 612 | static void flush_pending_writes(conf_t *conf) |
a35e63ef N |
613 | { |
614 | /* Any writes that have been queued but are awaiting | |
615 | * bitmap updates get flushed here. | |
a35e63ef | 616 | */ |
a35e63ef N |
617 | spin_lock_irq(&conf->device_lock); |
618 | ||
619 | if (conf->pending_bio_list.head) { | |
620 | struct bio *bio; | |
621 | bio = bio_list_get(&conf->pending_bio_list); | |
a35e63ef N |
622 | spin_unlock_irq(&conf->device_lock); |
623 | /* flush any pending bitmap writes to disk | |
624 | * before proceeding w/ I/O */ | |
625 | bitmap_unplug(conf->mddev->bitmap); | |
626 | ||
627 | while (bio) { /* submit pending writes */ | |
628 | struct bio *next = bio->bi_next; | |
629 | bio->bi_next = NULL; | |
630 | generic_make_request(bio); | |
631 | bio = next; | |
632 | } | |
a35e63ef N |
633 | } else |
634 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 635 | } |
7eaceacc JA |
636 | |
637 | static void md_kick_device(mddev_t *mddev) | |
638 | { | |
639 | blk_flush_plug(current); | |
640 | md_wakeup_thread(mddev->thread); | |
641 | } | |
642 | ||
0a27ec96 N |
643 | /* Barriers.... |
644 | * Sometimes we need to suspend IO while we do something else, | |
645 | * either some resync/recovery, or reconfigure the array. | |
646 | * To do this we raise a 'barrier'. | |
647 | * The 'barrier' is a counter that can be raised multiple times | |
648 | * to count how many activities are happening which preclude | |
649 | * normal IO. | |
650 | * We can only raise the barrier if there is no pending IO. | |
651 | * i.e. if nr_pending == 0. | |
652 | * We choose only to raise the barrier if no-one is waiting for the | |
653 | * barrier to go down. This means that as soon as an IO request | |
654 | * is ready, no other operations which require a barrier will start | |
655 | * until the IO request has had a chance. | |
656 | * | |
657 | * So: regular IO calls 'wait_barrier'. When that returns there | |
658 | * is no backgroup IO happening, It must arrange to call | |
659 | * allow_barrier when it has finished its IO. | |
660 | * backgroup IO calls must call raise_barrier. Once that returns | |
661 | * there is no normal IO happeing. It must arrange to call | |
662 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 663 | */ |
1da177e4 | 664 | |
6cce3b23 | 665 | static void raise_barrier(conf_t *conf, int force) |
1da177e4 | 666 | { |
6cce3b23 | 667 | BUG_ON(force && !conf->barrier); |
1da177e4 | 668 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 669 | |
6cce3b23 N |
670 | /* Wait until no block IO is waiting (unless 'force') */ |
671 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
7eaceacc | 672 | conf->resync_lock, md_kick_device(conf->mddev)); |
0a27ec96 N |
673 | |
674 | /* block any new IO from starting */ | |
675 | conf->barrier++; | |
676 | ||
677 | /* No wait for all pending IO to complete */ | |
678 | wait_event_lock_irq(conf->wait_barrier, | |
679 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
7eaceacc | 680 | conf->resync_lock, md_kick_device(conf->mddev)); |
0a27ec96 N |
681 | |
682 | spin_unlock_irq(&conf->resync_lock); | |
683 | } | |
684 | ||
685 | static void lower_barrier(conf_t *conf) | |
686 | { | |
687 | unsigned long flags; | |
688 | spin_lock_irqsave(&conf->resync_lock, flags); | |
689 | conf->barrier--; | |
690 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
691 | wake_up(&conf->wait_barrier); | |
692 | } | |
693 | ||
694 | static void wait_barrier(conf_t *conf) | |
695 | { | |
696 | spin_lock_irq(&conf->resync_lock); | |
697 | if (conf->barrier) { | |
698 | conf->nr_waiting++; | |
699 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, | |
700 | conf->resync_lock, | |
7eaceacc | 701 | md_kick_device(conf->mddev)); |
0a27ec96 | 702 | conf->nr_waiting--; |
1da177e4 | 703 | } |
0a27ec96 | 704 | conf->nr_pending++; |
1da177e4 LT |
705 | spin_unlock_irq(&conf->resync_lock); |
706 | } | |
707 | ||
0a27ec96 N |
708 | static void allow_barrier(conf_t *conf) |
709 | { | |
710 | unsigned long flags; | |
711 | spin_lock_irqsave(&conf->resync_lock, flags); | |
712 | conf->nr_pending--; | |
713 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
714 | wake_up(&conf->wait_barrier); | |
715 | } | |
716 | ||
4443ae10 N |
717 | static void freeze_array(conf_t *conf) |
718 | { | |
719 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 720 | * go quiet. |
4443ae10 | 721 | * We increment barrier and nr_waiting, and then |
1c830532 N |
722 | * wait until nr_pending match nr_queued+1 |
723 | * This is called in the context of one normal IO request | |
724 | * that has failed. Thus any sync request that might be pending | |
725 | * will be blocked by nr_pending, and we need to wait for | |
726 | * pending IO requests to complete or be queued for re-try. | |
727 | * Thus the number queued (nr_queued) plus this request (1) | |
728 | * must match the number of pending IOs (nr_pending) before | |
729 | * we continue. | |
4443ae10 N |
730 | */ |
731 | spin_lock_irq(&conf->resync_lock); | |
732 | conf->barrier++; | |
733 | conf->nr_waiting++; | |
734 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 735 | conf->nr_pending == conf->nr_queued+1, |
4443ae10 | 736 | conf->resync_lock, |
a35e63ef | 737 | ({ flush_pending_writes(conf); |
7eaceacc | 738 | md_kick_device(conf->mddev); })); |
4443ae10 N |
739 | spin_unlock_irq(&conf->resync_lock); |
740 | } | |
741 | ||
742 | static void unfreeze_array(conf_t *conf) | |
743 | { | |
744 | /* reverse the effect of the freeze */ | |
745 | spin_lock_irq(&conf->resync_lock); | |
746 | conf->barrier--; | |
747 | conf->nr_waiting--; | |
748 | wake_up(&conf->wait_barrier); | |
749 | spin_unlock_irq(&conf->resync_lock); | |
750 | } | |
751 | ||
21a52c6d | 752 | static int make_request(mddev_t *mddev, struct bio * bio) |
1da177e4 | 753 | { |
070ec55d | 754 | conf_t *conf = mddev->private; |
1da177e4 LT |
755 | mirror_info_t *mirror; |
756 | r10bio_t *r10_bio; | |
757 | struct bio *read_bio; | |
758 | int i; | |
759 | int chunk_sects = conf->chunk_mask + 1; | |
a362357b | 760 | const int rw = bio_data_dir(bio); |
2c7d46ec | 761 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 762 | const unsigned long do_fua = (bio->bi_rw & REQ_FUA); |
6cce3b23 | 763 | unsigned long flags; |
6bfe0b49 | 764 | mdk_rdev_t *blocked_rdev; |
1da177e4 | 765 | |
e9c7469b TH |
766 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
767 | md_flush_request(mddev, bio); | |
e5dcdd80 N |
768 | return 0; |
769 | } | |
770 | ||
1da177e4 LT |
771 | /* If this request crosses a chunk boundary, we need to |
772 | * split it. This will only happen for 1 PAGE (or less) requests. | |
773 | */ | |
774 | if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9) | |
775 | > chunk_sects && | |
776 | conf->near_copies < conf->raid_disks)) { | |
777 | struct bio_pair *bp; | |
778 | /* Sanity check -- queue functions should prevent this happening */ | |
779 | if (bio->bi_vcnt != 1 || | |
780 | bio->bi_idx != 0) | |
781 | goto bad_map; | |
782 | /* This is a one page bio that upper layers | |
783 | * refuse to split for us, so we need to split it. | |
784 | */ | |
6feef531 | 785 | bp = bio_split(bio, |
1da177e4 | 786 | chunk_sects - (bio->bi_sector & (chunk_sects - 1)) ); |
51e9ac77 N |
787 | |
788 | /* Each of these 'make_request' calls will call 'wait_barrier'. | |
789 | * If the first succeeds but the second blocks due to the resync | |
790 | * thread raising the barrier, we will deadlock because the | |
791 | * IO to the underlying device will be queued in generic_make_request | |
792 | * and will never complete, so will never reduce nr_pending. | |
793 | * So increment nr_waiting here so no new raise_barriers will | |
794 | * succeed, and so the second wait_barrier cannot block. | |
795 | */ | |
796 | spin_lock_irq(&conf->resync_lock); | |
797 | conf->nr_waiting++; | |
798 | spin_unlock_irq(&conf->resync_lock); | |
799 | ||
21a52c6d | 800 | if (make_request(mddev, &bp->bio1)) |
1da177e4 | 801 | generic_make_request(&bp->bio1); |
21a52c6d | 802 | if (make_request(mddev, &bp->bio2)) |
1da177e4 LT |
803 | generic_make_request(&bp->bio2); |
804 | ||
51e9ac77 N |
805 | spin_lock_irq(&conf->resync_lock); |
806 | conf->nr_waiting--; | |
807 | wake_up(&conf->wait_barrier); | |
808 | spin_unlock_irq(&conf->resync_lock); | |
809 | ||
1da177e4 LT |
810 | bio_pair_release(bp); |
811 | return 0; | |
812 | bad_map: | |
128595ed N |
813 | printk("md/raid10:%s: make_request bug: can't convert block across chunks" |
814 | " or bigger than %dk %llu %d\n", mdname(mddev), chunk_sects/2, | |
1da177e4 LT |
815 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); |
816 | ||
6712ecf8 | 817 | bio_io_error(bio); |
1da177e4 LT |
818 | return 0; |
819 | } | |
820 | ||
3d310eb7 | 821 | md_write_start(mddev, bio); |
06d91a5f | 822 | |
1da177e4 LT |
823 | /* |
824 | * Register the new request and wait if the reconstruction | |
825 | * thread has put up a bar for new requests. | |
826 | * Continue immediately if no resync is active currently. | |
827 | */ | |
0a27ec96 | 828 | wait_barrier(conf); |
1da177e4 | 829 | |
1da177e4 LT |
830 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); |
831 | ||
832 | r10_bio->master_bio = bio; | |
833 | r10_bio->sectors = bio->bi_size >> 9; | |
834 | ||
835 | r10_bio->mddev = mddev; | |
836 | r10_bio->sector = bio->bi_sector; | |
6cce3b23 | 837 | r10_bio->state = 0; |
1da177e4 | 838 | |
a362357b | 839 | if (rw == READ) { |
1da177e4 LT |
840 | /* |
841 | * read balancing logic: | |
842 | */ | |
843 | int disk = read_balance(conf, r10_bio); | |
844 | int slot = r10_bio->read_slot; | |
845 | if (disk < 0) { | |
846 | raid_end_bio_io(r10_bio); | |
847 | return 0; | |
848 | } | |
849 | mirror = conf->mirrors + disk; | |
850 | ||
a167f663 | 851 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1da177e4 LT |
852 | |
853 | r10_bio->devs[slot].bio = read_bio; | |
854 | ||
855 | read_bio->bi_sector = r10_bio->devs[slot].addr + | |
856 | mirror->rdev->data_offset; | |
857 | read_bio->bi_bdev = mirror->rdev->bdev; | |
858 | read_bio->bi_end_io = raid10_end_read_request; | |
7b6d91da | 859 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
860 | read_bio->bi_private = r10_bio; |
861 | ||
862 | generic_make_request(read_bio); | |
863 | return 0; | |
864 | } | |
865 | ||
866 | /* | |
867 | * WRITE: | |
868 | */ | |
6bfe0b49 | 869 | /* first select target devices under rcu_lock and |
1da177e4 LT |
870 | * inc refcount on their rdev. Record them by setting |
871 | * bios[x] to bio | |
872 | */ | |
873 | raid10_find_phys(conf, r10_bio); | |
6bfe0b49 | 874 | retry_write: |
cb6969e8 | 875 | blocked_rdev = NULL; |
1da177e4 LT |
876 | rcu_read_lock(); |
877 | for (i = 0; i < conf->copies; i++) { | |
878 | int d = r10_bio->devs[i].devnum; | |
d6065f7b | 879 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[d].rdev); |
6bfe0b49 DW |
880 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
881 | atomic_inc(&rdev->nr_pending); | |
882 | blocked_rdev = rdev; | |
883 | break; | |
884 | } | |
885 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
d6065f7b | 886 | atomic_inc(&rdev->nr_pending); |
1da177e4 | 887 | r10_bio->devs[i].bio = bio; |
6cce3b23 | 888 | } else { |
1da177e4 | 889 | r10_bio->devs[i].bio = NULL; |
6cce3b23 N |
890 | set_bit(R10BIO_Degraded, &r10_bio->state); |
891 | } | |
1da177e4 LT |
892 | } |
893 | rcu_read_unlock(); | |
894 | ||
6bfe0b49 DW |
895 | if (unlikely(blocked_rdev)) { |
896 | /* Have to wait for this device to get unblocked, then retry */ | |
897 | int j; | |
898 | int d; | |
899 | ||
900 | for (j = 0; j < i; j++) | |
901 | if (r10_bio->devs[j].bio) { | |
902 | d = r10_bio->devs[j].devnum; | |
903 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
904 | } | |
905 | allow_barrier(conf); | |
906 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
907 | wait_barrier(conf); | |
908 | goto retry_write; | |
909 | } | |
910 | ||
4e78064f N |
911 | atomic_set(&r10_bio->remaining, 1); |
912 | bitmap_startwrite(mddev->bitmap, bio->bi_sector, r10_bio->sectors, 0); | |
06d91a5f | 913 | |
1da177e4 LT |
914 | for (i = 0; i < conf->copies; i++) { |
915 | struct bio *mbio; | |
916 | int d = r10_bio->devs[i].devnum; | |
917 | if (!r10_bio->devs[i].bio) | |
918 | continue; | |
919 | ||
a167f663 | 920 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1da177e4 LT |
921 | r10_bio->devs[i].bio = mbio; |
922 | ||
923 | mbio->bi_sector = r10_bio->devs[i].addr+ | |
924 | conf->mirrors[d].rdev->data_offset; | |
925 | mbio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
926 | mbio->bi_end_io = raid10_end_write_request; | |
e9c7469b | 927 | mbio->bi_rw = WRITE | do_sync | do_fua; |
1da177e4 LT |
928 | mbio->bi_private = r10_bio; |
929 | ||
930 | atomic_inc(&r10_bio->remaining); | |
4e78064f N |
931 | spin_lock_irqsave(&conf->device_lock, flags); |
932 | bio_list_add(&conf->pending_bio_list, mbio); | |
4e78064f | 933 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
934 | } |
935 | ||
4e78064f N |
936 | if (atomic_dec_and_test(&r10_bio->remaining)) { |
937 | /* This matches the end of raid10_end_write_request() */ | |
938 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
939 | r10_bio->sectors, | |
940 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
941 | 0); | |
f6f953aa AR |
942 | md_write_end(mddev); |
943 | raid_end_bio_io(r10_bio); | |
f6f953aa AR |
944 | } |
945 | ||
a35e63ef N |
946 | /* In case raid10d snuck in to freeze_array */ |
947 | wake_up(&conf->wait_barrier); | |
948 | ||
7eaceacc | 949 | if (do_sync || !mddev->bitmap) |
e3881a68 LE |
950 | md_wakeup_thread(mddev->thread); |
951 | ||
1da177e4 LT |
952 | return 0; |
953 | } | |
954 | ||
955 | static void status(struct seq_file *seq, mddev_t *mddev) | |
956 | { | |
070ec55d | 957 | conf_t *conf = mddev->private; |
1da177e4 LT |
958 | int i; |
959 | ||
960 | if (conf->near_copies < conf->raid_disks) | |
9d8f0363 | 961 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
1da177e4 LT |
962 | if (conf->near_copies > 1) |
963 | seq_printf(seq, " %d near-copies", conf->near_copies); | |
c93983bf N |
964 | if (conf->far_copies > 1) { |
965 | if (conf->far_offset) | |
966 | seq_printf(seq, " %d offset-copies", conf->far_copies); | |
967 | else | |
968 | seq_printf(seq, " %d far-copies", conf->far_copies); | |
969 | } | |
1da177e4 | 970 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
76186dd8 | 971 | conf->raid_disks - mddev->degraded); |
1da177e4 LT |
972 | for (i = 0; i < conf->raid_disks; i++) |
973 | seq_printf(seq, "%s", | |
974 | conf->mirrors[i].rdev && | |
b2d444d7 | 975 | test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
976 | seq_printf(seq, "]"); |
977 | } | |
978 | ||
979 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
980 | { | |
981 | char b[BDEVNAME_SIZE]; | |
070ec55d | 982 | conf_t *conf = mddev->private; |
1da177e4 LT |
983 | |
984 | /* | |
985 | * If it is not operational, then we have already marked it as dead | |
986 | * else if it is the last working disks, ignore the error, let the | |
987 | * next level up know. | |
988 | * else mark the drive as failed | |
989 | */ | |
b2d444d7 | 990 | if (test_bit(In_sync, &rdev->flags) |
76186dd8 | 991 | && conf->raid_disks-mddev->degraded == 1) |
1da177e4 LT |
992 | /* |
993 | * Don't fail the drive, just return an IO error. | |
994 | * The test should really be more sophisticated than | |
995 | * "working_disks == 1", but it isn't critical, and | |
996 | * can wait until we do more sophisticated "is the drive | |
997 | * really dead" tests... | |
998 | */ | |
999 | return; | |
c04be0aa N |
1000 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1001 | unsigned long flags; | |
1002 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1003 | mddev->degraded++; |
c04be0aa | 1004 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1005 | /* |
1006 | * if recovery is running, make sure it aborts. | |
1007 | */ | |
dfc70645 | 1008 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1009 | } |
b2d444d7 | 1010 | set_bit(Faulty, &rdev->flags); |
850b2b42 | 1011 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
1012 | printk(KERN_ALERT |
1013 | "md/raid10:%s: Disk failure on %s, disabling device.\n" | |
1014 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
128595ed N |
1015 | mdname(mddev), bdevname(rdev->bdev, b), |
1016 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1017 | } |
1018 | ||
1019 | static void print_conf(conf_t *conf) | |
1020 | { | |
1021 | int i; | |
1022 | mirror_info_t *tmp; | |
1023 | ||
128595ed | 1024 | printk(KERN_DEBUG "RAID10 conf printout:\n"); |
1da177e4 | 1025 | if (!conf) { |
128595ed | 1026 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1027 | return; |
1028 | } | |
128595ed | 1029 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1030 | conf->raid_disks); |
1031 | ||
1032 | for (i = 0; i < conf->raid_disks; i++) { | |
1033 | char b[BDEVNAME_SIZE]; | |
1034 | tmp = conf->mirrors + i; | |
1035 | if (tmp->rdev) | |
128595ed | 1036 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
b2d444d7 N |
1037 | i, !test_bit(In_sync, &tmp->rdev->flags), |
1038 | !test_bit(Faulty, &tmp->rdev->flags), | |
1da177e4 LT |
1039 | bdevname(tmp->rdev->bdev,b)); |
1040 | } | |
1041 | } | |
1042 | ||
1043 | static void close_sync(conf_t *conf) | |
1044 | { | |
0a27ec96 N |
1045 | wait_barrier(conf); |
1046 | allow_barrier(conf); | |
1da177e4 LT |
1047 | |
1048 | mempool_destroy(conf->r10buf_pool); | |
1049 | conf->r10buf_pool = NULL; | |
1050 | } | |
1051 | ||
6d508242 N |
1052 | /* check if there are enough drives for |
1053 | * every block to appear on atleast one | |
1054 | */ | |
1055 | static int enough(conf_t *conf) | |
1056 | { | |
1057 | int first = 0; | |
1058 | ||
1059 | do { | |
1060 | int n = conf->copies; | |
1061 | int cnt = 0; | |
1062 | while (n--) { | |
1063 | if (conf->mirrors[first].rdev) | |
1064 | cnt++; | |
1065 | first = (first+1) % conf->raid_disks; | |
1066 | } | |
1067 | if (cnt == 0) | |
1068 | return 0; | |
1069 | } while (first != 0); | |
1070 | return 1; | |
1071 | } | |
1072 | ||
1da177e4 LT |
1073 | static int raid10_spare_active(mddev_t *mddev) |
1074 | { | |
1075 | int i; | |
1076 | conf_t *conf = mddev->private; | |
1077 | mirror_info_t *tmp; | |
6b965620 N |
1078 | int count = 0; |
1079 | unsigned long flags; | |
1da177e4 LT |
1080 | |
1081 | /* | |
1082 | * Find all non-in_sync disks within the RAID10 configuration | |
1083 | * and mark them in_sync | |
1084 | */ | |
1085 | for (i = 0; i < conf->raid_disks; i++) { | |
1086 | tmp = conf->mirrors + i; | |
1087 | if (tmp->rdev | |
b2d444d7 | 1088 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 1089 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 1090 | count++; |
e6ffbcb6 | 1091 | sysfs_notify_dirent(tmp->rdev->sysfs_state); |
1da177e4 LT |
1092 | } |
1093 | } | |
6b965620 N |
1094 | spin_lock_irqsave(&conf->device_lock, flags); |
1095 | mddev->degraded -= count; | |
1096 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1097 | |
1098 | print_conf(conf); | |
6b965620 | 1099 | return count; |
1da177e4 LT |
1100 | } |
1101 | ||
1102 | ||
1103 | static int raid10_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
1104 | { | |
1105 | conf_t *conf = mddev->private; | |
199050ea | 1106 | int err = -EEXIST; |
1da177e4 LT |
1107 | int mirror; |
1108 | mirror_info_t *p; | |
6c2fce2e | 1109 | int first = 0; |
84707f38 | 1110 | int last = conf->raid_disks - 1; |
1da177e4 LT |
1111 | |
1112 | if (mddev->recovery_cp < MaxSector) | |
1113 | /* only hot-add to in-sync arrays, as recovery is | |
1114 | * very different from resync | |
1115 | */ | |
199050ea | 1116 | return -EBUSY; |
6d508242 | 1117 | if (!enough(conf)) |
199050ea | 1118 | return -EINVAL; |
1da177e4 | 1119 | |
a53a6c85 | 1120 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1121 | first = last = rdev->raid_disk; |
1da177e4 | 1122 | |
6cce3b23 | 1123 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 1124 | rdev->saved_raid_disk >= first && |
6cce3b23 N |
1125 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1126 | mirror = rdev->saved_raid_disk; | |
1127 | else | |
6c2fce2e NB |
1128 | mirror = first; |
1129 | for ( ; mirror <= last ; mirror++) | |
1da177e4 LT |
1130 | if ( !(p=conf->mirrors+mirror)->rdev) { |
1131 | ||
8f6c2e4b MP |
1132 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1133 | rdev->data_offset << 9); | |
627a2d3c N |
1134 | /* as we don't honour merge_bvec_fn, we must |
1135 | * never risk violating it, so limit | |
1136 | * ->max_segments to one lying with a single | |
1137 | * page, as a one page request is never in | |
1138 | * violation. | |
1da177e4 | 1139 | */ |
627a2d3c N |
1140 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
1141 | blk_queue_max_segments(mddev->queue, 1); | |
1142 | blk_queue_segment_boundary(mddev->queue, | |
1143 | PAGE_CACHE_SIZE - 1); | |
1144 | } | |
1da177e4 LT |
1145 | |
1146 | p->head_position = 0; | |
1147 | rdev->raid_disk = mirror; | |
199050ea | 1148 | err = 0; |
6cce3b23 N |
1149 | if (rdev->saved_raid_disk != mirror) |
1150 | conf->fullsync = 1; | |
d6065f7b | 1151 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1152 | break; |
1153 | } | |
1154 | ||
ac5e7113 | 1155 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1156 | print_conf(conf); |
199050ea | 1157 | return err; |
1da177e4 LT |
1158 | } |
1159 | ||
1160 | static int raid10_remove_disk(mddev_t *mddev, int number) | |
1161 | { | |
1162 | conf_t *conf = mddev->private; | |
1163 | int err = 0; | |
1164 | mdk_rdev_t *rdev; | |
1165 | mirror_info_t *p = conf->mirrors+ number; | |
1166 | ||
1167 | print_conf(conf); | |
1168 | rdev = p->rdev; | |
1169 | if (rdev) { | |
b2d444d7 | 1170 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1171 | atomic_read(&rdev->nr_pending)) { |
1172 | err = -EBUSY; | |
1173 | goto abort; | |
1174 | } | |
dfc70645 N |
1175 | /* Only remove faulty devices in recovery |
1176 | * is not possible. | |
1177 | */ | |
1178 | if (!test_bit(Faulty, &rdev->flags) && | |
1179 | enough(conf)) { | |
1180 | err = -EBUSY; | |
1181 | goto abort; | |
1182 | } | |
1da177e4 | 1183 | p->rdev = NULL; |
fbd568a3 | 1184 | synchronize_rcu(); |
1da177e4 LT |
1185 | if (atomic_read(&rdev->nr_pending)) { |
1186 | /* lost the race, try later */ | |
1187 | err = -EBUSY; | |
1188 | p->rdev = rdev; | |
ac5e7113 | 1189 | goto abort; |
1da177e4 | 1190 | } |
a91a2785 | 1191 | err = md_integrity_register(mddev); |
1da177e4 LT |
1192 | } |
1193 | abort: | |
1194 | ||
1195 | print_conf(conf); | |
1196 | return err; | |
1197 | } | |
1198 | ||
1199 | ||
6712ecf8 | 1200 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1201 | { |
7b92813c | 1202 | r10bio_t *r10_bio = bio->bi_private; |
070ec55d | 1203 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
1204 | int i,d; |
1205 | ||
1da177e4 LT |
1206 | for (i=0; i<conf->copies; i++) |
1207 | if (r10_bio->devs[i].bio == bio) | |
1208 | break; | |
b6385483 | 1209 | BUG_ON(i == conf->copies); |
1da177e4 LT |
1210 | update_head_pos(i, r10_bio); |
1211 | d = r10_bio->devs[i].devnum; | |
0eb3ff12 N |
1212 | |
1213 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1214 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
4dbcdc75 N |
1215 | else { |
1216 | atomic_add(r10_bio->sectors, | |
1217 | &conf->mirrors[d].rdev->corrected_errors); | |
1218 | if (!test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | |
1219 | md_error(r10_bio->mddev, | |
1220 | conf->mirrors[d].rdev); | |
1221 | } | |
1da177e4 LT |
1222 | |
1223 | /* for reconstruct, we always reschedule after a read. | |
1224 | * for resync, only after all reads | |
1225 | */ | |
73d5c38a | 1226 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1227 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1228 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1229 | /* we have read all the blocks, | |
1230 | * do the comparison in process context in raid10d | |
1231 | */ | |
1232 | reschedule_retry(r10_bio); | |
1233 | } | |
1da177e4 LT |
1234 | } |
1235 | ||
6712ecf8 | 1236 | static void end_sync_write(struct bio *bio, int error) |
1da177e4 LT |
1237 | { |
1238 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 1239 | r10bio_t *r10_bio = bio->bi_private; |
1da177e4 | 1240 | mddev_t *mddev = r10_bio->mddev; |
070ec55d | 1241 | conf_t *conf = mddev->private; |
1da177e4 LT |
1242 | int i,d; |
1243 | ||
1da177e4 LT |
1244 | for (i = 0; i < conf->copies; i++) |
1245 | if (r10_bio->devs[i].bio == bio) | |
1246 | break; | |
1247 | d = r10_bio->devs[i].devnum; | |
1248 | ||
1249 | if (!uptodate) | |
1250 | md_error(mddev, conf->mirrors[d].rdev); | |
dfc70645 | 1251 | |
1da177e4 LT |
1252 | update_head_pos(i, r10_bio); |
1253 | ||
73d5c38a | 1254 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); |
1da177e4 LT |
1255 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1256 | if (r10_bio->master_bio == NULL) { | |
1257 | /* the primary of several recovery bios */ | |
73d5c38a | 1258 | sector_t s = r10_bio->sectors; |
1da177e4 | 1259 | put_buf(r10_bio); |
73d5c38a | 1260 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1261 | break; |
1262 | } else { | |
1263 | r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio; | |
1264 | put_buf(r10_bio); | |
1265 | r10_bio = r10_bio2; | |
1266 | } | |
1267 | } | |
1da177e4 LT |
1268 | } |
1269 | ||
1270 | /* | |
1271 | * Note: sync and recover and handled very differently for raid10 | |
1272 | * This code is for resync. | |
1273 | * For resync, we read through virtual addresses and read all blocks. | |
1274 | * If there is any error, we schedule a write. The lowest numbered | |
1275 | * drive is authoritative. | |
1276 | * However requests come for physical address, so we need to map. | |
1277 | * For every physical address there are raid_disks/copies virtual addresses, | |
1278 | * which is always are least one, but is not necessarly an integer. | |
1279 | * This means that a physical address can span multiple chunks, so we may | |
1280 | * have to submit multiple io requests for a single sync request. | |
1281 | */ | |
1282 | /* | |
1283 | * We check if all blocks are in-sync and only write to blocks that | |
1284 | * aren't in sync | |
1285 | */ | |
1286 | static void sync_request_write(mddev_t *mddev, r10bio_t *r10_bio) | |
1287 | { | |
070ec55d | 1288 | conf_t *conf = mddev->private; |
1da177e4 LT |
1289 | int i, first; |
1290 | struct bio *tbio, *fbio; | |
1291 | ||
1292 | atomic_set(&r10_bio->remaining, 1); | |
1293 | ||
1294 | /* find the first device with a block */ | |
1295 | for (i=0; i<conf->copies; i++) | |
1296 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) | |
1297 | break; | |
1298 | ||
1299 | if (i == conf->copies) | |
1300 | goto done; | |
1301 | ||
1302 | first = i; | |
1303 | fbio = r10_bio->devs[i].bio; | |
1304 | ||
1305 | /* now find blocks with errors */ | |
0eb3ff12 N |
1306 | for (i=0 ; i < conf->copies ; i++) { |
1307 | int j, d; | |
1308 | int vcnt = r10_bio->sectors >> (PAGE_SHIFT-9); | |
1da177e4 | 1309 | |
1da177e4 | 1310 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
1311 | |
1312 | if (tbio->bi_end_io != end_sync_read) | |
1313 | continue; | |
1314 | if (i == first) | |
1da177e4 | 1315 | continue; |
0eb3ff12 N |
1316 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) { |
1317 | /* We know that the bi_io_vec layout is the same for | |
1318 | * both 'first' and 'i', so we just compare them. | |
1319 | * All vec entries are PAGE_SIZE; | |
1320 | */ | |
1321 | for (j = 0; j < vcnt; j++) | |
1322 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), | |
1323 | page_address(tbio->bi_io_vec[j].bv_page), | |
1324 | PAGE_SIZE)) | |
1325 | break; | |
1326 | if (j == vcnt) | |
1327 | continue; | |
1328 | mddev->resync_mismatches += r10_bio->sectors; | |
1329 | } | |
18f08819 N |
1330 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
1331 | /* Don't fix anything. */ | |
1332 | continue; | |
1da177e4 LT |
1333 | /* Ok, we need to write this bio |
1334 | * First we need to fixup bv_offset, bv_len and | |
1335 | * bi_vecs, as the read request might have corrupted these | |
1336 | */ | |
1337 | tbio->bi_vcnt = vcnt; | |
1338 | tbio->bi_size = r10_bio->sectors << 9; | |
1339 | tbio->bi_idx = 0; | |
1340 | tbio->bi_phys_segments = 0; | |
1da177e4 LT |
1341 | tbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
1342 | tbio->bi_flags |= 1 << BIO_UPTODATE; | |
1343 | tbio->bi_next = NULL; | |
1344 | tbio->bi_rw = WRITE; | |
1345 | tbio->bi_private = r10_bio; | |
1346 | tbio->bi_sector = r10_bio->devs[i].addr; | |
1347 | ||
1348 | for (j=0; j < vcnt ; j++) { | |
1349 | tbio->bi_io_vec[j].bv_offset = 0; | |
1350 | tbio->bi_io_vec[j].bv_len = PAGE_SIZE; | |
1351 | ||
1352 | memcpy(page_address(tbio->bi_io_vec[j].bv_page), | |
1353 | page_address(fbio->bi_io_vec[j].bv_page), | |
1354 | PAGE_SIZE); | |
1355 | } | |
1356 | tbio->bi_end_io = end_sync_write; | |
1357 | ||
1358 | d = r10_bio->devs[i].devnum; | |
1359 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1360 | atomic_inc(&r10_bio->remaining); | |
1361 | md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9); | |
1362 | ||
1363 | tbio->bi_sector += conf->mirrors[d].rdev->data_offset; | |
1364 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
1365 | generic_make_request(tbio); | |
1366 | } | |
1367 | ||
1368 | done: | |
1369 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
1370 | md_done_sync(mddev, r10_bio->sectors, 1); | |
1371 | put_buf(r10_bio); | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Now for the recovery code. | |
1377 | * Recovery happens across physical sectors. | |
1378 | * We recover all non-is_sync drives by finding the virtual address of | |
1379 | * each, and then choose a working drive that also has that virt address. | |
1380 | * There is a separate r10_bio for each non-in_sync drive. | |
1381 | * Only the first two slots are in use. The first for reading, | |
1382 | * The second for writing. | |
1383 | * | |
1384 | */ | |
1385 | ||
1386 | static void recovery_request_write(mddev_t *mddev, r10bio_t *r10_bio) | |
1387 | { | |
070ec55d | 1388 | conf_t *conf = mddev->private; |
1da177e4 LT |
1389 | int i, d; |
1390 | struct bio *bio, *wbio; | |
1391 | ||
1392 | ||
1393 | /* move the pages across to the second bio | |
1394 | * and submit the write request | |
1395 | */ | |
1396 | bio = r10_bio->devs[0].bio; | |
1397 | wbio = r10_bio->devs[1].bio; | |
1398 | for (i=0; i < wbio->bi_vcnt; i++) { | |
1399 | struct page *p = bio->bi_io_vec[i].bv_page; | |
1400 | bio->bi_io_vec[i].bv_page = wbio->bi_io_vec[i].bv_page; | |
1401 | wbio->bi_io_vec[i].bv_page = p; | |
1402 | } | |
1403 | d = r10_bio->devs[1].devnum; | |
1404 | ||
1405 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1406 | md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9); | |
0eb3ff12 N |
1407 | if (test_bit(R10BIO_Uptodate, &r10_bio->state)) |
1408 | generic_make_request(wbio); | |
1409 | else | |
6712ecf8 | 1410 | bio_endio(wbio, -EIO); |
1da177e4 LT |
1411 | } |
1412 | ||
1413 | ||
1e50915f RB |
1414 | /* |
1415 | * Used by fix_read_error() to decay the per rdev read_errors. | |
1416 | * We halve the read error count for every hour that has elapsed | |
1417 | * since the last recorded read error. | |
1418 | * | |
1419 | */ | |
1420 | static void check_decay_read_errors(mddev_t *mddev, mdk_rdev_t *rdev) | |
1421 | { | |
1422 | struct timespec cur_time_mon; | |
1423 | unsigned long hours_since_last; | |
1424 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
1425 | ||
1426 | ktime_get_ts(&cur_time_mon); | |
1427 | ||
1428 | if (rdev->last_read_error.tv_sec == 0 && | |
1429 | rdev->last_read_error.tv_nsec == 0) { | |
1430 | /* first time we've seen a read error */ | |
1431 | rdev->last_read_error = cur_time_mon; | |
1432 | return; | |
1433 | } | |
1434 | ||
1435 | hours_since_last = (cur_time_mon.tv_sec - | |
1436 | rdev->last_read_error.tv_sec) / 3600; | |
1437 | ||
1438 | rdev->last_read_error = cur_time_mon; | |
1439 | ||
1440 | /* | |
1441 | * if hours_since_last is > the number of bits in read_errors | |
1442 | * just set read errors to 0. We do this to avoid | |
1443 | * overflowing the shift of read_errors by hours_since_last. | |
1444 | */ | |
1445 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
1446 | atomic_set(&rdev->read_errors, 0); | |
1447 | else | |
1448 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
1449 | } | |
1450 | ||
1da177e4 LT |
1451 | /* |
1452 | * This is a kernel thread which: | |
1453 | * | |
1454 | * 1. Retries failed read operations on working mirrors. | |
1455 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 1456 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
1457 | */ |
1458 | ||
6814d536 N |
1459 | static void fix_read_error(conf_t *conf, mddev_t *mddev, r10bio_t *r10_bio) |
1460 | { | |
1461 | int sect = 0; /* Offset from r10_bio->sector */ | |
1462 | int sectors = r10_bio->sectors; | |
1463 | mdk_rdev_t*rdev; | |
1e50915f | 1464 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 1465 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f RB |
1466 | |
1467 | rcu_read_lock(); | |
0544a21d PP |
1468 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
1469 | if (rdev) { /* If rdev is not NULL */ | |
1e50915f RB |
1470 | char b[BDEVNAME_SIZE]; |
1471 | int cur_read_error_count = 0; | |
1472 | ||
1e50915f RB |
1473 | bdevname(rdev->bdev, b); |
1474 | ||
1475 | if (test_bit(Faulty, &rdev->flags)) { | |
1476 | rcu_read_unlock(); | |
1477 | /* drive has already been failed, just ignore any | |
1478 | more fix_read_error() attempts */ | |
1479 | return; | |
1480 | } | |
1481 | ||
1482 | check_decay_read_errors(mddev, rdev); | |
1483 | atomic_inc(&rdev->read_errors); | |
1484 | cur_read_error_count = atomic_read(&rdev->read_errors); | |
1485 | if (cur_read_error_count > max_read_errors) { | |
1486 | rcu_read_unlock(); | |
1487 | printk(KERN_NOTICE | |
128595ed | 1488 | "md/raid10:%s: %s: Raid device exceeded " |
1e50915f RB |
1489 | "read_error threshold " |
1490 | "[cur %d:max %d]\n", | |
128595ed | 1491 | mdname(mddev), |
1e50915f RB |
1492 | b, cur_read_error_count, max_read_errors); |
1493 | printk(KERN_NOTICE | |
128595ed N |
1494 | "md/raid10:%s: %s: Failing raid " |
1495 | "device\n", mdname(mddev), b); | |
1e50915f RB |
1496 | md_error(mddev, conf->mirrors[d].rdev); |
1497 | return; | |
1498 | } | |
1499 | } | |
1500 | rcu_read_unlock(); | |
1501 | ||
6814d536 N |
1502 | while(sectors) { |
1503 | int s = sectors; | |
1504 | int sl = r10_bio->read_slot; | |
1505 | int success = 0; | |
1506 | int start; | |
1507 | ||
1508 | if (s > (PAGE_SIZE>>9)) | |
1509 | s = PAGE_SIZE >> 9; | |
1510 | ||
1511 | rcu_read_lock(); | |
1512 | do { | |
0544a21d | 1513 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
1514 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
1515 | if (rdev && | |
1516 | test_bit(In_sync, &rdev->flags)) { | |
1517 | atomic_inc(&rdev->nr_pending); | |
1518 | rcu_read_unlock(); | |
2b193363 | 1519 | success = sync_page_io(rdev, |
6814d536 | 1520 | r10_bio->devs[sl].addr + |
ccebd4c4 | 1521 | sect, |
6814d536 | 1522 | s<<9, |
ccebd4c4 | 1523 | conf->tmppage, READ, false); |
6814d536 N |
1524 | rdev_dec_pending(rdev, mddev); |
1525 | rcu_read_lock(); | |
1526 | if (success) | |
1527 | break; | |
1528 | } | |
1529 | sl++; | |
1530 | if (sl == conf->copies) | |
1531 | sl = 0; | |
1532 | } while (!success && sl != r10_bio->read_slot); | |
1533 | rcu_read_unlock(); | |
1534 | ||
1535 | if (!success) { | |
1536 | /* Cannot read from anywhere -- bye bye array */ | |
1537 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; | |
1538 | md_error(mddev, conf->mirrors[dn].rdev); | |
1539 | break; | |
1540 | } | |
1541 | ||
1542 | start = sl; | |
1543 | /* write it back and re-read */ | |
1544 | rcu_read_lock(); | |
1545 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 1546 | char b[BDEVNAME_SIZE]; |
0544a21d | 1547 | |
6814d536 N |
1548 | if (sl==0) |
1549 | sl = conf->copies; | |
1550 | sl--; | |
1551 | d = r10_bio->devs[sl].devnum; | |
1552 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1553 | if (rdev && | |
1554 | test_bit(In_sync, &rdev->flags)) { | |
1555 | atomic_inc(&rdev->nr_pending); | |
1556 | rcu_read_unlock(); | |
1557 | atomic_add(s, &rdev->corrected_errors); | |
2b193363 | 1558 | if (sync_page_io(rdev, |
6814d536 | 1559 | r10_bio->devs[sl].addr + |
ccebd4c4 JB |
1560 | sect, |
1561 | s<<9, conf->tmppage, WRITE, false) | |
67b8dc4b | 1562 | == 0) { |
6814d536 | 1563 | /* Well, this device is dead */ |
67b8dc4b | 1564 | printk(KERN_NOTICE |
128595ed | 1565 | "md/raid10:%s: read correction " |
67b8dc4b RB |
1566 | "write failed" |
1567 | " (%d sectors at %llu on %s)\n", | |
1568 | mdname(mddev), s, | |
1569 | (unsigned long long)(sect+ | |
1570 | rdev->data_offset), | |
1571 | bdevname(rdev->bdev, b)); | |
128595ed | 1572 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " |
67b8dc4b | 1573 | "drive\n", |
128595ed | 1574 | mdname(mddev), |
67b8dc4b | 1575 | bdevname(rdev->bdev, b)); |
6814d536 | 1576 | md_error(mddev, rdev); |
67b8dc4b | 1577 | } |
6814d536 N |
1578 | rdev_dec_pending(rdev, mddev); |
1579 | rcu_read_lock(); | |
1580 | } | |
1581 | } | |
1582 | sl = start; | |
1583 | while (sl != r10_bio->read_slot) { | |
0544a21d | 1584 | |
6814d536 N |
1585 | if (sl==0) |
1586 | sl = conf->copies; | |
1587 | sl--; | |
1588 | d = r10_bio->devs[sl].devnum; | |
1589 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1590 | if (rdev && | |
1591 | test_bit(In_sync, &rdev->flags)) { | |
1592 | char b[BDEVNAME_SIZE]; | |
1593 | atomic_inc(&rdev->nr_pending); | |
1594 | rcu_read_unlock(); | |
2b193363 | 1595 | if (sync_page_io(rdev, |
6814d536 | 1596 | r10_bio->devs[sl].addr + |
ccebd4c4 | 1597 | sect, |
67b8dc4b | 1598 | s<<9, conf->tmppage, |
ccebd4c4 | 1599 | READ, false) == 0) { |
6814d536 | 1600 | /* Well, this device is dead */ |
67b8dc4b | 1601 | printk(KERN_NOTICE |
128595ed | 1602 | "md/raid10:%s: unable to read back " |
67b8dc4b RB |
1603 | "corrected sectors" |
1604 | " (%d sectors at %llu on %s)\n", | |
1605 | mdname(mddev), s, | |
1606 | (unsigned long long)(sect+ | |
1607 | rdev->data_offset), | |
1608 | bdevname(rdev->bdev, b)); | |
128595ed N |
1609 | printk(KERN_NOTICE "md/raid10:%s: %s: failing drive\n", |
1610 | mdname(mddev), | |
67b8dc4b RB |
1611 | bdevname(rdev->bdev, b)); |
1612 | ||
6814d536 | 1613 | md_error(mddev, rdev); |
67b8dc4b | 1614 | } else { |
6814d536 | 1615 | printk(KERN_INFO |
128595ed | 1616 | "md/raid10:%s: read error corrected" |
6814d536 N |
1617 | " (%d sectors at %llu on %s)\n", |
1618 | mdname(mddev), s, | |
969b755a RD |
1619 | (unsigned long long)(sect+ |
1620 | rdev->data_offset), | |
6814d536 | 1621 | bdevname(rdev->bdev, b)); |
67b8dc4b | 1622 | } |
6814d536 N |
1623 | |
1624 | rdev_dec_pending(rdev, mddev); | |
1625 | rcu_read_lock(); | |
1626 | } | |
1627 | } | |
1628 | rcu_read_unlock(); | |
1629 | ||
1630 | sectors -= s; | |
1631 | sect += s; | |
1632 | } | |
1633 | } | |
1634 | ||
1da177e4 LT |
1635 | static void raid10d(mddev_t *mddev) |
1636 | { | |
1637 | r10bio_t *r10_bio; | |
1638 | struct bio *bio; | |
1639 | unsigned long flags; | |
070ec55d | 1640 | conf_t *conf = mddev->private; |
1da177e4 | 1641 | struct list_head *head = &conf->retry_list; |
1da177e4 | 1642 | mdk_rdev_t *rdev; |
e1dfa0a2 | 1643 | struct blk_plug plug; |
1da177e4 LT |
1644 | |
1645 | md_check_recovery(mddev); | |
1da177e4 | 1646 | |
e1dfa0a2 | 1647 | blk_start_plug(&plug); |
1da177e4 LT |
1648 | for (;;) { |
1649 | char b[BDEVNAME_SIZE]; | |
6cce3b23 | 1650 | |
7eaceacc | 1651 | flush_pending_writes(conf); |
6cce3b23 | 1652 | |
a35e63ef N |
1653 | spin_lock_irqsave(&conf->device_lock, flags); |
1654 | if (list_empty(head)) { | |
1655 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 1656 | break; |
a35e63ef | 1657 | } |
1da177e4 LT |
1658 | r10_bio = list_entry(head->prev, r10bio_t, retry_list); |
1659 | list_del(head->prev); | |
4443ae10 | 1660 | conf->nr_queued--; |
1da177e4 LT |
1661 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1662 | ||
1663 | mddev = r10_bio->mddev; | |
070ec55d | 1664 | conf = mddev->private; |
7eaceacc | 1665 | if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 1666 | sync_request_write(mddev, r10_bio); |
7eaceacc | 1667 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 1668 | recovery_request_write(mddev, r10_bio); |
7eaceacc | 1669 | else { |
1da177e4 | 1670 | int mirror; |
4443ae10 N |
1671 | /* we got a read error. Maybe the drive is bad. Maybe just |
1672 | * the block and we can fix it. | |
1673 | * We freeze all other IO, and try reading the block from | |
1674 | * other devices. When we find one, we re-write | |
1675 | * and check it that fixes the read error. | |
1676 | * This is all done synchronously while the array is | |
1677 | * frozen. | |
1678 | */ | |
6814d536 N |
1679 | if (mddev->ro == 0) { |
1680 | freeze_array(conf); | |
1681 | fix_read_error(conf, mddev, r10_bio); | |
1682 | unfreeze_array(conf); | |
4443ae10 N |
1683 | } |
1684 | ||
1da177e4 | 1685 | bio = r10_bio->devs[r10_bio->read_slot].bio; |
0eb3ff12 N |
1686 | r10_bio->devs[r10_bio->read_slot].bio = |
1687 | mddev->ro ? IO_BLOCKED : NULL; | |
1da177e4 LT |
1688 | mirror = read_balance(conf, r10_bio); |
1689 | if (mirror == -1) { | |
128595ed | 1690 | printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O" |
1da177e4 | 1691 | " read error for block %llu\n", |
128595ed | 1692 | mdname(mddev), |
1da177e4 LT |
1693 | bdevname(bio->bi_bdev,b), |
1694 | (unsigned long long)r10_bio->sector); | |
1695 | raid_end_bio_io(r10_bio); | |
14e71344 | 1696 | bio_put(bio); |
1da177e4 | 1697 | } else { |
2c7d46ec | 1698 | const unsigned long do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC); |
14e71344 | 1699 | bio_put(bio); |
1da177e4 LT |
1700 | rdev = conf->mirrors[mirror].rdev; |
1701 | if (printk_ratelimit()) | |
128595ed | 1702 | printk(KERN_ERR "md/raid10:%s: %s: redirecting sector %llu to" |
1da177e4 | 1703 | " another mirror\n", |
128595ed | 1704 | mdname(mddev), |
1da177e4 LT |
1705 | bdevname(rdev->bdev,b), |
1706 | (unsigned long long)r10_bio->sector); | |
a167f663 N |
1707 | bio = bio_clone_mddev(r10_bio->master_bio, |
1708 | GFP_NOIO, mddev); | |
1da177e4 LT |
1709 | r10_bio->devs[r10_bio->read_slot].bio = bio; |
1710 | bio->bi_sector = r10_bio->devs[r10_bio->read_slot].addr | |
1711 | + rdev->data_offset; | |
1712 | bio->bi_bdev = rdev->bdev; | |
7b6d91da | 1713 | bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1714 | bio->bi_private = r10_bio; |
1715 | bio->bi_end_io = raid10_end_read_request; | |
1da177e4 LT |
1716 | generic_make_request(bio); |
1717 | } | |
1718 | } | |
1d9d5241 | 1719 | cond_resched(); |
1da177e4 | 1720 | } |
e1dfa0a2 | 1721 | blk_finish_plug(&plug); |
1da177e4 LT |
1722 | } |
1723 | ||
1724 | ||
1725 | static int init_resync(conf_t *conf) | |
1726 | { | |
1727 | int buffs; | |
1728 | ||
1729 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 1730 | BUG_ON(conf->r10buf_pool); |
1da177e4 LT |
1731 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
1732 | if (!conf->r10buf_pool) | |
1733 | return -ENOMEM; | |
1734 | conf->next_resync = 0; | |
1735 | return 0; | |
1736 | } | |
1737 | ||
1738 | /* | |
1739 | * perform a "sync" on one "block" | |
1740 | * | |
1741 | * We need to make sure that no normal I/O request - particularly write | |
1742 | * requests - conflict with active sync requests. | |
1743 | * | |
1744 | * This is achieved by tracking pending requests and a 'barrier' concept | |
1745 | * that can be installed to exclude normal IO requests. | |
1746 | * | |
1747 | * Resync and recovery are handled very differently. | |
1748 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
1749 | * | |
1750 | * For resync, we iterate over virtual addresses, read all copies, | |
1751 | * and update if there are differences. If only one copy is live, | |
1752 | * skip it. | |
1753 | * For recovery, we iterate over physical addresses, read a good | |
1754 | * value for each non-in_sync drive, and over-write. | |
1755 | * | |
1756 | * So, for recovery we may have several outstanding complex requests for a | |
1757 | * given address, one for each out-of-sync device. We model this by allocating | |
1758 | * a number of r10_bio structures, one for each out-of-sync device. | |
1759 | * As we setup these structures, we collect all bio's together into a list | |
1760 | * which we then process collectively to add pages, and then process again | |
1761 | * to pass to generic_make_request. | |
1762 | * | |
1763 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
1764 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
1765 | * has its remaining count decremented to 0, the whole complex operation | |
1766 | * is complete. | |
1767 | * | |
1768 | */ | |
1769 | ||
57afd89f | 1770 | static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 | 1771 | { |
070ec55d | 1772 | conf_t *conf = mddev->private; |
1da177e4 LT |
1773 | r10bio_t *r10_bio; |
1774 | struct bio *biolist = NULL, *bio; | |
1775 | sector_t max_sector, nr_sectors; | |
1776 | int disk; | |
1777 | int i; | |
6cce3b23 | 1778 | int max_sync; |
57dab0bd | 1779 | sector_t sync_blocks; |
1da177e4 LT |
1780 | |
1781 | sector_t sectors_skipped = 0; | |
1782 | int chunks_skipped = 0; | |
1783 | ||
1784 | if (!conf->r10buf_pool) | |
1785 | if (init_resync(conf)) | |
57afd89f | 1786 | return 0; |
1da177e4 LT |
1787 | |
1788 | skipped: | |
58c0fed4 | 1789 | max_sector = mddev->dev_sectors; |
1da177e4 LT |
1790 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
1791 | max_sector = mddev->resync_max_sectors; | |
1792 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
1793 | /* If we aborted, we need to abort the |
1794 | * sync on the 'current' bitmap chucks (there can | |
1795 | * be several when recovering multiple devices). | |
1796 | * as we may have started syncing it but not finished. | |
1797 | * We can find the current address in | |
1798 | * mddev->curr_resync, but for recovery, | |
1799 | * we need to convert that to several | |
1800 | * virtual addresses. | |
1801 | */ | |
1802 | if (mddev->curr_resync < max_sector) { /* aborted */ | |
1803 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
1804 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
1805 | &sync_blocks, 1); | |
1806 | else for (i=0; i<conf->raid_disks; i++) { | |
1807 | sector_t sect = | |
1808 | raid10_find_virt(conf, mddev->curr_resync, i); | |
1809 | bitmap_end_sync(mddev->bitmap, sect, | |
1810 | &sync_blocks, 1); | |
1811 | } | |
1812 | } else /* completed sync */ | |
1813 | conf->fullsync = 0; | |
1814 | ||
1815 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 1816 | close_sync(conf); |
57afd89f | 1817 | *skipped = 1; |
1da177e4 LT |
1818 | return sectors_skipped; |
1819 | } | |
1820 | if (chunks_skipped >= conf->raid_disks) { | |
1821 | /* if there has been nothing to do on any drive, | |
1822 | * then there is nothing to do at all.. | |
1823 | */ | |
57afd89f N |
1824 | *skipped = 1; |
1825 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
1826 | } |
1827 | ||
c6207277 N |
1828 | if (max_sector > mddev->resync_max) |
1829 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
1830 | ||
1da177e4 LT |
1831 | /* make sure whole request will fit in a chunk - if chunks |
1832 | * are meaningful | |
1833 | */ | |
1834 | if (conf->near_copies < conf->raid_disks && | |
1835 | max_sector > (sector_nr | conf->chunk_mask)) | |
1836 | max_sector = (sector_nr | conf->chunk_mask) + 1; | |
1837 | /* | |
1838 | * If there is non-resync activity waiting for us then | |
1839 | * put in a delay to throttle resync. | |
1840 | */ | |
0a27ec96 | 1841 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 1842 | msleep_interruptible(1000); |
1da177e4 LT |
1843 | |
1844 | /* Again, very different code for resync and recovery. | |
1845 | * Both must result in an r10bio with a list of bios that | |
1846 | * have bi_end_io, bi_sector, bi_bdev set, | |
1847 | * and bi_private set to the r10bio. | |
1848 | * For recovery, we may actually create several r10bios | |
1849 | * with 2 bios in each, that correspond to the bios in the main one. | |
1850 | * In this case, the subordinate r10bios link back through a | |
1851 | * borrowed master_bio pointer, and the counter in the master | |
1852 | * includes a ref from each subordinate. | |
1853 | */ | |
1854 | /* First, we decide what to do and set ->bi_end_io | |
1855 | * To end_sync_read if we want to read, and | |
1856 | * end_sync_write if we will want to write. | |
1857 | */ | |
1858 | ||
6cce3b23 | 1859 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
1860 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1861 | /* recovery... the complicated one */ | |
a9f326eb | 1862 | int j, k; |
1da177e4 LT |
1863 | r10_bio = NULL; |
1864 | ||
1865 | for (i=0 ; i<conf->raid_disks; i++) | |
1866 | if (conf->mirrors[i].rdev && | |
b2d444d7 | 1867 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags)) { |
6cce3b23 | 1868 | int still_degraded = 0; |
1da177e4 LT |
1869 | /* want to reconstruct this device */ |
1870 | r10bio_t *rb2 = r10_bio; | |
6cce3b23 N |
1871 | sector_t sect = raid10_find_virt(conf, sector_nr, i); |
1872 | int must_sync; | |
1873 | /* Unless we are doing a full sync, we only need | |
1874 | * to recover the block if it is set in the bitmap | |
1875 | */ | |
1876 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
1877 | &sync_blocks, 1); | |
1878 | if (sync_blocks < max_sync) | |
1879 | max_sync = sync_blocks; | |
1880 | if (!must_sync && | |
1881 | !conf->fullsync) { | |
1882 | /* yep, skip the sync_blocks here, but don't assume | |
1883 | * that there will never be anything to do here | |
1884 | */ | |
1885 | chunks_skipped = -1; | |
1886 | continue; | |
1887 | } | |
1da177e4 LT |
1888 | |
1889 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); | |
6cce3b23 | 1890 | raise_barrier(conf, rb2 != NULL); |
1da177e4 LT |
1891 | atomic_set(&r10_bio->remaining, 0); |
1892 | ||
1893 | r10_bio->master_bio = (struct bio*)rb2; | |
1894 | if (rb2) | |
1895 | atomic_inc(&rb2->remaining); | |
1896 | r10_bio->mddev = mddev; | |
1897 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
6cce3b23 N |
1898 | r10_bio->sector = sect; |
1899 | ||
1da177e4 | 1900 | raid10_find_phys(conf, r10_bio); |
18055569 N |
1901 | |
1902 | /* Need to check if the array will still be | |
6cce3b23 N |
1903 | * degraded |
1904 | */ | |
18055569 N |
1905 | for (j=0; j<conf->raid_disks; j++) |
1906 | if (conf->mirrors[j].rdev == NULL || | |
1907 | test_bit(Faulty, &conf->mirrors[j].rdev->flags)) { | |
6cce3b23 | 1908 | still_degraded = 1; |
a24a8dd8 N |
1909 | break; |
1910 | } | |
18055569 | 1911 | |
6cce3b23 N |
1912 | must_sync = bitmap_start_sync(mddev->bitmap, sect, |
1913 | &sync_blocks, still_degraded); | |
1914 | ||
1da177e4 LT |
1915 | for (j=0; j<conf->copies;j++) { |
1916 | int d = r10_bio->devs[j].devnum; | |
1917 | if (conf->mirrors[d].rdev && | |
b2d444d7 | 1918 | test_bit(In_sync, &conf->mirrors[d].rdev->flags)) { |
1da177e4 LT |
1919 | /* This is where we read from */ |
1920 | bio = r10_bio->devs[0].bio; | |
1921 | bio->bi_next = biolist; | |
1922 | biolist = bio; | |
1923 | bio->bi_private = r10_bio; | |
1924 | bio->bi_end_io = end_sync_read; | |
802ba064 | 1925 | bio->bi_rw = READ; |
1da177e4 LT |
1926 | bio->bi_sector = r10_bio->devs[j].addr + |
1927 | conf->mirrors[d].rdev->data_offset; | |
1928 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
1929 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1930 | atomic_inc(&r10_bio->remaining); | |
1931 | /* and we write to 'i' */ | |
1932 | ||
1933 | for (k=0; k<conf->copies; k++) | |
1934 | if (r10_bio->devs[k].devnum == i) | |
1935 | break; | |
64a742bc | 1936 | BUG_ON(k == conf->copies); |
1da177e4 LT |
1937 | bio = r10_bio->devs[1].bio; |
1938 | bio->bi_next = biolist; | |
1939 | biolist = bio; | |
1940 | bio->bi_private = r10_bio; | |
1941 | bio->bi_end_io = end_sync_write; | |
802ba064 | 1942 | bio->bi_rw = WRITE; |
1da177e4 LT |
1943 | bio->bi_sector = r10_bio->devs[k].addr + |
1944 | conf->mirrors[i].rdev->data_offset; | |
1945 | bio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1946 | ||
1947 | r10_bio->devs[0].devnum = d; | |
1948 | r10_bio->devs[1].devnum = i; | |
1949 | ||
1950 | break; | |
1951 | } | |
1952 | } | |
1953 | if (j == conf->copies) { | |
87fc767b N |
1954 | /* Cannot recover, so abort the recovery */ |
1955 | put_buf(r10_bio); | |
a07e6ab4 T |
1956 | if (rb2) |
1957 | atomic_dec(&rb2->remaining); | |
87fc767b | 1958 | r10_bio = rb2; |
dfc70645 N |
1959 | if (!test_and_set_bit(MD_RECOVERY_INTR, |
1960 | &mddev->recovery)) | |
128595ed N |
1961 | printk(KERN_INFO "md/raid10:%s: insufficient " |
1962 | "working devices for recovery.\n", | |
87fc767b N |
1963 | mdname(mddev)); |
1964 | break; | |
1da177e4 LT |
1965 | } |
1966 | } | |
1967 | if (biolist == NULL) { | |
1968 | while (r10_bio) { | |
1969 | r10bio_t *rb2 = r10_bio; | |
1970 | r10_bio = (r10bio_t*) rb2->master_bio; | |
1971 | rb2->master_bio = NULL; | |
1972 | put_buf(rb2); | |
1973 | } | |
1974 | goto giveup; | |
1975 | } | |
1976 | } else { | |
1977 | /* resync. Schedule a read for every block at this virt offset */ | |
1978 | int count = 0; | |
6cce3b23 | 1979 | |
78200d45 N |
1980 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
1981 | ||
6cce3b23 N |
1982 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
1983 | &sync_blocks, mddev->degraded) && | |
1984 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
1985 | /* We can skip this block */ | |
1986 | *skipped = 1; | |
1987 | return sync_blocks + sectors_skipped; | |
1988 | } | |
1989 | if (sync_blocks < max_sync) | |
1990 | max_sync = sync_blocks; | |
1da177e4 LT |
1991 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
1992 | ||
1da177e4 LT |
1993 | r10_bio->mddev = mddev; |
1994 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
1995 | raise_barrier(conf, 0); |
1996 | conf->next_resync = sector_nr; | |
1da177e4 LT |
1997 | |
1998 | r10_bio->master_bio = NULL; | |
1999 | r10_bio->sector = sector_nr; | |
2000 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
2001 | raid10_find_phys(conf, r10_bio); | |
2002 | r10_bio->sectors = (sector_nr | conf->chunk_mask) - sector_nr +1; | |
2003 | ||
2004 | for (i=0; i<conf->copies; i++) { | |
2005 | int d = r10_bio->devs[i].devnum; | |
2006 | bio = r10_bio->devs[i].bio; | |
2007 | bio->bi_end_io = NULL; | |
af03b8e4 | 2008 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
1da177e4 | 2009 | if (conf->mirrors[d].rdev == NULL || |
b2d444d7 | 2010 | test_bit(Faulty, &conf->mirrors[d].rdev->flags)) |
1da177e4 LT |
2011 | continue; |
2012 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
2013 | atomic_inc(&r10_bio->remaining); | |
2014 | bio->bi_next = biolist; | |
2015 | biolist = bio; | |
2016 | bio->bi_private = r10_bio; | |
2017 | bio->bi_end_io = end_sync_read; | |
802ba064 | 2018 | bio->bi_rw = READ; |
1da177e4 LT |
2019 | bio->bi_sector = r10_bio->devs[i].addr + |
2020 | conf->mirrors[d].rdev->data_offset; | |
2021 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
2022 | count++; | |
2023 | } | |
2024 | ||
2025 | if (count < 2) { | |
2026 | for (i=0; i<conf->copies; i++) { | |
2027 | int d = r10_bio->devs[i].devnum; | |
2028 | if (r10_bio->devs[i].bio->bi_end_io) | |
2029 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
2030 | } | |
2031 | put_buf(r10_bio); | |
2032 | biolist = NULL; | |
2033 | goto giveup; | |
2034 | } | |
2035 | } | |
2036 | ||
2037 | for (bio = biolist; bio ; bio=bio->bi_next) { | |
2038 | ||
2039 | bio->bi_flags &= ~(BIO_POOL_MASK - 1); | |
2040 | if (bio->bi_end_io) | |
2041 | bio->bi_flags |= 1 << BIO_UPTODATE; | |
2042 | bio->bi_vcnt = 0; | |
2043 | bio->bi_idx = 0; | |
2044 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
2045 | bio->bi_size = 0; |
2046 | } | |
2047 | ||
2048 | nr_sectors = 0; | |
6cce3b23 N |
2049 | if (sector_nr + max_sync < max_sector) |
2050 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
2051 | do { |
2052 | struct page *page; | |
2053 | int len = PAGE_SIZE; | |
2054 | disk = 0; | |
2055 | if (sector_nr + (len>>9) > max_sector) | |
2056 | len = (max_sector - sector_nr) << 9; | |
2057 | if (len == 0) | |
2058 | break; | |
2059 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
2060 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; | |
2061 | if (bio_add_page(bio, page, len, 0) == 0) { | |
2062 | /* stop here */ | |
2063 | struct bio *bio2; | |
2064 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
2065 | for (bio2 = biolist; bio2 && bio2 != bio; bio2 = bio2->bi_next) { | |
2066 | /* remove last page from this bio */ | |
2067 | bio2->bi_vcnt--; | |
2068 | bio2->bi_size -= len; | |
2069 | bio2->bi_flags &= ~(1<< BIO_SEG_VALID); | |
2070 | } | |
2071 | goto bio_full; | |
2072 | } | |
2073 | disk = i; | |
2074 | } | |
2075 | nr_sectors += len>>9; | |
2076 | sector_nr += len>>9; | |
2077 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
2078 | bio_full: | |
2079 | r10_bio->sectors = nr_sectors; | |
2080 | ||
2081 | while (biolist) { | |
2082 | bio = biolist; | |
2083 | biolist = biolist->bi_next; | |
2084 | ||
2085 | bio->bi_next = NULL; | |
2086 | r10_bio = bio->bi_private; | |
2087 | r10_bio->sectors = nr_sectors; | |
2088 | ||
2089 | if (bio->bi_end_io == end_sync_read) { | |
2090 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
2091 | generic_make_request(bio); | |
2092 | } | |
2093 | } | |
2094 | ||
57afd89f N |
2095 | if (sectors_skipped) |
2096 | /* pretend they weren't skipped, it makes | |
2097 | * no important difference in this case | |
2098 | */ | |
2099 | md_done_sync(mddev, sectors_skipped, 1); | |
2100 | ||
1da177e4 LT |
2101 | return sectors_skipped + nr_sectors; |
2102 | giveup: | |
2103 | /* There is nowhere to write, so all non-sync | |
2104 | * drives must be failed, so try the next chunk... | |
2105 | */ | |
09b4068a N |
2106 | if (sector_nr + max_sync < max_sector) |
2107 | max_sector = sector_nr + max_sync; | |
2108 | ||
2109 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
2110 | chunks_skipped ++; |
2111 | sector_nr = max_sector; | |
1da177e4 | 2112 | goto skipped; |
1da177e4 LT |
2113 | } |
2114 | ||
80c3a6ce DW |
2115 | static sector_t |
2116 | raid10_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
2117 | { | |
2118 | sector_t size; | |
070ec55d | 2119 | conf_t *conf = mddev->private; |
80c3a6ce DW |
2120 | |
2121 | if (!raid_disks) | |
84707f38 | 2122 | raid_disks = conf->raid_disks; |
80c3a6ce | 2123 | if (!sectors) |
dab8b292 | 2124 | sectors = conf->dev_sectors; |
80c3a6ce DW |
2125 | |
2126 | size = sectors >> conf->chunk_shift; | |
2127 | sector_div(size, conf->far_copies); | |
2128 | size = size * raid_disks; | |
2129 | sector_div(size, conf->near_copies); | |
2130 | ||
2131 | return size << conf->chunk_shift; | |
2132 | } | |
2133 | ||
dab8b292 TM |
2134 | |
2135 | static conf_t *setup_conf(mddev_t *mddev) | |
1da177e4 | 2136 | { |
dab8b292 | 2137 | conf_t *conf = NULL; |
c93983bf | 2138 | int nc, fc, fo; |
1da177e4 | 2139 | sector_t stride, size; |
dab8b292 | 2140 | int err = -EINVAL; |
1da177e4 | 2141 | |
f73ea873 MT |
2142 | if (mddev->new_chunk_sectors < (PAGE_SIZE >> 9) || |
2143 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
128595ed N |
2144 | printk(KERN_ERR "md/raid10:%s: chunk size must be " |
2145 | "at least PAGE_SIZE(%ld) and be a power of 2.\n", | |
2146 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 2147 | goto out; |
1da177e4 | 2148 | } |
2604b703 | 2149 | |
f73ea873 MT |
2150 | nc = mddev->new_layout & 255; |
2151 | fc = (mddev->new_layout >> 8) & 255; | |
2152 | fo = mddev->new_layout & (1<<16); | |
dab8b292 | 2153 | |
1da177e4 | 2154 | if ((nc*fc) <2 || (nc*fc) > mddev->raid_disks || |
f73ea873 | 2155 | (mddev->new_layout >> 17)) { |
128595ed | 2156 | printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
f73ea873 | 2157 | mdname(mddev), mddev->new_layout); |
1da177e4 LT |
2158 | goto out; |
2159 | } | |
dab8b292 TM |
2160 | |
2161 | err = -ENOMEM; | |
4443ae10 | 2162 | conf = kzalloc(sizeof(conf_t), GFP_KERNEL); |
dab8b292 | 2163 | if (!conf) |
1da177e4 | 2164 | goto out; |
dab8b292 | 2165 | |
4443ae10 | 2166 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
dab8b292 TM |
2167 | GFP_KERNEL); |
2168 | if (!conf->mirrors) | |
2169 | goto out; | |
4443ae10 N |
2170 | |
2171 | conf->tmppage = alloc_page(GFP_KERNEL); | |
2172 | if (!conf->tmppage) | |
dab8b292 TM |
2173 | goto out; |
2174 | ||
1da177e4 | 2175 | |
64a742bc | 2176 | conf->raid_disks = mddev->raid_disks; |
1da177e4 LT |
2177 | conf->near_copies = nc; |
2178 | conf->far_copies = fc; | |
2179 | conf->copies = nc*fc; | |
c93983bf | 2180 | conf->far_offset = fo; |
dab8b292 TM |
2181 | conf->chunk_mask = mddev->new_chunk_sectors - 1; |
2182 | conf->chunk_shift = ffz(~mddev->new_chunk_sectors); | |
2183 | ||
2184 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, | |
2185 | r10bio_pool_free, conf); | |
2186 | if (!conf->r10bio_pool) | |
2187 | goto out; | |
2188 | ||
58c0fed4 | 2189 | size = mddev->dev_sectors >> conf->chunk_shift; |
64a742bc N |
2190 | sector_div(size, fc); |
2191 | size = size * conf->raid_disks; | |
2192 | sector_div(size, nc); | |
2193 | /* 'size' is now the number of chunks in the array */ | |
2194 | /* calculate "used chunks per device" in 'stride' */ | |
2195 | stride = size * conf->copies; | |
af03b8e4 N |
2196 | |
2197 | /* We need to round up when dividing by raid_disks to | |
2198 | * get the stride size. | |
2199 | */ | |
2200 | stride += conf->raid_disks - 1; | |
64a742bc | 2201 | sector_div(stride, conf->raid_disks); |
dab8b292 TM |
2202 | |
2203 | conf->dev_sectors = stride << conf->chunk_shift; | |
64a742bc | 2204 | |
c93983bf | 2205 | if (fo) |
64a742bc N |
2206 | stride = 1; |
2207 | else | |
c93983bf | 2208 | sector_div(stride, fc); |
64a742bc N |
2209 | conf->stride = stride << conf->chunk_shift; |
2210 | ||
1da177e4 | 2211 | |
e7e72bf6 | 2212 | spin_lock_init(&conf->device_lock); |
dab8b292 TM |
2213 | INIT_LIST_HEAD(&conf->retry_list); |
2214 | ||
2215 | spin_lock_init(&conf->resync_lock); | |
2216 | init_waitqueue_head(&conf->wait_barrier); | |
2217 | ||
2218 | conf->thread = md_register_thread(raid10d, mddev, NULL); | |
2219 | if (!conf->thread) | |
2220 | goto out; | |
2221 | ||
dab8b292 TM |
2222 | conf->mddev = mddev; |
2223 | return conf; | |
2224 | ||
2225 | out: | |
128595ed | 2226 | printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n", |
dab8b292 TM |
2227 | mdname(mddev)); |
2228 | if (conf) { | |
2229 | if (conf->r10bio_pool) | |
2230 | mempool_destroy(conf->r10bio_pool); | |
2231 | kfree(conf->mirrors); | |
2232 | safe_put_page(conf->tmppage); | |
2233 | kfree(conf); | |
2234 | } | |
2235 | return ERR_PTR(err); | |
2236 | } | |
2237 | ||
2238 | static int run(mddev_t *mddev) | |
2239 | { | |
2240 | conf_t *conf; | |
2241 | int i, disk_idx, chunk_size; | |
2242 | mirror_info_t *disk; | |
2243 | mdk_rdev_t *rdev; | |
2244 | sector_t size; | |
2245 | ||
2246 | /* | |
2247 | * copy the already verified devices into our private RAID10 | |
2248 | * bookkeeping area. [whatever we allocate in run(), | |
2249 | * should be freed in stop()] | |
2250 | */ | |
2251 | ||
2252 | if (mddev->private == NULL) { | |
2253 | conf = setup_conf(mddev); | |
2254 | if (IS_ERR(conf)) | |
2255 | return PTR_ERR(conf); | |
2256 | mddev->private = conf; | |
2257 | } | |
2258 | conf = mddev->private; | |
2259 | if (!conf) | |
2260 | goto out; | |
2261 | ||
dab8b292 TM |
2262 | mddev->thread = conf->thread; |
2263 | conf->thread = NULL; | |
2264 | ||
8f6c2e4b MP |
2265 | chunk_size = mddev->chunk_sectors << 9; |
2266 | blk_queue_io_min(mddev->queue, chunk_size); | |
2267 | if (conf->raid_disks % conf->near_copies) | |
2268 | blk_queue_io_opt(mddev->queue, chunk_size * conf->raid_disks); | |
2269 | else | |
2270 | blk_queue_io_opt(mddev->queue, chunk_size * | |
2271 | (conf->raid_disks / conf->near_copies)); | |
2272 | ||
159ec1fc | 2273 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 2274 | disk_idx = rdev->raid_disk; |
84707f38 | 2275 | if (disk_idx >= conf->raid_disks |
1da177e4 LT |
2276 | || disk_idx < 0) |
2277 | continue; | |
2278 | disk = conf->mirrors + disk_idx; | |
2279 | ||
2280 | disk->rdev = rdev; | |
8f6c2e4b MP |
2281 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2282 | rdev->data_offset << 9); | |
1da177e4 | 2283 | /* as we don't honour merge_bvec_fn, we must never risk |
627a2d3c N |
2284 | * violating it, so limit max_segments to 1 lying |
2285 | * within a single page. | |
1da177e4 | 2286 | */ |
627a2d3c N |
2287 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
2288 | blk_queue_max_segments(mddev->queue, 1); | |
2289 | blk_queue_segment_boundary(mddev->queue, | |
2290 | PAGE_CACHE_SIZE - 1); | |
2291 | } | |
1da177e4 LT |
2292 | |
2293 | disk->head_position = 0; | |
1da177e4 | 2294 | } |
6d508242 N |
2295 | /* need to check that every block has at least one working mirror */ |
2296 | if (!enough(conf)) { | |
128595ed | 2297 | printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 2298 | mdname(mddev)); |
1da177e4 LT |
2299 | goto out_free_conf; |
2300 | } | |
2301 | ||
2302 | mddev->degraded = 0; | |
2303 | for (i = 0; i < conf->raid_disks; i++) { | |
2304 | ||
2305 | disk = conf->mirrors + i; | |
2306 | ||
5fd6c1dc | 2307 | if (!disk->rdev || |
2e333e89 | 2308 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
2309 | disk->head_position = 0; |
2310 | mddev->degraded++; | |
8c2e870a NB |
2311 | if (disk->rdev) |
2312 | conf->fullsync = 1; | |
1da177e4 LT |
2313 | } |
2314 | } | |
2315 | ||
8c6ac868 | 2316 | if (mddev->recovery_cp != MaxSector) |
128595ed | 2317 | printk(KERN_NOTICE "md/raid10:%s: not clean" |
8c6ac868 AN |
2318 | " -- starting background reconstruction\n", |
2319 | mdname(mddev)); | |
1da177e4 | 2320 | printk(KERN_INFO |
128595ed | 2321 | "md/raid10:%s: active with %d out of %d devices\n", |
84707f38 N |
2322 | mdname(mddev), conf->raid_disks - mddev->degraded, |
2323 | conf->raid_disks); | |
1da177e4 LT |
2324 | /* |
2325 | * Ok, everything is just fine now | |
2326 | */ | |
dab8b292 TM |
2327 | mddev->dev_sectors = conf->dev_sectors; |
2328 | size = raid10_size(mddev, 0, 0); | |
2329 | md_set_array_sectors(mddev, size); | |
2330 | mddev->resync_max_sectors = size; | |
1da177e4 | 2331 | |
0d129228 N |
2332 | mddev->queue->backing_dev_info.congested_fn = raid10_congested; |
2333 | mddev->queue->backing_dev_info.congested_data = mddev; | |
7a5febe9 | 2334 | |
1da177e4 LT |
2335 | /* Calculate max read-ahead size. |
2336 | * We need to readahead at least twice a whole stripe.... | |
2337 | * maybe... | |
2338 | */ | |
2339 | { | |
9d8f0363 AN |
2340 | int stripe = conf->raid_disks * |
2341 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
1da177e4 LT |
2342 | stripe /= conf->near_copies; |
2343 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) | |
2344 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
2345 | } | |
2346 | ||
84707f38 | 2347 | if (conf->near_copies < conf->raid_disks) |
1da177e4 | 2348 | blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec); |
a91a2785 MP |
2349 | |
2350 | if (md_integrity_register(mddev)) | |
2351 | goto out_free_conf; | |
2352 | ||
1da177e4 LT |
2353 | return 0; |
2354 | ||
2355 | out_free_conf: | |
589a594b | 2356 | md_unregister_thread(mddev->thread); |
1da177e4 LT |
2357 | if (conf->r10bio_pool) |
2358 | mempool_destroy(conf->r10bio_pool); | |
1345b1d8 | 2359 | safe_put_page(conf->tmppage); |
990a8baf | 2360 | kfree(conf->mirrors); |
1da177e4 LT |
2361 | kfree(conf); |
2362 | mddev->private = NULL; | |
2363 | out: | |
2364 | return -EIO; | |
2365 | } | |
2366 | ||
2367 | static int stop(mddev_t *mddev) | |
2368 | { | |
070ec55d | 2369 | conf_t *conf = mddev->private; |
1da177e4 | 2370 | |
409c57f3 N |
2371 | raise_barrier(conf, 0); |
2372 | lower_barrier(conf); | |
2373 | ||
1da177e4 LT |
2374 | md_unregister_thread(mddev->thread); |
2375 | mddev->thread = NULL; | |
2376 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
2377 | if (conf->r10bio_pool) | |
2378 | mempool_destroy(conf->r10bio_pool); | |
990a8baf | 2379 | kfree(conf->mirrors); |
1da177e4 LT |
2380 | kfree(conf); |
2381 | mddev->private = NULL; | |
2382 | return 0; | |
2383 | } | |
2384 | ||
6cce3b23 N |
2385 | static void raid10_quiesce(mddev_t *mddev, int state) |
2386 | { | |
070ec55d | 2387 | conf_t *conf = mddev->private; |
6cce3b23 N |
2388 | |
2389 | switch(state) { | |
2390 | case 1: | |
2391 | raise_barrier(conf, 0); | |
2392 | break; | |
2393 | case 0: | |
2394 | lower_barrier(conf); | |
2395 | break; | |
2396 | } | |
6cce3b23 | 2397 | } |
1da177e4 | 2398 | |
dab8b292 TM |
2399 | static void *raid10_takeover_raid0(mddev_t *mddev) |
2400 | { | |
2401 | mdk_rdev_t *rdev; | |
2402 | conf_t *conf; | |
2403 | ||
2404 | if (mddev->degraded > 0) { | |
128595ed N |
2405 | printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n", |
2406 | mdname(mddev)); | |
dab8b292 TM |
2407 | return ERR_PTR(-EINVAL); |
2408 | } | |
2409 | ||
dab8b292 TM |
2410 | /* Set new parameters */ |
2411 | mddev->new_level = 10; | |
2412 | /* new layout: far_copies = 1, near_copies = 2 */ | |
2413 | mddev->new_layout = (1<<8) + 2; | |
2414 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
2415 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
2416 | mddev->raid_disks *= 2; |
2417 | /* make sure it will be not marked as dirty */ | |
2418 | mddev->recovery_cp = MaxSector; | |
2419 | ||
2420 | conf = setup_conf(mddev); | |
02214dc5 | 2421 | if (!IS_ERR(conf)) { |
e93f68a1 N |
2422 | list_for_each_entry(rdev, &mddev->disks, same_set) |
2423 | if (rdev->raid_disk >= 0) | |
2424 | rdev->new_raid_disk = rdev->raid_disk * 2; | |
02214dc5 KW |
2425 | conf->barrier = 1; |
2426 | } | |
2427 | ||
dab8b292 TM |
2428 | return conf; |
2429 | } | |
2430 | ||
2431 | static void *raid10_takeover(mddev_t *mddev) | |
2432 | { | |
2433 | struct raid0_private_data *raid0_priv; | |
2434 | ||
2435 | /* raid10 can take over: | |
2436 | * raid0 - providing it has only two drives | |
2437 | */ | |
2438 | if (mddev->level == 0) { | |
2439 | /* for raid0 takeover only one zone is supported */ | |
2440 | raid0_priv = mddev->private; | |
2441 | if (raid0_priv->nr_strip_zones > 1) { | |
128595ed N |
2442 | printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0" |
2443 | " with more than one zone.\n", | |
2444 | mdname(mddev)); | |
dab8b292 TM |
2445 | return ERR_PTR(-EINVAL); |
2446 | } | |
2447 | return raid10_takeover_raid0(mddev); | |
2448 | } | |
2449 | return ERR_PTR(-EINVAL); | |
2450 | } | |
2451 | ||
2604b703 | 2452 | static struct mdk_personality raid10_personality = |
1da177e4 LT |
2453 | { |
2454 | .name = "raid10", | |
2604b703 | 2455 | .level = 10, |
1da177e4 LT |
2456 | .owner = THIS_MODULE, |
2457 | .make_request = make_request, | |
2458 | .run = run, | |
2459 | .stop = stop, | |
2460 | .status = status, | |
2461 | .error_handler = error, | |
2462 | .hot_add_disk = raid10_add_disk, | |
2463 | .hot_remove_disk= raid10_remove_disk, | |
2464 | .spare_active = raid10_spare_active, | |
2465 | .sync_request = sync_request, | |
6cce3b23 | 2466 | .quiesce = raid10_quiesce, |
80c3a6ce | 2467 | .size = raid10_size, |
dab8b292 | 2468 | .takeover = raid10_takeover, |
1da177e4 LT |
2469 | }; |
2470 | ||
2471 | static int __init raid_init(void) | |
2472 | { | |
2604b703 | 2473 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
2474 | } |
2475 | ||
2476 | static void raid_exit(void) | |
2477 | { | |
2604b703 | 2478 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
2479 | } |
2480 | ||
2481 | module_init(raid_init); | |
2482 | module_exit(raid_exit); | |
2483 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 2484 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 2485 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 2486 | MODULE_ALIAS("md-raid10"); |
2604b703 | 2487 | MODULE_ALIAS("md-level-10"); |