]>
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 | * | |
8 | * Base on code in raid1.c. See raid1.c for futher copyright information. | |
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 | ||
60 | static void unplug_slaves(mddev_t *mddev); | |
61 | ||
0a27ec96 N |
62 | static void allow_barrier(conf_t *conf); |
63 | static void lower_barrier(conf_t *conf); | |
64 | ||
dd0fc66f | 65 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
66 | { |
67 | conf_t *conf = data; | |
68 | r10bio_t *r10_bio; | |
69 | int size = offsetof(struct r10bio_s, devs[conf->copies]); | |
70 | ||
71 | /* allocate a r10bio with room for raid_disks entries in the bios array */ | |
9ffae0cf | 72 | r10_bio = kzalloc(size, gfp_flags); |
ed9bfdf1 | 73 | if (!r10_bio && conf->mddev) |
1da177e4 LT |
74 | unplug_slaves(conf->mddev); |
75 | ||
76 | return r10_bio; | |
77 | } | |
78 | ||
79 | static void r10bio_pool_free(void *r10_bio, void *data) | |
80 | { | |
81 | kfree(r10_bio); | |
82 | } | |
83 | ||
0310fa21 | 84 | /* Maximum size of each resync request */ |
1da177e4 | 85 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 86 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
87 | /* amount of memory to reserve for resync requests */ |
88 | #define RESYNC_WINDOW (1024*1024) | |
89 | /* maximum number of concurrent requests, memory permitting */ | |
90 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
91 | |
92 | /* | |
93 | * When performing a resync, we need to read and compare, so | |
94 | * we need as many pages are there are copies. | |
95 | * When performing a recovery, we need 2 bios, one for read, | |
96 | * one for write (we recover only one drive per r10buf) | |
97 | * | |
98 | */ | |
dd0fc66f | 99 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
100 | { |
101 | conf_t *conf = data; | |
102 | struct page *page; | |
103 | r10bio_t *r10_bio; | |
104 | struct bio *bio; | |
105 | int i, j; | |
106 | int nalloc; | |
107 | ||
108 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
109 | if (!r10_bio) { | |
110 | unplug_slaves(conf->mddev); | |
111 | return NULL; | |
112 | } | |
113 | ||
114 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | |
115 | nalloc = conf->copies; /* resync */ | |
116 | else | |
117 | nalloc = 2; /* recovery */ | |
118 | ||
119 | /* | |
120 | * Allocate bios. | |
121 | */ | |
122 | for (j = nalloc ; j-- ; ) { | |
6746557f | 123 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
124 | if (!bio) |
125 | goto out_free_bio; | |
126 | r10_bio->devs[j].bio = bio; | |
127 | } | |
128 | /* | |
129 | * Allocate RESYNC_PAGES data pages and attach them | |
130 | * where needed. | |
131 | */ | |
132 | for (j = 0 ; j < nalloc; j++) { | |
133 | bio = r10_bio->devs[j].bio; | |
134 | for (i = 0; i < RESYNC_PAGES; i++) { | |
135 | page = alloc_page(gfp_flags); | |
136 | if (unlikely(!page)) | |
137 | goto out_free_pages; | |
138 | ||
139 | bio->bi_io_vec[i].bv_page = page; | |
140 | } | |
141 | } | |
142 | ||
143 | return r10_bio; | |
144 | ||
145 | out_free_pages: | |
146 | for ( ; i > 0 ; i--) | |
1345b1d8 | 147 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
148 | while (j--) |
149 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 150 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
151 | j = -1; |
152 | out_free_bio: | |
153 | while ( ++j < nalloc ) | |
154 | bio_put(r10_bio->devs[j].bio); | |
155 | r10bio_pool_free(r10_bio, conf); | |
156 | return NULL; | |
157 | } | |
158 | ||
159 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
160 | { | |
161 | int i; | |
162 | conf_t *conf = data; | |
163 | r10bio_t *r10bio = __r10_bio; | |
164 | int j; | |
165 | ||
166 | for (j=0; j < conf->copies; j++) { | |
167 | struct bio *bio = r10bio->devs[j].bio; | |
168 | if (bio) { | |
169 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 170 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
171 | bio->bi_io_vec[i].bv_page = NULL; |
172 | } | |
173 | bio_put(bio); | |
174 | } | |
175 | } | |
176 | r10bio_pool_free(r10bio, conf); | |
177 | } | |
178 | ||
179 | static void put_all_bios(conf_t *conf, r10bio_t *r10_bio) | |
180 | { | |
181 | int i; | |
182 | ||
183 | for (i = 0; i < conf->copies; i++) { | |
184 | struct bio **bio = & r10_bio->devs[i].bio; | |
0eb3ff12 | 185 | if (*bio && *bio != IO_BLOCKED) |
1da177e4 LT |
186 | bio_put(*bio); |
187 | *bio = NULL; | |
188 | } | |
189 | } | |
190 | ||
858119e1 | 191 | static void free_r10bio(r10bio_t *r10_bio) |
1da177e4 | 192 | { |
070ec55d | 193 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
194 | |
195 | /* | |
196 | * Wake up any possible resync thread that waits for the device | |
197 | * to go idle. | |
198 | */ | |
0a27ec96 | 199 | allow_barrier(conf); |
1da177e4 LT |
200 | |
201 | put_all_bios(conf, r10_bio); | |
202 | mempool_free(r10_bio, conf->r10bio_pool); | |
203 | } | |
204 | ||
858119e1 | 205 | static void put_buf(r10bio_t *r10_bio) |
1da177e4 | 206 | { |
070ec55d | 207 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
208 | |
209 | mempool_free(r10_bio, conf->r10buf_pool); | |
210 | ||
0a27ec96 | 211 | lower_barrier(conf); |
1da177e4 LT |
212 | } |
213 | ||
214 | static void reschedule_retry(r10bio_t *r10_bio) | |
215 | { | |
216 | unsigned long flags; | |
217 | mddev_t *mddev = r10_bio->mddev; | |
070ec55d | 218 | conf_t *conf = mddev->private; |
1da177e4 LT |
219 | |
220 | spin_lock_irqsave(&conf->device_lock, flags); | |
221 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 222 | conf->nr_queued ++; |
1da177e4 LT |
223 | spin_unlock_irqrestore(&conf->device_lock, flags); |
224 | ||
388667be AJ |
225 | /* wake up frozen array... */ |
226 | wake_up(&conf->wait_barrier); | |
227 | ||
1da177e4 LT |
228 | md_wakeup_thread(mddev->thread); |
229 | } | |
230 | ||
231 | /* | |
232 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
233 | * operation and are ready to return a success/failure code to the buffer | |
234 | * cache layer. | |
235 | */ | |
236 | static void raid_end_bio_io(r10bio_t *r10_bio) | |
237 | { | |
238 | struct bio *bio = r10_bio->master_bio; | |
239 | ||
6712ecf8 | 240 | bio_endio(bio, |
1da177e4 LT |
241 | test_bit(R10BIO_Uptodate, &r10_bio->state) ? 0 : -EIO); |
242 | free_r10bio(r10_bio); | |
243 | } | |
244 | ||
245 | /* | |
246 | * Update disk head position estimator based on IRQ completion info. | |
247 | */ | |
248 | static inline void update_head_pos(int slot, r10bio_t *r10_bio) | |
249 | { | |
070ec55d | 250 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
251 | |
252 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
253 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
254 | } | |
255 | ||
6712ecf8 | 256 | static void raid10_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
257 | { |
258 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 259 | r10bio_t *r10_bio = bio->bi_private; |
1da177e4 | 260 | int slot, dev; |
070ec55d | 261 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 | 262 | |
1da177e4 LT |
263 | |
264 | slot = r10_bio->read_slot; | |
265 | dev = r10_bio->devs[slot].devnum; | |
266 | /* | |
267 | * this branch is our 'one mirror IO has finished' event handler: | |
268 | */ | |
4443ae10 N |
269 | update_head_pos(slot, r10_bio); |
270 | ||
271 | if (uptodate) { | |
1da177e4 LT |
272 | /* |
273 | * Set R10BIO_Uptodate in our master bio, so that | |
274 | * we will return a good error code to the higher | |
275 | * levels even if IO on some other mirrored buffer fails. | |
276 | * | |
277 | * The 'master' represents the composite IO operation to | |
278 | * user-side. So if something waits for IO, then it will | |
279 | * wait for the 'master' bio. | |
280 | */ | |
281 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 282 | raid_end_bio_io(r10_bio); |
4443ae10 | 283 | } else { |
1da177e4 LT |
284 | /* |
285 | * oops, read error: | |
286 | */ | |
287 | char b[BDEVNAME_SIZE]; | |
288 | if (printk_ratelimit()) | |
128595ed N |
289 | printk(KERN_ERR "md/raid10:%s: %s: rescheduling sector %llu\n", |
290 | mdname(conf->mddev), | |
1da177e4 LT |
291 | bdevname(conf->mirrors[dev].rdev->bdev,b), (unsigned long long)r10_bio->sector); |
292 | reschedule_retry(r10_bio); | |
293 | } | |
294 | ||
295 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); | |
1da177e4 LT |
296 | } |
297 | ||
6712ecf8 | 298 | static void raid10_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
299 | { |
300 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 301 | r10bio_t *r10_bio = bio->bi_private; |
1da177e4 | 302 | int slot, dev; |
070ec55d | 303 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 | 304 | |
1da177e4 LT |
305 | for (slot = 0; slot < conf->copies; slot++) |
306 | if (r10_bio->devs[slot].bio == bio) | |
307 | break; | |
308 | dev = r10_bio->devs[slot].devnum; | |
309 | ||
310 | /* | |
311 | * this branch is our 'one mirror IO has finished' event handler: | |
312 | */ | |
6cce3b23 | 313 | if (!uptodate) { |
1da177e4 | 314 | md_error(r10_bio->mddev, conf->mirrors[dev].rdev); |
6cce3b23 N |
315 | /* an I/O failed, we can't clear the bitmap */ |
316 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
317 | } else | |
1da177e4 LT |
318 | /* |
319 | * Set R10BIO_Uptodate in our master bio, so that | |
320 | * we will return a good error code for to the higher | |
321 | * levels even if IO on some other mirrored buffer fails. | |
322 | * | |
323 | * The 'master' represents the composite IO operation to | |
324 | * user-side. So if something waits for IO, then it will | |
325 | * wait for the 'master' bio. | |
326 | */ | |
327 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
328 | ||
329 | update_head_pos(slot, r10_bio); | |
330 | ||
331 | /* | |
332 | * | |
333 | * Let's see if all mirrored write operations have finished | |
334 | * already. | |
335 | */ | |
336 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
6cce3b23 N |
337 | /* clear the bitmap if all writes complete successfully */ |
338 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
339 | r10_bio->sectors, | |
340 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
341 | 0); | |
1da177e4 LT |
342 | md_write_end(r10_bio->mddev); |
343 | raid_end_bio_io(r10_bio); | |
344 | } | |
345 | ||
346 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); | |
1da177e4 LT |
347 | } |
348 | ||
349 | ||
350 | /* | |
351 | * RAID10 layout manager | |
352 | * Aswell as the chunksize and raid_disks count, there are two | |
353 | * parameters: near_copies and far_copies. | |
354 | * near_copies * far_copies must be <= raid_disks. | |
355 | * Normally one of these will be 1. | |
356 | * If both are 1, we get raid0. | |
357 | * If near_copies == raid_disks, we get raid1. | |
358 | * | |
359 | * Chunks are layed out in raid0 style with near_copies copies of the | |
360 | * first chunk, followed by near_copies copies of the next chunk and | |
361 | * so on. | |
362 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
363 | * as described above, we start again with a device offset of near_copies. | |
364 | * So we effectively have another copy of the whole array further down all | |
365 | * the drives, but with blocks on different drives. | |
366 | * With this layout, and block is never stored twice on the one device. | |
367 | * | |
368 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 369 | * on each device that it is on. |
1da177e4 LT |
370 | * |
371 | * raid10_find_virt does the reverse mapping, from a device and a | |
372 | * sector offset to a virtual address | |
373 | */ | |
374 | ||
375 | static void raid10_find_phys(conf_t *conf, r10bio_t *r10bio) | |
376 | { | |
377 | int n,f; | |
378 | sector_t sector; | |
379 | sector_t chunk; | |
380 | sector_t stripe; | |
381 | int dev; | |
382 | ||
383 | int slot = 0; | |
384 | ||
385 | /* now calculate first sector/dev */ | |
386 | chunk = r10bio->sector >> conf->chunk_shift; | |
387 | sector = r10bio->sector & conf->chunk_mask; | |
388 | ||
389 | chunk *= conf->near_copies; | |
390 | stripe = chunk; | |
391 | dev = sector_div(stripe, conf->raid_disks); | |
c93983bf N |
392 | if (conf->far_offset) |
393 | stripe *= conf->far_copies; | |
1da177e4 LT |
394 | |
395 | sector += stripe << conf->chunk_shift; | |
396 | ||
397 | /* and calculate all the others */ | |
398 | for (n=0; n < conf->near_copies; n++) { | |
399 | int d = dev; | |
400 | sector_t s = sector; | |
401 | r10bio->devs[slot].addr = sector; | |
402 | r10bio->devs[slot].devnum = d; | |
403 | slot++; | |
404 | ||
405 | for (f = 1; f < conf->far_copies; f++) { | |
406 | d += conf->near_copies; | |
407 | if (d >= conf->raid_disks) | |
408 | d -= conf->raid_disks; | |
409 | s += conf->stride; | |
410 | r10bio->devs[slot].devnum = d; | |
411 | r10bio->devs[slot].addr = s; | |
412 | slot++; | |
413 | } | |
414 | dev++; | |
415 | if (dev >= conf->raid_disks) { | |
416 | dev = 0; | |
417 | sector += (conf->chunk_mask + 1); | |
418 | } | |
419 | } | |
420 | BUG_ON(slot != conf->copies); | |
421 | } | |
422 | ||
423 | static sector_t raid10_find_virt(conf_t *conf, sector_t sector, int dev) | |
424 | { | |
425 | sector_t offset, chunk, vchunk; | |
426 | ||
1da177e4 | 427 | offset = sector & conf->chunk_mask; |
c93983bf N |
428 | if (conf->far_offset) { |
429 | int fc; | |
430 | chunk = sector >> conf->chunk_shift; | |
431 | fc = sector_div(chunk, conf->far_copies); | |
432 | dev -= fc * conf->near_copies; | |
433 | if (dev < 0) | |
434 | dev += conf->raid_disks; | |
435 | } else { | |
64a742bc | 436 | while (sector >= conf->stride) { |
c93983bf N |
437 | sector -= conf->stride; |
438 | if (dev < conf->near_copies) | |
439 | dev += conf->raid_disks - conf->near_copies; | |
440 | else | |
441 | dev -= conf->near_copies; | |
442 | } | |
443 | chunk = sector >> conf->chunk_shift; | |
444 | } | |
1da177e4 LT |
445 | vchunk = chunk * conf->raid_disks + dev; |
446 | sector_div(vchunk, conf->near_copies); | |
447 | return (vchunk << conf->chunk_shift) + offset; | |
448 | } | |
449 | ||
450 | /** | |
451 | * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged | |
452 | * @q: request queue | |
cc371e66 | 453 | * @bvm: properties of new bio |
1da177e4 LT |
454 | * @biovec: the request that could be merged to it. |
455 | * | |
456 | * Return amount of bytes we can accept at this offset | |
457 | * If near_copies == raid_disk, there are no striping issues, | |
458 | * but in that case, the function isn't called at all. | |
459 | */ | |
cc371e66 AK |
460 | static int raid10_mergeable_bvec(struct request_queue *q, |
461 | struct bvec_merge_data *bvm, | |
462 | struct bio_vec *biovec) | |
1da177e4 LT |
463 | { |
464 | mddev_t *mddev = q->queuedata; | |
cc371e66 | 465 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
1da177e4 | 466 | int max; |
9d8f0363 | 467 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 468 | unsigned int bio_sectors = bvm->bi_size >> 9; |
1da177e4 LT |
469 | |
470 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
471 | if (max < 0) max = 0; /* bio_add cannot handle a negative return */ | |
cc371e66 AK |
472 | if (max <= biovec->bv_len && bio_sectors == 0) |
473 | return biovec->bv_len; | |
1da177e4 LT |
474 | else |
475 | return max; | |
476 | } | |
477 | ||
478 | /* | |
479 | * This routine returns the disk from which the requested read should | |
480 | * be done. There is a per-array 'next expected sequential IO' sector | |
481 | * number - if this matches on the next IO then we use the last disk. | |
482 | * There is also a per-disk 'last know head position' sector that is | |
483 | * maintained from IRQ contexts, both the normal and the resync IO | |
484 | * completion handlers update this position correctly. If there is no | |
485 | * perfect sequential match then we pick the disk whose head is closest. | |
486 | * | |
487 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
488 | * because position is mirror, not device based. | |
489 | * | |
490 | * The rdev for the device selected will have nr_pending incremented. | |
491 | */ | |
492 | ||
493 | /* | |
494 | * FIXME: possibly should rethink readbalancing and do it differently | |
495 | * depending on near_copies / far_copies geometry. | |
496 | */ | |
497 | static int read_balance(conf_t *conf, r10bio_t *r10_bio) | |
498 | { | |
af3a2cd6 | 499 | const sector_t this_sector = r10_bio->sector; |
1da177e4 LT |
500 | int disk, slot, nslot; |
501 | const int sectors = r10_bio->sectors; | |
502 | sector_t new_distance, current_distance; | |
d6065f7b | 503 | mdk_rdev_t *rdev; |
1da177e4 LT |
504 | |
505 | raid10_find_phys(conf, r10_bio); | |
506 | rcu_read_lock(); | |
507 | /* | |
508 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
509 | * device if no resync is going on (recovery is ok), or below |
510 | * the resync window. We take the first readable disk when | |
511 | * above the resync window. | |
1da177e4 LT |
512 | */ |
513 | if (conf->mddev->recovery_cp < MaxSector | |
514 | && (this_sector + sectors >= conf->next_resync)) { | |
515 | /* make sure that disk is operational */ | |
516 | slot = 0; | |
517 | disk = r10_bio->devs[slot].devnum; | |
518 | ||
d6065f7b | 519 | while ((rdev = rcu_dereference(conf->mirrors[disk].rdev)) == NULL || |
0eb3ff12 | 520 | r10_bio->devs[slot].bio == IO_BLOCKED || |
b2d444d7 | 521 | !test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
522 | slot++; |
523 | if (slot == conf->copies) { | |
524 | slot = 0; | |
525 | disk = -1; | |
526 | break; | |
527 | } | |
528 | disk = r10_bio->devs[slot].devnum; | |
529 | } | |
530 | goto rb_out; | |
531 | } | |
532 | ||
533 | ||
534 | /* make sure the disk is operational */ | |
535 | slot = 0; | |
536 | disk = r10_bio->devs[slot].devnum; | |
d6065f7b | 537 | while ((rdev=rcu_dereference(conf->mirrors[disk].rdev)) == NULL || |
0eb3ff12 | 538 | r10_bio->devs[slot].bio == IO_BLOCKED || |
b2d444d7 | 539 | !test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
540 | slot ++; |
541 | if (slot == conf->copies) { | |
542 | disk = -1; | |
543 | goto rb_out; | |
544 | } | |
545 | disk = r10_bio->devs[slot].devnum; | |
546 | } | |
547 | ||
548 | ||
3ec67ac1 N |
549 | current_distance = abs(r10_bio->devs[slot].addr - |
550 | conf->mirrors[disk].head_position); | |
1da177e4 | 551 | |
8ed3a195 KS |
552 | /* Find the disk whose head is closest, |
553 | * or - for far > 1 - find the closest to partition beginning */ | |
1da177e4 LT |
554 | |
555 | for (nslot = slot; nslot < conf->copies; nslot++) { | |
556 | int ndisk = r10_bio->devs[nslot].devnum; | |
557 | ||
558 | ||
d6065f7b | 559 | if ((rdev=rcu_dereference(conf->mirrors[ndisk].rdev)) == NULL || |
0eb3ff12 | 560 | r10_bio->devs[nslot].bio == IO_BLOCKED || |
b2d444d7 | 561 | !test_bit(In_sync, &rdev->flags)) |
1da177e4 LT |
562 | continue; |
563 | ||
22dfdf52 N |
564 | /* This optimisation is debatable, and completely destroys |
565 | * sequential read speed for 'far copies' arrays. So only | |
566 | * keep it for 'near' arrays, and review those later. | |
567 | */ | |
568 | if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending)) { | |
1da177e4 LT |
569 | disk = ndisk; |
570 | slot = nslot; | |
571 | break; | |
572 | } | |
8ed3a195 KS |
573 | |
574 | /* for far > 1 always use the lowest address */ | |
575 | if (conf->far_copies > 1) | |
576 | new_distance = r10_bio->devs[nslot].addr; | |
577 | else | |
578 | new_distance = abs(r10_bio->devs[nslot].addr - | |
579 | conf->mirrors[ndisk].head_position); | |
1da177e4 LT |
580 | if (new_distance < current_distance) { |
581 | current_distance = new_distance; | |
582 | disk = ndisk; | |
583 | slot = nslot; | |
584 | } | |
585 | } | |
586 | ||
587 | rb_out: | |
588 | r10_bio->read_slot = slot; | |
589 | /* conf->next_seq_sect = this_sector + sectors;*/ | |
590 | ||
d6065f7b | 591 | if (disk >= 0 && (rdev=rcu_dereference(conf->mirrors[disk].rdev))!= NULL) |
1da177e4 | 592 | atomic_inc(&conf->mirrors[disk].rdev->nr_pending); |
29fc7e3e N |
593 | else |
594 | disk = -1; | |
1da177e4 LT |
595 | rcu_read_unlock(); |
596 | ||
597 | return disk; | |
598 | } | |
599 | ||
600 | static void unplug_slaves(mddev_t *mddev) | |
601 | { | |
070ec55d | 602 | conf_t *conf = mddev->private; |
1da177e4 LT |
603 | int i; |
604 | ||
605 | rcu_read_lock(); | |
84707f38 | 606 | for (i=0; i < conf->raid_disks; i++) { |
d6065f7b | 607 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
b2d444d7 | 608 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { |
165125e1 | 609 | struct request_queue *r_queue = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
610 | |
611 | atomic_inc(&rdev->nr_pending); | |
612 | rcu_read_unlock(); | |
613 | ||
2ad8b1ef | 614 | blk_unplug(r_queue); |
1da177e4 LT |
615 | |
616 | rdev_dec_pending(rdev, mddev); | |
617 | rcu_read_lock(); | |
618 | } | |
619 | } | |
620 | rcu_read_unlock(); | |
621 | } | |
622 | ||
165125e1 | 623 | static void raid10_unplug(struct request_queue *q) |
1da177e4 | 624 | { |
6cce3b23 N |
625 | mddev_t *mddev = q->queuedata; |
626 | ||
1da177e4 | 627 | unplug_slaves(q->queuedata); |
6cce3b23 | 628 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
629 | } |
630 | ||
0d129228 N |
631 | static int raid10_congested(void *data, int bits) |
632 | { | |
633 | mddev_t *mddev = data; | |
070ec55d | 634 | conf_t *conf = mddev->private; |
0d129228 N |
635 | int i, ret = 0; |
636 | ||
3fa841d7 N |
637 | if (mddev_congested(mddev, bits)) |
638 | return 1; | |
0d129228 | 639 | rcu_read_lock(); |
84707f38 | 640 | for (i = 0; i < conf->raid_disks && ret == 0; i++) { |
0d129228 N |
641 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); |
642 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
165125e1 | 643 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
644 | |
645 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
646 | } | |
647 | } | |
648 | rcu_read_unlock(); | |
649 | return ret; | |
650 | } | |
651 | ||
a35e63ef N |
652 | static int flush_pending_writes(conf_t *conf) |
653 | { | |
654 | /* Any writes that have been queued but are awaiting | |
655 | * bitmap updates get flushed here. | |
656 | * We return 1 if any requests were actually submitted. | |
657 | */ | |
658 | int rv = 0; | |
659 | ||
660 | spin_lock_irq(&conf->device_lock); | |
661 | ||
662 | if (conf->pending_bio_list.head) { | |
663 | struct bio *bio; | |
664 | bio = bio_list_get(&conf->pending_bio_list); | |
665 | blk_remove_plug(conf->mddev->queue); | |
666 | spin_unlock_irq(&conf->device_lock); | |
667 | /* flush any pending bitmap writes to disk | |
668 | * before proceeding w/ I/O */ | |
669 | bitmap_unplug(conf->mddev->bitmap); | |
670 | ||
671 | while (bio) { /* submit pending writes */ | |
672 | struct bio *next = bio->bi_next; | |
673 | bio->bi_next = NULL; | |
674 | generic_make_request(bio); | |
675 | bio = next; | |
676 | } | |
677 | rv = 1; | |
678 | } else | |
679 | spin_unlock_irq(&conf->device_lock); | |
680 | return rv; | |
681 | } | |
0a27ec96 N |
682 | /* Barriers.... |
683 | * Sometimes we need to suspend IO while we do something else, | |
684 | * either some resync/recovery, or reconfigure the array. | |
685 | * To do this we raise a 'barrier'. | |
686 | * The 'barrier' is a counter that can be raised multiple times | |
687 | * to count how many activities are happening which preclude | |
688 | * normal IO. | |
689 | * We can only raise the barrier if there is no pending IO. | |
690 | * i.e. if nr_pending == 0. | |
691 | * We choose only to raise the barrier if no-one is waiting for the | |
692 | * barrier to go down. This means that as soon as an IO request | |
693 | * is ready, no other operations which require a barrier will start | |
694 | * until the IO request has had a chance. | |
695 | * | |
696 | * So: regular IO calls 'wait_barrier'. When that returns there | |
697 | * is no backgroup IO happening, It must arrange to call | |
698 | * allow_barrier when it has finished its IO. | |
699 | * backgroup IO calls must call raise_barrier. Once that returns | |
700 | * there is no normal IO happeing. It must arrange to call | |
701 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 702 | */ |
1da177e4 | 703 | |
6cce3b23 | 704 | static void raise_barrier(conf_t *conf, int force) |
1da177e4 | 705 | { |
6cce3b23 | 706 | BUG_ON(force && !conf->barrier); |
1da177e4 | 707 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 708 | |
6cce3b23 N |
709 | /* Wait until no block IO is waiting (unless 'force') */ |
710 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
0a27ec96 N |
711 | conf->resync_lock, |
712 | raid10_unplug(conf->mddev->queue)); | |
713 | ||
714 | /* block any new IO from starting */ | |
715 | conf->barrier++; | |
716 | ||
717 | /* No wait for all pending IO to complete */ | |
718 | wait_event_lock_irq(conf->wait_barrier, | |
719 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
720 | conf->resync_lock, | |
721 | raid10_unplug(conf->mddev->queue)); | |
722 | ||
723 | spin_unlock_irq(&conf->resync_lock); | |
724 | } | |
725 | ||
726 | static void lower_barrier(conf_t *conf) | |
727 | { | |
728 | unsigned long flags; | |
729 | spin_lock_irqsave(&conf->resync_lock, flags); | |
730 | conf->barrier--; | |
731 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
732 | wake_up(&conf->wait_barrier); | |
733 | } | |
734 | ||
735 | static void wait_barrier(conf_t *conf) | |
736 | { | |
737 | spin_lock_irq(&conf->resync_lock); | |
738 | if (conf->barrier) { | |
739 | conf->nr_waiting++; | |
740 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, | |
741 | conf->resync_lock, | |
742 | raid10_unplug(conf->mddev->queue)); | |
743 | conf->nr_waiting--; | |
1da177e4 | 744 | } |
0a27ec96 | 745 | conf->nr_pending++; |
1da177e4 LT |
746 | spin_unlock_irq(&conf->resync_lock); |
747 | } | |
748 | ||
0a27ec96 N |
749 | static void allow_barrier(conf_t *conf) |
750 | { | |
751 | unsigned long flags; | |
752 | spin_lock_irqsave(&conf->resync_lock, flags); | |
753 | conf->nr_pending--; | |
754 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
755 | wake_up(&conf->wait_barrier); | |
756 | } | |
757 | ||
4443ae10 N |
758 | static void freeze_array(conf_t *conf) |
759 | { | |
760 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 761 | * go quiet. |
4443ae10 | 762 | * We increment barrier and nr_waiting, and then |
1c830532 N |
763 | * wait until nr_pending match nr_queued+1 |
764 | * This is called in the context of one normal IO request | |
765 | * that has failed. Thus any sync request that might be pending | |
766 | * will be blocked by nr_pending, and we need to wait for | |
767 | * pending IO requests to complete or be queued for re-try. | |
768 | * Thus the number queued (nr_queued) plus this request (1) | |
769 | * must match the number of pending IOs (nr_pending) before | |
770 | * we continue. | |
4443ae10 N |
771 | */ |
772 | spin_lock_irq(&conf->resync_lock); | |
773 | conf->barrier++; | |
774 | conf->nr_waiting++; | |
775 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 776 | conf->nr_pending == conf->nr_queued+1, |
4443ae10 | 777 | conf->resync_lock, |
a35e63ef N |
778 | ({ flush_pending_writes(conf); |
779 | raid10_unplug(conf->mddev->queue); })); | |
4443ae10 N |
780 | spin_unlock_irq(&conf->resync_lock); |
781 | } | |
782 | ||
783 | static void unfreeze_array(conf_t *conf) | |
784 | { | |
785 | /* reverse the effect of the freeze */ | |
786 | spin_lock_irq(&conf->resync_lock); | |
787 | conf->barrier--; | |
788 | conf->nr_waiting--; | |
789 | wake_up(&conf->wait_barrier); | |
790 | spin_unlock_irq(&conf->resync_lock); | |
791 | } | |
792 | ||
21a52c6d | 793 | static int make_request(mddev_t *mddev, struct bio * bio) |
1da177e4 | 794 | { |
070ec55d | 795 | conf_t *conf = mddev->private; |
1da177e4 LT |
796 | mirror_info_t *mirror; |
797 | r10bio_t *r10_bio; | |
798 | struct bio *read_bio; | |
799 | int i; | |
800 | int chunk_sects = conf->chunk_mask + 1; | |
a362357b | 801 | const int rw = bio_data_dir(bio); |
2c7d46ec | 802 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 803 | const unsigned long do_fua = (bio->bi_rw & REQ_FUA); |
6cce3b23 | 804 | unsigned long flags; |
6bfe0b49 | 805 | mdk_rdev_t *blocked_rdev; |
1da177e4 | 806 | |
e9c7469b TH |
807 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
808 | md_flush_request(mddev, bio); | |
e5dcdd80 N |
809 | return 0; |
810 | } | |
811 | ||
1da177e4 LT |
812 | /* If this request crosses a chunk boundary, we need to |
813 | * split it. This will only happen for 1 PAGE (or less) requests. | |
814 | */ | |
815 | if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9) | |
816 | > chunk_sects && | |
817 | conf->near_copies < conf->raid_disks)) { | |
818 | struct bio_pair *bp; | |
819 | /* Sanity check -- queue functions should prevent this happening */ | |
820 | if (bio->bi_vcnt != 1 || | |
821 | bio->bi_idx != 0) | |
822 | goto bad_map; | |
823 | /* This is a one page bio that upper layers | |
824 | * refuse to split for us, so we need to split it. | |
825 | */ | |
6feef531 | 826 | bp = bio_split(bio, |
1da177e4 | 827 | chunk_sects - (bio->bi_sector & (chunk_sects - 1)) ); |
51e9ac77 N |
828 | |
829 | /* Each of these 'make_request' calls will call 'wait_barrier'. | |
830 | * If the first succeeds but the second blocks due to the resync | |
831 | * thread raising the barrier, we will deadlock because the | |
832 | * IO to the underlying device will be queued in generic_make_request | |
833 | * and will never complete, so will never reduce nr_pending. | |
834 | * So increment nr_waiting here so no new raise_barriers will | |
835 | * succeed, and so the second wait_barrier cannot block. | |
836 | */ | |
837 | spin_lock_irq(&conf->resync_lock); | |
838 | conf->nr_waiting++; | |
839 | spin_unlock_irq(&conf->resync_lock); | |
840 | ||
21a52c6d | 841 | if (make_request(mddev, &bp->bio1)) |
1da177e4 | 842 | generic_make_request(&bp->bio1); |
21a52c6d | 843 | if (make_request(mddev, &bp->bio2)) |
1da177e4 LT |
844 | generic_make_request(&bp->bio2); |
845 | ||
51e9ac77 N |
846 | spin_lock_irq(&conf->resync_lock); |
847 | conf->nr_waiting--; | |
848 | wake_up(&conf->wait_barrier); | |
849 | spin_unlock_irq(&conf->resync_lock); | |
850 | ||
1da177e4 LT |
851 | bio_pair_release(bp); |
852 | return 0; | |
853 | bad_map: | |
128595ed N |
854 | printk("md/raid10:%s: make_request bug: can't convert block across chunks" |
855 | " or bigger than %dk %llu %d\n", mdname(mddev), chunk_sects/2, | |
1da177e4 LT |
856 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); |
857 | ||
6712ecf8 | 858 | bio_io_error(bio); |
1da177e4 LT |
859 | return 0; |
860 | } | |
861 | ||
3d310eb7 | 862 | md_write_start(mddev, bio); |
06d91a5f | 863 | |
1da177e4 LT |
864 | /* |
865 | * Register the new request and wait if the reconstruction | |
866 | * thread has put up a bar for new requests. | |
867 | * Continue immediately if no resync is active currently. | |
868 | */ | |
0a27ec96 | 869 | wait_barrier(conf); |
1da177e4 | 870 | |
1da177e4 LT |
871 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); |
872 | ||
873 | r10_bio->master_bio = bio; | |
874 | r10_bio->sectors = bio->bi_size >> 9; | |
875 | ||
876 | r10_bio->mddev = mddev; | |
877 | r10_bio->sector = bio->bi_sector; | |
6cce3b23 | 878 | r10_bio->state = 0; |
1da177e4 | 879 | |
a362357b | 880 | if (rw == READ) { |
1da177e4 LT |
881 | /* |
882 | * read balancing logic: | |
883 | */ | |
884 | int disk = read_balance(conf, r10_bio); | |
885 | int slot = r10_bio->read_slot; | |
886 | if (disk < 0) { | |
887 | raid_end_bio_io(r10_bio); | |
888 | return 0; | |
889 | } | |
890 | mirror = conf->mirrors + disk; | |
891 | ||
a167f663 | 892 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1da177e4 LT |
893 | |
894 | r10_bio->devs[slot].bio = read_bio; | |
895 | ||
896 | read_bio->bi_sector = r10_bio->devs[slot].addr + | |
897 | mirror->rdev->data_offset; | |
898 | read_bio->bi_bdev = mirror->rdev->bdev; | |
899 | read_bio->bi_end_io = raid10_end_read_request; | |
7b6d91da | 900 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
901 | read_bio->bi_private = r10_bio; |
902 | ||
903 | generic_make_request(read_bio); | |
904 | return 0; | |
905 | } | |
906 | ||
907 | /* | |
908 | * WRITE: | |
909 | */ | |
6bfe0b49 | 910 | /* first select target devices under rcu_lock and |
1da177e4 LT |
911 | * inc refcount on their rdev. Record them by setting |
912 | * bios[x] to bio | |
913 | */ | |
914 | raid10_find_phys(conf, r10_bio); | |
6bfe0b49 | 915 | retry_write: |
cb6969e8 | 916 | blocked_rdev = NULL; |
1da177e4 LT |
917 | rcu_read_lock(); |
918 | for (i = 0; i < conf->copies; i++) { | |
919 | int d = r10_bio->devs[i].devnum; | |
d6065f7b | 920 | mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[d].rdev); |
6bfe0b49 DW |
921 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
922 | atomic_inc(&rdev->nr_pending); | |
923 | blocked_rdev = rdev; | |
924 | break; | |
925 | } | |
926 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
d6065f7b | 927 | atomic_inc(&rdev->nr_pending); |
1da177e4 | 928 | r10_bio->devs[i].bio = bio; |
6cce3b23 | 929 | } else { |
1da177e4 | 930 | r10_bio->devs[i].bio = NULL; |
6cce3b23 N |
931 | set_bit(R10BIO_Degraded, &r10_bio->state); |
932 | } | |
1da177e4 LT |
933 | } |
934 | rcu_read_unlock(); | |
935 | ||
6bfe0b49 DW |
936 | if (unlikely(blocked_rdev)) { |
937 | /* Have to wait for this device to get unblocked, then retry */ | |
938 | int j; | |
939 | int d; | |
940 | ||
941 | for (j = 0; j < i; j++) | |
942 | if (r10_bio->devs[j].bio) { | |
943 | d = r10_bio->devs[j].devnum; | |
944 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
945 | } | |
946 | allow_barrier(conf); | |
947 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
948 | wait_barrier(conf); | |
949 | goto retry_write; | |
950 | } | |
951 | ||
4e78064f N |
952 | atomic_set(&r10_bio->remaining, 1); |
953 | bitmap_startwrite(mddev->bitmap, bio->bi_sector, r10_bio->sectors, 0); | |
06d91a5f | 954 | |
1da177e4 LT |
955 | for (i = 0; i < conf->copies; i++) { |
956 | struct bio *mbio; | |
957 | int d = r10_bio->devs[i].devnum; | |
958 | if (!r10_bio->devs[i].bio) | |
959 | continue; | |
960 | ||
a167f663 | 961 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1da177e4 LT |
962 | r10_bio->devs[i].bio = mbio; |
963 | ||
964 | mbio->bi_sector = r10_bio->devs[i].addr+ | |
965 | conf->mirrors[d].rdev->data_offset; | |
966 | mbio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
967 | mbio->bi_end_io = raid10_end_write_request; | |
e9c7469b | 968 | mbio->bi_rw = WRITE | do_sync | do_fua; |
1da177e4 LT |
969 | mbio->bi_private = r10_bio; |
970 | ||
971 | atomic_inc(&r10_bio->remaining); | |
4e78064f N |
972 | spin_lock_irqsave(&conf->device_lock, flags); |
973 | bio_list_add(&conf->pending_bio_list, mbio); | |
974 | blk_plug_device(mddev->queue); | |
975 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
976 | } |
977 | ||
4e78064f N |
978 | if (atomic_dec_and_test(&r10_bio->remaining)) { |
979 | /* This matches the end of raid10_end_write_request() */ | |
980 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
981 | r10_bio->sectors, | |
982 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
983 | 0); | |
f6f953aa AR |
984 | md_write_end(mddev); |
985 | raid_end_bio_io(r10_bio); | |
f6f953aa AR |
986 | } |
987 | ||
a35e63ef N |
988 | /* In case raid10d snuck in to freeze_array */ |
989 | wake_up(&conf->wait_barrier); | |
990 | ||
e3881a68 LE |
991 | if (do_sync) |
992 | md_wakeup_thread(mddev->thread); | |
993 | ||
1da177e4 LT |
994 | return 0; |
995 | } | |
996 | ||
997 | static void status(struct seq_file *seq, mddev_t *mddev) | |
998 | { | |
070ec55d | 999 | conf_t *conf = mddev->private; |
1da177e4 LT |
1000 | int i; |
1001 | ||
1002 | if (conf->near_copies < conf->raid_disks) | |
9d8f0363 | 1003 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
1da177e4 LT |
1004 | if (conf->near_copies > 1) |
1005 | seq_printf(seq, " %d near-copies", conf->near_copies); | |
c93983bf N |
1006 | if (conf->far_copies > 1) { |
1007 | if (conf->far_offset) | |
1008 | seq_printf(seq, " %d offset-copies", conf->far_copies); | |
1009 | else | |
1010 | seq_printf(seq, " %d far-copies", conf->far_copies); | |
1011 | } | |
1da177e4 | 1012 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
76186dd8 | 1013 | conf->raid_disks - mddev->degraded); |
1da177e4 LT |
1014 | for (i = 0; i < conf->raid_disks; i++) |
1015 | seq_printf(seq, "%s", | |
1016 | conf->mirrors[i].rdev && | |
b2d444d7 | 1017 | test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
1018 | seq_printf(seq, "]"); |
1019 | } | |
1020 | ||
1021 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
1022 | { | |
1023 | char b[BDEVNAME_SIZE]; | |
070ec55d | 1024 | conf_t *conf = mddev->private; |
1da177e4 LT |
1025 | |
1026 | /* | |
1027 | * If it is not operational, then we have already marked it as dead | |
1028 | * else if it is the last working disks, ignore the error, let the | |
1029 | * next level up know. | |
1030 | * else mark the drive as failed | |
1031 | */ | |
b2d444d7 | 1032 | if (test_bit(In_sync, &rdev->flags) |
76186dd8 | 1033 | && conf->raid_disks-mddev->degraded == 1) |
1da177e4 LT |
1034 | /* |
1035 | * Don't fail the drive, just return an IO error. | |
1036 | * The test should really be more sophisticated than | |
1037 | * "working_disks == 1", but it isn't critical, and | |
1038 | * can wait until we do more sophisticated "is the drive | |
1039 | * really dead" tests... | |
1040 | */ | |
1041 | return; | |
c04be0aa N |
1042 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1043 | unsigned long flags; | |
1044 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1045 | mddev->degraded++; |
c04be0aa | 1046 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1047 | /* |
1048 | * if recovery is running, make sure it aborts. | |
1049 | */ | |
dfc70645 | 1050 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1051 | } |
b2d444d7 | 1052 | set_bit(Faulty, &rdev->flags); |
850b2b42 | 1053 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
128595ed N |
1054 | printk(KERN_ALERT "md/raid10:%s: Disk failure on %s, disabling device.\n" |
1055 | KERN_ALERT "md/raid10:%s: Operation continuing on %d devices.\n", | |
1056 | mdname(mddev), bdevname(rdev->bdev, b), | |
1057 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1058 | } |
1059 | ||
1060 | static void print_conf(conf_t *conf) | |
1061 | { | |
1062 | int i; | |
1063 | mirror_info_t *tmp; | |
1064 | ||
128595ed | 1065 | printk(KERN_DEBUG "RAID10 conf printout:\n"); |
1da177e4 | 1066 | if (!conf) { |
128595ed | 1067 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1068 | return; |
1069 | } | |
128595ed | 1070 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1071 | conf->raid_disks); |
1072 | ||
1073 | for (i = 0; i < conf->raid_disks; i++) { | |
1074 | char b[BDEVNAME_SIZE]; | |
1075 | tmp = conf->mirrors + i; | |
1076 | if (tmp->rdev) | |
128595ed | 1077 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
b2d444d7 N |
1078 | i, !test_bit(In_sync, &tmp->rdev->flags), |
1079 | !test_bit(Faulty, &tmp->rdev->flags), | |
1da177e4 LT |
1080 | bdevname(tmp->rdev->bdev,b)); |
1081 | } | |
1082 | } | |
1083 | ||
1084 | static void close_sync(conf_t *conf) | |
1085 | { | |
0a27ec96 N |
1086 | wait_barrier(conf); |
1087 | allow_barrier(conf); | |
1da177e4 LT |
1088 | |
1089 | mempool_destroy(conf->r10buf_pool); | |
1090 | conf->r10buf_pool = NULL; | |
1091 | } | |
1092 | ||
6d508242 N |
1093 | /* check if there are enough drives for |
1094 | * every block to appear on atleast one | |
1095 | */ | |
1096 | static int enough(conf_t *conf) | |
1097 | { | |
1098 | int first = 0; | |
1099 | ||
1100 | do { | |
1101 | int n = conf->copies; | |
1102 | int cnt = 0; | |
1103 | while (n--) { | |
1104 | if (conf->mirrors[first].rdev) | |
1105 | cnt++; | |
1106 | first = (first+1) % conf->raid_disks; | |
1107 | } | |
1108 | if (cnt == 0) | |
1109 | return 0; | |
1110 | } while (first != 0); | |
1111 | return 1; | |
1112 | } | |
1113 | ||
1da177e4 LT |
1114 | static int raid10_spare_active(mddev_t *mddev) |
1115 | { | |
1116 | int i; | |
1117 | conf_t *conf = mddev->private; | |
1118 | mirror_info_t *tmp; | |
6b965620 N |
1119 | int count = 0; |
1120 | unsigned long flags; | |
1da177e4 LT |
1121 | |
1122 | /* | |
1123 | * Find all non-in_sync disks within the RAID10 configuration | |
1124 | * and mark them in_sync | |
1125 | */ | |
1126 | for (i = 0; i < conf->raid_disks; i++) { | |
1127 | tmp = conf->mirrors + i; | |
1128 | if (tmp->rdev | |
b2d444d7 | 1129 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 1130 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 1131 | count++; |
e6ffbcb6 | 1132 | sysfs_notify_dirent(tmp->rdev->sysfs_state); |
1da177e4 LT |
1133 | } |
1134 | } | |
6b965620 N |
1135 | spin_lock_irqsave(&conf->device_lock, flags); |
1136 | mddev->degraded -= count; | |
1137 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1138 | |
1139 | print_conf(conf); | |
6b965620 | 1140 | return count; |
1da177e4 LT |
1141 | } |
1142 | ||
1143 | ||
1144 | static int raid10_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
1145 | { | |
1146 | conf_t *conf = mddev->private; | |
199050ea | 1147 | int err = -EEXIST; |
1da177e4 LT |
1148 | int mirror; |
1149 | mirror_info_t *p; | |
6c2fce2e | 1150 | int first = 0; |
84707f38 | 1151 | int last = conf->raid_disks - 1; |
1da177e4 LT |
1152 | |
1153 | if (mddev->recovery_cp < MaxSector) | |
1154 | /* only hot-add to in-sync arrays, as recovery is | |
1155 | * very different from resync | |
1156 | */ | |
199050ea | 1157 | return -EBUSY; |
6d508242 | 1158 | if (!enough(conf)) |
199050ea | 1159 | return -EINVAL; |
1da177e4 | 1160 | |
a53a6c85 | 1161 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1162 | first = last = rdev->raid_disk; |
1da177e4 | 1163 | |
6cce3b23 | 1164 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 1165 | rdev->saved_raid_disk >= first && |
6cce3b23 N |
1166 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1167 | mirror = rdev->saved_raid_disk; | |
1168 | else | |
6c2fce2e NB |
1169 | mirror = first; |
1170 | for ( ; mirror <= last ; mirror++) | |
1da177e4 LT |
1171 | if ( !(p=conf->mirrors+mirror)->rdev) { |
1172 | ||
8f6c2e4b MP |
1173 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1174 | rdev->data_offset << 9); | |
627a2d3c N |
1175 | /* as we don't honour merge_bvec_fn, we must |
1176 | * never risk violating it, so limit | |
1177 | * ->max_segments to one lying with a single | |
1178 | * page, as a one page request is never in | |
1179 | * violation. | |
1da177e4 | 1180 | */ |
627a2d3c N |
1181 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
1182 | blk_queue_max_segments(mddev->queue, 1); | |
1183 | blk_queue_segment_boundary(mddev->queue, | |
1184 | PAGE_CACHE_SIZE - 1); | |
1185 | } | |
1da177e4 LT |
1186 | |
1187 | p->head_position = 0; | |
1188 | rdev->raid_disk = mirror; | |
199050ea | 1189 | err = 0; |
6cce3b23 N |
1190 | if (rdev->saved_raid_disk != mirror) |
1191 | conf->fullsync = 1; | |
d6065f7b | 1192 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1193 | break; |
1194 | } | |
1195 | ||
ac5e7113 | 1196 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1197 | print_conf(conf); |
199050ea | 1198 | return err; |
1da177e4 LT |
1199 | } |
1200 | ||
1201 | static int raid10_remove_disk(mddev_t *mddev, int number) | |
1202 | { | |
1203 | conf_t *conf = mddev->private; | |
1204 | int err = 0; | |
1205 | mdk_rdev_t *rdev; | |
1206 | mirror_info_t *p = conf->mirrors+ number; | |
1207 | ||
1208 | print_conf(conf); | |
1209 | rdev = p->rdev; | |
1210 | if (rdev) { | |
b2d444d7 | 1211 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1212 | atomic_read(&rdev->nr_pending)) { |
1213 | err = -EBUSY; | |
1214 | goto abort; | |
1215 | } | |
dfc70645 N |
1216 | /* Only remove faulty devices in recovery |
1217 | * is not possible. | |
1218 | */ | |
1219 | if (!test_bit(Faulty, &rdev->flags) && | |
1220 | enough(conf)) { | |
1221 | err = -EBUSY; | |
1222 | goto abort; | |
1223 | } | |
1da177e4 | 1224 | p->rdev = NULL; |
fbd568a3 | 1225 | synchronize_rcu(); |
1da177e4 LT |
1226 | if (atomic_read(&rdev->nr_pending)) { |
1227 | /* lost the race, try later */ | |
1228 | err = -EBUSY; | |
1229 | p->rdev = rdev; | |
ac5e7113 | 1230 | goto abort; |
1da177e4 | 1231 | } |
ac5e7113 | 1232 | md_integrity_register(mddev); |
1da177e4 LT |
1233 | } |
1234 | abort: | |
1235 | ||
1236 | print_conf(conf); | |
1237 | return err; | |
1238 | } | |
1239 | ||
1240 | ||
6712ecf8 | 1241 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1242 | { |
7b92813c | 1243 | r10bio_t *r10_bio = bio->bi_private; |
070ec55d | 1244 | conf_t *conf = r10_bio->mddev->private; |
1da177e4 LT |
1245 | int i,d; |
1246 | ||
1da177e4 LT |
1247 | for (i=0; i<conf->copies; i++) |
1248 | if (r10_bio->devs[i].bio == bio) | |
1249 | break; | |
b6385483 | 1250 | BUG_ON(i == conf->copies); |
1da177e4 LT |
1251 | update_head_pos(i, r10_bio); |
1252 | d = r10_bio->devs[i].devnum; | |
0eb3ff12 N |
1253 | |
1254 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1255 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
4dbcdc75 N |
1256 | else { |
1257 | atomic_add(r10_bio->sectors, | |
1258 | &conf->mirrors[d].rdev->corrected_errors); | |
1259 | if (!test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | |
1260 | md_error(r10_bio->mddev, | |
1261 | conf->mirrors[d].rdev); | |
1262 | } | |
1da177e4 LT |
1263 | |
1264 | /* for reconstruct, we always reschedule after a read. | |
1265 | * for resync, only after all reads | |
1266 | */ | |
73d5c38a | 1267 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1268 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1269 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1270 | /* we have read all the blocks, | |
1271 | * do the comparison in process context in raid10d | |
1272 | */ | |
1273 | reschedule_retry(r10_bio); | |
1274 | } | |
1da177e4 LT |
1275 | } |
1276 | ||
6712ecf8 | 1277 | static void end_sync_write(struct bio *bio, int error) |
1da177e4 LT |
1278 | { |
1279 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
7b92813c | 1280 | r10bio_t *r10_bio = bio->bi_private; |
1da177e4 | 1281 | mddev_t *mddev = r10_bio->mddev; |
070ec55d | 1282 | conf_t *conf = mddev->private; |
1da177e4 LT |
1283 | int i,d; |
1284 | ||
1da177e4 LT |
1285 | for (i = 0; i < conf->copies; i++) |
1286 | if (r10_bio->devs[i].bio == bio) | |
1287 | break; | |
1288 | d = r10_bio->devs[i].devnum; | |
1289 | ||
1290 | if (!uptodate) | |
1291 | md_error(mddev, conf->mirrors[d].rdev); | |
dfc70645 | 1292 | |
1da177e4 LT |
1293 | update_head_pos(i, r10_bio); |
1294 | ||
73d5c38a | 1295 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); |
1da177e4 LT |
1296 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1297 | if (r10_bio->master_bio == NULL) { | |
1298 | /* the primary of several recovery bios */ | |
73d5c38a | 1299 | sector_t s = r10_bio->sectors; |
1da177e4 | 1300 | put_buf(r10_bio); |
73d5c38a | 1301 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1302 | break; |
1303 | } else { | |
1304 | r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio; | |
1305 | put_buf(r10_bio); | |
1306 | r10_bio = r10_bio2; | |
1307 | } | |
1308 | } | |
1da177e4 LT |
1309 | } |
1310 | ||
1311 | /* | |
1312 | * Note: sync and recover and handled very differently for raid10 | |
1313 | * This code is for resync. | |
1314 | * For resync, we read through virtual addresses and read all blocks. | |
1315 | * If there is any error, we schedule a write. The lowest numbered | |
1316 | * drive is authoritative. | |
1317 | * However requests come for physical address, so we need to map. | |
1318 | * For every physical address there are raid_disks/copies virtual addresses, | |
1319 | * which is always are least one, but is not necessarly an integer. | |
1320 | * This means that a physical address can span multiple chunks, so we may | |
1321 | * have to submit multiple io requests for a single sync request. | |
1322 | */ | |
1323 | /* | |
1324 | * We check if all blocks are in-sync and only write to blocks that | |
1325 | * aren't in sync | |
1326 | */ | |
1327 | static void sync_request_write(mddev_t *mddev, r10bio_t *r10_bio) | |
1328 | { | |
070ec55d | 1329 | conf_t *conf = mddev->private; |
1da177e4 LT |
1330 | int i, first; |
1331 | struct bio *tbio, *fbio; | |
1332 | ||
1333 | atomic_set(&r10_bio->remaining, 1); | |
1334 | ||
1335 | /* find the first device with a block */ | |
1336 | for (i=0; i<conf->copies; i++) | |
1337 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) | |
1338 | break; | |
1339 | ||
1340 | if (i == conf->copies) | |
1341 | goto done; | |
1342 | ||
1343 | first = i; | |
1344 | fbio = r10_bio->devs[i].bio; | |
1345 | ||
1346 | /* now find blocks with errors */ | |
0eb3ff12 N |
1347 | for (i=0 ; i < conf->copies ; i++) { |
1348 | int j, d; | |
1349 | int vcnt = r10_bio->sectors >> (PAGE_SHIFT-9); | |
1da177e4 | 1350 | |
1da177e4 | 1351 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
1352 | |
1353 | if (tbio->bi_end_io != end_sync_read) | |
1354 | continue; | |
1355 | if (i == first) | |
1da177e4 | 1356 | continue; |
0eb3ff12 N |
1357 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) { |
1358 | /* We know that the bi_io_vec layout is the same for | |
1359 | * both 'first' and 'i', so we just compare them. | |
1360 | * All vec entries are PAGE_SIZE; | |
1361 | */ | |
1362 | for (j = 0; j < vcnt; j++) | |
1363 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), | |
1364 | page_address(tbio->bi_io_vec[j].bv_page), | |
1365 | PAGE_SIZE)) | |
1366 | break; | |
1367 | if (j == vcnt) | |
1368 | continue; | |
1369 | mddev->resync_mismatches += r10_bio->sectors; | |
1370 | } | |
18f08819 N |
1371 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
1372 | /* Don't fix anything. */ | |
1373 | continue; | |
1da177e4 LT |
1374 | /* Ok, we need to write this bio |
1375 | * First we need to fixup bv_offset, bv_len and | |
1376 | * bi_vecs, as the read request might have corrupted these | |
1377 | */ | |
1378 | tbio->bi_vcnt = vcnt; | |
1379 | tbio->bi_size = r10_bio->sectors << 9; | |
1380 | tbio->bi_idx = 0; | |
1381 | tbio->bi_phys_segments = 0; | |
1da177e4 LT |
1382 | tbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
1383 | tbio->bi_flags |= 1 << BIO_UPTODATE; | |
1384 | tbio->bi_next = NULL; | |
1385 | tbio->bi_rw = WRITE; | |
1386 | tbio->bi_private = r10_bio; | |
1387 | tbio->bi_sector = r10_bio->devs[i].addr; | |
1388 | ||
1389 | for (j=0; j < vcnt ; j++) { | |
1390 | tbio->bi_io_vec[j].bv_offset = 0; | |
1391 | tbio->bi_io_vec[j].bv_len = PAGE_SIZE; | |
1392 | ||
1393 | memcpy(page_address(tbio->bi_io_vec[j].bv_page), | |
1394 | page_address(fbio->bi_io_vec[j].bv_page), | |
1395 | PAGE_SIZE); | |
1396 | } | |
1397 | tbio->bi_end_io = end_sync_write; | |
1398 | ||
1399 | d = r10_bio->devs[i].devnum; | |
1400 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1401 | atomic_inc(&r10_bio->remaining); | |
1402 | md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9); | |
1403 | ||
1404 | tbio->bi_sector += conf->mirrors[d].rdev->data_offset; | |
1405 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
1406 | generic_make_request(tbio); | |
1407 | } | |
1408 | ||
1409 | done: | |
1410 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
1411 | md_done_sync(mddev, r10_bio->sectors, 1); | |
1412 | put_buf(r10_bio); | |
1413 | } | |
1414 | } | |
1415 | ||
1416 | /* | |
1417 | * Now for the recovery code. | |
1418 | * Recovery happens across physical sectors. | |
1419 | * We recover all non-is_sync drives by finding the virtual address of | |
1420 | * each, and then choose a working drive that also has that virt address. | |
1421 | * There is a separate r10_bio for each non-in_sync drive. | |
1422 | * Only the first two slots are in use. The first for reading, | |
1423 | * The second for writing. | |
1424 | * | |
1425 | */ | |
1426 | ||
1427 | static void recovery_request_write(mddev_t *mddev, r10bio_t *r10_bio) | |
1428 | { | |
070ec55d | 1429 | conf_t *conf = mddev->private; |
1da177e4 LT |
1430 | int i, d; |
1431 | struct bio *bio, *wbio; | |
1432 | ||
1433 | ||
1434 | /* move the pages across to the second bio | |
1435 | * and submit the write request | |
1436 | */ | |
1437 | bio = r10_bio->devs[0].bio; | |
1438 | wbio = r10_bio->devs[1].bio; | |
1439 | for (i=0; i < wbio->bi_vcnt; i++) { | |
1440 | struct page *p = bio->bi_io_vec[i].bv_page; | |
1441 | bio->bi_io_vec[i].bv_page = wbio->bi_io_vec[i].bv_page; | |
1442 | wbio->bi_io_vec[i].bv_page = p; | |
1443 | } | |
1444 | d = r10_bio->devs[1].devnum; | |
1445 | ||
1446 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1447 | md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9); | |
0eb3ff12 N |
1448 | if (test_bit(R10BIO_Uptodate, &r10_bio->state)) |
1449 | generic_make_request(wbio); | |
1450 | else | |
6712ecf8 | 1451 | bio_endio(wbio, -EIO); |
1da177e4 LT |
1452 | } |
1453 | ||
1454 | ||
1e50915f RB |
1455 | /* |
1456 | * Used by fix_read_error() to decay the per rdev read_errors. | |
1457 | * We halve the read error count for every hour that has elapsed | |
1458 | * since the last recorded read error. | |
1459 | * | |
1460 | */ | |
1461 | static void check_decay_read_errors(mddev_t *mddev, mdk_rdev_t *rdev) | |
1462 | { | |
1463 | struct timespec cur_time_mon; | |
1464 | unsigned long hours_since_last; | |
1465 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
1466 | ||
1467 | ktime_get_ts(&cur_time_mon); | |
1468 | ||
1469 | if (rdev->last_read_error.tv_sec == 0 && | |
1470 | rdev->last_read_error.tv_nsec == 0) { | |
1471 | /* first time we've seen a read error */ | |
1472 | rdev->last_read_error = cur_time_mon; | |
1473 | return; | |
1474 | } | |
1475 | ||
1476 | hours_since_last = (cur_time_mon.tv_sec - | |
1477 | rdev->last_read_error.tv_sec) / 3600; | |
1478 | ||
1479 | rdev->last_read_error = cur_time_mon; | |
1480 | ||
1481 | /* | |
1482 | * if hours_since_last is > the number of bits in read_errors | |
1483 | * just set read errors to 0. We do this to avoid | |
1484 | * overflowing the shift of read_errors by hours_since_last. | |
1485 | */ | |
1486 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
1487 | atomic_set(&rdev->read_errors, 0); | |
1488 | else | |
1489 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
1490 | } | |
1491 | ||
1da177e4 LT |
1492 | /* |
1493 | * This is a kernel thread which: | |
1494 | * | |
1495 | * 1. Retries failed read operations on working mirrors. | |
1496 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 1497 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
1498 | */ |
1499 | ||
6814d536 N |
1500 | static void fix_read_error(conf_t *conf, mddev_t *mddev, r10bio_t *r10_bio) |
1501 | { | |
1502 | int sect = 0; /* Offset from r10_bio->sector */ | |
1503 | int sectors = r10_bio->sectors; | |
1504 | mdk_rdev_t*rdev; | |
1e50915f | 1505 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 1506 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f RB |
1507 | |
1508 | rcu_read_lock(); | |
0544a21d PP |
1509 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
1510 | if (rdev) { /* If rdev is not NULL */ | |
1e50915f RB |
1511 | char b[BDEVNAME_SIZE]; |
1512 | int cur_read_error_count = 0; | |
1513 | ||
1e50915f RB |
1514 | bdevname(rdev->bdev, b); |
1515 | ||
1516 | if (test_bit(Faulty, &rdev->flags)) { | |
1517 | rcu_read_unlock(); | |
1518 | /* drive has already been failed, just ignore any | |
1519 | more fix_read_error() attempts */ | |
1520 | return; | |
1521 | } | |
1522 | ||
1523 | check_decay_read_errors(mddev, rdev); | |
1524 | atomic_inc(&rdev->read_errors); | |
1525 | cur_read_error_count = atomic_read(&rdev->read_errors); | |
1526 | if (cur_read_error_count > max_read_errors) { | |
1527 | rcu_read_unlock(); | |
1528 | printk(KERN_NOTICE | |
128595ed | 1529 | "md/raid10:%s: %s: Raid device exceeded " |
1e50915f RB |
1530 | "read_error threshold " |
1531 | "[cur %d:max %d]\n", | |
128595ed | 1532 | mdname(mddev), |
1e50915f RB |
1533 | b, cur_read_error_count, max_read_errors); |
1534 | printk(KERN_NOTICE | |
128595ed N |
1535 | "md/raid10:%s: %s: Failing raid " |
1536 | "device\n", mdname(mddev), b); | |
1e50915f RB |
1537 | md_error(mddev, conf->mirrors[d].rdev); |
1538 | return; | |
1539 | } | |
1540 | } | |
1541 | rcu_read_unlock(); | |
1542 | ||
6814d536 N |
1543 | while(sectors) { |
1544 | int s = sectors; | |
1545 | int sl = r10_bio->read_slot; | |
1546 | int success = 0; | |
1547 | int start; | |
1548 | ||
1549 | if (s > (PAGE_SIZE>>9)) | |
1550 | s = PAGE_SIZE >> 9; | |
1551 | ||
1552 | rcu_read_lock(); | |
1553 | do { | |
0544a21d | 1554 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
1555 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
1556 | if (rdev && | |
1557 | test_bit(In_sync, &rdev->flags)) { | |
1558 | atomic_inc(&rdev->nr_pending); | |
1559 | rcu_read_unlock(); | |
2b193363 | 1560 | success = sync_page_io(rdev, |
6814d536 N |
1561 | r10_bio->devs[sl].addr + |
1562 | sect + rdev->data_offset, | |
1563 | s<<9, | |
1564 | conf->tmppage, READ); | |
1565 | rdev_dec_pending(rdev, mddev); | |
1566 | rcu_read_lock(); | |
1567 | if (success) | |
1568 | break; | |
1569 | } | |
1570 | sl++; | |
1571 | if (sl == conf->copies) | |
1572 | sl = 0; | |
1573 | } while (!success && sl != r10_bio->read_slot); | |
1574 | rcu_read_unlock(); | |
1575 | ||
1576 | if (!success) { | |
1577 | /* Cannot read from anywhere -- bye bye array */ | |
1578 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; | |
1579 | md_error(mddev, conf->mirrors[dn].rdev); | |
1580 | break; | |
1581 | } | |
1582 | ||
1583 | start = sl; | |
1584 | /* write it back and re-read */ | |
1585 | rcu_read_lock(); | |
1586 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 1587 | char b[BDEVNAME_SIZE]; |
0544a21d | 1588 | |
6814d536 N |
1589 | if (sl==0) |
1590 | sl = conf->copies; | |
1591 | sl--; | |
1592 | d = r10_bio->devs[sl].devnum; | |
1593 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1594 | if (rdev && | |
1595 | test_bit(In_sync, &rdev->flags)) { | |
1596 | atomic_inc(&rdev->nr_pending); | |
1597 | rcu_read_unlock(); | |
1598 | atomic_add(s, &rdev->corrected_errors); | |
2b193363 | 1599 | if (sync_page_io(rdev, |
6814d536 N |
1600 | r10_bio->devs[sl].addr + |
1601 | sect + rdev->data_offset, | |
1602 | s<<9, conf->tmppage, WRITE) | |
67b8dc4b | 1603 | == 0) { |
6814d536 | 1604 | /* Well, this device is dead */ |
67b8dc4b | 1605 | printk(KERN_NOTICE |
128595ed | 1606 | "md/raid10:%s: read correction " |
67b8dc4b RB |
1607 | "write failed" |
1608 | " (%d sectors at %llu on %s)\n", | |
1609 | mdname(mddev), s, | |
1610 | (unsigned long long)(sect+ | |
1611 | rdev->data_offset), | |
1612 | bdevname(rdev->bdev, b)); | |
128595ed | 1613 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " |
67b8dc4b | 1614 | "drive\n", |
128595ed | 1615 | mdname(mddev), |
67b8dc4b | 1616 | bdevname(rdev->bdev, b)); |
6814d536 | 1617 | md_error(mddev, rdev); |
67b8dc4b | 1618 | } |
6814d536 N |
1619 | rdev_dec_pending(rdev, mddev); |
1620 | rcu_read_lock(); | |
1621 | } | |
1622 | } | |
1623 | sl = start; | |
1624 | while (sl != r10_bio->read_slot) { | |
0544a21d | 1625 | |
6814d536 N |
1626 | if (sl==0) |
1627 | sl = conf->copies; | |
1628 | sl--; | |
1629 | d = r10_bio->devs[sl].devnum; | |
1630 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1631 | if (rdev && | |
1632 | test_bit(In_sync, &rdev->flags)) { | |
1633 | char b[BDEVNAME_SIZE]; | |
1634 | atomic_inc(&rdev->nr_pending); | |
1635 | rcu_read_unlock(); | |
2b193363 | 1636 | if (sync_page_io(rdev, |
6814d536 N |
1637 | r10_bio->devs[sl].addr + |
1638 | sect + rdev->data_offset, | |
67b8dc4b RB |
1639 | s<<9, conf->tmppage, |
1640 | READ) == 0) { | |
6814d536 | 1641 | /* Well, this device is dead */ |
67b8dc4b | 1642 | printk(KERN_NOTICE |
128595ed | 1643 | "md/raid10:%s: unable to read back " |
67b8dc4b RB |
1644 | "corrected sectors" |
1645 | " (%d sectors at %llu on %s)\n", | |
1646 | mdname(mddev), s, | |
1647 | (unsigned long long)(sect+ | |
1648 | rdev->data_offset), | |
1649 | bdevname(rdev->bdev, b)); | |
128595ed N |
1650 | printk(KERN_NOTICE "md/raid10:%s: %s: failing drive\n", |
1651 | mdname(mddev), | |
67b8dc4b RB |
1652 | bdevname(rdev->bdev, b)); |
1653 | ||
6814d536 | 1654 | md_error(mddev, rdev); |
67b8dc4b | 1655 | } else { |
6814d536 | 1656 | printk(KERN_INFO |
128595ed | 1657 | "md/raid10:%s: read error corrected" |
6814d536 N |
1658 | " (%d sectors at %llu on %s)\n", |
1659 | mdname(mddev), s, | |
969b755a RD |
1660 | (unsigned long long)(sect+ |
1661 | rdev->data_offset), | |
6814d536 | 1662 | bdevname(rdev->bdev, b)); |
67b8dc4b | 1663 | } |
6814d536 N |
1664 | |
1665 | rdev_dec_pending(rdev, mddev); | |
1666 | rcu_read_lock(); | |
1667 | } | |
1668 | } | |
1669 | rcu_read_unlock(); | |
1670 | ||
1671 | sectors -= s; | |
1672 | sect += s; | |
1673 | } | |
1674 | } | |
1675 | ||
1da177e4 LT |
1676 | static void raid10d(mddev_t *mddev) |
1677 | { | |
1678 | r10bio_t *r10_bio; | |
1679 | struct bio *bio; | |
1680 | unsigned long flags; | |
070ec55d | 1681 | conf_t *conf = mddev->private; |
1da177e4 LT |
1682 | struct list_head *head = &conf->retry_list; |
1683 | int unplug=0; | |
1684 | mdk_rdev_t *rdev; | |
1685 | ||
1686 | md_check_recovery(mddev); | |
1da177e4 LT |
1687 | |
1688 | for (;;) { | |
1689 | char b[BDEVNAME_SIZE]; | |
6cce3b23 | 1690 | |
a35e63ef | 1691 | unplug += flush_pending_writes(conf); |
6cce3b23 | 1692 | |
a35e63ef N |
1693 | spin_lock_irqsave(&conf->device_lock, flags); |
1694 | if (list_empty(head)) { | |
1695 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 1696 | break; |
a35e63ef | 1697 | } |
1da177e4 LT |
1698 | r10_bio = list_entry(head->prev, r10bio_t, retry_list); |
1699 | list_del(head->prev); | |
4443ae10 | 1700 | conf->nr_queued--; |
1da177e4 LT |
1701 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1702 | ||
1703 | mddev = r10_bio->mddev; | |
070ec55d | 1704 | conf = mddev->private; |
1da177e4 LT |
1705 | if (test_bit(R10BIO_IsSync, &r10_bio->state)) { |
1706 | sync_request_write(mddev, r10_bio); | |
1707 | unplug = 1; | |
1708 | } else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1709 | recovery_request_write(mddev, r10_bio); | |
1710 | unplug = 1; | |
1711 | } else { | |
1712 | int mirror; | |
4443ae10 N |
1713 | /* we got a read error. Maybe the drive is bad. Maybe just |
1714 | * the block and we can fix it. | |
1715 | * We freeze all other IO, and try reading the block from | |
1716 | * other devices. When we find one, we re-write | |
1717 | * and check it that fixes the read error. | |
1718 | * This is all done synchronously while the array is | |
1719 | * frozen. | |
1720 | */ | |
6814d536 N |
1721 | if (mddev->ro == 0) { |
1722 | freeze_array(conf); | |
1723 | fix_read_error(conf, mddev, r10_bio); | |
1724 | unfreeze_array(conf); | |
4443ae10 N |
1725 | } |
1726 | ||
1da177e4 | 1727 | bio = r10_bio->devs[r10_bio->read_slot].bio; |
0eb3ff12 N |
1728 | r10_bio->devs[r10_bio->read_slot].bio = |
1729 | mddev->ro ? IO_BLOCKED : NULL; | |
1da177e4 LT |
1730 | mirror = read_balance(conf, r10_bio); |
1731 | if (mirror == -1) { | |
128595ed | 1732 | printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O" |
1da177e4 | 1733 | " read error for block %llu\n", |
128595ed | 1734 | mdname(mddev), |
1da177e4 LT |
1735 | bdevname(bio->bi_bdev,b), |
1736 | (unsigned long long)r10_bio->sector); | |
1737 | raid_end_bio_io(r10_bio); | |
14e71344 | 1738 | bio_put(bio); |
1da177e4 | 1739 | } else { |
2c7d46ec | 1740 | const unsigned long do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC); |
14e71344 | 1741 | bio_put(bio); |
1da177e4 LT |
1742 | rdev = conf->mirrors[mirror].rdev; |
1743 | if (printk_ratelimit()) | |
128595ed | 1744 | printk(KERN_ERR "md/raid10:%s: %s: redirecting sector %llu to" |
1da177e4 | 1745 | " another mirror\n", |
128595ed | 1746 | mdname(mddev), |
1da177e4 LT |
1747 | bdevname(rdev->bdev,b), |
1748 | (unsigned long long)r10_bio->sector); | |
a167f663 N |
1749 | bio = bio_clone_mddev(r10_bio->master_bio, |
1750 | GFP_NOIO, mddev); | |
1da177e4 LT |
1751 | r10_bio->devs[r10_bio->read_slot].bio = bio; |
1752 | bio->bi_sector = r10_bio->devs[r10_bio->read_slot].addr | |
1753 | + rdev->data_offset; | |
1754 | bio->bi_bdev = rdev->bdev; | |
7b6d91da | 1755 | bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1756 | bio->bi_private = r10_bio; |
1757 | bio->bi_end_io = raid10_end_read_request; | |
1758 | unplug = 1; | |
1759 | generic_make_request(bio); | |
1760 | } | |
1761 | } | |
1d9d5241 | 1762 | cond_resched(); |
1da177e4 | 1763 | } |
1da177e4 LT |
1764 | if (unplug) |
1765 | unplug_slaves(mddev); | |
1766 | } | |
1767 | ||
1768 | ||
1769 | static int init_resync(conf_t *conf) | |
1770 | { | |
1771 | int buffs; | |
1772 | ||
1773 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 1774 | BUG_ON(conf->r10buf_pool); |
1da177e4 LT |
1775 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
1776 | if (!conf->r10buf_pool) | |
1777 | return -ENOMEM; | |
1778 | conf->next_resync = 0; | |
1779 | return 0; | |
1780 | } | |
1781 | ||
1782 | /* | |
1783 | * perform a "sync" on one "block" | |
1784 | * | |
1785 | * We need to make sure that no normal I/O request - particularly write | |
1786 | * requests - conflict with active sync requests. | |
1787 | * | |
1788 | * This is achieved by tracking pending requests and a 'barrier' concept | |
1789 | * that can be installed to exclude normal IO requests. | |
1790 | * | |
1791 | * Resync and recovery are handled very differently. | |
1792 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
1793 | * | |
1794 | * For resync, we iterate over virtual addresses, read all copies, | |
1795 | * and update if there are differences. If only one copy is live, | |
1796 | * skip it. | |
1797 | * For recovery, we iterate over physical addresses, read a good | |
1798 | * value for each non-in_sync drive, and over-write. | |
1799 | * | |
1800 | * So, for recovery we may have several outstanding complex requests for a | |
1801 | * given address, one for each out-of-sync device. We model this by allocating | |
1802 | * a number of r10_bio structures, one for each out-of-sync device. | |
1803 | * As we setup these structures, we collect all bio's together into a list | |
1804 | * which we then process collectively to add pages, and then process again | |
1805 | * to pass to generic_make_request. | |
1806 | * | |
1807 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
1808 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
1809 | * has its remaining count decremented to 0, the whole complex operation | |
1810 | * is complete. | |
1811 | * | |
1812 | */ | |
1813 | ||
57afd89f | 1814 | static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 | 1815 | { |
070ec55d | 1816 | conf_t *conf = mddev->private; |
1da177e4 LT |
1817 | r10bio_t *r10_bio; |
1818 | struct bio *biolist = NULL, *bio; | |
1819 | sector_t max_sector, nr_sectors; | |
1820 | int disk; | |
1821 | int i; | |
6cce3b23 | 1822 | int max_sync; |
57dab0bd | 1823 | sector_t sync_blocks; |
1da177e4 LT |
1824 | |
1825 | sector_t sectors_skipped = 0; | |
1826 | int chunks_skipped = 0; | |
1827 | ||
1828 | if (!conf->r10buf_pool) | |
1829 | if (init_resync(conf)) | |
57afd89f | 1830 | return 0; |
1da177e4 LT |
1831 | |
1832 | skipped: | |
58c0fed4 | 1833 | max_sector = mddev->dev_sectors; |
1da177e4 LT |
1834 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
1835 | max_sector = mddev->resync_max_sectors; | |
1836 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
1837 | /* If we aborted, we need to abort the |
1838 | * sync on the 'current' bitmap chucks (there can | |
1839 | * be several when recovering multiple devices). | |
1840 | * as we may have started syncing it but not finished. | |
1841 | * We can find the current address in | |
1842 | * mddev->curr_resync, but for recovery, | |
1843 | * we need to convert that to several | |
1844 | * virtual addresses. | |
1845 | */ | |
1846 | if (mddev->curr_resync < max_sector) { /* aborted */ | |
1847 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
1848 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
1849 | &sync_blocks, 1); | |
1850 | else for (i=0; i<conf->raid_disks; i++) { | |
1851 | sector_t sect = | |
1852 | raid10_find_virt(conf, mddev->curr_resync, i); | |
1853 | bitmap_end_sync(mddev->bitmap, sect, | |
1854 | &sync_blocks, 1); | |
1855 | } | |
1856 | } else /* completed sync */ | |
1857 | conf->fullsync = 0; | |
1858 | ||
1859 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 1860 | close_sync(conf); |
57afd89f | 1861 | *skipped = 1; |
1da177e4 LT |
1862 | return sectors_skipped; |
1863 | } | |
1864 | if (chunks_skipped >= conf->raid_disks) { | |
1865 | /* if there has been nothing to do on any drive, | |
1866 | * then there is nothing to do at all.. | |
1867 | */ | |
57afd89f N |
1868 | *skipped = 1; |
1869 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
1870 | } |
1871 | ||
c6207277 N |
1872 | if (max_sector > mddev->resync_max) |
1873 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
1874 | ||
1da177e4 LT |
1875 | /* make sure whole request will fit in a chunk - if chunks |
1876 | * are meaningful | |
1877 | */ | |
1878 | if (conf->near_copies < conf->raid_disks && | |
1879 | max_sector > (sector_nr | conf->chunk_mask)) | |
1880 | max_sector = (sector_nr | conf->chunk_mask) + 1; | |
1881 | /* | |
1882 | * If there is non-resync activity waiting for us then | |
1883 | * put in a delay to throttle resync. | |
1884 | */ | |
0a27ec96 | 1885 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 1886 | msleep_interruptible(1000); |
1da177e4 LT |
1887 | |
1888 | /* Again, very different code for resync and recovery. | |
1889 | * Both must result in an r10bio with a list of bios that | |
1890 | * have bi_end_io, bi_sector, bi_bdev set, | |
1891 | * and bi_private set to the r10bio. | |
1892 | * For recovery, we may actually create several r10bios | |
1893 | * with 2 bios in each, that correspond to the bios in the main one. | |
1894 | * In this case, the subordinate r10bios link back through a | |
1895 | * borrowed master_bio pointer, and the counter in the master | |
1896 | * includes a ref from each subordinate. | |
1897 | */ | |
1898 | /* First, we decide what to do and set ->bi_end_io | |
1899 | * To end_sync_read if we want to read, and | |
1900 | * end_sync_write if we will want to write. | |
1901 | */ | |
1902 | ||
6cce3b23 | 1903 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
1904 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
1905 | /* recovery... the complicated one */ | |
a9f326eb | 1906 | int j, k; |
1da177e4 LT |
1907 | r10_bio = NULL; |
1908 | ||
1909 | for (i=0 ; i<conf->raid_disks; i++) | |
1910 | if (conf->mirrors[i].rdev && | |
b2d444d7 | 1911 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags)) { |
6cce3b23 | 1912 | int still_degraded = 0; |
1da177e4 LT |
1913 | /* want to reconstruct this device */ |
1914 | r10bio_t *rb2 = r10_bio; | |
6cce3b23 N |
1915 | sector_t sect = raid10_find_virt(conf, sector_nr, i); |
1916 | int must_sync; | |
1917 | /* Unless we are doing a full sync, we only need | |
1918 | * to recover the block if it is set in the bitmap | |
1919 | */ | |
1920 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
1921 | &sync_blocks, 1); | |
1922 | if (sync_blocks < max_sync) | |
1923 | max_sync = sync_blocks; | |
1924 | if (!must_sync && | |
1925 | !conf->fullsync) { | |
1926 | /* yep, skip the sync_blocks here, but don't assume | |
1927 | * that there will never be anything to do here | |
1928 | */ | |
1929 | chunks_skipped = -1; | |
1930 | continue; | |
1931 | } | |
1da177e4 LT |
1932 | |
1933 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); | |
6cce3b23 | 1934 | raise_barrier(conf, rb2 != NULL); |
1da177e4 LT |
1935 | atomic_set(&r10_bio->remaining, 0); |
1936 | ||
1937 | r10_bio->master_bio = (struct bio*)rb2; | |
1938 | if (rb2) | |
1939 | atomic_inc(&rb2->remaining); | |
1940 | r10_bio->mddev = mddev; | |
1941 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
6cce3b23 N |
1942 | r10_bio->sector = sect; |
1943 | ||
1da177e4 | 1944 | raid10_find_phys(conf, r10_bio); |
18055569 N |
1945 | |
1946 | /* Need to check if the array will still be | |
6cce3b23 N |
1947 | * degraded |
1948 | */ | |
18055569 N |
1949 | for (j=0; j<conf->raid_disks; j++) |
1950 | if (conf->mirrors[j].rdev == NULL || | |
1951 | test_bit(Faulty, &conf->mirrors[j].rdev->flags)) { | |
6cce3b23 | 1952 | still_degraded = 1; |
a24a8dd8 N |
1953 | break; |
1954 | } | |
18055569 | 1955 | |
6cce3b23 N |
1956 | must_sync = bitmap_start_sync(mddev->bitmap, sect, |
1957 | &sync_blocks, still_degraded); | |
1958 | ||
1da177e4 LT |
1959 | for (j=0; j<conf->copies;j++) { |
1960 | int d = r10_bio->devs[j].devnum; | |
1961 | if (conf->mirrors[d].rdev && | |
b2d444d7 | 1962 | test_bit(In_sync, &conf->mirrors[d].rdev->flags)) { |
1da177e4 LT |
1963 | /* This is where we read from */ |
1964 | bio = r10_bio->devs[0].bio; | |
1965 | bio->bi_next = biolist; | |
1966 | biolist = bio; | |
1967 | bio->bi_private = r10_bio; | |
1968 | bio->bi_end_io = end_sync_read; | |
802ba064 | 1969 | bio->bi_rw = READ; |
1da177e4 LT |
1970 | bio->bi_sector = r10_bio->devs[j].addr + |
1971 | conf->mirrors[d].rdev->data_offset; | |
1972 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
1973 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1974 | atomic_inc(&r10_bio->remaining); | |
1975 | /* and we write to 'i' */ | |
1976 | ||
1977 | for (k=0; k<conf->copies; k++) | |
1978 | if (r10_bio->devs[k].devnum == i) | |
1979 | break; | |
64a742bc | 1980 | BUG_ON(k == conf->copies); |
1da177e4 LT |
1981 | bio = r10_bio->devs[1].bio; |
1982 | bio->bi_next = biolist; | |
1983 | biolist = bio; | |
1984 | bio->bi_private = r10_bio; | |
1985 | bio->bi_end_io = end_sync_write; | |
802ba064 | 1986 | bio->bi_rw = WRITE; |
1da177e4 LT |
1987 | bio->bi_sector = r10_bio->devs[k].addr + |
1988 | conf->mirrors[i].rdev->data_offset; | |
1989 | bio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1990 | ||
1991 | r10_bio->devs[0].devnum = d; | |
1992 | r10_bio->devs[1].devnum = i; | |
1993 | ||
1994 | break; | |
1995 | } | |
1996 | } | |
1997 | if (j == conf->copies) { | |
87fc767b N |
1998 | /* Cannot recover, so abort the recovery */ |
1999 | put_buf(r10_bio); | |
a07e6ab4 T |
2000 | if (rb2) |
2001 | atomic_dec(&rb2->remaining); | |
87fc767b | 2002 | r10_bio = rb2; |
dfc70645 N |
2003 | if (!test_and_set_bit(MD_RECOVERY_INTR, |
2004 | &mddev->recovery)) | |
128595ed N |
2005 | printk(KERN_INFO "md/raid10:%s: insufficient " |
2006 | "working devices for recovery.\n", | |
87fc767b N |
2007 | mdname(mddev)); |
2008 | break; | |
1da177e4 LT |
2009 | } |
2010 | } | |
2011 | if (biolist == NULL) { | |
2012 | while (r10_bio) { | |
2013 | r10bio_t *rb2 = r10_bio; | |
2014 | r10_bio = (r10bio_t*) rb2->master_bio; | |
2015 | rb2->master_bio = NULL; | |
2016 | put_buf(rb2); | |
2017 | } | |
2018 | goto giveup; | |
2019 | } | |
2020 | } else { | |
2021 | /* resync. Schedule a read for every block at this virt offset */ | |
2022 | int count = 0; | |
6cce3b23 | 2023 | |
78200d45 N |
2024 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
2025 | ||
6cce3b23 N |
2026 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
2027 | &sync_blocks, mddev->degraded) && | |
2028 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2029 | /* We can skip this block */ | |
2030 | *skipped = 1; | |
2031 | return sync_blocks + sectors_skipped; | |
2032 | } | |
2033 | if (sync_blocks < max_sync) | |
2034 | max_sync = sync_blocks; | |
1da177e4 LT |
2035 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
2036 | ||
1da177e4 LT |
2037 | r10_bio->mddev = mddev; |
2038 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
2039 | raise_barrier(conf, 0); |
2040 | conf->next_resync = sector_nr; | |
1da177e4 LT |
2041 | |
2042 | r10_bio->master_bio = NULL; | |
2043 | r10_bio->sector = sector_nr; | |
2044 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
2045 | raid10_find_phys(conf, r10_bio); | |
2046 | r10_bio->sectors = (sector_nr | conf->chunk_mask) - sector_nr +1; | |
2047 | ||
2048 | for (i=0; i<conf->copies; i++) { | |
2049 | int d = r10_bio->devs[i].devnum; | |
2050 | bio = r10_bio->devs[i].bio; | |
2051 | bio->bi_end_io = NULL; | |
af03b8e4 | 2052 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
1da177e4 | 2053 | if (conf->mirrors[d].rdev == NULL || |
b2d444d7 | 2054 | test_bit(Faulty, &conf->mirrors[d].rdev->flags)) |
1da177e4 LT |
2055 | continue; |
2056 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
2057 | atomic_inc(&r10_bio->remaining); | |
2058 | bio->bi_next = biolist; | |
2059 | biolist = bio; | |
2060 | bio->bi_private = r10_bio; | |
2061 | bio->bi_end_io = end_sync_read; | |
802ba064 | 2062 | bio->bi_rw = READ; |
1da177e4 LT |
2063 | bio->bi_sector = r10_bio->devs[i].addr + |
2064 | conf->mirrors[d].rdev->data_offset; | |
2065 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
2066 | count++; | |
2067 | } | |
2068 | ||
2069 | if (count < 2) { | |
2070 | for (i=0; i<conf->copies; i++) { | |
2071 | int d = r10_bio->devs[i].devnum; | |
2072 | if (r10_bio->devs[i].bio->bi_end_io) | |
2073 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
2074 | } | |
2075 | put_buf(r10_bio); | |
2076 | biolist = NULL; | |
2077 | goto giveup; | |
2078 | } | |
2079 | } | |
2080 | ||
2081 | for (bio = biolist; bio ; bio=bio->bi_next) { | |
2082 | ||
2083 | bio->bi_flags &= ~(BIO_POOL_MASK - 1); | |
2084 | if (bio->bi_end_io) | |
2085 | bio->bi_flags |= 1 << BIO_UPTODATE; | |
2086 | bio->bi_vcnt = 0; | |
2087 | bio->bi_idx = 0; | |
2088 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
2089 | bio->bi_size = 0; |
2090 | } | |
2091 | ||
2092 | nr_sectors = 0; | |
6cce3b23 N |
2093 | if (sector_nr + max_sync < max_sector) |
2094 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
2095 | do { |
2096 | struct page *page; | |
2097 | int len = PAGE_SIZE; | |
2098 | disk = 0; | |
2099 | if (sector_nr + (len>>9) > max_sector) | |
2100 | len = (max_sector - sector_nr) << 9; | |
2101 | if (len == 0) | |
2102 | break; | |
2103 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
2104 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; | |
2105 | if (bio_add_page(bio, page, len, 0) == 0) { | |
2106 | /* stop here */ | |
2107 | struct bio *bio2; | |
2108 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
2109 | for (bio2 = biolist; bio2 && bio2 != bio; bio2 = bio2->bi_next) { | |
2110 | /* remove last page from this bio */ | |
2111 | bio2->bi_vcnt--; | |
2112 | bio2->bi_size -= len; | |
2113 | bio2->bi_flags &= ~(1<< BIO_SEG_VALID); | |
2114 | } | |
2115 | goto bio_full; | |
2116 | } | |
2117 | disk = i; | |
2118 | } | |
2119 | nr_sectors += len>>9; | |
2120 | sector_nr += len>>9; | |
2121 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
2122 | bio_full: | |
2123 | r10_bio->sectors = nr_sectors; | |
2124 | ||
2125 | while (biolist) { | |
2126 | bio = biolist; | |
2127 | biolist = biolist->bi_next; | |
2128 | ||
2129 | bio->bi_next = NULL; | |
2130 | r10_bio = bio->bi_private; | |
2131 | r10_bio->sectors = nr_sectors; | |
2132 | ||
2133 | if (bio->bi_end_io == end_sync_read) { | |
2134 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
2135 | generic_make_request(bio); | |
2136 | } | |
2137 | } | |
2138 | ||
57afd89f N |
2139 | if (sectors_skipped) |
2140 | /* pretend they weren't skipped, it makes | |
2141 | * no important difference in this case | |
2142 | */ | |
2143 | md_done_sync(mddev, sectors_skipped, 1); | |
2144 | ||
1da177e4 LT |
2145 | return sectors_skipped + nr_sectors; |
2146 | giveup: | |
2147 | /* There is nowhere to write, so all non-sync | |
2148 | * drives must be failed, so try the next chunk... | |
2149 | */ | |
09b4068a N |
2150 | if (sector_nr + max_sync < max_sector) |
2151 | max_sector = sector_nr + max_sync; | |
2152 | ||
2153 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
2154 | chunks_skipped ++; |
2155 | sector_nr = max_sector; | |
1da177e4 | 2156 | goto skipped; |
1da177e4 LT |
2157 | } |
2158 | ||
80c3a6ce DW |
2159 | static sector_t |
2160 | raid10_size(mddev_t *mddev, sector_t sectors, int raid_disks) | |
2161 | { | |
2162 | sector_t size; | |
070ec55d | 2163 | conf_t *conf = mddev->private; |
80c3a6ce DW |
2164 | |
2165 | if (!raid_disks) | |
84707f38 | 2166 | raid_disks = conf->raid_disks; |
80c3a6ce | 2167 | if (!sectors) |
dab8b292 | 2168 | sectors = conf->dev_sectors; |
80c3a6ce DW |
2169 | |
2170 | size = sectors >> conf->chunk_shift; | |
2171 | sector_div(size, conf->far_copies); | |
2172 | size = size * raid_disks; | |
2173 | sector_div(size, conf->near_copies); | |
2174 | ||
2175 | return size << conf->chunk_shift; | |
2176 | } | |
2177 | ||
dab8b292 TM |
2178 | |
2179 | static conf_t *setup_conf(mddev_t *mddev) | |
1da177e4 | 2180 | { |
dab8b292 | 2181 | conf_t *conf = NULL; |
c93983bf | 2182 | int nc, fc, fo; |
1da177e4 | 2183 | sector_t stride, size; |
dab8b292 | 2184 | int err = -EINVAL; |
1da177e4 | 2185 | |
f73ea873 MT |
2186 | if (mddev->new_chunk_sectors < (PAGE_SIZE >> 9) || |
2187 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
128595ed N |
2188 | printk(KERN_ERR "md/raid10:%s: chunk size must be " |
2189 | "at least PAGE_SIZE(%ld) and be a power of 2.\n", | |
2190 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 2191 | goto out; |
1da177e4 | 2192 | } |
2604b703 | 2193 | |
f73ea873 MT |
2194 | nc = mddev->new_layout & 255; |
2195 | fc = (mddev->new_layout >> 8) & 255; | |
2196 | fo = mddev->new_layout & (1<<16); | |
dab8b292 | 2197 | |
1da177e4 | 2198 | if ((nc*fc) <2 || (nc*fc) > mddev->raid_disks || |
f73ea873 | 2199 | (mddev->new_layout >> 17)) { |
128595ed | 2200 | printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
f73ea873 | 2201 | mdname(mddev), mddev->new_layout); |
1da177e4 LT |
2202 | goto out; |
2203 | } | |
dab8b292 TM |
2204 | |
2205 | err = -ENOMEM; | |
4443ae10 | 2206 | conf = kzalloc(sizeof(conf_t), GFP_KERNEL); |
dab8b292 | 2207 | if (!conf) |
1da177e4 | 2208 | goto out; |
dab8b292 | 2209 | |
4443ae10 | 2210 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
dab8b292 TM |
2211 | GFP_KERNEL); |
2212 | if (!conf->mirrors) | |
2213 | goto out; | |
4443ae10 N |
2214 | |
2215 | conf->tmppage = alloc_page(GFP_KERNEL); | |
2216 | if (!conf->tmppage) | |
dab8b292 TM |
2217 | goto out; |
2218 | ||
1da177e4 | 2219 | |
64a742bc | 2220 | conf->raid_disks = mddev->raid_disks; |
1da177e4 LT |
2221 | conf->near_copies = nc; |
2222 | conf->far_copies = fc; | |
2223 | conf->copies = nc*fc; | |
c93983bf | 2224 | conf->far_offset = fo; |
dab8b292 TM |
2225 | conf->chunk_mask = mddev->new_chunk_sectors - 1; |
2226 | conf->chunk_shift = ffz(~mddev->new_chunk_sectors); | |
2227 | ||
2228 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, | |
2229 | r10bio_pool_free, conf); | |
2230 | if (!conf->r10bio_pool) | |
2231 | goto out; | |
2232 | ||
58c0fed4 | 2233 | size = mddev->dev_sectors >> conf->chunk_shift; |
64a742bc N |
2234 | sector_div(size, fc); |
2235 | size = size * conf->raid_disks; | |
2236 | sector_div(size, nc); | |
2237 | /* 'size' is now the number of chunks in the array */ | |
2238 | /* calculate "used chunks per device" in 'stride' */ | |
2239 | stride = size * conf->copies; | |
af03b8e4 N |
2240 | |
2241 | /* We need to round up when dividing by raid_disks to | |
2242 | * get the stride size. | |
2243 | */ | |
2244 | stride += conf->raid_disks - 1; | |
64a742bc | 2245 | sector_div(stride, conf->raid_disks); |
dab8b292 TM |
2246 | |
2247 | conf->dev_sectors = stride << conf->chunk_shift; | |
64a742bc | 2248 | |
c93983bf | 2249 | if (fo) |
64a742bc N |
2250 | stride = 1; |
2251 | else | |
c93983bf | 2252 | sector_div(stride, fc); |
64a742bc N |
2253 | conf->stride = stride << conf->chunk_shift; |
2254 | ||
1da177e4 | 2255 | |
e7e72bf6 | 2256 | spin_lock_init(&conf->device_lock); |
dab8b292 TM |
2257 | INIT_LIST_HEAD(&conf->retry_list); |
2258 | ||
2259 | spin_lock_init(&conf->resync_lock); | |
2260 | init_waitqueue_head(&conf->wait_barrier); | |
2261 | ||
2262 | conf->thread = md_register_thread(raid10d, mddev, NULL); | |
2263 | if (!conf->thread) | |
2264 | goto out; | |
2265 | ||
dab8b292 TM |
2266 | conf->mddev = mddev; |
2267 | return conf; | |
2268 | ||
2269 | out: | |
128595ed | 2270 | printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n", |
dab8b292 TM |
2271 | mdname(mddev)); |
2272 | if (conf) { | |
2273 | if (conf->r10bio_pool) | |
2274 | mempool_destroy(conf->r10bio_pool); | |
2275 | kfree(conf->mirrors); | |
2276 | safe_put_page(conf->tmppage); | |
2277 | kfree(conf); | |
2278 | } | |
2279 | return ERR_PTR(err); | |
2280 | } | |
2281 | ||
2282 | static int run(mddev_t *mddev) | |
2283 | { | |
2284 | conf_t *conf; | |
2285 | int i, disk_idx, chunk_size; | |
2286 | mirror_info_t *disk; | |
2287 | mdk_rdev_t *rdev; | |
2288 | sector_t size; | |
2289 | ||
2290 | /* | |
2291 | * copy the already verified devices into our private RAID10 | |
2292 | * bookkeeping area. [whatever we allocate in run(), | |
2293 | * should be freed in stop()] | |
2294 | */ | |
2295 | ||
2296 | if (mddev->private == NULL) { | |
2297 | conf = setup_conf(mddev); | |
2298 | if (IS_ERR(conf)) | |
2299 | return PTR_ERR(conf); | |
2300 | mddev->private = conf; | |
2301 | } | |
2302 | conf = mddev->private; | |
2303 | if (!conf) | |
2304 | goto out; | |
2305 | ||
e7e72bf6 NB |
2306 | mddev->queue->queue_lock = &conf->device_lock; |
2307 | ||
dab8b292 TM |
2308 | mddev->thread = conf->thread; |
2309 | conf->thread = NULL; | |
2310 | ||
8f6c2e4b MP |
2311 | chunk_size = mddev->chunk_sectors << 9; |
2312 | blk_queue_io_min(mddev->queue, chunk_size); | |
2313 | if (conf->raid_disks % conf->near_copies) | |
2314 | blk_queue_io_opt(mddev->queue, chunk_size * conf->raid_disks); | |
2315 | else | |
2316 | blk_queue_io_opt(mddev->queue, chunk_size * | |
2317 | (conf->raid_disks / conf->near_copies)); | |
2318 | ||
159ec1fc | 2319 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
1da177e4 | 2320 | disk_idx = rdev->raid_disk; |
84707f38 | 2321 | if (disk_idx >= conf->raid_disks |
1da177e4 LT |
2322 | || disk_idx < 0) |
2323 | continue; | |
2324 | disk = conf->mirrors + disk_idx; | |
2325 | ||
2326 | disk->rdev = rdev; | |
8f6c2e4b MP |
2327 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2328 | rdev->data_offset << 9); | |
1da177e4 | 2329 | /* as we don't honour merge_bvec_fn, we must never risk |
627a2d3c N |
2330 | * violating it, so limit max_segments to 1 lying |
2331 | * within a single page. | |
1da177e4 | 2332 | */ |
627a2d3c N |
2333 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
2334 | blk_queue_max_segments(mddev->queue, 1); | |
2335 | blk_queue_segment_boundary(mddev->queue, | |
2336 | PAGE_CACHE_SIZE - 1); | |
2337 | } | |
1da177e4 LT |
2338 | |
2339 | disk->head_position = 0; | |
1da177e4 | 2340 | } |
6d508242 N |
2341 | /* need to check that every block has at least one working mirror */ |
2342 | if (!enough(conf)) { | |
128595ed | 2343 | printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 2344 | mdname(mddev)); |
1da177e4 LT |
2345 | goto out_free_conf; |
2346 | } | |
2347 | ||
2348 | mddev->degraded = 0; | |
2349 | for (i = 0; i < conf->raid_disks; i++) { | |
2350 | ||
2351 | disk = conf->mirrors + i; | |
2352 | ||
5fd6c1dc | 2353 | if (!disk->rdev || |
2e333e89 | 2354 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
2355 | disk->head_position = 0; |
2356 | mddev->degraded++; | |
8c2e870a NB |
2357 | if (disk->rdev) |
2358 | conf->fullsync = 1; | |
1da177e4 LT |
2359 | } |
2360 | } | |
2361 | ||
8c6ac868 | 2362 | if (mddev->recovery_cp != MaxSector) |
128595ed | 2363 | printk(KERN_NOTICE "md/raid10:%s: not clean" |
8c6ac868 AN |
2364 | " -- starting background reconstruction\n", |
2365 | mdname(mddev)); | |
1da177e4 | 2366 | printk(KERN_INFO |
128595ed | 2367 | "md/raid10:%s: active with %d out of %d devices\n", |
84707f38 N |
2368 | mdname(mddev), conf->raid_disks - mddev->degraded, |
2369 | conf->raid_disks); | |
1da177e4 LT |
2370 | /* |
2371 | * Ok, everything is just fine now | |
2372 | */ | |
dab8b292 TM |
2373 | mddev->dev_sectors = conf->dev_sectors; |
2374 | size = raid10_size(mddev, 0, 0); | |
2375 | md_set_array_sectors(mddev, size); | |
2376 | mddev->resync_max_sectors = size; | |
1da177e4 | 2377 | |
7a5febe9 | 2378 | mddev->queue->unplug_fn = raid10_unplug; |
0d129228 N |
2379 | mddev->queue->backing_dev_info.congested_fn = raid10_congested; |
2380 | mddev->queue->backing_dev_info.congested_data = mddev; | |
7a5febe9 | 2381 | |
1da177e4 LT |
2382 | /* Calculate max read-ahead size. |
2383 | * We need to readahead at least twice a whole stripe.... | |
2384 | * maybe... | |
2385 | */ | |
2386 | { | |
9d8f0363 AN |
2387 | int stripe = conf->raid_disks * |
2388 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
1da177e4 LT |
2389 | stripe /= conf->near_copies; |
2390 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) | |
2391 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
2392 | } | |
2393 | ||
84707f38 | 2394 | if (conf->near_copies < conf->raid_disks) |
1da177e4 | 2395 | blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec); |
ac5e7113 | 2396 | md_integrity_register(mddev); |
1da177e4 LT |
2397 | return 0; |
2398 | ||
2399 | out_free_conf: | |
589a594b | 2400 | md_unregister_thread(mddev->thread); |
1da177e4 LT |
2401 | if (conf->r10bio_pool) |
2402 | mempool_destroy(conf->r10bio_pool); | |
1345b1d8 | 2403 | safe_put_page(conf->tmppage); |
990a8baf | 2404 | kfree(conf->mirrors); |
1da177e4 LT |
2405 | kfree(conf); |
2406 | mddev->private = NULL; | |
2407 | out: | |
2408 | return -EIO; | |
2409 | } | |
2410 | ||
2411 | static int stop(mddev_t *mddev) | |
2412 | { | |
070ec55d | 2413 | conf_t *conf = mddev->private; |
1da177e4 | 2414 | |
409c57f3 N |
2415 | raise_barrier(conf, 0); |
2416 | lower_barrier(conf); | |
2417 | ||
1da177e4 LT |
2418 | md_unregister_thread(mddev->thread); |
2419 | mddev->thread = NULL; | |
2420 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ | |
2421 | if (conf->r10bio_pool) | |
2422 | mempool_destroy(conf->r10bio_pool); | |
990a8baf | 2423 | kfree(conf->mirrors); |
1da177e4 LT |
2424 | kfree(conf); |
2425 | mddev->private = NULL; | |
2426 | return 0; | |
2427 | } | |
2428 | ||
6cce3b23 N |
2429 | static void raid10_quiesce(mddev_t *mddev, int state) |
2430 | { | |
070ec55d | 2431 | conf_t *conf = mddev->private; |
6cce3b23 N |
2432 | |
2433 | switch(state) { | |
2434 | case 1: | |
2435 | raise_barrier(conf, 0); | |
2436 | break; | |
2437 | case 0: | |
2438 | lower_barrier(conf); | |
2439 | break; | |
2440 | } | |
6cce3b23 | 2441 | } |
1da177e4 | 2442 | |
dab8b292 TM |
2443 | static void *raid10_takeover_raid0(mddev_t *mddev) |
2444 | { | |
2445 | mdk_rdev_t *rdev; | |
2446 | conf_t *conf; | |
2447 | ||
2448 | if (mddev->degraded > 0) { | |
128595ed N |
2449 | printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n", |
2450 | mdname(mddev)); | |
dab8b292 TM |
2451 | return ERR_PTR(-EINVAL); |
2452 | } | |
2453 | ||
dab8b292 TM |
2454 | /* Set new parameters */ |
2455 | mddev->new_level = 10; | |
2456 | /* new layout: far_copies = 1, near_copies = 2 */ | |
2457 | mddev->new_layout = (1<<8) + 2; | |
2458 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
2459 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
2460 | mddev->raid_disks *= 2; |
2461 | /* make sure it will be not marked as dirty */ | |
2462 | mddev->recovery_cp = MaxSector; | |
2463 | ||
2464 | conf = setup_conf(mddev); | |
e93f68a1 N |
2465 | if (!IS_ERR(conf)) |
2466 | list_for_each_entry(rdev, &mddev->disks, same_set) | |
2467 | if (rdev->raid_disk >= 0) | |
2468 | rdev->new_raid_disk = rdev->raid_disk * 2; | |
2469 | ||
dab8b292 TM |
2470 | return conf; |
2471 | } | |
2472 | ||
2473 | static void *raid10_takeover(mddev_t *mddev) | |
2474 | { | |
2475 | struct raid0_private_data *raid0_priv; | |
2476 | ||
2477 | /* raid10 can take over: | |
2478 | * raid0 - providing it has only two drives | |
2479 | */ | |
2480 | if (mddev->level == 0) { | |
2481 | /* for raid0 takeover only one zone is supported */ | |
2482 | raid0_priv = mddev->private; | |
2483 | if (raid0_priv->nr_strip_zones > 1) { | |
128595ed N |
2484 | printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0" |
2485 | " with more than one zone.\n", | |
2486 | mdname(mddev)); | |
dab8b292 TM |
2487 | return ERR_PTR(-EINVAL); |
2488 | } | |
2489 | return raid10_takeover_raid0(mddev); | |
2490 | } | |
2491 | return ERR_PTR(-EINVAL); | |
2492 | } | |
2493 | ||
2604b703 | 2494 | static struct mdk_personality raid10_personality = |
1da177e4 LT |
2495 | { |
2496 | .name = "raid10", | |
2604b703 | 2497 | .level = 10, |
1da177e4 LT |
2498 | .owner = THIS_MODULE, |
2499 | .make_request = make_request, | |
2500 | .run = run, | |
2501 | .stop = stop, | |
2502 | .status = status, | |
2503 | .error_handler = error, | |
2504 | .hot_add_disk = raid10_add_disk, | |
2505 | .hot_remove_disk= raid10_remove_disk, | |
2506 | .spare_active = raid10_spare_active, | |
2507 | .sync_request = sync_request, | |
6cce3b23 | 2508 | .quiesce = raid10_quiesce, |
80c3a6ce | 2509 | .size = raid10_size, |
dab8b292 | 2510 | .takeover = raid10_takeover, |
1da177e4 LT |
2511 | }; |
2512 | ||
2513 | static int __init raid_init(void) | |
2514 | { | |
2604b703 | 2515 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
2516 | } |
2517 | ||
2518 | static void raid_exit(void) | |
2519 | { | |
2604b703 | 2520 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
2521 | } |
2522 | ||
2523 | module_init(raid_init); | |
2524 | module_exit(raid_exit); | |
2525 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 2526 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 2527 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 2528 | MODULE_ALIAS("md-raid10"); |
2604b703 | 2529 | MODULE_ALIAS("md-level-10"); |