blob: b4ba41ecbd2030fa9e8d0e4ed765b5a8d133ee84 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Stephen Rothwell25570722008-10-15 09:09:21 +110021#include <linux/delay.h>
NeilBrownbff61972009-03-31 14:33:13 +110022#include <linux/blkdev.h>
NeilBrownbff61972009-03-31 14:33:13 +110023#include <linux/seq_file.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110024#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110025#include "raid10.h"
26#include "bitmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28/*
29 * RAID10 provides a combination of RAID0 and RAID1 functionality.
30 * The layout of data is defined by
31 * chunk_size
32 * raid_disks
33 * near_copies (stored in low byte of layout)
34 * far_copies (stored in second byte of layout)
NeilBrownc93983b2006-06-26 00:27:41 -070035 * far_offset (stored in bit 16 of layout )
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 *
37 * The data to be stored is divided into chunks using chunksize.
38 * Each device is divided into far_copies sections.
39 * In each section, chunks are laid out in a style similar to raid0, but
40 * near_copies copies of each chunk is stored (each on a different drive).
41 * The starting device for each section is offset near_copies from the starting
42 * device of the previous section.
NeilBrownc93983b2006-06-26 00:27:41 -070043 * Thus they are (near_copies*far_copies) of each chunk, and each is on a different
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * drive.
45 * near_copies and far_copies must be at least one, and their product is at most
46 * raid_disks.
NeilBrownc93983b2006-06-26 00:27:41 -070047 *
48 * If far_offset is true, then the far_copies are handled a bit differently.
49 * The copies are still in different stripes, but instead of be very far apart
50 * on disk, there are adjacent stripes.
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 */
52
53/*
54 * Number of guaranteed r10bios in case of extreme VM load:
55 */
56#define NR_RAID10_BIOS 256
57
58static void unplug_slaves(mddev_t *mddev);
59
NeilBrown0a27ec92006-01-06 00:20:13 -080060static void allow_barrier(conf_t *conf);
61static void lower_barrier(conf_t *conf);
62
Al Virodd0fc662005-10-07 07:46:04 +010063static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064{
65 conf_t *conf = data;
66 r10bio_t *r10_bio;
67 int size = offsetof(struct r10bio_s, devs[conf->copies]);
68
69 /* allocate a r10bio with room for raid_disks entries in the bios array */
NeilBrown9ffae0c2006-01-06 00:20:32 -080070 r10_bio = kzalloc(size, gfp_flags);
NeilBrowned9bfdf2009-10-16 15:55:44 +110071 if (!r10_bio && conf->mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 unplug_slaves(conf->mddev);
73
74 return r10_bio;
75}
76
77static void r10bio_pool_free(void *r10_bio, void *data)
78{
79 kfree(r10_bio);
80}
81
NeilBrown0310fa22008-08-05 15:54:14 +100082/* Maximum size of each resync request */
Linus Torvalds1da177e2005-04-16 15:20:36 -070083#define RESYNC_BLOCK_SIZE (64*1024)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
NeilBrown0310fa22008-08-05 15:54:14 +100085/* amount of memory to reserve for resync requests */
86#define RESYNC_WINDOW (1024*1024)
87/* maximum number of concurrent requests, memory permitting */
88#define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
90/*
91 * When performing a resync, we need to read and compare, so
92 * we need as many pages are there are copies.
93 * When performing a recovery, we need 2 bios, one for read,
94 * one for write (we recover only one drive per r10buf)
95 *
96 */
Al Virodd0fc662005-10-07 07:46:04 +010097static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
99 conf_t *conf = data;
100 struct page *page;
101 r10bio_t *r10_bio;
102 struct bio *bio;
103 int i, j;
104 int nalloc;
105
106 r10_bio = r10bio_pool_alloc(gfp_flags, conf);
107 if (!r10_bio) {
108 unplug_slaves(conf->mddev);
109 return NULL;
110 }
111
112 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery))
113 nalloc = conf->copies; /* resync */
114 else
115 nalloc = 2; /* recovery */
116
117 /*
118 * Allocate bios.
119 */
120 for (j = nalloc ; j-- ; ) {
121 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
122 if (!bio)
123 goto out_free_bio;
124 r10_bio->devs[j].bio = bio;
125 }
126 /*
127 * Allocate RESYNC_PAGES data pages and attach them
128 * where needed.
129 */
130 for (j = 0 ; j < nalloc; j++) {
131 bio = r10_bio->devs[j].bio;
132 for (i = 0; i < RESYNC_PAGES; i++) {
133 page = alloc_page(gfp_flags);
134 if (unlikely(!page))
135 goto out_free_pages;
136
137 bio->bi_io_vec[i].bv_page = page;
138 }
139 }
140
141 return r10_bio;
142
143out_free_pages:
144 for ( ; i > 0 ; i--)
NeilBrown1345b1d2006-01-06 00:20:40 -0800145 safe_put_page(bio->bi_io_vec[i-1].bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 while (j--)
147 for (i = 0; i < RESYNC_PAGES ; i++)
NeilBrown1345b1d2006-01-06 00:20:40 -0800148 safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 j = -1;
150out_free_bio:
151 while ( ++j < nalloc )
152 bio_put(r10_bio->devs[j].bio);
153 r10bio_pool_free(r10_bio, conf);
154 return NULL;
155}
156
157static void r10buf_pool_free(void *__r10_bio, void *data)
158{
159 int i;
160 conf_t *conf = data;
161 r10bio_t *r10bio = __r10_bio;
162 int j;
163
164 for (j=0; j < conf->copies; j++) {
165 struct bio *bio = r10bio->devs[j].bio;
166 if (bio) {
167 for (i = 0; i < RESYNC_PAGES; i++) {
NeilBrown1345b1d2006-01-06 00:20:40 -0800168 safe_put_page(bio->bi_io_vec[i].bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 bio->bi_io_vec[i].bv_page = NULL;
170 }
171 bio_put(bio);
172 }
173 }
174 r10bio_pool_free(r10bio, conf);
175}
176
177static void put_all_bios(conf_t *conf, r10bio_t *r10_bio)
178{
179 int i;
180
181 for (i = 0; i < conf->copies; i++) {
182 struct bio **bio = & r10_bio->devs[i].bio;
NeilBrown0eb3ff12006-01-06 00:20:29 -0800183 if (*bio && *bio != IO_BLOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 bio_put(*bio);
185 *bio = NULL;
186 }
187}
188
Arjan van de Ven858119e2006-01-14 13:20:43 -0800189static void free_r10bio(r10bio_t *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
NeilBrown070ec552009-06-16 16:54:21 +1000191 conf_t *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
193 /*
194 * Wake up any possible resync thread that waits for the device
195 * to go idle.
196 */
NeilBrown0a27ec92006-01-06 00:20:13 -0800197 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
199 put_all_bios(conf, r10_bio);
200 mempool_free(r10_bio, conf->r10bio_pool);
201}
202
Arjan van de Ven858119e2006-01-14 13:20:43 -0800203static void put_buf(r10bio_t *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
NeilBrown070ec552009-06-16 16:54:21 +1000205 conf_t *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207 mempool_free(r10_bio, conf->r10buf_pool);
208
NeilBrown0a27ec92006-01-06 00:20:13 -0800209 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
211
212static void reschedule_retry(r10bio_t *r10_bio)
213{
214 unsigned long flags;
215 mddev_t *mddev = r10_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +1000216 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218 spin_lock_irqsave(&conf->device_lock, flags);
219 list_add(&r10_bio->retry_list, &conf->retry_list);
NeilBrown4443ae12006-01-06 00:20:28 -0800220 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 spin_unlock_irqrestore(&conf->device_lock, flags);
222
Arthur Jones388667b2008-07-25 12:03:38 -0700223 /* wake up frozen array... */
224 wake_up(&conf->wait_barrier);
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 md_wakeup_thread(mddev->thread);
227}
228
229/*
230 * raid_end_bio_io() is called when we have finished servicing a mirrored
231 * operation and are ready to return a success/failure code to the buffer
232 * cache layer.
233 */
234static void raid_end_bio_io(r10bio_t *r10_bio)
235{
236 struct bio *bio = r10_bio->master_bio;
237
NeilBrown6712ecf2007-09-27 12:47:43 +0200238 bio_endio(bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 test_bit(R10BIO_Uptodate, &r10_bio->state) ? 0 : -EIO);
240 free_r10bio(r10_bio);
241}
242
243/*
244 * Update disk head position estimator based on IRQ completion info.
245 */
246static inline void update_head_pos(int slot, r10bio_t *r10_bio)
247{
NeilBrown070ec552009-06-16 16:54:21 +1000248 conf_t *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 conf->mirrors[r10_bio->devs[slot].devnum].head_position =
251 r10_bio->devs[slot].addr + (r10_bio->sectors);
252}
253
NeilBrown6712ecf2007-09-27 12:47:43 +0200254static void raid10_end_read_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
257 r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private);
258 int slot, dev;
NeilBrown070ec552009-06-16 16:54:21 +1000259 conf_t *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262 slot = r10_bio->read_slot;
263 dev = r10_bio->devs[slot].devnum;
264 /*
265 * this branch is our 'one mirror IO has finished' event handler:
266 */
NeilBrown4443ae12006-01-06 00:20:28 -0800267 update_head_pos(slot, r10_bio);
268
269 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 /*
271 * Set R10BIO_Uptodate in our master bio, so that
272 * we will return a good error code to the higher
273 * levels even if IO on some other mirrored buffer fails.
274 *
275 * The 'master' represents the composite IO operation to
276 * user-side. So if something waits for IO, then it will
277 * wait for the 'master' bio.
278 */
279 set_bit(R10BIO_Uptodate, &r10_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 raid_end_bio_io(r10_bio);
NeilBrown4443ae12006-01-06 00:20:28 -0800281 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 /*
283 * oops, read error:
284 */
285 char b[BDEVNAME_SIZE];
286 if (printk_ratelimit())
287 printk(KERN_ERR "raid10: %s: rescheduling sector %llu\n",
288 bdevname(conf->mirrors[dev].rdev->bdev,b), (unsigned long long)r10_bio->sector);
289 reschedule_retry(r10_bio);
290 }
291
292 rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293}
294
NeilBrown6712ecf2007-09-27 12:47:43 +0200295static void raid10_end_write_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
297 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
298 r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private);
299 int slot, dev;
NeilBrown070ec552009-06-16 16:54:21 +1000300 conf_t *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 for (slot = 0; slot < conf->copies; slot++)
303 if (r10_bio->devs[slot].bio == bio)
304 break;
305 dev = r10_bio->devs[slot].devnum;
306
307 /*
308 * this branch is our 'one mirror IO has finished' event handler:
309 */
NeilBrown6cce3b22006-01-06 00:20:16 -0800310 if (!uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 md_error(r10_bio->mddev, conf->mirrors[dev].rdev);
NeilBrown6cce3b22006-01-06 00:20:16 -0800312 /* an I/O failed, we can't clear the bitmap */
313 set_bit(R10BIO_Degraded, &r10_bio->state);
314 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 /*
316 * Set R10BIO_Uptodate in our master bio, so that
317 * we will return a good error code for to the higher
318 * levels even if IO on some other mirrored buffer fails.
319 *
320 * The 'master' represents the composite IO operation to
321 * user-side. So if something waits for IO, then it will
322 * wait for the 'master' bio.
323 */
324 set_bit(R10BIO_Uptodate, &r10_bio->state);
325
326 update_head_pos(slot, r10_bio);
327
328 /*
329 *
330 * Let's see if all mirrored write operations have finished
331 * already.
332 */
333 if (atomic_dec_and_test(&r10_bio->remaining)) {
NeilBrown6cce3b22006-01-06 00:20:16 -0800334 /* clear the bitmap if all writes complete successfully */
335 bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector,
336 r10_bio->sectors,
337 !test_bit(R10BIO_Degraded, &r10_bio->state),
338 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 md_write_end(r10_bio->mddev);
340 raid_end_bio_io(r10_bio);
341 }
342
343 rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
346
347/*
348 * RAID10 layout manager
349 * Aswell as the chunksize and raid_disks count, there are two
350 * parameters: near_copies and far_copies.
351 * near_copies * far_copies must be <= raid_disks.
352 * Normally one of these will be 1.
353 * If both are 1, we get raid0.
354 * If near_copies == raid_disks, we get raid1.
355 *
356 * Chunks are layed out in raid0 style with near_copies copies of the
357 * first chunk, followed by near_copies copies of the next chunk and
358 * so on.
359 * If far_copies > 1, then after 1/far_copies of the array has been assigned
360 * as described above, we start again with a device offset of near_copies.
361 * So we effectively have another copy of the whole array further down all
362 * the drives, but with blocks on different drives.
363 * With this layout, and block is never stored twice on the one device.
364 *
365 * raid10_find_phys finds the sector offset of a given virtual sector
NeilBrownc93983b2006-06-26 00:27:41 -0700366 * on each device that it is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 *
368 * raid10_find_virt does the reverse mapping, from a device and a
369 * sector offset to a virtual address
370 */
371
372static void raid10_find_phys(conf_t *conf, r10bio_t *r10bio)
373{
374 int n,f;
375 sector_t sector;
376 sector_t chunk;
377 sector_t stripe;
378 int dev;
379
380 int slot = 0;
381
382 /* now calculate first sector/dev */
383 chunk = r10bio->sector >> conf->chunk_shift;
384 sector = r10bio->sector & conf->chunk_mask;
385
386 chunk *= conf->near_copies;
387 stripe = chunk;
388 dev = sector_div(stripe, conf->raid_disks);
NeilBrownc93983b2006-06-26 00:27:41 -0700389 if (conf->far_offset)
390 stripe *= conf->far_copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 sector += stripe << conf->chunk_shift;
393
394 /* and calculate all the others */
395 for (n=0; n < conf->near_copies; n++) {
396 int d = dev;
397 sector_t s = sector;
398 r10bio->devs[slot].addr = sector;
399 r10bio->devs[slot].devnum = d;
400 slot++;
401
402 for (f = 1; f < conf->far_copies; f++) {
403 d += conf->near_copies;
404 if (d >= conf->raid_disks)
405 d -= conf->raid_disks;
406 s += conf->stride;
407 r10bio->devs[slot].devnum = d;
408 r10bio->devs[slot].addr = s;
409 slot++;
410 }
411 dev++;
412 if (dev >= conf->raid_disks) {
413 dev = 0;
414 sector += (conf->chunk_mask + 1);
415 }
416 }
417 BUG_ON(slot != conf->copies);
418}
419
420static sector_t raid10_find_virt(conf_t *conf, sector_t sector, int dev)
421{
422 sector_t offset, chunk, vchunk;
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 offset = sector & conf->chunk_mask;
NeilBrownc93983b2006-06-26 00:27:41 -0700425 if (conf->far_offset) {
426 int fc;
427 chunk = sector >> conf->chunk_shift;
428 fc = sector_div(chunk, conf->far_copies);
429 dev -= fc * conf->near_copies;
430 if (dev < 0)
431 dev += conf->raid_disks;
432 } else {
NeilBrown64a742b2007-02-28 20:11:18 -0800433 while (sector >= conf->stride) {
NeilBrownc93983b2006-06-26 00:27:41 -0700434 sector -= conf->stride;
435 if (dev < conf->near_copies)
436 dev += conf->raid_disks - conf->near_copies;
437 else
438 dev -= conf->near_copies;
439 }
440 chunk = sector >> conf->chunk_shift;
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 vchunk = chunk * conf->raid_disks + dev;
443 sector_div(vchunk, conf->near_copies);
444 return (vchunk << conf->chunk_shift) + offset;
445}
446
447/**
448 * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
449 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200450 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 * @biovec: the request that could be merged to it.
452 *
453 * Return amount of bytes we can accept at this offset
454 * If near_copies == raid_disk, there are no striping issues,
455 * but in that case, the function isn't called at all.
456 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200457static int raid10_mergeable_bvec(struct request_queue *q,
458 struct bvec_merge_data *bvm,
459 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
461 mddev_t *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200462 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +1000464 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200465 unsigned int bio_sectors = bvm->bi_size >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467 max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
468 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200469 if (max <= biovec->bv_len && bio_sectors == 0)
470 return biovec->bv_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 else
472 return max;
473}
474
475/*
476 * This routine returns the disk from which the requested read should
477 * be done. There is a per-array 'next expected sequential IO' sector
478 * number - if this matches on the next IO then we use the last disk.
479 * There is also a per-disk 'last know head position' sector that is
480 * maintained from IRQ contexts, both the normal and the resync IO
481 * completion handlers update this position correctly. If there is no
482 * perfect sequential match then we pick the disk whose head is closest.
483 *
484 * If there are 2 mirrors in the same 2 devices, performance degrades
485 * because position is mirror, not device based.
486 *
487 * The rdev for the device selected will have nr_pending incremented.
488 */
489
490/*
491 * FIXME: possibly should rethink readbalancing and do it differently
492 * depending on near_copies / far_copies geometry.
493 */
494static int read_balance(conf_t *conf, r10bio_t *r10_bio)
495{
496 const unsigned long this_sector = r10_bio->sector;
497 int disk, slot, nslot;
498 const int sectors = r10_bio->sectors;
499 sector_t new_distance, current_distance;
Suzanne Woodd6065f72005-11-08 21:39:27 -0800500 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
502 raid10_find_phys(conf, r10_bio);
503 rcu_read_lock();
504 /*
505 * Check if we can balance. We can balance on the whole
NeilBrown6cce3b22006-01-06 00:20:16 -0800506 * device if no resync is going on (recovery is ok), or below
507 * the resync window. We take the first readable disk when
508 * above the resync window.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 */
510 if (conf->mddev->recovery_cp < MaxSector
511 && (this_sector + sectors >= conf->next_resync)) {
512 /* make sure that disk is operational */
513 slot = 0;
514 disk = r10_bio->devs[slot].devnum;
515
Suzanne Woodd6065f72005-11-08 21:39:27 -0800516 while ((rdev = rcu_dereference(conf->mirrors[disk].rdev)) == NULL ||
NeilBrown0eb3ff12006-01-06 00:20:29 -0800517 r10_bio->devs[slot].bio == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800518 !test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 slot++;
520 if (slot == conf->copies) {
521 slot = 0;
522 disk = -1;
523 break;
524 }
525 disk = r10_bio->devs[slot].devnum;
526 }
527 goto rb_out;
528 }
529
530
531 /* make sure the disk is operational */
532 slot = 0;
533 disk = r10_bio->devs[slot].devnum;
Suzanne Woodd6065f72005-11-08 21:39:27 -0800534 while ((rdev=rcu_dereference(conf->mirrors[disk].rdev)) == NULL ||
NeilBrown0eb3ff12006-01-06 00:20:29 -0800535 r10_bio->devs[slot].bio == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800536 !test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 slot ++;
538 if (slot == conf->copies) {
539 disk = -1;
540 goto rb_out;
541 }
542 disk = r10_bio->devs[slot].devnum;
543 }
544
545
NeilBrown3ec67ac2005-09-09 16:23:40 -0700546 current_distance = abs(r10_bio->devs[slot].addr -
547 conf->mirrors[disk].head_position);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
Keld Simonsen8ed3a192008-03-04 14:29:34 -0800549 /* Find the disk whose head is closest,
550 * or - for far > 1 - find the closest to partition beginning */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552 for (nslot = slot; nslot < conf->copies; nslot++) {
553 int ndisk = r10_bio->devs[nslot].devnum;
554
555
Suzanne Woodd6065f72005-11-08 21:39:27 -0800556 if ((rdev=rcu_dereference(conf->mirrors[ndisk].rdev)) == NULL ||
NeilBrown0eb3ff12006-01-06 00:20:29 -0800557 r10_bio->devs[nslot].bio == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800558 !test_bit(In_sync, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 continue;
560
NeilBrown22dfdf52005-11-28 13:44:09 -0800561 /* This optimisation is debatable, and completely destroys
562 * sequential read speed for 'far copies' arrays. So only
563 * keep it for 'near' arrays, and review those later.
564 */
565 if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 disk = ndisk;
567 slot = nslot;
568 break;
569 }
Keld Simonsen8ed3a192008-03-04 14:29:34 -0800570
571 /* for far > 1 always use the lowest address */
572 if (conf->far_copies > 1)
573 new_distance = r10_bio->devs[nslot].addr;
574 else
575 new_distance = abs(r10_bio->devs[nslot].addr -
576 conf->mirrors[ndisk].head_position);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 if (new_distance < current_distance) {
578 current_distance = new_distance;
579 disk = ndisk;
580 slot = nslot;
581 }
582 }
583
584rb_out:
585 r10_bio->read_slot = slot;
586/* conf->next_seq_sect = this_sector + sectors;*/
587
Suzanne Woodd6065f72005-11-08 21:39:27 -0800588 if (disk >= 0 && (rdev=rcu_dereference(conf->mirrors[disk].rdev))!= NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 atomic_inc(&conf->mirrors[disk].rdev->nr_pending);
NeilBrown29fc7e32006-02-03 03:03:41 -0800590 else
591 disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 rcu_read_unlock();
593
594 return disk;
595}
596
597static void unplug_slaves(mddev_t *mddev)
598{
NeilBrown070ec552009-06-16 16:54:21 +1000599 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 int i;
601
602 rcu_read_lock();
603 for (i=0; i<mddev->raid_disks; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800604 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -0800605 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200606 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 atomic_inc(&rdev->nr_pending);
609 rcu_read_unlock();
610
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500611 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
613 rdev_dec_pending(rdev, mddev);
614 rcu_read_lock();
615 }
616 }
617 rcu_read_unlock();
618}
619
Jens Axboe165125e2007-07-24 09:28:11 +0200620static void raid10_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
NeilBrown6cce3b22006-01-06 00:20:16 -0800622 mddev_t *mddev = q->queuedata;
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 unplug_slaves(q->queuedata);
NeilBrown6cce3b22006-01-06 00:20:16 -0800625 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
627
NeilBrown0d129222006-10-03 01:15:54 -0700628static int raid10_congested(void *data, int bits)
629{
630 mddev_t *mddev = data;
NeilBrown070ec552009-06-16 16:54:21 +1000631 conf_t *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700632 int i, ret = 0;
633
NeilBrown3fa841d2009-09-23 18:10:29 +1000634 if (mddev_congested(mddev, bits))
635 return 1;
NeilBrown0d129222006-10-03 01:15:54 -0700636 rcu_read_lock();
637 for (i = 0; i < mddev->raid_disks && ret == 0; i++) {
638 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
639 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200640 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700641
642 ret |= bdi_congested(&q->backing_dev_info, bits);
643 }
644 }
645 rcu_read_unlock();
646 return ret;
647}
648
NeilBrowna35e63e2008-03-04 14:29:29 -0800649static int flush_pending_writes(conf_t *conf)
650{
651 /* Any writes that have been queued but are awaiting
652 * bitmap updates get flushed here.
653 * We return 1 if any requests were actually submitted.
654 */
655 int rv = 0;
NeilBrown0d129222006-10-03 01:15:54 -0700656
NeilBrowna35e63e2008-03-04 14:29:29 -0800657 spin_lock_irq(&conf->device_lock);
658
659 if (conf->pending_bio_list.head) {
660 struct bio *bio;
661 bio = bio_list_get(&conf->pending_bio_list);
662 blk_remove_plug(conf->mddev->queue);
663 spin_unlock_irq(&conf->device_lock);
664 /* flush any pending bitmap writes to disk
665 * before proceeding w/ I/O */
666 bitmap_unplug(conf->mddev->bitmap);
667
668 while (bio) { /* submit pending writes */
669 struct bio *next = bio->bi_next;
670 bio->bi_next = NULL;
671 generic_make_request(bio);
672 bio = next;
673 }
674 rv = 1;
675 } else
676 spin_unlock_irq(&conf->device_lock);
677 return rv;
678}
NeilBrown0a27ec92006-01-06 00:20:13 -0800679/* Barriers....
680 * Sometimes we need to suspend IO while we do something else,
681 * either some resync/recovery, or reconfigure the array.
682 * To do this we raise a 'barrier'.
683 * The 'barrier' is a counter that can be raised multiple times
684 * to count how many activities are happening which preclude
685 * normal IO.
686 * We can only raise the barrier if there is no pending IO.
687 * i.e. if nr_pending == 0.
688 * We choose only to raise the barrier if no-one is waiting for the
689 * barrier to go down. This means that as soon as an IO request
690 * is ready, no other operations which require a barrier will start
691 * until the IO request has had a chance.
692 *
693 * So: regular IO calls 'wait_barrier'. When that returns there
694 * is no backgroup IO happening, It must arrange to call
695 * allow_barrier when it has finished its IO.
696 * backgroup IO calls must call raise_barrier. Once that returns
697 * there is no normal IO happeing. It must arrange to call
698 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
NeilBrown6cce3b22006-01-06 00:20:16 -0800701static void raise_barrier(conf_t *conf, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
NeilBrown6cce3b22006-01-06 00:20:16 -0800703 BUG_ON(force && !conf->barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 spin_lock_irq(&conf->resync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
NeilBrown6cce3b22006-01-06 00:20:16 -0800706 /* Wait until no block IO is waiting (unless 'force') */
707 wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting,
NeilBrown0a27ec92006-01-06 00:20:13 -0800708 conf->resync_lock,
709 raid10_unplug(conf->mddev->queue));
710
711 /* block any new IO from starting */
712 conf->barrier++;
713
714 /* No wait for all pending IO to complete */
715 wait_event_lock_irq(conf->wait_barrier,
716 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
717 conf->resync_lock,
718 raid10_unplug(conf->mddev->queue));
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 spin_unlock_irq(&conf->resync_lock);
721}
722
NeilBrown0a27ec92006-01-06 00:20:13 -0800723static void lower_barrier(conf_t *conf)
724{
725 unsigned long flags;
726 spin_lock_irqsave(&conf->resync_lock, flags);
727 conf->barrier--;
728 spin_unlock_irqrestore(&conf->resync_lock, flags);
729 wake_up(&conf->wait_barrier);
730}
731
732static void wait_barrier(conf_t *conf)
733{
734 spin_lock_irq(&conf->resync_lock);
735 if (conf->barrier) {
736 conf->nr_waiting++;
737 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
738 conf->resync_lock,
739 raid10_unplug(conf->mddev->queue));
740 conf->nr_waiting--;
741 }
742 conf->nr_pending++;
743 spin_unlock_irq(&conf->resync_lock);
744}
745
746static void allow_barrier(conf_t *conf)
747{
748 unsigned long flags;
749 spin_lock_irqsave(&conf->resync_lock, flags);
750 conf->nr_pending--;
751 spin_unlock_irqrestore(&conf->resync_lock, flags);
752 wake_up(&conf->wait_barrier);
753}
754
NeilBrown4443ae12006-01-06 00:20:28 -0800755static void freeze_array(conf_t *conf)
756{
757 /* stop syncio and normal IO and wait for everything to
NeilBrownf1885932006-01-06 00:20:42 -0800758 * go quiet.
NeilBrown4443ae12006-01-06 00:20:28 -0800759 * We increment barrier and nr_waiting, and then
NeilBrown1c830532008-03-04 14:29:35 -0800760 * wait until nr_pending match nr_queued+1
761 * This is called in the context of one normal IO request
762 * that has failed. Thus any sync request that might be pending
763 * will be blocked by nr_pending, and we need to wait for
764 * pending IO requests to complete or be queued for re-try.
765 * Thus the number queued (nr_queued) plus this request (1)
766 * must match the number of pending IOs (nr_pending) before
767 * we continue.
NeilBrown4443ae12006-01-06 00:20:28 -0800768 */
769 spin_lock_irq(&conf->resync_lock);
770 conf->barrier++;
771 conf->nr_waiting++;
772 wait_event_lock_irq(conf->wait_barrier,
NeilBrown1c830532008-03-04 14:29:35 -0800773 conf->nr_pending == conf->nr_queued+1,
NeilBrown4443ae12006-01-06 00:20:28 -0800774 conf->resync_lock,
NeilBrowna35e63e2008-03-04 14:29:29 -0800775 ({ flush_pending_writes(conf);
776 raid10_unplug(conf->mddev->queue); }));
NeilBrown4443ae12006-01-06 00:20:28 -0800777 spin_unlock_irq(&conf->resync_lock);
778}
779
780static void unfreeze_array(conf_t *conf)
781{
782 /* reverse the effect of the freeze */
783 spin_lock_irq(&conf->resync_lock);
784 conf->barrier--;
785 conf->nr_waiting--;
786 wake_up(&conf->wait_barrier);
787 spin_unlock_irq(&conf->resync_lock);
788}
789
Jens Axboe165125e2007-07-24 09:28:11 +0200790static int make_request(struct request_queue *q, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
792 mddev_t *mddev = q->queuedata;
NeilBrown070ec552009-06-16 16:54:21 +1000793 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 mirror_info_t *mirror;
795 r10bio_t *r10_bio;
796 struct bio *read_bio;
Tejun Heoc9959052008-08-25 19:47:21 +0900797 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 int i;
799 int chunk_sects = conf->chunk_mask + 1;
Jens Axboea3623572005-11-01 09:26:16 +0100800 const int rw = bio_data_dir(bio);
Jens Axboe1f98a132009-09-11 14:32:04 +0200801 const bool do_sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
NeilBrown6cce3b22006-01-06 00:20:16 -0800802 struct bio_list bl;
803 unsigned long flags;
Dan Williams6bfe0b42008-04-30 00:52:32 -0700804 mdk_rdev_t *blocked_rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Jens Axboe1f98a132009-09-11 14:32:04 +0200806 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
NeilBrowna2826aa2009-12-14 12:49:49 +1100807 md_barrier_request(mddev, bio);
NeilBrowne5dcdd82005-09-09 16:23:41 -0700808 return 0;
809 }
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 /* If this request crosses a chunk boundary, we need to
812 * split it. This will only happen for 1 PAGE (or less) requests.
813 */
814 if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9)
815 > chunk_sects &&
816 conf->near_copies < conf->raid_disks)) {
817 struct bio_pair *bp;
818 /* Sanity check -- queue functions should prevent this happening */
819 if (bio->bi_vcnt != 1 ||
820 bio->bi_idx != 0)
821 goto bad_map;
822 /* This is a one page bio that upper layers
823 * refuse to split for us, so we need to split it.
824 */
Denis ChengRq6feef532008-10-09 08:57:05 +0200825 bp = bio_split(bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 chunk_sects - (bio->bi_sector & (chunk_sects - 1)) );
827 if (make_request(q, &bp->bio1))
828 generic_make_request(&bp->bio1);
829 if (make_request(q, &bp->bio2))
830 generic_make_request(&bp->bio2);
831
832 bio_pair_release(bp);
833 return 0;
834 bad_map:
835 printk("raid10_make_request bug: can't convert block across chunks"
836 " or bigger than %dk %llu %d\n", chunk_sects/2,
837 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
838
NeilBrown6712ecf2007-09-27 12:47:43 +0200839 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 return 0;
841 }
842
NeilBrown3d310eb2005-06-21 17:17:26 -0700843 md_write_start(mddev, bio);
NeilBrown06d91a52005-06-21 17:17:12 -0700844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 /*
846 * Register the new request and wait if the reconstruction
847 * thread has put up a bar for new requests.
848 * Continue immediately if no resync is active currently.
849 */
NeilBrown0a27ec92006-01-06 00:20:13 -0800850 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Tejun Heo074a7ac2008-08-25 19:56:14 +0900852 cpu = part_stat_lock();
853 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
854 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
855 bio_sectors(bio));
856 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
859
860 r10_bio->master_bio = bio;
861 r10_bio->sectors = bio->bi_size >> 9;
862
863 r10_bio->mddev = mddev;
864 r10_bio->sector = bio->bi_sector;
NeilBrown6cce3b22006-01-06 00:20:16 -0800865 r10_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
Jens Axboea3623572005-11-01 09:26:16 +0100867 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 /*
869 * read balancing logic:
870 */
871 int disk = read_balance(conf, r10_bio);
872 int slot = r10_bio->read_slot;
873 if (disk < 0) {
874 raid_end_bio_io(r10_bio);
875 return 0;
876 }
877 mirror = conf->mirrors + disk;
878
879 read_bio = bio_clone(bio, GFP_NOIO);
880
881 r10_bio->devs[slot].bio = read_bio;
882
883 read_bio->bi_sector = r10_bio->devs[slot].addr +
884 mirror->rdev->data_offset;
885 read_bio->bi_bdev = mirror->rdev->bdev;
886 read_bio->bi_end_io = raid10_end_read_request;
Dmitry Monakhov1ef04fe2009-09-20 05:52:25 +0400887 read_bio->bi_rw = READ | (do_sync << BIO_RW_SYNCIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 read_bio->bi_private = r10_bio;
889
890 generic_make_request(read_bio);
891 return 0;
892 }
893
894 /*
895 * WRITE:
896 */
Dan Williams6bfe0b42008-04-30 00:52:32 -0700897 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 * inc refcount on their rdev. Record them by setting
899 * bios[x] to bio
900 */
901 raid10_find_phys(conf, r10_bio);
Dan Williams6bfe0b42008-04-30 00:52:32 -0700902 retry_write:
Harvey Harrisoncb6969e2008-05-06 20:42:32 -0700903 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 rcu_read_lock();
905 for (i = 0; i < conf->copies; i++) {
906 int d = r10_bio->devs[i].devnum;
Suzanne Woodd6065f72005-11-08 21:39:27 -0800907 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[d].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -0700908 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
909 atomic_inc(&rdev->nr_pending);
910 blocked_rdev = rdev;
911 break;
912 }
913 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800914 atomic_inc(&rdev->nr_pending);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 r10_bio->devs[i].bio = bio;
NeilBrown6cce3b22006-01-06 00:20:16 -0800916 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 r10_bio->devs[i].bio = NULL;
NeilBrown6cce3b22006-01-06 00:20:16 -0800918 set_bit(R10BIO_Degraded, &r10_bio->state);
919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 }
921 rcu_read_unlock();
922
Dan Williams6bfe0b42008-04-30 00:52:32 -0700923 if (unlikely(blocked_rdev)) {
924 /* Have to wait for this device to get unblocked, then retry */
925 int j;
926 int d;
927
928 for (j = 0; j < i; j++)
929 if (r10_bio->devs[j].bio) {
930 d = r10_bio->devs[j].devnum;
931 rdev_dec_pending(conf->mirrors[d].rdev, mddev);
932 }
933 allow_barrier(conf);
934 md_wait_for_blocked_rdev(blocked_rdev, mddev);
935 wait_barrier(conf);
936 goto retry_write;
937 }
938
NeilBrown6cce3b22006-01-06 00:20:16 -0800939 atomic_set(&r10_bio->remaining, 0);
NeilBrown06d91a52005-06-21 17:17:12 -0700940
NeilBrown6cce3b22006-01-06 00:20:16 -0800941 bio_list_init(&bl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 for (i = 0; i < conf->copies; i++) {
943 struct bio *mbio;
944 int d = r10_bio->devs[i].devnum;
945 if (!r10_bio->devs[i].bio)
946 continue;
947
948 mbio = bio_clone(bio, GFP_NOIO);
949 r10_bio->devs[i].bio = mbio;
950
951 mbio->bi_sector = r10_bio->devs[i].addr+
952 conf->mirrors[d].rdev->data_offset;
953 mbio->bi_bdev = conf->mirrors[d].rdev->bdev;
954 mbio->bi_end_io = raid10_end_write_request;
Dmitry Monakhov1ef04fe2009-09-20 05:52:25 +0400955 mbio->bi_rw = WRITE | (do_sync << BIO_RW_SYNCIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 mbio->bi_private = r10_bio;
957
958 atomic_inc(&r10_bio->remaining);
NeilBrown6cce3b22006-01-06 00:20:16 -0800959 bio_list_add(&bl, mbio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
961
Arne Redlichf6f953a2007-07-31 00:37:57 -0700962 if (unlikely(!atomic_read(&r10_bio->remaining))) {
963 /* the array is dead */
964 md_write_end(mddev);
965 raid_end_bio_io(r10_bio);
966 return 0;
967 }
968
NeilBrown6cce3b22006-01-06 00:20:16 -0800969 bitmap_startwrite(mddev->bitmap, bio->bi_sector, r10_bio->sectors, 0);
970 spin_lock_irqsave(&conf->device_lock, flags);
971 bio_list_merge(&conf->pending_bio_list, &bl);
972 blk_plug_device(mddev->queue);
973 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
NeilBrowna35e63e2008-03-04 14:29:29 -0800975 /* In case raid10d snuck in to freeze_array */
976 wake_up(&conf->wait_barrier);
977
Lars Ellenberge3881a62007-01-10 23:15:37 -0800978 if (do_sync)
979 md_wakeup_thread(mddev->thread);
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 return 0;
982}
983
984static void status(struct seq_file *seq, mddev_t *mddev)
985{
NeilBrown070ec552009-06-16 16:54:21 +1000986 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 int i;
988
989 if (conf->near_copies < conf->raid_disks)
Andre Noll9d8f0362009-06-18 08:45:01 +1000990 seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 if (conf->near_copies > 1)
992 seq_printf(seq, " %d near-copies", conf->near_copies);
NeilBrownc93983b2006-06-26 00:27:41 -0700993 if (conf->far_copies > 1) {
994 if (conf->far_offset)
995 seq_printf(seq, " %d offset-copies", conf->far_copies);
996 else
997 seq_printf(seq, " %d far-copies", conf->far_copies);
998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown76186dd2006-10-03 01:15:48 -07001000 conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 for (i = 0; i < conf->raid_disks; i++)
1002 seq_printf(seq, "%s",
1003 conf->mirrors[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08001004 test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 seq_printf(seq, "]");
1006}
1007
1008static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1009{
1010 char b[BDEVNAME_SIZE];
NeilBrown070ec552009-06-16 16:54:21 +10001011 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
1013 /*
1014 * If it is not operational, then we have already marked it as dead
1015 * else if it is the last working disks, ignore the error, let the
1016 * next level up know.
1017 * else mark the drive as failed
1018 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001019 if (test_bit(In_sync, &rdev->flags)
NeilBrown76186dd2006-10-03 01:15:48 -07001020 && conf->raid_disks-mddev->degraded == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 /*
1022 * Don't fail the drive, just return an IO error.
1023 * The test should really be more sophisticated than
1024 * "working_disks == 1", but it isn't critical, and
1025 * can wait until we do more sophisticated "is the drive
1026 * really dead" tests...
1027 */
1028 return;
NeilBrownc04be0a2006-10-03 01:15:53 -07001029 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1030 unsigned long flags;
1031 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 mddev->degraded++;
NeilBrownc04be0a2006-10-03 01:15:53 -07001033 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 /*
1035 * if recovery is running, make sure it aborts.
1036 */
NeilBrowndfc70642008-05-23 13:04:39 -07001037 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
NeilBrownb2d444d2005-11-08 21:39:31 -08001039 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001040 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Nick Andrewd7a420c2008-04-28 02:15:55 -07001041 printk(KERN_ALERT "raid10: Disk failure on %s, disabling device.\n"
1042 "raid10: Operation continuing on %d devices.\n",
NeilBrown76186dd2006-10-03 01:15:48 -07001043 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044}
1045
1046static void print_conf(conf_t *conf)
1047{
1048 int i;
1049 mirror_info_t *tmp;
1050
1051 printk("RAID10 conf printout:\n");
1052 if (!conf) {
1053 printk("(!conf)\n");
1054 return;
1055 }
NeilBrown76186dd2006-10-03 01:15:48 -07001056 printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 conf->raid_disks);
1058
1059 for (i = 0; i < conf->raid_disks; i++) {
1060 char b[BDEVNAME_SIZE];
1061 tmp = conf->mirrors + i;
1062 if (tmp->rdev)
1063 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownb2d444d2005-11-08 21:39:31 -08001064 i, !test_bit(In_sync, &tmp->rdev->flags),
1065 !test_bit(Faulty, &tmp->rdev->flags),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 bdevname(tmp->rdev->bdev,b));
1067 }
1068}
1069
1070static void close_sync(conf_t *conf)
1071{
NeilBrown0a27ec92006-01-06 00:20:13 -08001072 wait_barrier(conf);
1073 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075 mempool_destroy(conf->r10buf_pool);
1076 conf->r10buf_pool = NULL;
1077}
1078
NeilBrown6d508242005-09-09 16:24:03 -07001079/* check if there are enough drives for
1080 * every block to appear on atleast one
1081 */
1082static int enough(conf_t *conf)
1083{
1084 int first = 0;
1085
1086 do {
1087 int n = conf->copies;
1088 int cnt = 0;
1089 while (n--) {
1090 if (conf->mirrors[first].rdev)
1091 cnt++;
1092 first = (first+1) % conf->raid_disks;
1093 }
1094 if (cnt == 0)
1095 return 0;
1096 } while (first != 0);
1097 return 1;
1098}
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100static int raid10_spare_active(mddev_t *mddev)
1101{
1102 int i;
1103 conf_t *conf = mddev->private;
1104 mirror_info_t *tmp;
1105
1106 /*
1107 * Find all non-in_sync disks within the RAID10 configuration
1108 * and mark them in_sync
1109 */
1110 for (i = 0; i < conf->raid_disks; i++) {
1111 tmp = conf->mirrors + i;
1112 if (tmp->rdev
NeilBrownb2d444d2005-11-08 21:39:31 -08001113 && !test_bit(Faulty, &tmp->rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001114 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
1115 unsigned long flags;
1116 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 mddev->degraded--;
NeilBrownc04be0a2006-10-03 01:15:53 -07001118 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120 }
1121
1122 print_conf(conf);
1123 return 0;
1124}
1125
1126
1127static int raid10_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1128{
1129 conf_t *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001130 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 int mirror;
1132 mirror_info_t *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001133 int first = 0;
1134 int last = mddev->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 if (mddev->recovery_cp < MaxSector)
1137 /* only hot-add to in-sync arrays, as recovery is
1138 * very different from resync
1139 */
Neil Brown199050e2008-06-28 08:31:33 +10001140 return -EBUSY;
NeilBrown6d508242005-09-09 16:24:03 -07001141 if (!enough(conf))
Neil Brown199050e2008-06-28 08:31:33 +10001142 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
NeilBrowna53a6c82008-11-06 17:28:20 +11001144 if (rdev->raid_disk >= 0)
Neil Brown6c2fce22008-06-28 08:31:31 +10001145 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
NeilBrown6cce3b22006-01-06 00:20:16 -08001147 if (rdev->saved_raid_disk >= 0 &&
Neil Brown6c2fce22008-06-28 08:31:31 +10001148 rdev->saved_raid_disk >= first &&
NeilBrown6cce3b22006-01-06 00:20:16 -08001149 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1150 mirror = rdev->saved_raid_disk;
1151 else
Neil Brown6c2fce22008-06-28 08:31:31 +10001152 mirror = first;
1153 for ( ; mirror <= last ; mirror++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if ( !(p=conf->mirrors+mirror)->rdev) {
1155
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10001156 disk_stack_limits(mddev->gendisk, rdev->bdev,
1157 rdev->data_offset << 9);
NeilBrown627a2d32010-03-08 16:44:38 +11001158 /* as we don't honour merge_bvec_fn, we must
1159 * never risk violating it, so limit
1160 * ->max_segments to one lying with a single
1161 * page, as a one page request is never in
1162 * violation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 */
NeilBrown627a2d32010-03-08 16:44:38 +11001164 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
1165 blk_queue_max_segments(mddev->queue, 1);
1166 blk_queue_segment_boundary(mddev->queue,
1167 PAGE_CACHE_SIZE - 1);
1168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 p->head_position = 0;
1171 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001172 err = 0;
NeilBrown6cce3b22006-01-06 00:20:16 -08001173 if (rdev->saved_raid_disk != mirror)
1174 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001175 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 break;
1177 }
1178
Andre Nollac5e7112009-08-03 10:59:47 +10001179 md_integrity_add_rdev(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001181 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182}
1183
1184static int raid10_remove_disk(mddev_t *mddev, int number)
1185{
1186 conf_t *conf = mddev->private;
1187 int err = 0;
1188 mdk_rdev_t *rdev;
1189 mirror_info_t *p = conf->mirrors+ number;
1190
1191 print_conf(conf);
1192 rdev = p->rdev;
1193 if (rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001194 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 atomic_read(&rdev->nr_pending)) {
1196 err = -EBUSY;
1197 goto abort;
1198 }
NeilBrowndfc70642008-05-23 13:04:39 -07001199 /* Only remove faulty devices in recovery
1200 * is not possible.
1201 */
1202 if (!test_bit(Faulty, &rdev->flags) &&
1203 enough(conf)) {
1204 err = -EBUSY;
1205 goto abort;
1206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001208 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (atomic_read(&rdev->nr_pending)) {
1210 /* lost the race, try later */
1211 err = -EBUSY;
1212 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +10001213 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
Andre Nollac5e7112009-08-03 10:59:47 +10001215 md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
1217abort:
1218
1219 print_conf(conf);
1220 return err;
1221}
1222
1223
NeilBrown6712ecf2007-09-27 12:47:43 +02001224static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private);
NeilBrown070ec552009-06-16 16:54:21 +10001227 conf_t *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 int i,d;
1229
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 for (i=0; i<conf->copies; i++)
1231 if (r10_bio->devs[i].bio == bio)
1232 break;
Eric Sesterhennb6385482006-04-02 13:34:29 +02001233 BUG_ON(i == conf->copies);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 update_head_pos(i, r10_bio);
1235 d = r10_bio->devs[i].devnum;
NeilBrown0eb3ff12006-01-06 00:20:29 -08001236
1237 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1238 set_bit(R10BIO_Uptodate, &r10_bio->state);
NeilBrown4dbcdc72006-01-06 00:20:52 -08001239 else {
1240 atomic_add(r10_bio->sectors,
1241 &conf->mirrors[d].rdev->corrected_errors);
1242 if (!test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery))
1243 md_error(r10_bio->mddev,
1244 conf->mirrors[d].rdev);
1245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247 /* for reconstruct, we always reschedule after a read.
1248 * for resync, only after all reads
1249 */
NeilBrown73d5c382009-02-25 13:18:47 +11001250 rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
1252 atomic_dec_and_test(&r10_bio->remaining)) {
1253 /* we have read all the blocks,
1254 * do the comparison in process context in raid10d
1255 */
1256 reschedule_retry(r10_bio);
1257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258}
1259
NeilBrown6712ecf2007-09-27 12:47:43 +02001260static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
1262 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1263 r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private);
1264 mddev_t *mddev = r10_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10001265 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 int i,d;
1267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 for (i = 0; i < conf->copies; i++)
1269 if (r10_bio->devs[i].bio == bio)
1270 break;
1271 d = r10_bio->devs[i].devnum;
1272
1273 if (!uptodate)
1274 md_error(mddev, conf->mirrors[d].rdev);
NeilBrowndfc70642008-05-23 13:04:39 -07001275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 update_head_pos(i, r10_bio);
1277
NeilBrown73d5c382009-02-25 13:18:47 +11001278 rdev_dec_pending(conf->mirrors[d].rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 while (atomic_dec_and_test(&r10_bio->remaining)) {
1280 if (r10_bio->master_bio == NULL) {
1281 /* the primary of several recovery bios */
NeilBrown73d5c382009-02-25 13:18:47 +11001282 sector_t s = r10_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 put_buf(r10_bio);
NeilBrown73d5c382009-02-25 13:18:47 +11001284 md_done_sync(mddev, s, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 break;
1286 } else {
1287 r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio;
1288 put_buf(r10_bio);
1289 r10_bio = r10_bio2;
1290 }
1291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292}
1293
1294/*
1295 * Note: sync and recover and handled very differently for raid10
1296 * This code is for resync.
1297 * For resync, we read through virtual addresses and read all blocks.
1298 * If there is any error, we schedule a write. The lowest numbered
1299 * drive is authoritative.
1300 * However requests come for physical address, so we need to map.
1301 * For every physical address there are raid_disks/copies virtual addresses,
1302 * which is always are least one, but is not necessarly an integer.
1303 * This means that a physical address can span multiple chunks, so we may
1304 * have to submit multiple io requests for a single sync request.
1305 */
1306/*
1307 * We check if all blocks are in-sync and only write to blocks that
1308 * aren't in sync
1309 */
1310static void sync_request_write(mddev_t *mddev, r10bio_t *r10_bio)
1311{
NeilBrown070ec552009-06-16 16:54:21 +10001312 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 int i, first;
1314 struct bio *tbio, *fbio;
1315
1316 atomic_set(&r10_bio->remaining, 1);
1317
1318 /* find the first device with a block */
1319 for (i=0; i<conf->copies; i++)
1320 if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags))
1321 break;
1322
1323 if (i == conf->copies)
1324 goto done;
1325
1326 first = i;
1327 fbio = r10_bio->devs[i].bio;
1328
1329 /* now find blocks with errors */
NeilBrown0eb3ff12006-01-06 00:20:29 -08001330 for (i=0 ; i < conf->copies ; i++) {
1331 int j, d;
1332 int vcnt = r10_bio->sectors >> (PAGE_SHIFT-9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 tbio = r10_bio->devs[i].bio;
NeilBrown0eb3ff12006-01-06 00:20:29 -08001335
1336 if (tbio->bi_end_io != end_sync_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 continue;
NeilBrown0eb3ff12006-01-06 00:20:29 -08001338 if (i == first)
1339 continue;
1340 if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) {
1341 /* We know that the bi_io_vec layout is the same for
1342 * both 'first' and 'i', so we just compare them.
1343 * All vec entries are PAGE_SIZE;
1344 */
1345 for (j = 0; j < vcnt; j++)
1346 if (memcmp(page_address(fbio->bi_io_vec[j].bv_page),
1347 page_address(tbio->bi_io_vec[j].bv_page),
1348 PAGE_SIZE))
1349 break;
1350 if (j == vcnt)
1351 continue;
1352 mddev->resync_mismatches += r10_bio->sectors;
1353 }
NeilBrown18f08812006-01-06 00:20:25 -08001354 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1355 /* Don't fix anything. */
1356 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 /* Ok, we need to write this bio
1358 * First we need to fixup bv_offset, bv_len and
1359 * bi_vecs, as the read request might have corrupted these
1360 */
1361 tbio->bi_vcnt = vcnt;
1362 tbio->bi_size = r10_bio->sectors << 9;
1363 tbio->bi_idx = 0;
1364 tbio->bi_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 tbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1366 tbio->bi_flags |= 1 << BIO_UPTODATE;
1367 tbio->bi_next = NULL;
1368 tbio->bi_rw = WRITE;
1369 tbio->bi_private = r10_bio;
1370 tbio->bi_sector = r10_bio->devs[i].addr;
1371
1372 for (j=0; j < vcnt ; j++) {
1373 tbio->bi_io_vec[j].bv_offset = 0;
1374 tbio->bi_io_vec[j].bv_len = PAGE_SIZE;
1375
1376 memcpy(page_address(tbio->bi_io_vec[j].bv_page),
1377 page_address(fbio->bi_io_vec[j].bv_page),
1378 PAGE_SIZE);
1379 }
1380 tbio->bi_end_io = end_sync_write;
1381
1382 d = r10_bio->devs[i].devnum;
1383 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
1384 atomic_inc(&r10_bio->remaining);
1385 md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9);
1386
1387 tbio->bi_sector += conf->mirrors[d].rdev->data_offset;
1388 tbio->bi_bdev = conf->mirrors[d].rdev->bdev;
1389 generic_make_request(tbio);
1390 }
1391
1392done:
1393 if (atomic_dec_and_test(&r10_bio->remaining)) {
1394 md_done_sync(mddev, r10_bio->sectors, 1);
1395 put_buf(r10_bio);
1396 }
1397}
1398
1399/*
1400 * Now for the recovery code.
1401 * Recovery happens across physical sectors.
1402 * We recover all non-is_sync drives by finding the virtual address of
1403 * each, and then choose a working drive that also has that virt address.
1404 * There is a separate r10_bio for each non-in_sync drive.
1405 * Only the first two slots are in use. The first for reading,
1406 * The second for writing.
1407 *
1408 */
1409
1410static void recovery_request_write(mddev_t *mddev, r10bio_t *r10_bio)
1411{
NeilBrown070ec552009-06-16 16:54:21 +10001412 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 int i, d;
1414 struct bio *bio, *wbio;
1415
1416
1417 /* move the pages across to the second bio
1418 * and submit the write request
1419 */
1420 bio = r10_bio->devs[0].bio;
1421 wbio = r10_bio->devs[1].bio;
1422 for (i=0; i < wbio->bi_vcnt; i++) {
1423 struct page *p = bio->bi_io_vec[i].bv_page;
1424 bio->bi_io_vec[i].bv_page = wbio->bi_io_vec[i].bv_page;
1425 wbio->bi_io_vec[i].bv_page = p;
1426 }
1427 d = r10_bio->devs[1].devnum;
1428
1429 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
1430 md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9);
NeilBrown0eb3ff12006-01-06 00:20:29 -08001431 if (test_bit(R10BIO_Uptodate, &r10_bio->state))
1432 generic_make_request(wbio);
1433 else
NeilBrown6712ecf2007-09-27 12:47:43 +02001434 bio_endio(wbio, -EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435}
1436
1437
1438/*
Robert Becker1e509152009-12-14 12:49:58 +11001439 * Used by fix_read_error() to decay the per rdev read_errors.
1440 * We halve the read error count for every hour that has elapsed
1441 * since the last recorded read error.
1442 *
1443 */
1444static void check_decay_read_errors(mddev_t *mddev, mdk_rdev_t *rdev)
1445{
1446 struct timespec cur_time_mon;
1447 unsigned long hours_since_last;
1448 unsigned int read_errors = atomic_read(&rdev->read_errors);
1449
1450 ktime_get_ts(&cur_time_mon);
1451
1452 if (rdev->last_read_error.tv_sec == 0 &&
1453 rdev->last_read_error.tv_nsec == 0) {
1454 /* first time we've seen a read error */
1455 rdev->last_read_error = cur_time_mon;
1456 return;
1457 }
1458
1459 hours_since_last = (cur_time_mon.tv_sec -
1460 rdev->last_read_error.tv_sec) / 3600;
1461
1462 rdev->last_read_error = cur_time_mon;
1463
1464 /*
1465 * if hours_since_last is > the number of bits in read_errors
1466 * just set read errors to 0. We do this to avoid
1467 * overflowing the shift of read_errors by hours_since_last.
1468 */
1469 if (hours_since_last >= 8 * sizeof(read_errors))
1470 atomic_set(&rdev->read_errors, 0);
1471 else
1472 atomic_set(&rdev->read_errors, read_errors >> hours_since_last);
1473}
1474
1475/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 * This is a kernel thread which:
1477 *
1478 * 1. Retries failed read operations on working mirrors.
1479 * 2. Updates the raid superblock when problems encounter.
NeilBrown6814d532006-10-03 01:15:45 -07001480 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 */
1482
NeilBrown6814d532006-10-03 01:15:45 -07001483static void fix_read_error(conf_t *conf, mddev_t *mddev, r10bio_t *r10_bio)
1484{
1485 int sect = 0; /* Offset from r10_bio->sector */
1486 int sectors = r10_bio->sectors;
1487 mdk_rdev_t*rdev;
Robert Becker1e509152009-12-14 12:49:58 +11001488 int max_read_errors = atomic_read(&mddev->max_corr_read_errors);
1489
1490 rcu_read_lock();
1491 {
1492 int d = r10_bio->devs[r10_bio->read_slot].devnum;
1493 char b[BDEVNAME_SIZE];
1494 int cur_read_error_count = 0;
1495
1496 rdev = rcu_dereference(conf->mirrors[d].rdev);
1497 bdevname(rdev->bdev, b);
1498
1499 if (test_bit(Faulty, &rdev->flags)) {
1500 rcu_read_unlock();
1501 /* drive has already been failed, just ignore any
1502 more fix_read_error() attempts */
1503 return;
1504 }
1505
1506 check_decay_read_errors(mddev, rdev);
1507 atomic_inc(&rdev->read_errors);
1508 cur_read_error_count = atomic_read(&rdev->read_errors);
1509 if (cur_read_error_count > max_read_errors) {
1510 rcu_read_unlock();
1511 printk(KERN_NOTICE
1512 "raid10: %s: Raid device exceeded "
1513 "read_error threshold "
1514 "[cur %d:max %d]\n",
1515 b, cur_read_error_count, max_read_errors);
1516 printk(KERN_NOTICE
1517 "raid10: %s: Failing raid "
1518 "device\n", b);
1519 md_error(mddev, conf->mirrors[d].rdev);
1520 return;
1521 }
1522 }
1523 rcu_read_unlock();
1524
NeilBrown6814d532006-10-03 01:15:45 -07001525 while(sectors) {
1526 int s = sectors;
1527 int sl = r10_bio->read_slot;
1528 int success = 0;
1529 int start;
1530
1531 if (s > (PAGE_SIZE>>9))
1532 s = PAGE_SIZE >> 9;
1533
1534 rcu_read_lock();
1535 do {
1536 int d = r10_bio->devs[sl].devnum;
1537 rdev = rcu_dereference(conf->mirrors[d].rdev);
1538 if (rdev &&
1539 test_bit(In_sync, &rdev->flags)) {
1540 atomic_inc(&rdev->nr_pending);
1541 rcu_read_unlock();
1542 success = sync_page_io(rdev->bdev,
1543 r10_bio->devs[sl].addr +
1544 sect + rdev->data_offset,
1545 s<<9,
1546 conf->tmppage, READ);
1547 rdev_dec_pending(rdev, mddev);
1548 rcu_read_lock();
1549 if (success)
1550 break;
1551 }
1552 sl++;
1553 if (sl == conf->copies)
1554 sl = 0;
1555 } while (!success && sl != r10_bio->read_slot);
1556 rcu_read_unlock();
1557
1558 if (!success) {
1559 /* Cannot read from anywhere -- bye bye array */
1560 int dn = r10_bio->devs[r10_bio->read_slot].devnum;
1561 md_error(mddev, conf->mirrors[dn].rdev);
1562 break;
1563 }
1564
1565 start = sl;
1566 /* write it back and re-read */
1567 rcu_read_lock();
1568 while (sl != r10_bio->read_slot) {
Robert Becker67b8dc42009-12-14 12:49:57 +11001569 char b[BDEVNAME_SIZE];
NeilBrown6814d532006-10-03 01:15:45 -07001570 int d;
1571 if (sl==0)
1572 sl = conf->copies;
1573 sl--;
1574 d = r10_bio->devs[sl].devnum;
1575 rdev = rcu_dereference(conf->mirrors[d].rdev);
1576 if (rdev &&
1577 test_bit(In_sync, &rdev->flags)) {
1578 atomic_inc(&rdev->nr_pending);
1579 rcu_read_unlock();
1580 atomic_add(s, &rdev->corrected_errors);
1581 if (sync_page_io(rdev->bdev,
1582 r10_bio->devs[sl].addr +
1583 sect + rdev->data_offset,
1584 s<<9, conf->tmppage, WRITE)
Robert Becker67b8dc42009-12-14 12:49:57 +11001585 == 0) {
NeilBrown6814d532006-10-03 01:15:45 -07001586 /* Well, this device is dead */
Robert Becker67b8dc42009-12-14 12:49:57 +11001587 printk(KERN_NOTICE
1588 "raid10:%s: read correction "
1589 "write failed"
1590 " (%d sectors at %llu on %s)\n",
1591 mdname(mddev), s,
1592 (unsigned long long)(sect+
1593 rdev->data_offset),
1594 bdevname(rdev->bdev, b));
1595 printk(KERN_NOTICE "raid10:%s: failing "
1596 "drive\n",
1597 bdevname(rdev->bdev, b));
NeilBrown6814d532006-10-03 01:15:45 -07001598 md_error(mddev, rdev);
Robert Becker67b8dc42009-12-14 12:49:57 +11001599 }
NeilBrown6814d532006-10-03 01:15:45 -07001600 rdev_dec_pending(rdev, mddev);
1601 rcu_read_lock();
1602 }
1603 }
1604 sl = start;
1605 while (sl != r10_bio->read_slot) {
1606 int d;
1607 if (sl==0)
1608 sl = conf->copies;
1609 sl--;
1610 d = r10_bio->devs[sl].devnum;
1611 rdev = rcu_dereference(conf->mirrors[d].rdev);
1612 if (rdev &&
1613 test_bit(In_sync, &rdev->flags)) {
1614 char b[BDEVNAME_SIZE];
1615 atomic_inc(&rdev->nr_pending);
1616 rcu_read_unlock();
1617 if (sync_page_io(rdev->bdev,
1618 r10_bio->devs[sl].addr +
1619 sect + rdev->data_offset,
Robert Becker67b8dc42009-12-14 12:49:57 +11001620 s<<9, conf->tmppage,
1621 READ) == 0) {
NeilBrown6814d532006-10-03 01:15:45 -07001622 /* Well, this device is dead */
Robert Becker67b8dc42009-12-14 12:49:57 +11001623 printk(KERN_NOTICE
1624 "raid10:%s: unable to read back "
1625 "corrected sectors"
1626 " (%d sectors at %llu on %s)\n",
1627 mdname(mddev), s,
1628 (unsigned long long)(sect+
1629 rdev->data_offset),
1630 bdevname(rdev->bdev, b));
1631 printk(KERN_NOTICE "raid10:%s: failing drive\n",
1632 bdevname(rdev->bdev, b));
1633
NeilBrown6814d532006-10-03 01:15:45 -07001634 md_error(mddev, rdev);
Robert Becker67b8dc42009-12-14 12:49:57 +11001635 } else {
NeilBrown6814d532006-10-03 01:15:45 -07001636 printk(KERN_INFO
1637 "raid10:%s: read error corrected"
1638 " (%d sectors at %llu on %s)\n",
1639 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07001640 (unsigned long long)(sect+
1641 rdev->data_offset),
NeilBrown6814d532006-10-03 01:15:45 -07001642 bdevname(rdev->bdev, b));
Robert Becker67b8dc42009-12-14 12:49:57 +11001643 }
NeilBrown6814d532006-10-03 01:15:45 -07001644
1645 rdev_dec_pending(rdev, mddev);
1646 rcu_read_lock();
1647 }
1648 }
1649 rcu_read_unlock();
1650
1651 sectors -= s;
1652 sect += s;
1653 }
1654}
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656static void raid10d(mddev_t *mddev)
1657{
1658 r10bio_t *r10_bio;
1659 struct bio *bio;
1660 unsigned long flags;
NeilBrown070ec552009-06-16 16:54:21 +10001661 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 struct list_head *head = &conf->retry_list;
1663 int unplug=0;
1664 mdk_rdev_t *rdev;
1665
1666 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
1668 for (;;) {
1669 char b[BDEVNAME_SIZE];
NeilBrowna35e63e2008-03-04 14:29:29 -08001670
1671 unplug += flush_pending_writes(conf);
1672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08001674 if (list_empty(head)) {
NeilBrown6cce3b22006-01-06 00:20:16 -08001675 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08001677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 r10_bio = list_entry(head->prev, r10bio_t, retry_list);
1679 list_del(head->prev);
NeilBrown4443ae12006-01-06 00:20:28 -08001680 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 spin_unlock_irqrestore(&conf->device_lock, flags);
1682
1683 mddev = r10_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10001684 conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 if (test_bit(R10BIO_IsSync, &r10_bio->state)) {
1686 sync_request_write(mddev, r10_bio);
1687 unplug = 1;
1688 } else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) {
1689 recovery_request_write(mddev, r10_bio);
1690 unplug = 1;
1691 } else {
1692 int mirror;
NeilBrown4443ae12006-01-06 00:20:28 -08001693 /* we got a read error. Maybe the drive is bad. Maybe just
1694 * the block and we can fix it.
1695 * We freeze all other IO, and try reading the block from
1696 * other devices. When we find one, we re-write
1697 * and check it that fixes the read error.
1698 * This is all done synchronously while the array is
1699 * frozen.
1700 */
NeilBrown6814d532006-10-03 01:15:45 -07001701 if (mddev->ro == 0) {
1702 freeze_array(conf);
1703 fix_read_error(conf, mddev, r10_bio);
1704 unfreeze_array(conf);
NeilBrown4443ae12006-01-06 00:20:28 -08001705 }
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 bio = r10_bio->devs[r10_bio->read_slot].bio;
NeilBrown0eb3ff12006-01-06 00:20:29 -08001708 r10_bio->devs[r10_bio->read_slot].bio =
1709 mddev->ro ? IO_BLOCKED : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 mirror = read_balance(conf, r10_bio);
1711 if (mirror == -1) {
1712 printk(KERN_ALERT "raid10: %s: unrecoverable I/O"
1713 " read error for block %llu\n",
1714 bdevname(bio->bi_bdev,b),
1715 (unsigned long long)r10_bio->sector);
1716 raid_end_bio_io(r10_bio);
Maik Hampel14e71342007-07-31 00:37:57 -07001717 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 } else {
Jens Axboe1f98a132009-09-11 14:32:04 +02001719 const bool do_sync = bio_rw_flagged(r10_bio->master_bio, BIO_RW_SYNCIO);
Maik Hampel14e71342007-07-31 00:37:57 -07001720 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 rdev = conf->mirrors[mirror].rdev;
1722 if (printk_ratelimit())
1723 printk(KERN_ERR "raid10: %s: redirecting sector %llu to"
1724 " another mirror\n",
1725 bdevname(rdev->bdev,b),
1726 (unsigned long long)r10_bio->sector);
1727 bio = bio_clone(r10_bio->master_bio, GFP_NOIO);
1728 r10_bio->devs[r10_bio->read_slot].bio = bio;
1729 bio->bi_sector = r10_bio->devs[r10_bio->read_slot].addr
1730 + rdev->data_offset;
1731 bio->bi_bdev = rdev->bdev;
Dmitry Monakhov1ef04fe2009-09-20 05:52:25 +04001732 bio->bi_rw = READ | (do_sync << BIO_RW_SYNCIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 bio->bi_private = r10_bio;
1734 bio->bi_end_io = raid10_end_read_request;
1735 unplug = 1;
1736 generic_make_request(bio);
1737 }
1738 }
NeilBrown1d9d5242009-10-16 15:55:32 +11001739 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 if (unplug)
1742 unplug_slaves(mddev);
1743}
1744
1745
1746static int init_resync(conf_t *conf)
1747{
1748 int buffs;
1749
1750 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhennb6385482006-04-02 13:34:29 +02001751 BUG_ON(conf->r10buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf);
1753 if (!conf->r10buf_pool)
1754 return -ENOMEM;
1755 conf->next_resync = 0;
1756 return 0;
1757}
1758
1759/*
1760 * perform a "sync" on one "block"
1761 *
1762 * We need to make sure that no normal I/O request - particularly write
1763 * requests - conflict with active sync requests.
1764 *
1765 * This is achieved by tracking pending requests and a 'barrier' concept
1766 * that can be installed to exclude normal IO requests.
1767 *
1768 * Resync and recovery are handled very differently.
1769 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
1770 *
1771 * For resync, we iterate over virtual addresses, read all copies,
1772 * and update if there are differences. If only one copy is live,
1773 * skip it.
1774 * For recovery, we iterate over physical addresses, read a good
1775 * value for each non-in_sync drive, and over-write.
1776 *
1777 * So, for recovery we may have several outstanding complex requests for a
1778 * given address, one for each out-of-sync device. We model this by allocating
1779 * a number of r10_bio structures, one for each out-of-sync device.
1780 * As we setup these structures, we collect all bio's together into a list
1781 * which we then process collectively to add pages, and then process again
1782 * to pass to generic_make_request.
1783 *
1784 * The r10_bio structures are linked using a borrowed master_bio pointer.
1785 * This link is counted in ->remaining. When the r10_bio that points to NULL
1786 * has its remaining count decremented to 0, the whole complex operation
1787 * is complete.
1788 *
1789 */
1790
NeilBrown57afd892005-06-21 17:17:13 -07001791static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
NeilBrown070ec552009-06-16 16:54:21 +10001793 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 r10bio_t *r10_bio;
1795 struct bio *biolist = NULL, *bio;
1796 sector_t max_sector, nr_sectors;
1797 int disk;
1798 int i;
NeilBrown6cce3b22006-01-06 00:20:16 -08001799 int max_sync;
1800 int sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
1802 sector_t sectors_skipped = 0;
1803 int chunks_skipped = 0;
1804
1805 if (!conf->r10buf_pool)
1806 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07001807 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
1809 skipped:
Andre Noll58c0fed2009-03-31 14:33:13 +11001810 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
1812 max_sector = mddev->resync_max_sectors;
1813 if (sector_nr >= max_sector) {
NeilBrown6cce3b22006-01-06 00:20:16 -08001814 /* If we aborted, we need to abort the
1815 * sync on the 'current' bitmap chucks (there can
1816 * be several when recovering multiple devices).
1817 * as we may have started syncing it but not finished.
1818 * We can find the current address in
1819 * mddev->curr_resync, but for recovery,
1820 * we need to convert that to several
1821 * virtual addresses.
1822 */
1823 if (mddev->curr_resync < max_sector) { /* aborted */
1824 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
1825 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
1826 &sync_blocks, 1);
1827 else for (i=0; i<conf->raid_disks; i++) {
1828 sector_t sect =
1829 raid10_find_virt(conf, mddev->curr_resync, i);
1830 bitmap_end_sync(mddev->bitmap, sect,
1831 &sync_blocks, 1);
1832 }
1833 } else /* completed sync */
1834 conf->fullsync = 0;
1835
1836 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 close_sync(conf);
NeilBrown57afd892005-06-21 17:17:13 -07001838 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 return sectors_skipped;
1840 }
1841 if (chunks_skipped >= conf->raid_disks) {
1842 /* if there has been nothing to do on any drive,
1843 * then there is nothing to do at all..
1844 */
NeilBrown57afd892005-06-21 17:17:13 -07001845 *skipped = 1;
1846 return (max_sector - sector_nr) + sectors_skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 }
1848
NeilBrownc6207272008-02-06 01:39:52 -08001849 if (max_sector > mddev->resync_max)
1850 max_sector = mddev->resync_max; /* Don't do IO beyond here */
1851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 /* make sure whole request will fit in a chunk - if chunks
1853 * are meaningful
1854 */
1855 if (conf->near_copies < conf->raid_disks &&
1856 max_sector > (sector_nr | conf->chunk_mask))
1857 max_sector = (sector_nr | conf->chunk_mask) + 1;
1858 /*
1859 * If there is non-resync activity waiting for us then
1860 * put in a delay to throttle resync.
1861 */
NeilBrown0a27ec92006-01-06 00:20:13 -08001862 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 msleep_interruptible(1000);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
1865 /* Again, very different code for resync and recovery.
1866 * Both must result in an r10bio with a list of bios that
1867 * have bi_end_io, bi_sector, bi_bdev set,
1868 * and bi_private set to the r10bio.
1869 * For recovery, we may actually create several r10bios
1870 * with 2 bios in each, that correspond to the bios in the main one.
1871 * In this case, the subordinate r10bios link back through a
1872 * borrowed master_bio pointer, and the counter in the master
1873 * includes a ref from each subordinate.
1874 */
1875 /* First, we decide what to do and set ->bi_end_io
1876 * To end_sync_read if we want to read, and
1877 * end_sync_write if we will want to write.
1878 */
1879
NeilBrown6cce3b22006-01-06 00:20:16 -08001880 max_sync = RESYNC_PAGES << (PAGE_SHIFT-9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1882 /* recovery... the complicated one */
NeilBrowna9f326e2009-09-23 18:06:41 +10001883 int j, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 r10_bio = NULL;
1885
1886 for (i=0 ; i<conf->raid_disks; i++)
1887 if (conf->mirrors[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08001888 !test_bit(In_sync, &conf->mirrors[i].rdev->flags)) {
NeilBrown6cce3b22006-01-06 00:20:16 -08001889 int still_degraded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 /* want to reconstruct this device */
1891 r10bio_t *rb2 = r10_bio;
NeilBrown6cce3b22006-01-06 00:20:16 -08001892 sector_t sect = raid10_find_virt(conf, sector_nr, i);
1893 int must_sync;
1894 /* Unless we are doing a full sync, we only need
1895 * to recover the block if it is set in the bitmap
1896 */
1897 must_sync = bitmap_start_sync(mddev->bitmap, sect,
1898 &sync_blocks, 1);
1899 if (sync_blocks < max_sync)
1900 max_sync = sync_blocks;
1901 if (!must_sync &&
1902 !conf->fullsync) {
1903 /* yep, skip the sync_blocks here, but don't assume
1904 * that there will never be anything to do here
1905 */
1906 chunks_skipped = -1;
1907 continue;
1908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
1910 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
NeilBrown6cce3b22006-01-06 00:20:16 -08001911 raise_barrier(conf, rb2 != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 atomic_set(&r10_bio->remaining, 0);
1913
1914 r10_bio->master_bio = (struct bio*)rb2;
1915 if (rb2)
1916 atomic_inc(&rb2->remaining);
1917 r10_bio->mddev = mddev;
1918 set_bit(R10BIO_IsRecover, &r10_bio->state);
NeilBrown6cce3b22006-01-06 00:20:16 -08001919 r10_bio->sector = sect;
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 raid10_find_phys(conf, r10_bio);
NeilBrown18055562009-05-07 12:48:10 +10001922
1923 /* Need to check if the array will still be
NeilBrown6cce3b22006-01-06 00:20:16 -08001924 * degraded
1925 */
NeilBrown18055562009-05-07 12:48:10 +10001926 for (j=0; j<conf->raid_disks; j++)
1927 if (conf->mirrors[j].rdev == NULL ||
1928 test_bit(Faulty, &conf->mirrors[j].rdev->flags)) {
NeilBrown6cce3b22006-01-06 00:20:16 -08001929 still_degraded = 1;
NeilBrowna24a8dd2006-01-06 00:20:35 -08001930 break;
1931 }
NeilBrown18055562009-05-07 12:48:10 +10001932
NeilBrown6cce3b22006-01-06 00:20:16 -08001933 must_sync = bitmap_start_sync(mddev->bitmap, sect,
1934 &sync_blocks, still_degraded);
1935
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 for (j=0; j<conf->copies;j++) {
1937 int d = r10_bio->devs[j].devnum;
1938 if (conf->mirrors[d].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08001939 test_bit(In_sync, &conf->mirrors[d].rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 /* This is where we read from */
1941 bio = r10_bio->devs[0].bio;
1942 bio->bi_next = biolist;
1943 biolist = bio;
1944 bio->bi_private = r10_bio;
1945 bio->bi_end_io = end_sync_read;
NeilBrown802ba062006-12-13 00:34:13 -08001946 bio->bi_rw = READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 bio->bi_sector = r10_bio->devs[j].addr +
1948 conf->mirrors[d].rdev->data_offset;
1949 bio->bi_bdev = conf->mirrors[d].rdev->bdev;
1950 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
1951 atomic_inc(&r10_bio->remaining);
1952 /* and we write to 'i' */
1953
1954 for (k=0; k<conf->copies; k++)
1955 if (r10_bio->devs[k].devnum == i)
1956 break;
NeilBrown64a742b2007-02-28 20:11:18 -08001957 BUG_ON(k == conf->copies);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 bio = r10_bio->devs[1].bio;
1959 bio->bi_next = biolist;
1960 biolist = bio;
1961 bio->bi_private = r10_bio;
1962 bio->bi_end_io = end_sync_write;
NeilBrown802ba062006-12-13 00:34:13 -08001963 bio->bi_rw = WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 bio->bi_sector = r10_bio->devs[k].addr +
1965 conf->mirrors[i].rdev->data_offset;
1966 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1967
1968 r10_bio->devs[0].devnum = d;
1969 r10_bio->devs[1].devnum = i;
1970
1971 break;
1972 }
1973 }
1974 if (j == conf->copies) {
NeilBrown87fc7672005-09-09 16:24:04 -07001975 /* Cannot recover, so abort the recovery */
1976 put_buf(r10_bio);
K.Tanakaa07e6ab2008-03-04 14:29:37 -08001977 if (rb2)
1978 atomic_dec(&rb2->remaining);
NeilBrown87fc7672005-09-09 16:24:04 -07001979 r10_bio = rb2;
NeilBrowndfc70642008-05-23 13:04:39 -07001980 if (!test_and_set_bit(MD_RECOVERY_INTR,
1981 &mddev->recovery))
NeilBrown87fc7672005-09-09 16:24:04 -07001982 printk(KERN_INFO "raid10: %s: insufficient working devices for recovery.\n",
1983 mdname(mddev));
1984 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 }
1986 }
1987 if (biolist == NULL) {
1988 while (r10_bio) {
1989 r10bio_t *rb2 = r10_bio;
1990 r10_bio = (r10bio_t*) rb2->master_bio;
1991 rb2->master_bio = NULL;
1992 put_buf(rb2);
1993 }
1994 goto giveup;
1995 }
1996 } else {
1997 /* resync. Schedule a read for every block at this virt offset */
1998 int count = 0;
NeilBrown6cce3b22006-01-06 00:20:16 -08001999
NeilBrown78200d42009-02-25 13:18:47 +11002000 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
2001
NeilBrown6cce3b22006-01-06 00:20:16 -08002002 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
2003 &sync_blocks, mddev->degraded) &&
2004 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2005 /* We can skip this block */
2006 *skipped = 1;
2007 return sync_blocks + sectors_skipped;
2008 }
2009 if (sync_blocks < max_sync)
2010 max_sync = sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
2012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 r10_bio->mddev = mddev;
2014 atomic_set(&r10_bio->remaining, 0);
NeilBrown6cce3b22006-01-06 00:20:16 -08002015 raise_barrier(conf, 0);
2016 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017
2018 r10_bio->master_bio = NULL;
2019 r10_bio->sector = sector_nr;
2020 set_bit(R10BIO_IsSync, &r10_bio->state);
2021 raid10_find_phys(conf, r10_bio);
2022 r10_bio->sectors = (sector_nr | conf->chunk_mask) - sector_nr +1;
2023
2024 for (i=0; i<conf->copies; i++) {
2025 int d = r10_bio->devs[i].devnum;
2026 bio = r10_bio->devs[i].bio;
2027 bio->bi_end_io = NULL;
NeilBrownaf03b8e2007-06-16 10:16:06 -07002028 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 if (conf->mirrors[d].rdev == NULL ||
NeilBrownb2d444d2005-11-08 21:39:31 -08002030 test_bit(Faulty, &conf->mirrors[d].rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 continue;
2032 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
2033 atomic_inc(&r10_bio->remaining);
2034 bio->bi_next = biolist;
2035 biolist = bio;
2036 bio->bi_private = r10_bio;
2037 bio->bi_end_io = end_sync_read;
NeilBrown802ba062006-12-13 00:34:13 -08002038 bio->bi_rw = READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 bio->bi_sector = r10_bio->devs[i].addr +
2040 conf->mirrors[d].rdev->data_offset;
2041 bio->bi_bdev = conf->mirrors[d].rdev->bdev;
2042 count++;
2043 }
2044
2045 if (count < 2) {
2046 for (i=0; i<conf->copies; i++) {
2047 int d = r10_bio->devs[i].devnum;
2048 if (r10_bio->devs[i].bio->bi_end_io)
2049 rdev_dec_pending(conf->mirrors[d].rdev, mddev);
2050 }
2051 put_buf(r10_bio);
2052 biolist = NULL;
2053 goto giveup;
2054 }
2055 }
2056
2057 for (bio = biolist; bio ; bio=bio->bi_next) {
2058
2059 bio->bi_flags &= ~(BIO_POOL_MASK - 1);
2060 if (bio->bi_end_io)
2061 bio->bi_flags |= 1 << BIO_UPTODATE;
2062 bio->bi_vcnt = 0;
2063 bio->bi_idx = 0;
2064 bio->bi_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 bio->bi_size = 0;
2066 }
2067
2068 nr_sectors = 0;
NeilBrown6cce3b22006-01-06 00:20:16 -08002069 if (sector_nr + max_sync < max_sector)
2070 max_sector = sector_nr + max_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 do {
2072 struct page *page;
2073 int len = PAGE_SIZE;
2074 disk = 0;
2075 if (sector_nr + (len>>9) > max_sector)
2076 len = (max_sector - sector_nr) << 9;
2077 if (len == 0)
2078 break;
2079 for (bio= biolist ; bio ; bio=bio->bi_next) {
2080 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
2081 if (bio_add_page(bio, page, len, 0) == 0) {
2082 /* stop here */
2083 struct bio *bio2;
2084 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
2085 for (bio2 = biolist; bio2 && bio2 != bio; bio2 = bio2->bi_next) {
2086 /* remove last page from this bio */
2087 bio2->bi_vcnt--;
2088 bio2->bi_size -= len;
2089 bio2->bi_flags &= ~(1<< BIO_SEG_VALID);
2090 }
2091 goto bio_full;
2092 }
2093 disk = i;
2094 }
2095 nr_sectors += len>>9;
2096 sector_nr += len>>9;
2097 } while (biolist->bi_vcnt < RESYNC_PAGES);
2098 bio_full:
2099 r10_bio->sectors = nr_sectors;
2100
2101 while (biolist) {
2102 bio = biolist;
2103 biolist = biolist->bi_next;
2104
2105 bio->bi_next = NULL;
2106 r10_bio = bio->bi_private;
2107 r10_bio->sectors = nr_sectors;
2108
2109 if (bio->bi_end_io == end_sync_read) {
2110 md_sync_acct(bio->bi_bdev, nr_sectors);
2111 generic_make_request(bio);
2112 }
2113 }
2114
NeilBrown57afd892005-06-21 17:17:13 -07002115 if (sectors_skipped)
2116 /* pretend they weren't skipped, it makes
2117 * no important difference in this case
2118 */
2119 md_done_sync(mddev, sectors_skipped, 1);
2120
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 return sectors_skipped + nr_sectors;
2122 giveup:
2123 /* There is nowhere to write, so all non-sync
2124 * drives must be failed, so try the next chunk...
2125 */
NeilBrown09b40682009-02-25 13:18:47 +11002126 if (sector_nr + max_sync < max_sector)
2127 max_sector = sector_nr + max_sync;
2128
2129 sectors_skipped += (max_sector - sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 chunks_skipped ++;
2131 sector_nr = max_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 goto skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133}
2134
Dan Williams80c3a6c2009-03-17 18:10:40 -07002135static sector_t
2136raid10_size(mddev_t *mddev, sector_t sectors, int raid_disks)
2137{
2138 sector_t size;
NeilBrown070ec552009-06-16 16:54:21 +10002139 conf_t *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07002140
2141 if (!raid_disks)
2142 raid_disks = mddev->raid_disks;
2143 if (!sectors)
2144 sectors = mddev->dev_sectors;
2145
2146 size = sectors >> conf->chunk_shift;
2147 sector_div(size, conf->far_copies);
2148 size = size * raid_disks;
2149 sector_div(size, conf->near_copies);
2150
2151 return size << conf->chunk_shift;
2152}
2153
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154static int run(mddev_t *mddev)
2155{
2156 conf_t *conf;
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10002157 int i, disk_idx, chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 mirror_info_t *disk;
2159 mdk_rdev_t *rdev;
NeilBrownc93983b2006-06-26 00:27:41 -07002160 int nc, fc, fo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 sector_t stride, size;
2162
Andre Noll9d8f0362009-06-18 08:45:01 +10002163 if (mddev->chunk_sectors < (PAGE_SIZE >> 9) ||
2164 !is_power_of_2(mddev->chunk_sectors)) {
NeilBrown4bbf3772008-10-13 11:55:12 +11002165 printk(KERN_ERR "md/raid10: chunk size must be "
raz ben yehuda964e7912009-06-16 17:01:22 +10002166 "at least PAGE_SIZE(%ld) and be a power of 2.\n", PAGE_SIZE);
NeilBrown2604b702006-01-06 00:20:36 -08002167 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 }
NeilBrown2604b702006-01-06 00:20:36 -08002169
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 nc = mddev->layout & 255;
2171 fc = (mddev->layout >> 8) & 255;
NeilBrownc93983b2006-06-26 00:27:41 -07002172 fo = mddev->layout & (1<<16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 if ((nc*fc) <2 || (nc*fc) > mddev->raid_disks ||
NeilBrownc93983b2006-06-26 00:27:41 -07002174 (mddev->layout >> 17)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 printk(KERN_ERR "raid10: %s: unsupported raid10 layout: 0x%8x\n",
2176 mdname(mddev), mddev->layout);
2177 goto out;
2178 }
2179 /*
2180 * copy the already verified devices into our private RAID10
2181 * bookkeeping area. [whatever we allocate in run(),
2182 * should be freed in stop()]
2183 */
NeilBrown4443ae12006-01-06 00:20:28 -08002184 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 mddev->private = conf;
2186 if (!conf) {
2187 printk(KERN_ERR "raid10: couldn't allocate memory for %s\n",
2188 mdname(mddev));
2189 goto out;
2190 }
NeilBrown4443ae12006-01-06 00:20:28 -08002191 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 GFP_KERNEL);
2193 if (!conf->mirrors) {
2194 printk(KERN_ERR "raid10: couldn't allocate memory for %s\n",
2195 mdname(mddev));
2196 goto out_free_conf;
2197 }
NeilBrown4443ae12006-01-06 00:20:28 -08002198
2199 conf->tmppage = alloc_page(GFP_KERNEL);
2200 if (!conf->tmppage)
2201 goto out_free_conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202
NeilBrown64a742b2007-02-28 20:11:18 -08002203 conf->raid_disks = mddev->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 conf->near_copies = nc;
2205 conf->far_copies = fc;
2206 conf->copies = nc*fc;
NeilBrownc93983b2006-06-26 00:27:41 -07002207 conf->far_offset = fo;
Andre Noll9d8f0362009-06-18 08:45:01 +10002208 conf->chunk_mask = mddev->chunk_sectors - 1;
2209 conf->chunk_shift = ffz(~mddev->chunk_sectors);
Andre Noll58c0fed2009-03-31 14:33:13 +11002210 size = mddev->dev_sectors >> conf->chunk_shift;
NeilBrown64a742b2007-02-28 20:11:18 -08002211 sector_div(size, fc);
2212 size = size * conf->raid_disks;
2213 sector_div(size, nc);
2214 /* 'size' is now the number of chunks in the array */
2215 /* calculate "used chunks per device" in 'stride' */
2216 stride = size * conf->copies;
NeilBrownaf03b8e2007-06-16 10:16:06 -07002217
2218 /* We need to round up when dividing by raid_disks to
2219 * get the stride size.
2220 */
2221 stride += conf->raid_disks - 1;
NeilBrown64a742b2007-02-28 20:11:18 -08002222 sector_div(stride, conf->raid_disks);
Andre Noll58c0fed2009-03-31 14:33:13 +11002223 mddev->dev_sectors = stride << conf->chunk_shift;
NeilBrown64a742b2007-02-28 20:11:18 -08002224
NeilBrownc93983b2006-06-26 00:27:41 -07002225 if (fo)
NeilBrown64a742b2007-02-28 20:11:18 -08002226 stride = 1;
2227 else
NeilBrownc93983b2006-06-26 00:27:41 -07002228 sector_div(stride, fc);
NeilBrown64a742b2007-02-28 20:11:18 -08002229 conf->stride = stride << conf->chunk_shift;
2230
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc,
2232 r10bio_pool_free, conf);
2233 if (!conf->r10bio_pool) {
2234 printk(KERN_ERR "raid10: couldn't allocate memory for %s\n",
2235 mdname(mddev));
2236 goto out_free_conf;
2237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238
NeilBrowned9bfdf2009-10-16 15:55:44 +11002239 conf->mddev = mddev;
Neil Browne7e72bf2008-05-14 16:05:54 -07002240 spin_lock_init(&conf->device_lock);
2241 mddev->queue->queue_lock = &conf->device_lock;
2242
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10002243 chunk_size = mddev->chunk_sectors << 9;
2244 blk_queue_io_min(mddev->queue, chunk_size);
2245 if (conf->raid_disks % conf->near_copies)
2246 blk_queue_io_opt(mddev->queue, chunk_size * conf->raid_disks);
2247 else
2248 blk_queue_io_opt(mddev->queue, chunk_size *
2249 (conf->raid_disks / conf->near_copies));
2250
Cheng Renquan159ec1f2009-01-09 08:31:08 +11002251 list_for_each_entry(rdev, &mddev->disks, same_set) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 disk_idx = rdev->raid_disk;
2253 if (disk_idx >= mddev->raid_disks
2254 || disk_idx < 0)
2255 continue;
2256 disk = conf->mirrors + disk_idx;
2257
2258 disk->rdev = rdev;
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10002259 disk_stack_limits(mddev->gendisk, rdev->bdev,
2260 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 /* as we don't honour merge_bvec_fn, we must never risk
NeilBrown627a2d32010-03-08 16:44:38 +11002262 * violating it, so limit max_segments to 1 lying
2263 * within a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 */
NeilBrown627a2d32010-03-08 16:44:38 +11002265 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
2266 blk_queue_max_segments(mddev->queue, 1);
2267 blk_queue_segment_boundary(mddev->queue,
2268 PAGE_CACHE_SIZE - 1);
2269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
2271 disk->head_position = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 INIT_LIST_HEAD(&conf->retry_list);
2274
2275 spin_lock_init(&conf->resync_lock);
NeilBrown0a27ec92006-01-06 00:20:13 -08002276 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
NeilBrown6d508242005-09-09 16:24:03 -07002278 /* need to check that every block has at least one working mirror */
2279 if (!enough(conf)) {
2280 printk(KERN_ERR "raid10: not enough operational mirrors for %s\n",
2281 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 goto out_free_conf;
2283 }
2284
2285 mddev->degraded = 0;
2286 for (i = 0; i < conf->raid_disks; i++) {
2287
2288 disk = conf->mirrors + i;
2289
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002290 if (!disk->rdev ||
NeilBrown2e333e82006-10-21 10:24:07 -07002291 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 disk->head_position = 0;
2293 mddev->degraded++;
Neil Brown8c2e8702008-06-28 08:30:52 +10002294 if (disk->rdev)
2295 conf->fullsync = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 }
2297 }
2298
2299
NeilBrown0da3c612009-09-23 18:09:45 +10002300 mddev->thread = md_register_thread(raid10d, mddev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 if (!mddev->thread) {
2302 printk(KERN_ERR
2303 "raid10: couldn't allocate thread for %s\n",
2304 mdname(mddev));
2305 goto out_free_conf;
2306 }
2307
Andre Noll8c6ac862009-06-18 08:48:06 +10002308 if (mddev->recovery_cp != MaxSector)
2309 printk(KERN_NOTICE "raid10: %s is not clean"
2310 " -- starting background reconstruction\n",
2311 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 printk(KERN_INFO
2313 "raid10: raid set %s active with %d out of %d devices\n",
2314 mdname(mddev), mddev->raid_disks - mddev->degraded,
2315 mddev->raid_disks);
2316 /*
2317 * Ok, everything is just fine now
2318 */
Dan Williams1f403622009-03-31 14:59:03 +11002319 md_set_array_sectors(mddev, raid10_size(mddev, 0, 0));
Dan Williamsb522adc2009-03-31 15:00:31 +11002320 mddev->resync_max_sectors = raid10_size(mddev, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
NeilBrown7a5febe2005-05-16 21:53:16 -07002322 mddev->queue->unplug_fn = raid10_unplug;
NeilBrown0d129222006-10-03 01:15:54 -07002323 mddev->queue->backing_dev_info.congested_fn = raid10_congested;
2324 mddev->queue->backing_dev_info.congested_data = mddev;
NeilBrown7a5febe2005-05-16 21:53:16 -07002325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 /* Calculate max read-ahead size.
2327 * We need to readahead at least twice a whole stripe....
2328 * maybe...
2329 */
2330 {
Andre Noll9d8f0362009-06-18 08:45:01 +10002331 int stripe = conf->raid_disks *
2332 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 stripe /= conf->near_copies;
2334 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
2335 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
2336 }
2337
2338 if (conf->near_copies < mddev->raid_disks)
2339 blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec);
Andre Nollac5e7112009-08-03 10:59:47 +10002340 md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 return 0;
2342
2343out_free_conf:
2344 if (conf->r10bio_pool)
2345 mempool_destroy(conf->r10bio_pool);
NeilBrown1345b1d2006-01-06 00:20:40 -08002346 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002347 kfree(conf->mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 kfree(conf);
2349 mddev->private = NULL;
2350out:
2351 return -EIO;
2352}
2353
2354static int stop(mddev_t *mddev)
2355{
NeilBrown070ec552009-06-16 16:54:21 +10002356 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
NeilBrown409c57f2009-03-31 14:39:39 +11002358 raise_barrier(conf, 0);
2359 lower_barrier(conf);
2360
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 md_unregister_thread(mddev->thread);
2362 mddev->thread = NULL;
2363 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
2364 if (conf->r10bio_pool)
2365 mempool_destroy(conf->r10bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002366 kfree(conf->mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 kfree(conf);
2368 mddev->private = NULL;
2369 return 0;
2370}
2371
NeilBrown6cce3b22006-01-06 00:20:16 -08002372static void raid10_quiesce(mddev_t *mddev, int state)
2373{
NeilBrown070ec552009-06-16 16:54:21 +10002374 conf_t *conf = mddev->private;
NeilBrown6cce3b22006-01-06 00:20:16 -08002375
2376 switch(state) {
2377 case 1:
2378 raise_barrier(conf, 0);
2379 break;
2380 case 0:
2381 lower_barrier(conf);
2382 break;
2383 }
NeilBrown6cce3b22006-01-06 00:20:16 -08002384}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
NeilBrown2604b702006-01-06 00:20:36 -08002386static struct mdk_personality raid10_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387{
2388 .name = "raid10",
NeilBrown2604b702006-01-06 00:20:36 -08002389 .level = 10,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 .owner = THIS_MODULE,
2391 .make_request = make_request,
2392 .run = run,
2393 .stop = stop,
2394 .status = status,
2395 .error_handler = error,
2396 .hot_add_disk = raid10_add_disk,
2397 .hot_remove_disk= raid10_remove_disk,
2398 .spare_active = raid10_spare_active,
2399 .sync_request = sync_request,
NeilBrown6cce3b22006-01-06 00:20:16 -08002400 .quiesce = raid10_quiesce,
Dan Williams80c3a6c2009-03-17 18:10:40 -07002401 .size = raid10_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402};
2403
2404static int __init raid_init(void)
2405{
NeilBrown2604b702006-01-06 00:20:36 -08002406 return register_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407}
2408
2409static void raid_exit(void)
2410{
NeilBrown2604b702006-01-06 00:20:36 -08002411 unregister_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412}
2413
2414module_init(raid_init);
2415module_exit(raid_exit);
2416MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11002417MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418MODULE_ALIAS("md-personality-9"); /* RAID10 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002419MODULE_ALIAS("md-raid10");
NeilBrown2604b702006-01-06 00:20:36 -08002420MODULE_ALIAS("md-level-10");