blob: 957a719e8c2f1545073d2d897a00ffdf17321878 [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 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -03008 * Base on code in raid1.c. See raid1.c for further copyright information.
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
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
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090021#include <linux/slab.h>
Stephen Rothwell25570722008-10-15 09:09:21 +110022#include <linux/delay.h>
NeilBrownbff61972009-03-31 14:33:13 +110023#include <linux/blkdev.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040024#include <linux/module.h>
NeilBrownbff61972009-03-31 14:33:13 +110025#include <linux/seq_file.h>
Christian Dietrich8bda4702011-07-27 11:00:36 +100026#include <linux/ratelimit.h>
NeilBrown3ea7daa2012-05-22 13:53:47 +100027#include <linux/kthread.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110028#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110029#include "raid10.h"
Trela, Maciejdab8b292010-03-08 16:02:45 +110030#include "raid0.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110031#include "bitmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33/*
34 * RAID10 provides a combination of RAID0 and RAID1 functionality.
35 * The layout of data is defined by
36 * chunk_size
37 * raid_disks
38 * near_copies (stored in low byte of layout)
39 * far_copies (stored in second byte of layout)
NeilBrownc93983b2006-06-26 00:27:41 -070040 * far_offset (stored in bit 16 of layout )
Jonathan Brassow475901a2013-02-21 13:28:10 +110041 * use_far_sets (stored in bit 17 of layout )
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 *
Jonathan Brassow475901a2013-02-21 13:28:10 +110043 * The data to be stored is divided into chunks using chunksize. Each device
44 * is divided into far_copies sections. In each section, chunks are laid out
45 * in a style similar to raid0, but near_copies copies of each chunk is stored
46 * (each on a different drive). The starting device for each section is offset
47 * near_copies from the starting device of the previous section. Thus there
48 * are (near_copies * far_copies) of each chunk, and each is on a different
49 * drive. near_copies and far_copies must be at least one, and their product
50 * is at most raid_disks.
NeilBrownc93983b2006-06-26 00:27:41 -070051 *
52 * If far_offset is true, then the far_copies are handled a bit differently.
Jonathan Brassow475901a2013-02-21 13:28:10 +110053 * The copies are still in different stripes, but instead of being very far
54 * apart on disk, there are adjacent stripes.
55 *
56 * The far and offset algorithms are handled slightly differently if
57 * 'use_far_sets' is true. In this case, the array's devices are grouped into
58 * sets that are (near_copies * far_copies) in size. The far copied stripes
59 * are still shifted by 'near_copies' devices, but this shifting stays confined
60 * to the set rather than the entire array. This is done to improve the number
61 * of device combinations that can fail without causing the array to fail.
62 * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk
63 * on a device):
64 * A B C D A B C D E
65 * ... ...
66 * D A B C E A B C D
67 * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s):
68 * [A B] [C D] [A B] [C D E]
69 * |...| |...| |...| | ... |
70 * [B A] [D C] [B A] [E C D]
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 */
72
73/*
74 * Number of guaranteed r10bios in case of extreme VM load:
75 */
76#define NR_RAID10_BIOS 256
77
Jonathan Brassow473e87c2012-07-31 10:03:52 +100078/* when we get a read error on a read-only array, we redirect to another
79 * device without failing the first device, or trying to over-write to
80 * correct the read error. To keep track of bad blocks on a per-bio
81 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
82 */
83#define IO_BLOCKED ((struct bio *)1)
84/* When we successfully write to a known bad-block, we need to remove the
85 * bad-block marking which must be done from process context. So we record
86 * the success by setting devs[n].bio to IO_MADE_GOOD
87 */
88#define IO_MADE_GOOD ((struct bio *)2)
89
90#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
91
92/* When there are this many requests queued to be written by
NeilBrown34db0cd2011-10-11 16:50:01 +110093 * the raid10 thread, we become 'congested' to provide back-pressure
94 * for writeback.
95 */
96static int max_queued_requests = 1024;
97
NeilBrowne879a872011-10-11 16:49:02 +110098static void allow_barrier(struct r10conf *conf);
99static void lower_barrier(struct r10conf *conf);
NeilBrown635f6412013-06-11 14:57:09 +1000100static int _enough(struct r10conf *conf, int previous, int ignore);
NeilBrown3ea7daa2012-05-22 13:53:47 +1000101static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
102 int *skipped);
103static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio);
104static void end_reshape_write(struct bio *bio, int error);
105static void end_reshape(struct r10conf *conf);
NeilBrown0a27ec92006-01-06 00:20:13 -0800106
Al Virodd0fc662005-10-07 07:46:04 +0100107static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
NeilBrowne879a872011-10-11 16:49:02 +1100109 struct r10conf *conf = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100110 int size = offsetof(struct r10bio, devs[conf->copies]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
NeilBrown69335ef2011-12-23 10:17:54 +1100112 /* allocate a r10bio with room for raid_disks entries in the
113 * bios array */
Jens Axboe7eaceac2011-03-10 08:52:07 +0100114 return kzalloc(size, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
117static void r10bio_pool_free(void *r10_bio, void *data)
118{
119 kfree(r10_bio);
120}
121
NeilBrown0310fa22008-08-05 15:54:14 +1000122/* Maximum size of each resync request */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123#define RESYNC_BLOCK_SIZE (64*1024)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
NeilBrown0310fa22008-08-05 15:54:14 +1000125/* amount of memory to reserve for resync requests */
126#define RESYNC_WINDOW (1024*1024)
127/* maximum number of concurrent requests, memory permitting */
128#define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130/*
131 * When performing a resync, we need to read and compare, so
132 * we need as many pages are there are copies.
133 * When performing a recovery, we need 2 bios, one for read,
134 * one for write (we recover only one drive per r10buf)
135 *
136 */
Al Virodd0fc662005-10-07 07:46:04 +0100137static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
NeilBrowne879a872011-10-11 16:49:02 +1100139 struct r10conf *conf = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 struct page *page;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100141 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 struct bio *bio;
143 int i, j;
144 int nalloc;
145
146 r10_bio = r10bio_pool_alloc(gfp_flags, conf);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100147 if (!r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
NeilBrown3ea7daa2012-05-22 13:53:47 +1000150 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) ||
151 test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 nalloc = conf->copies; /* resync */
153 else
154 nalloc = 2; /* recovery */
155
156 /*
157 * Allocate bios.
158 */
159 for (j = nalloc ; j-- ; ) {
NeilBrown67465572010-10-26 17:33:54 +1100160 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 if (!bio)
162 goto out_free_bio;
163 r10_bio->devs[j].bio = bio;
NeilBrown69335ef2011-12-23 10:17:54 +1100164 if (!conf->have_replacement)
165 continue;
166 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
167 if (!bio)
168 goto out_free_bio;
169 r10_bio->devs[j].repl_bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 }
171 /*
172 * Allocate RESYNC_PAGES data pages and attach them
173 * where needed.
174 */
175 for (j = 0 ; j < nalloc; j++) {
NeilBrown69335ef2011-12-23 10:17:54 +1100176 struct bio *rbio = r10_bio->devs[j].repl_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 bio = r10_bio->devs[j].bio;
178 for (i = 0; i < RESYNC_PAGES; i++) {
NeilBrown3ea7daa2012-05-22 13:53:47 +1000179 if (j > 0 && !test_bit(MD_RECOVERY_SYNC,
180 &conf->mddev->recovery)) {
181 /* we can share bv_page's during recovery
182 * and reshape */
Namhyung Kimc65060a2011-07-18 17:38:49 +1000183 struct bio *rbio = r10_bio->devs[0].bio;
184 page = rbio->bi_io_vec[i].bv_page;
185 get_page(page);
186 } else
187 page = alloc_page(gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 if (unlikely(!page))
189 goto out_free_pages;
190
191 bio->bi_io_vec[i].bv_page = page;
NeilBrown69335ef2011-12-23 10:17:54 +1100192 if (rbio)
193 rbio->bi_io_vec[i].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 }
195 }
196
197 return r10_bio;
198
199out_free_pages:
200 for ( ; i > 0 ; i--)
NeilBrown1345b1d2006-01-06 00:20:40 -0800201 safe_put_page(bio->bi_io_vec[i-1].bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 while (j--)
203 for (i = 0; i < RESYNC_PAGES ; i++)
NeilBrown1345b1d2006-01-06 00:20:40 -0800204 safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page);
majianpeng5fdd2cf2012-05-22 13:55:03 +1000205 j = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206out_free_bio:
majianpeng5fdd2cf2012-05-22 13:55:03 +1000207 for ( ; j < nalloc; j++) {
208 if (r10_bio->devs[j].bio)
209 bio_put(r10_bio->devs[j].bio);
NeilBrown69335ef2011-12-23 10:17:54 +1100210 if (r10_bio->devs[j].repl_bio)
211 bio_put(r10_bio->devs[j].repl_bio);
212 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 r10bio_pool_free(r10_bio, conf);
214 return NULL;
215}
216
217static void r10buf_pool_free(void *__r10_bio, void *data)
218{
219 int i;
NeilBrowne879a872011-10-11 16:49:02 +1100220 struct r10conf *conf = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100221 struct r10bio *r10bio = __r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 int j;
223
224 for (j=0; j < conf->copies; j++) {
225 struct bio *bio = r10bio->devs[j].bio;
226 if (bio) {
227 for (i = 0; i < RESYNC_PAGES; i++) {
NeilBrown1345b1d2006-01-06 00:20:40 -0800228 safe_put_page(bio->bi_io_vec[i].bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 bio->bi_io_vec[i].bv_page = NULL;
230 }
231 bio_put(bio);
232 }
NeilBrown69335ef2011-12-23 10:17:54 +1100233 bio = r10bio->devs[j].repl_bio;
234 if (bio)
235 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 }
237 r10bio_pool_free(r10bio, conf);
238}
239
NeilBrowne879a872011-10-11 16:49:02 +1100240static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
242 int i;
243
244 for (i = 0; i < conf->copies; i++) {
245 struct bio **bio = & r10_bio->devs[i].bio;
NeilBrown749c55e2011-07-28 11:39:24 +1000246 if (!BIO_SPECIAL(*bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 bio_put(*bio);
248 *bio = NULL;
NeilBrown69335ef2011-12-23 10:17:54 +1100249 bio = &r10_bio->devs[i].repl_bio;
250 if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio))
251 bio_put(*bio);
252 *bio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 }
254}
255
NeilBrown9f2c9d12011-10-11 16:48:43 +1100256static void free_r10bio(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
NeilBrowne879a872011-10-11 16:49:02 +1100258 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 put_all_bios(conf, r10_bio);
261 mempool_free(r10_bio, conf->r10bio_pool);
262}
263
NeilBrown9f2c9d12011-10-11 16:48:43 +1100264static void put_buf(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
NeilBrowne879a872011-10-11 16:49:02 +1100266 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268 mempool_free(r10_bio, conf->r10buf_pool);
269
NeilBrown0a27ec92006-01-06 00:20:13 -0800270 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
NeilBrown9f2c9d12011-10-11 16:48:43 +1100273static void reschedule_retry(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
275 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +1100276 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +1100277 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279 spin_lock_irqsave(&conf->device_lock, flags);
280 list_add(&r10_bio->retry_list, &conf->retry_list);
NeilBrown4443ae12006-01-06 00:20:28 -0800281 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 spin_unlock_irqrestore(&conf->device_lock, flags);
283
Arthur Jones388667b2008-07-25 12:03:38 -0700284 /* wake up frozen array... */
285 wake_up(&conf->wait_barrier);
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 md_wakeup_thread(mddev->thread);
288}
289
290/*
291 * raid_end_bio_io() is called when we have finished servicing a mirrored
292 * operation and are ready to return a success/failure code to the buffer
293 * cache layer.
294 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100295static void raid_end_bio_io(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
297 struct bio *bio = r10_bio->master_bio;
NeilBrown856e08e2011-07-28 11:39:23 +1000298 int done;
NeilBrowne879a872011-10-11 16:49:02 +1100299 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
NeilBrown856e08e2011-07-28 11:39:23 +1000301 if (bio->bi_phys_segments) {
302 unsigned long flags;
303 spin_lock_irqsave(&conf->device_lock, flags);
304 bio->bi_phys_segments--;
305 done = (bio->bi_phys_segments == 0);
306 spin_unlock_irqrestore(&conf->device_lock, flags);
307 } else
308 done = 1;
309 if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
310 clear_bit(BIO_UPTODATE, &bio->bi_flags);
311 if (done) {
312 bio_endio(bio, 0);
313 /*
314 * Wake up any possible resync thread that waits for the device
315 * to go idle.
316 */
317 allow_barrier(conf);
318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 free_r10bio(r10_bio);
320}
321
322/*
323 * Update disk head position estimator based on IRQ completion info.
324 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100325static inline void update_head_pos(int slot, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326{
NeilBrowne879a872011-10-11 16:49:02 +1100327 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
329 conf->mirrors[r10_bio->devs[slot].devnum].head_position =
330 r10_bio->devs[slot].addr + (r10_bio->sectors);
331}
332
Namhyung Kim778ca012011-07-18 17:38:47 +1000333/*
334 * Find the disk number which triggered given bio
335 */
NeilBrowne879a872011-10-11 16:49:02 +1100336static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio,
NeilBrown69335ef2011-12-23 10:17:54 +1100337 struct bio *bio, int *slotp, int *replp)
Namhyung Kim778ca012011-07-18 17:38:47 +1000338{
339 int slot;
NeilBrown69335ef2011-12-23 10:17:54 +1100340 int repl = 0;
Namhyung Kim778ca012011-07-18 17:38:47 +1000341
NeilBrown69335ef2011-12-23 10:17:54 +1100342 for (slot = 0; slot < conf->copies; slot++) {
Namhyung Kim778ca012011-07-18 17:38:47 +1000343 if (r10_bio->devs[slot].bio == bio)
344 break;
NeilBrown69335ef2011-12-23 10:17:54 +1100345 if (r10_bio->devs[slot].repl_bio == bio) {
346 repl = 1;
347 break;
348 }
349 }
Namhyung Kim778ca012011-07-18 17:38:47 +1000350
351 BUG_ON(slot == conf->copies);
352 update_head_pos(slot, r10_bio);
353
NeilBrown749c55e2011-07-28 11:39:24 +1000354 if (slotp)
355 *slotp = slot;
NeilBrown69335ef2011-12-23 10:17:54 +1100356 if (replp)
357 *replp = repl;
Namhyung Kim778ca012011-07-18 17:38:47 +1000358 return r10_bio->devs[slot].devnum;
359}
360
NeilBrown6712ecf2007-09-27 12:47:43 +0200361static void raid10_end_read_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
363 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100364 struct r10bio *r10_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 int slot, dev;
NeilBrownabbf0982011-12-23 10:17:54 +1100366 struct md_rdev *rdev;
NeilBrowne879a872011-10-11 16:49:02 +1100367 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
370 slot = r10_bio->read_slot;
371 dev = r10_bio->devs[slot].devnum;
NeilBrownabbf0982011-12-23 10:17:54 +1100372 rdev = r10_bio->devs[slot].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 /*
374 * this branch is our 'one mirror IO has finished' event handler:
375 */
NeilBrown4443ae12006-01-06 00:20:28 -0800376 update_head_pos(slot, r10_bio);
377
378 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 /*
380 * Set R10BIO_Uptodate in our master bio, so that
381 * we will return a good error code to the higher
382 * levels even if IO on some other mirrored buffer fails.
383 *
384 * The 'master' represents the composite IO operation to
385 * user-side. So if something waits for IO, then it will
386 * wait for the 'master' bio.
387 */
388 set_bit(R10BIO_Uptodate, &r10_bio->state);
NeilBrownfae8cc52012-02-14 11:10:10 +1100389 } else {
390 /* If all other devices that store this block have
391 * failed, we want to return the error upwards rather
392 * than fail the last device. Here we redefine
393 * "uptodate" to mean "Don't want to retry"
394 */
NeilBrown635f6412013-06-11 14:57:09 +1000395 if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state),
396 rdev->raid_disk))
NeilBrownfae8cc52012-02-14 11:10:10 +1100397 uptodate = 1;
NeilBrownfae8cc52012-02-14 11:10:10 +1100398 }
399 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 raid_end_bio_io(r10_bio);
NeilBrownabbf0982011-12-23 10:17:54 +1100401 rdev_dec_pending(rdev, conf->mddev);
NeilBrown4443ae12006-01-06 00:20:28 -0800402 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 /*
NeilBrown7c4e06f2011-05-11 14:53:17 +1000404 * oops, read error - keep the refcount on the rdev
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 */
406 char b[BDEVNAME_SIZE];
Christian Dietrich8bda4702011-07-27 11:00:36 +1000407 printk_ratelimited(KERN_ERR
408 "md/raid10:%s: %s: rescheduling sector %llu\n",
409 mdname(conf->mddev),
NeilBrownabbf0982011-12-23 10:17:54 +1100410 bdevname(rdev->bdev, b),
Christian Dietrich8bda4702011-07-27 11:00:36 +1000411 (unsigned long long)r10_bio->sector);
NeilBrown856e08e2011-07-28 11:39:23 +1000412 set_bit(R10BIO_ReadError, &r10_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 reschedule_retry(r10_bio);
414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
NeilBrown9f2c9d12011-10-11 16:48:43 +1100417static void close_write(struct r10bio *r10_bio)
NeilBrownbd870a12011-07-28 11:39:24 +1000418{
419 /* clear the bitmap if all writes complete successfully */
420 bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector,
421 r10_bio->sectors,
422 !test_bit(R10BIO_Degraded, &r10_bio->state),
423 0);
424 md_write_end(r10_bio->mddev);
425}
426
NeilBrown9f2c9d12011-10-11 16:48:43 +1100427static void one_write_done(struct r10bio *r10_bio)
NeilBrown19d5f832011-09-10 17:21:17 +1000428{
429 if (atomic_dec_and_test(&r10_bio->remaining)) {
430 if (test_bit(R10BIO_WriteError, &r10_bio->state))
431 reschedule_retry(r10_bio);
432 else {
433 close_write(r10_bio);
434 if (test_bit(R10BIO_MadeGood, &r10_bio->state))
435 reschedule_retry(r10_bio);
436 else
437 raid_end_bio_io(r10_bio);
438 }
439 }
440}
441
NeilBrown6712ecf2007-09-27 12:47:43 +0200442static void raid10_end_write_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
444 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100445 struct r10bio *r10_bio = bio->bi_private;
Namhyung Kim778ca012011-07-18 17:38:47 +1000446 int dev;
NeilBrown749c55e2011-07-28 11:39:24 +1000447 int dec_rdev = 1;
NeilBrowne879a872011-10-11 16:49:02 +1100448 struct r10conf *conf = r10_bio->mddev->private;
NeilBrown475b0322011-12-23 10:17:55 +1100449 int slot, repl;
NeilBrown4ca40c22011-12-23 10:17:55 +1100450 struct md_rdev *rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
NeilBrown475b0322011-12-23 10:17:55 +1100452 dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
NeilBrown475b0322011-12-23 10:17:55 +1100454 if (repl)
455 rdev = conf->mirrors[dev].replacement;
NeilBrown4ca40c22011-12-23 10:17:55 +1100456 if (!rdev) {
457 smp_rmb();
458 repl = 0;
NeilBrown475b0322011-12-23 10:17:55 +1100459 rdev = conf->mirrors[dev].rdev;
NeilBrown4ca40c22011-12-23 10:17:55 +1100460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 /*
462 * this branch is our 'one mirror IO has finished' event handler:
463 */
NeilBrown6cce3b22006-01-06 00:20:16 -0800464 if (!uptodate) {
NeilBrown475b0322011-12-23 10:17:55 +1100465 if (repl)
466 /* Never record new bad blocks to replacement,
467 * just fail it.
468 */
469 md_error(rdev->mddev, rdev);
470 else {
471 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +1100472 if (!test_and_set_bit(WantReplacement, &rdev->flags))
473 set_bit(MD_RECOVERY_NEEDED,
474 &rdev->mddev->recovery);
NeilBrown475b0322011-12-23 10:17:55 +1100475 set_bit(R10BIO_WriteError, &r10_bio->state);
476 dec_rdev = 0;
477 }
NeilBrown749c55e2011-07-28 11:39:24 +1000478 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 /*
480 * Set R10BIO_Uptodate in our master bio, so that
481 * we will return a good error code for to the higher
482 * levels even if IO on some other mirrored buffer fails.
483 *
484 * The 'master' represents the composite IO operation to
485 * user-side. So if something waits for IO, then it will
486 * wait for the 'master' bio.
487 */
NeilBrown749c55e2011-07-28 11:39:24 +1000488 sector_t first_bad;
489 int bad_sectors;
490
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300491 /*
492 * Do not set R10BIO_Uptodate if the current device is
493 * rebuilding or Faulty. This is because we cannot use
494 * such device for properly reading the data back (we could
495 * potentially use it, if the current write would have felt
496 * before rdev->recovery_offset, but for simplicity we don't
497 * check this here.
498 */
499 if (test_bit(In_sync, &rdev->flags) &&
500 !test_bit(Faulty, &rdev->flags))
501 set_bit(R10BIO_Uptodate, &r10_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
NeilBrown749c55e2011-07-28 11:39:24 +1000503 /* Maybe we can clear some bad blocks. */
NeilBrown475b0322011-12-23 10:17:55 +1100504 if (is_badblock(rdev,
NeilBrown749c55e2011-07-28 11:39:24 +1000505 r10_bio->devs[slot].addr,
506 r10_bio->sectors,
507 &first_bad, &bad_sectors)) {
508 bio_put(bio);
NeilBrown475b0322011-12-23 10:17:55 +1100509 if (repl)
510 r10_bio->devs[slot].repl_bio = IO_MADE_GOOD;
511 else
512 r10_bio->devs[slot].bio = IO_MADE_GOOD;
NeilBrown749c55e2011-07-28 11:39:24 +1000513 dec_rdev = 0;
514 set_bit(R10BIO_MadeGood, &r10_bio->state);
515 }
516 }
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 /*
519 *
520 * Let's see if all mirrored write operations have finished
521 * already.
522 */
NeilBrown19d5f832011-09-10 17:21:17 +1000523 one_write_done(r10_bio);
NeilBrown749c55e2011-07-28 11:39:24 +1000524 if (dec_rdev)
NeilBrown884162d2012-11-22 15:12:09 +1100525 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528/*
529 * RAID10 layout manager
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300530 * As well as the chunksize and raid_disks count, there are two
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 * parameters: near_copies and far_copies.
532 * near_copies * far_copies must be <= raid_disks.
533 * Normally one of these will be 1.
534 * If both are 1, we get raid0.
535 * If near_copies == raid_disks, we get raid1.
536 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300537 * Chunks are laid out in raid0 style with near_copies copies of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 * first chunk, followed by near_copies copies of the next chunk and
539 * so on.
540 * If far_copies > 1, then after 1/far_copies of the array has been assigned
541 * as described above, we start again with a device offset of near_copies.
542 * So we effectively have another copy of the whole array further down all
543 * the drives, but with blocks on different drives.
544 * With this layout, and block is never stored twice on the one device.
545 *
546 * raid10_find_phys finds the sector offset of a given virtual sector
NeilBrownc93983b2006-06-26 00:27:41 -0700547 * on each device that it is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 *
549 * raid10_find_virt does the reverse mapping, from a device and a
550 * sector offset to a virtual address
551 */
552
NeilBrownf8c9e742012-05-21 09:28:33 +1000553static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 int n,f;
556 sector_t sector;
557 sector_t chunk;
558 sector_t stripe;
559 int dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 int slot = 0;
Jonathan Brassow9a3152a2013-02-21 13:28:10 +1100561 int last_far_set_start, last_far_set_size;
562
563 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
564 last_far_set_start *= geo->far_set_size;
565
566 last_far_set_size = geo->far_set_size;
567 last_far_set_size += (geo->raid_disks % geo->far_set_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
569 /* now calculate first sector/dev */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000570 chunk = r10bio->sector >> geo->chunk_shift;
571 sector = r10bio->sector & geo->chunk_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
NeilBrown5cf00fc2012-05-21 09:28:20 +1000573 chunk *= geo->near_copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 stripe = chunk;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000575 dev = sector_div(stripe, geo->raid_disks);
576 if (geo->far_offset)
577 stripe *= geo->far_copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
NeilBrown5cf00fc2012-05-21 09:28:20 +1000579 sector += stripe << geo->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 /* and calculate all the others */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000582 for (n = 0; n < geo->near_copies; n++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 int d = dev;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100584 int set;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 sector_t s = sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 r10bio->devs[slot].devnum = d;
Jonathan Brassow4c0ca262013-02-21 13:28:09 +1100587 r10bio->devs[slot].addr = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 slot++;
589
NeilBrown5cf00fc2012-05-21 09:28:20 +1000590 for (f = 1; f < geo->far_copies; f++) {
Jonathan Brassow475901a2013-02-21 13:28:10 +1100591 set = d / geo->far_set_size;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000592 d += geo->near_copies;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100593
Jonathan Brassow9a3152a2013-02-21 13:28:10 +1100594 if ((geo->raid_disks % geo->far_set_size) &&
595 (d > last_far_set_start)) {
596 d -= last_far_set_start;
597 d %= last_far_set_size;
598 d += last_far_set_start;
599 } else {
600 d %= geo->far_set_size;
601 d += geo->far_set_size * set;
602 }
NeilBrown5cf00fc2012-05-21 09:28:20 +1000603 s += geo->stride;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 r10bio->devs[slot].devnum = d;
605 r10bio->devs[slot].addr = s;
606 slot++;
607 }
608 dev++;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000609 if (dev >= geo->raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 dev = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000611 sector += (geo->chunk_mask + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 }
613 }
NeilBrownf8c9e742012-05-21 09:28:33 +1000614}
615
616static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio)
617{
618 struct geom *geo = &conf->geo;
619
620 if (conf->reshape_progress != MaxSector &&
621 ((r10bio->sector >= conf->reshape_progress) !=
622 conf->mddev->reshape_backwards)) {
623 set_bit(R10BIO_Previous, &r10bio->state);
624 geo = &conf->prev;
625 } else
626 clear_bit(R10BIO_Previous, &r10bio->state);
627
628 __raid10_find_phys(geo, r10bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
NeilBrowne879a872011-10-11 16:49:02 +1100631static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 sector_t offset, chunk, vchunk;
NeilBrownf8c9e742012-05-21 09:28:33 +1000634 /* Never use conf->prev as this is only called during resync
635 * or recovery, so reshape isn't happening
636 */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000637 struct geom *geo = &conf->geo;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100638 int far_set_start = (dev / geo->far_set_size) * geo->far_set_size;
639 int far_set_size = geo->far_set_size;
Jonathan Brassow9a3152a2013-02-21 13:28:10 +1100640 int last_far_set_start;
641
642 if (geo->raid_disks % geo->far_set_size) {
643 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
644 last_far_set_start *= geo->far_set_size;
645
646 if (dev >= last_far_set_start) {
647 far_set_size = geo->far_set_size;
648 far_set_size += (geo->raid_disks % geo->far_set_size);
649 far_set_start = last_far_set_start;
650 }
651 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
NeilBrown5cf00fc2012-05-21 09:28:20 +1000653 offset = sector & geo->chunk_mask;
654 if (geo->far_offset) {
NeilBrownc93983b2006-06-26 00:27:41 -0700655 int fc;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000656 chunk = sector >> geo->chunk_shift;
657 fc = sector_div(chunk, geo->far_copies);
658 dev -= fc * geo->near_copies;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100659 if (dev < far_set_start)
660 dev += far_set_size;
NeilBrownc93983b2006-06-26 00:27:41 -0700661 } else {
NeilBrown5cf00fc2012-05-21 09:28:20 +1000662 while (sector >= geo->stride) {
663 sector -= geo->stride;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100664 if (dev < (geo->near_copies + far_set_start))
665 dev += far_set_size - geo->near_copies;
NeilBrownc93983b2006-06-26 00:27:41 -0700666 else
NeilBrown5cf00fc2012-05-21 09:28:20 +1000667 dev -= geo->near_copies;
NeilBrownc93983b2006-06-26 00:27:41 -0700668 }
NeilBrown5cf00fc2012-05-21 09:28:20 +1000669 chunk = sector >> geo->chunk_shift;
NeilBrownc93983b2006-06-26 00:27:41 -0700670 }
NeilBrown5cf00fc2012-05-21 09:28:20 +1000671 vchunk = chunk * geo->raid_disks + dev;
672 sector_div(vchunk, geo->near_copies);
673 return (vchunk << geo->chunk_shift) + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676/**
677 * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
678 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200679 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 * @biovec: the request that could be merged to it.
681 *
682 * Return amount of bytes we can accept at this offset
NeilBrown050b6612012-03-19 12:46:39 +1100683 * This requires checking for end-of-chunk if near_copies != raid_disks,
684 * and for subordinate merge_bvec_fns if merge_check_needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200686static int raid10_mergeable_bvec(struct request_queue *q,
687 struct bvec_merge_data *bvm,
688 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
NeilBrownfd01b882011-10-11 16:47:53 +1100690 struct mddev *mddev = q->queuedata;
NeilBrown050b6612012-03-19 12:46:39 +1100691 struct r10conf *conf = mddev->private;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200692 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 int max;
NeilBrown3ea7daa2012-05-22 13:53:47 +1000694 unsigned int chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200695 unsigned int bio_sectors = bvm->bi_size >> 9;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000696 struct geom *geo = &conf->geo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
NeilBrown3ea7daa2012-05-22 13:53:47 +1000698 chunk_sectors = (conf->geo.chunk_mask & conf->prev.chunk_mask) + 1;
NeilBrownf8c9e742012-05-21 09:28:33 +1000699 if (conf->reshape_progress != MaxSector &&
700 ((sector >= conf->reshape_progress) !=
701 conf->mddev->reshape_backwards))
702 geo = &conf->prev;
703
NeilBrown5cf00fc2012-05-21 09:28:20 +1000704 if (geo->near_copies < geo->raid_disks) {
NeilBrown050b6612012-03-19 12:46:39 +1100705 max = (chunk_sectors - ((sector & (chunk_sectors - 1))
706 + bio_sectors)) << 9;
707 if (max < 0)
708 /* bio_add cannot handle a negative return */
709 max = 0;
710 if (max <= biovec->bv_len && bio_sectors == 0)
711 return biovec->bv_len;
712 } else
713 max = biovec->bv_len;
714
715 if (mddev->merge_check_needed) {
NeilBrowne0ee7782012-08-18 09:51:42 +1000716 struct {
717 struct r10bio r10_bio;
718 struct r10dev devs[conf->copies];
719 } on_stack;
720 struct r10bio *r10_bio = &on_stack.r10_bio;
NeilBrown050b6612012-03-19 12:46:39 +1100721 int s;
NeilBrownf8c9e742012-05-21 09:28:33 +1000722 if (conf->reshape_progress != MaxSector) {
723 /* Cannot give any guidance during reshape */
724 if (max <= biovec->bv_len && bio_sectors == 0)
725 return biovec->bv_len;
726 return 0;
727 }
NeilBrowne0ee7782012-08-18 09:51:42 +1000728 r10_bio->sector = sector;
729 raid10_find_phys(conf, r10_bio);
NeilBrown050b6612012-03-19 12:46:39 +1100730 rcu_read_lock();
731 for (s = 0; s < conf->copies; s++) {
NeilBrowne0ee7782012-08-18 09:51:42 +1000732 int disk = r10_bio->devs[s].devnum;
NeilBrown050b6612012-03-19 12:46:39 +1100733 struct md_rdev *rdev = rcu_dereference(
734 conf->mirrors[disk].rdev);
735 if (rdev && !test_bit(Faulty, &rdev->flags)) {
736 struct request_queue *q =
737 bdev_get_queue(rdev->bdev);
738 if (q->merge_bvec_fn) {
NeilBrowne0ee7782012-08-18 09:51:42 +1000739 bvm->bi_sector = r10_bio->devs[s].addr
NeilBrown050b6612012-03-19 12:46:39 +1100740 + rdev->data_offset;
741 bvm->bi_bdev = rdev->bdev;
742 max = min(max, q->merge_bvec_fn(
743 q, bvm, biovec));
744 }
745 }
746 rdev = rcu_dereference(conf->mirrors[disk].replacement);
747 if (rdev && !test_bit(Faulty, &rdev->flags)) {
748 struct request_queue *q =
749 bdev_get_queue(rdev->bdev);
750 if (q->merge_bvec_fn) {
NeilBrowne0ee7782012-08-18 09:51:42 +1000751 bvm->bi_sector = r10_bio->devs[s].addr
NeilBrown050b6612012-03-19 12:46:39 +1100752 + rdev->data_offset;
753 bvm->bi_bdev = rdev->bdev;
754 max = min(max, q->merge_bvec_fn(
755 q, bvm, biovec));
756 }
757 }
758 }
759 rcu_read_unlock();
760 }
761 return max;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
764/*
765 * This routine returns the disk from which the requested read should
766 * be done. There is a per-array 'next expected sequential IO' sector
767 * number - if this matches on the next IO then we use the last disk.
768 * There is also a per-disk 'last know head position' sector that is
769 * maintained from IRQ contexts, both the normal and the resync IO
770 * completion handlers update this position correctly. If there is no
771 * perfect sequential match then we pick the disk whose head is closest.
772 *
773 * If there are 2 mirrors in the same 2 devices, performance degrades
774 * because position is mirror, not device based.
775 *
776 * The rdev for the device selected will have nr_pending incremented.
777 */
778
779/*
780 * FIXME: possibly should rethink readbalancing and do it differently
781 * depending on near_copies / far_copies geometry.
782 */
NeilBrown96c3fd12011-12-23 10:17:54 +1100783static struct md_rdev *read_balance(struct r10conf *conf,
784 struct r10bio *r10_bio,
785 int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000787 const sector_t this_sector = r10_bio->sector;
NeilBrown56d99122011-05-11 14:27:03 +1000788 int disk, slot;
NeilBrown856e08e2011-07-28 11:39:23 +1000789 int sectors = r10_bio->sectors;
790 int best_good_sectors;
NeilBrown56d99122011-05-11 14:27:03 +1000791 sector_t new_distance, best_dist;
Jonathan Brassow3bbae042012-07-31 10:03:52 +1000792 struct md_rdev *best_rdev, *rdev = NULL;
NeilBrown56d99122011-05-11 14:27:03 +1000793 int do_balance;
794 int best_slot;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000795 struct geom *geo = &conf->geo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
797 raid10_find_phys(conf, r10_bio);
798 rcu_read_lock();
NeilBrown56d99122011-05-11 14:27:03 +1000799retry:
NeilBrown856e08e2011-07-28 11:39:23 +1000800 sectors = r10_bio->sectors;
NeilBrown56d99122011-05-11 14:27:03 +1000801 best_slot = -1;
NeilBrownabbf0982011-12-23 10:17:54 +1100802 best_rdev = NULL;
NeilBrown56d99122011-05-11 14:27:03 +1000803 best_dist = MaxSector;
NeilBrown856e08e2011-07-28 11:39:23 +1000804 best_good_sectors = 0;
NeilBrown56d99122011-05-11 14:27:03 +1000805 do_balance = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 /*
807 * Check if we can balance. We can balance on the whole
NeilBrown6cce3b22006-01-06 00:20:16 -0800808 * device if no resync is going on (recovery is ok), or below
809 * the resync window. We take the first readable disk when
810 * above the resync window.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 */
812 if (conf->mddev->recovery_cp < MaxSector
NeilBrown56d99122011-05-11 14:27:03 +1000813 && (this_sector + sectors >= conf->next_resync))
814 do_balance = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
NeilBrown56d99122011-05-11 14:27:03 +1000816 for (slot = 0; slot < conf->copies ; slot++) {
NeilBrown856e08e2011-07-28 11:39:23 +1000817 sector_t first_bad;
818 int bad_sectors;
819 sector_t dev_sector;
820
NeilBrown56d99122011-05-11 14:27:03 +1000821 if (r10_bio->devs[slot].bio == IO_BLOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 continue;
NeilBrown56d99122011-05-11 14:27:03 +1000823 disk = r10_bio->devs[slot].devnum;
NeilBrownabbf0982011-12-23 10:17:54 +1100824 rdev = rcu_dereference(conf->mirrors[disk].replacement);
825 if (rdev == NULL || test_bit(Faulty, &rdev->flags) ||
NeilBrown050b6612012-03-19 12:46:39 +1100826 test_bit(Unmerged, &rdev->flags) ||
NeilBrownabbf0982011-12-23 10:17:54 +1100827 r10_bio->devs[slot].addr + sectors > rdev->recovery_offset)
828 rdev = rcu_dereference(conf->mirrors[disk].rdev);
NeilBrown050b6612012-03-19 12:46:39 +1100829 if (rdev == NULL ||
830 test_bit(Faulty, &rdev->flags) ||
831 test_bit(Unmerged, &rdev->flags))
NeilBrownabbf0982011-12-23 10:17:54 +1100832 continue;
833 if (!test_bit(In_sync, &rdev->flags) &&
834 r10_bio->devs[slot].addr + sectors > rdev->recovery_offset)
NeilBrown56d99122011-05-11 14:27:03 +1000835 continue;
836
NeilBrown856e08e2011-07-28 11:39:23 +1000837 dev_sector = r10_bio->devs[slot].addr;
838 if (is_badblock(rdev, dev_sector, sectors,
839 &first_bad, &bad_sectors)) {
840 if (best_dist < MaxSector)
841 /* Already have a better slot */
842 continue;
843 if (first_bad <= dev_sector) {
844 /* Cannot read here. If this is the
845 * 'primary' device, then we must not read
846 * beyond 'bad_sectors' from another device.
847 */
848 bad_sectors -= (dev_sector - first_bad);
849 if (!do_balance && sectors > bad_sectors)
850 sectors = bad_sectors;
851 if (best_good_sectors > sectors)
852 best_good_sectors = sectors;
853 } else {
854 sector_t good_sectors =
855 first_bad - dev_sector;
856 if (good_sectors > best_good_sectors) {
857 best_good_sectors = good_sectors;
858 best_slot = slot;
NeilBrownabbf0982011-12-23 10:17:54 +1100859 best_rdev = rdev;
NeilBrown856e08e2011-07-28 11:39:23 +1000860 }
861 if (!do_balance)
862 /* Must read from here */
863 break;
864 }
865 continue;
866 } else
867 best_good_sectors = sectors;
868
NeilBrown56d99122011-05-11 14:27:03 +1000869 if (!do_balance)
870 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
NeilBrown22dfdf52005-11-28 13:44:09 -0800872 /* This optimisation is debatable, and completely destroys
873 * sequential read speed for 'far copies' arrays. So only
874 * keep it for 'near' arrays, and review those later.
875 */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000876 if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 break;
Keld Simonsen8ed3a192008-03-04 14:29:34 -0800878
879 /* for far > 1 always use the lowest address */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000880 if (geo->far_copies > 1)
NeilBrown56d99122011-05-11 14:27:03 +1000881 new_distance = r10_bio->devs[slot].addr;
Keld Simonsen8ed3a192008-03-04 14:29:34 -0800882 else
NeilBrown56d99122011-05-11 14:27:03 +1000883 new_distance = abs(r10_bio->devs[slot].addr -
884 conf->mirrors[disk].head_position);
885 if (new_distance < best_dist) {
886 best_dist = new_distance;
887 best_slot = slot;
NeilBrownabbf0982011-12-23 10:17:54 +1100888 best_rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890 }
NeilBrownabbf0982011-12-23 10:17:54 +1100891 if (slot >= conf->copies) {
NeilBrown56d99122011-05-11 14:27:03 +1000892 slot = best_slot;
NeilBrownabbf0982011-12-23 10:17:54 +1100893 rdev = best_rdev;
894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
NeilBrown56d99122011-05-11 14:27:03 +1000896 if (slot >= 0) {
NeilBrown56d99122011-05-11 14:27:03 +1000897 atomic_inc(&rdev->nr_pending);
898 if (test_bit(Faulty, &rdev->flags)) {
899 /* Cannot risk returning a device that failed
900 * before we inc'ed nr_pending
901 */
902 rdev_dec_pending(rdev, conf->mddev);
903 goto retry;
904 }
905 r10_bio->read_slot = slot;
906 } else
NeilBrown96c3fd12011-12-23 10:17:54 +1100907 rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 rcu_read_unlock();
NeilBrown856e08e2011-07-28 11:39:23 +1000909 *max_sectors = best_good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
NeilBrown96c3fd12011-12-23 10:17:54 +1100911 return rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912}
913
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +1000914int md_raid10_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700915{
NeilBrowne879a872011-10-11 16:49:02 +1100916 struct r10conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700917 int i, ret = 0;
918
NeilBrown34db0cd2011-10-11 16:50:01 +1100919 if ((bits & (1 << BDI_async_congested)) &&
920 conf->pending_count >= max_queued_requests)
921 return 1;
922
NeilBrown0d129222006-10-03 01:15:54 -0700923 rcu_read_lock();
NeilBrownf8c9e742012-05-21 09:28:33 +1000924 for (i = 0;
925 (i < conf->geo.raid_disks || i < conf->prev.raid_disks)
926 && ret == 0;
927 i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100928 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700929 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200930 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700931
932 ret |= bdi_congested(&q->backing_dev_info, bits);
933 }
934 }
935 rcu_read_unlock();
936 return ret;
937}
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +1000938EXPORT_SYMBOL_GPL(md_raid10_congested);
939
940static int raid10_congested(void *data, int bits)
941{
942 struct mddev *mddev = data;
943
944 return mddev_congested(mddev, bits) ||
945 md_raid10_congested(mddev, bits);
946}
NeilBrown0d129222006-10-03 01:15:54 -0700947
NeilBrowne879a872011-10-11 16:49:02 +1100948static void flush_pending_writes(struct r10conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800949{
950 /* Any writes that have been queued but are awaiting
951 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800952 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800953 spin_lock_irq(&conf->device_lock);
954
955 if (conf->pending_bio_list.head) {
956 struct bio *bio;
957 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100958 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800959 spin_unlock_irq(&conf->device_lock);
960 /* flush any pending bitmap writes to disk
961 * before proceeding w/ I/O */
962 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100963 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800964
965 while (bio) { /* submit pending writes */
966 struct bio *next = bio->bi_next;
967 bio->bi_next = NULL;
Shaohua Li532a2a32012-10-11 13:30:52 +1100968 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
969 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
970 /* Just ignore it */
971 bio_endio(bio, 0);
972 else
973 generic_make_request(bio);
NeilBrowna35e63e2008-03-04 14:29:29 -0800974 bio = next;
975 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800976 } else
977 spin_unlock_irq(&conf->device_lock);
NeilBrowna35e63e2008-03-04 14:29:29 -0800978}
Jens Axboe7eaceac2011-03-10 08:52:07 +0100979
NeilBrown0a27ec92006-01-06 00:20:13 -0800980/* Barriers....
981 * Sometimes we need to suspend IO while we do something else,
982 * either some resync/recovery, or reconfigure the array.
983 * To do this we raise a 'barrier'.
984 * The 'barrier' is a counter that can be raised multiple times
985 * to count how many activities are happening which preclude
986 * normal IO.
987 * We can only raise the barrier if there is no pending IO.
988 * i.e. if nr_pending == 0.
989 * We choose only to raise the barrier if no-one is waiting for the
990 * barrier to go down. This means that as soon as an IO request
991 * is ready, no other operations which require a barrier will start
992 * until the IO request has had a chance.
993 *
994 * So: regular IO calls 'wait_barrier'. When that returns there
995 * is no backgroup IO happening, It must arrange to call
996 * allow_barrier when it has finished its IO.
997 * backgroup IO calls must call raise_barrier. Once that returns
998 * there is no normal IO happeing. It must arrange to call
999 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
NeilBrowne879a872011-10-11 16:49:02 +11001002static void raise_barrier(struct r10conf *conf, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003{
NeilBrown6cce3b22006-01-06 00:20:16 -08001004 BUG_ON(force && !conf->barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 spin_lock_irq(&conf->resync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
NeilBrown6cce3b22006-01-06 00:20:16 -08001007 /* Wait until no block IO is waiting (unless 'force') */
1008 wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting,
Lukas Czernereed8c022012-11-30 11:42:40 +01001009 conf->resync_lock);
NeilBrown0a27ec92006-01-06 00:20:13 -08001010
1011 /* block any new IO from starting */
1012 conf->barrier++;
1013
NeilBrownc3b328a2011-04-18 18:25:43 +10001014 /* Now wait for all pending IO to complete */
NeilBrown0a27ec92006-01-06 00:20:13 -08001015 wait_event_lock_irq(conf->wait_barrier,
1016 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
Lukas Czernereed8c022012-11-30 11:42:40 +01001017 conf->resync_lock);
NeilBrown0a27ec92006-01-06 00:20:13 -08001018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 spin_unlock_irq(&conf->resync_lock);
1020}
1021
NeilBrowne879a872011-10-11 16:49:02 +11001022static void lower_barrier(struct r10conf *conf)
NeilBrown0a27ec92006-01-06 00:20:13 -08001023{
1024 unsigned long flags;
1025 spin_lock_irqsave(&conf->resync_lock, flags);
1026 conf->barrier--;
1027 spin_unlock_irqrestore(&conf->resync_lock, flags);
1028 wake_up(&conf->wait_barrier);
1029}
1030
NeilBrowne879a872011-10-11 16:49:02 +11001031static void wait_barrier(struct r10conf *conf)
NeilBrown0a27ec92006-01-06 00:20:13 -08001032{
1033 spin_lock_irq(&conf->resync_lock);
1034 if (conf->barrier) {
1035 conf->nr_waiting++;
NeilBrownd6b42dc2012-03-19 12:46:38 +11001036 /* Wait for the barrier to drop.
1037 * However if there are already pending
1038 * requests (preventing the barrier from
1039 * rising completely), and the
1040 * pre-process bio queue isn't empty,
1041 * then don't wait, as we need to empty
1042 * that queue to get the nr_pending
1043 * count down.
1044 */
1045 wait_event_lock_irq(conf->wait_barrier,
1046 !conf->barrier ||
1047 (conf->nr_pending &&
1048 current->bio_list &&
1049 !bio_list_empty(current->bio_list)),
Lukas Czernereed8c022012-11-30 11:42:40 +01001050 conf->resync_lock);
NeilBrown0a27ec92006-01-06 00:20:13 -08001051 conf->nr_waiting--;
1052 }
1053 conf->nr_pending++;
1054 spin_unlock_irq(&conf->resync_lock);
1055}
1056
NeilBrowne879a872011-10-11 16:49:02 +11001057static void allow_barrier(struct r10conf *conf)
NeilBrown0a27ec92006-01-06 00:20:13 -08001058{
1059 unsigned long flags;
1060 spin_lock_irqsave(&conf->resync_lock, flags);
1061 conf->nr_pending--;
1062 spin_unlock_irqrestore(&conf->resync_lock, flags);
1063 wake_up(&conf->wait_barrier);
1064}
1065
NeilBrowne2d59922013-06-12 11:01:22 +10001066static void freeze_array(struct r10conf *conf, int extra)
NeilBrown4443ae12006-01-06 00:20:28 -08001067{
1068 /* stop syncio and normal IO and wait for everything to
NeilBrownf1885932006-01-06 00:20:42 -08001069 * go quiet.
NeilBrown4443ae12006-01-06 00:20:28 -08001070 * We increment barrier and nr_waiting, and then
NeilBrowne2d59922013-06-12 11:01:22 +10001071 * wait until nr_pending match nr_queued+extra
NeilBrown1c830532008-03-04 14:29:35 -08001072 * This is called in the context of one normal IO request
1073 * that has failed. Thus any sync request that might be pending
1074 * will be blocked by nr_pending, and we need to wait for
1075 * pending IO requests to complete or be queued for re-try.
NeilBrowne2d59922013-06-12 11:01:22 +10001076 * Thus the number queued (nr_queued) plus this request (extra)
NeilBrown1c830532008-03-04 14:29:35 -08001077 * must match the number of pending IOs (nr_pending) before
1078 * we continue.
NeilBrown4443ae12006-01-06 00:20:28 -08001079 */
1080 spin_lock_irq(&conf->resync_lock);
1081 conf->barrier++;
1082 conf->nr_waiting++;
Lukas Czernereed8c022012-11-30 11:42:40 +01001083 wait_event_lock_irq_cmd(conf->wait_barrier,
NeilBrowne2d59922013-06-12 11:01:22 +10001084 conf->nr_pending == conf->nr_queued+extra,
Lukas Czernereed8c022012-11-30 11:42:40 +01001085 conf->resync_lock,
1086 flush_pending_writes(conf));
NeilBrownc3b328a2011-04-18 18:25:43 +10001087
NeilBrown4443ae12006-01-06 00:20:28 -08001088 spin_unlock_irq(&conf->resync_lock);
1089}
1090
NeilBrowne879a872011-10-11 16:49:02 +11001091static void unfreeze_array(struct r10conf *conf)
NeilBrown4443ae12006-01-06 00:20:28 -08001092{
1093 /* reverse the effect of the freeze */
1094 spin_lock_irq(&conf->resync_lock);
1095 conf->barrier--;
1096 conf->nr_waiting--;
1097 wake_up(&conf->wait_barrier);
1098 spin_unlock_irq(&conf->resync_lock);
1099}
1100
NeilBrownf8c9e742012-05-21 09:28:33 +10001101static sector_t choose_data_offset(struct r10bio *r10_bio,
1102 struct md_rdev *rdev)
1103{
1104 if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) ||
1105 test_bit(R10BIO_Previous, &r10_bio->state))
1106 return rdev->data_offset;
1107 else
1108 return rdev->new_data_offset;
1109}
1110
NeilBrown57c67df2012-10-11 13:32:13 +11001111struct raid10_plug_cb {
1112 struct blk_plug_cb cb;
1113 struct bio_list pending;
1114 int pending_cnt;
1115};
1116
1117static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule)
1118{
1119 struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb,
1120 cb);
1121 struct mddev *mddev = plug->cb.data;
1122 struct r10conf *conf = mddev->private;
1123 struct bio *bio;
1124
NeilBrown874807a2012-11-27 12:14:40 +11001125 if (from_schedule || current->bio_list) {
NeilBrown57c67df2012-10-11 13:32:13 +11001126 spin_lock_irq(&conf->device_lock);
1127 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1128 conf->pending_count += plug->pending_cnt;
1129 spin_unlock_irq(&conf->device_lock);
NeilBrownee0b0242013-02-25 12:38:29 +11001130 wake_up(&conf->wait_barrier);
NeilBrown57c67df2012-10-11 13:32:13 +11001131 md_wakeup_thread(mddev->thread);
1132 kfree(plug);
1133 return;
1134 }
1135
1136 /* we aren't scheduling, so we can do the write-out directly. */
1137 bio = bio_list_get(&plug->pending);
1138 bitmap_unplug(mddev->bitmap);
1139 wake_up(&conf->wait_barrier);
1140
1141 while (bio) { /* submit pending writes */
1142 struct bio *next = bio->bi_next;
1143 bio->bi_next = NULL;
Shaohua Li32f9f572013-04-28 18:26:38 +08001144 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
1145 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1146 /* Just ignore it */
1147 bio_endio(bio, 0);
1148 else
1149 generic_make_request(bio);
NeilBrown57c67df2012-10-11 13:32:13 +11001150 bio = next;
1151 }
1152 kfree(plug);
1153}
1154
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -07001155static void make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
NeilBrowne879a872011-10-11 16:49:02 +11001157 struct r10conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11001158 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 struct bio *read_bio;
1160 int i;
NeilBrownf8c9e742012-05-21 09:28:33 +10001161 sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask);
NeilBrown5cf00fc2012-05-21 09:28:20 +10001162 int chunk_sects = chunk_mask + 1;
Jens Axboea3623572005-11-01 09:26:16 +01001163 const int rw = bio_data_dir(bio);
NeilBrown2c7d46e2010-08-18 16:16:05 +10001164 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
Tejun Heoe9c74692010-09-03 11:56:18 +02001165 const unsigned long do_fua = (bio->bi_rw & REQ_FUA);
Shaohua Li532a2a32012-10-11 13:30:52 +11001166 const unsigned long do_discard = (bio->bi_rw
1167 & (REQ_DISCARD | REQ_SECURE));
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001168 const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME);
NeilBrown6cce3b22006-01-06 00:20:16 -08001169 unsigned long flags;
NeilBrown3cb03002011-10-11 16:45:26 +11001170 struct md_rdev *blocked_rdev;
NeilBrown57c67df2012-10-11 13:32:13 +11001171 struct blk_plug_cb *cb;
1172 struct raid10_plug_cb *plug = NULL;
NeilBrownd4432c22011-07-28 11:39:24 +10001173 int sectors_handled;
1174 int max_sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10001175 int sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Tejun Heoe9c74692010-09-03 11:56:18 +02001177 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
1178 md_flush_request(mddev, bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001179 return;
NeilBrowne5dcdd82005-09-09 16:23:41 -07001180 }
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 /* If this request crosses a chunk boundary, we need to
1183 * split it. This will only happen for 1 PAGE (or less) requests.
1184 */
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001185 if (unlikely((bio->bi_sector & chunk_mask) + bio_sectors(bio)
NeilBrown5cf00fc2012-05-21 09:28:20 +10001186 > chunk_sects
NeilBrownf8c9e742012-05-21 09:28:33 +10001187 && (conf->geo.near_copies < conf->geo.raid_disks
1188 || conf->prev.near_copies < conf->prev.raid_disks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 struct bio_pair *bp;
1190 /* Sanity check -- queue functions should prevent this happening */
Kent Overstreet5b836362012-09-04 15:20:38 -07001191 if (bio_segments(bio) > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 goto bad_map;
1193 /* This is a one page bio that upper layers
1194 * refuse to split for us, so we need to split it.
1195 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001196 bp = bio_split(bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 chunk_sects - (bio->bi_sector & (chunk_sects - 1)) );
NeilBrown51e9ac72010-08-07 21:17:00 +10001198
1199 /* Each of these 'make_request' calls will call 'wait_barrier'.
1200 * If the first succeeds but the second blocks due to the resync
1201 * thread raising the barrier, we will deadlock because the
1202 * IO to the underlying device will be queued in generic_make_request
1203 * and will never complete, so will never reduce nr_pending.
1204 * So increment nr_waiting here so no new raise_barriers will
1205 * succeed, and so the second wait_barrier cannot block.
1206 */
1207 spin_lock_irq(&conf->resync_lock);
1208 conf->nr_waiting++;
1209 spin_unlock_irq(&conf->resync_lock);
1210
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001211 make_request(mddev, &bp->bio1);
1212 make_request(mddev, &bp->bio2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
NeilBrown51e9ac72010-08-07 21:17:00 +10001214 spin_lock_irq(&conf->resync_lock);
1215 conf->nr_waiting--;
1216 wake_up(&conf->wait_barrier);
1217 spin_unlock_irq(&conf->resync_lock);
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 bio_pair_release(bp);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001220 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 bad_map:
NeilBrown128595e2010-05-03 14:47:14 +10001222 printk("md/raid10:%s: make_request bug: can't convert block across chunks"
1223 " or bigger than %dk %llu %d\n", mdname(mddev), chunk_sects/2,
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001224 (unsigned long long)bio->bi_sector, bio_sectors(bio) / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
NeilBrown6712ecf2007-09-27 12:47:43 +02001226 bio_io_error(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001227 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 }
1229
NeilBrown3d310eb2005-06-21 17:17:26 -07001230 md_write_start(mddev, bio);
NeilBrown06d91a52005-06-21 17:17:12 -07001231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 /*
1233 * Register the new request and wait if the reconstruction
1234 * thread has put up a bar for new requests.
1235 * Continue immediately if no resync is active currently.
1236 */
NeilBrown0a27ec92006-01-06 00:20:13 -08001237 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001239 sectors = bio_sectors(bio);
NeilBrown3ea7daa2012-05-22 13:53:47 +10001240 while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
1241 bio->bi_sector < conf->reshape_progress &&
1242 bio->bi_sector + sectors > conf->reshape_progress) {
1243 /* IO spans the reshape position. Need to wait for
1244 * reshape to pass
1245 */
1246 allow_barrier(conf);
1247 wait_event(conf->wait_barrier,
1248 conf->reshape_progress <= bio->bi_sector ||
1249 conf->reshape_progress >= bio->bi_sector + sectors);
1250 wait_barrier(conf);
1251 }
1252 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
1253 bio_data_dir(bio) == WRITE &&
1254 (mddev->reshape_backwards
1255 ? (bio->bi_sector < conf->reshape_safe &&
1256 bio->bi_sector + sectors > conf->reshape_progress)
1257 : (bio->bi_sector + sectors > conf->reshape_safe &&
1258 bio->bi_sector < conf->reshape_progress))) {
1259 /* Need to update reshape_position in metadata */
1260 mddev->reshape_position = conf->reshape_progress;
1261 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1262 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1263 md_wakeup_thread(mddev->thread);
1264 wait_event(mddev->sb_wait,
1265 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1266
1267 conf->reshape_safe = mddev->reshape_position;
1268 }
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
1271
1272 r10_bio->master_bio = bio;
NeilBrown3ea7daa2012-05-22 13:53:47 +10001273 r10_bio->sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
1275 r10_bio->mddev = mddev;
1276 r10_bio->sector = bio->bi_sector;
NeilBrown6cce3b22006-01-06 00:20:16 -08001277 r10_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
NeilBrown856e08e2011-07-28 11:39:23 +10001279 /* We might need to issue multiple reads to different
1280 * devices if there are bad blocks around, so we keep
1281 * track of the number of reads in bio->bi_phys_segments.
1282 * If this is 0, there is only one r10_bio and no locking
1283 * will be needed when the request completes. If it is
1284 * non-zero, then it is the number of not-completed requests.
1285 */
1286 bio->bi_phys_segments = 0;
1287 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1288
Jens Axboea3623572005-11-01 09:26:16 +01001289 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 /*
1291 * read balancing logic:
1292 */
NeilBrown96c3fd12011-12-23 10:17:54 +11001293 struct md_rdev *rdev;
NeilBrown856e08e2011-07-28 11:39:23 +10001294 int slot;
1295
1296read_again:
NeilBrown96c3fd12011-12-23 10:17:54 +11001297 rdev = read_balance(conf, r10_bio, &max_sectors);
1298 if (!rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 raid_end_bio_io(r10_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001300 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 }
NeilBrown96c3fd12011-12-23 10:17:54 +11001302 slot = r10_bio->read_slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
NeilBrowna167f662010-10-26 18:31:13 +11001304 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrown856e08e2011-07-28 11:39:23 +10001305 md_trim_bio(read_bio, r10_bio->sector - bio->bi_sector,
1306 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 r10_bio->devs[slot].bio = read_bio;
NeilBrownabbf0982011-12-23 10:17:54 +11001309 r10_bio->devs[slot].rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311 read_bio->bi_sector = r10_bio->devs[slot].addr +
NeilBrownf8c9e742012-05-21 09:28:33 +10001312 choose_data_offset(r10_bio, rdev);
NeilBrown96c3fd12011-12-23 10:17:54 +11001313 read_bio->bi_bdev = rdev->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 read_bio->bi_end_io = raid10_end_read_request;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001315 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 read_bio->bi_private = r10_bio;
1317
NeilBrown856e08e2011-07-28 11:39:23 +10001318 if (max_sectors < r10_bio->sectors) {
1319 /* Could not read all from this device, so we will
1320 * need another r10_bio.
1321 */
NeilBrown856e08e2011-07-28 11:39:23 +10001322 sectors_handled = (r10_bio->sectors + max_sectors
1323 - bio->bi_sector);
1324 r10_bio->sectors = max_sectors;
1325 spin_lock_irq(&conf->device_lock);
1326 if (bio->bi_phys_segments == 0)
1327 bio->bi_phys_segments = 2;
1328 else
1329 bio->bi_phys_segments++;
1330 spin_unlock(&conf->device_lock);
1331 /* Cannot call generic_make_request directly
1332 * as that will be queued in __generic_make_request
1333 * and subsequent mempool_alloc might block
1334 * waiting for it. so hand bio over to raid10d.
1335 */
1336 reschedule_retry(r10_bio);
1337
1338 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
1339
1340 r10_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001341 r10_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrown856e08e2011-07-28 11:39:23 +10001342 r10_bio->state = 0;
1343 r10_bio->mddev = mddev;
1344 r10_bio->sector = bio->bi_sector + sectors_handled;
1345 goto read_again;
1346 } else
1347 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001348 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
1350
1351 /*
1352 * WRITE:
1353 */
NeilBrown34db0cd2011-10-11 16:50:01 +11001354 if (conf->pending_count >= max_queued_requests) {
1355 md_wakeup_thread(mddev->thread);
1356 wait_event(conf->wait_barrier,
1357 conf->pending_count < max_queued_requests);
1358 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07001359 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 * inc refcount on their rdev. Record them by setting
1361 * bios[x] to bio
NeilBrownd4432c22011-07-28 11:39:24 +10001362 * If there are known/acknowledged bad blocks on any device
1363 * on which we have seen a write error, we want to avoid
1364 * writing to those blocks. This potentially requires several
1365 * writes to write around the bad blocks. Each set of writes
1366 * gets its own r10_bio with a set of bios attached. The number
1367 * of r10_bios is recored in bio->bi_phys_segments just as with
1368 * the read case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001370
NeilBrown69335ef2011-12-23 10:17:54 +11001371 r10_bio->read_slot = -1; /* make sure repl_bio gets freed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 raid10_find_phys(conf, r10_bio);
NeilBrownd4432c22011-07-28 11:39:24 +10001373retry_write:
Harvey Harrisoncb6969e2008-05-06 20:42:32 -07001374 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 rcu_read_lock();
NeilBrownd4432c22011-07-28 11:39:24 +10001376 max_sectors = r10_bio->sectors;
1377
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 for (i = 0; i < conf->copies; i++) {
1379 int d = r10_bio->devs[i].devnum;
NeilBrown3cb03002011-10-11 16:45:26 +11001380 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown475b0322011-12-23 10:17:55 +11001381 struct md_rdev *rrdev = rcu_dereference(
1382 conf->mirrors[d].replacement);
NeilBrown4ca40c22011-12-23 10:17:55 +11001383 if (rdev == rrdev)
1384 rrdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001385 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1386 atomic_inc(&rdev->nr_pending);
1387 blocked_rdev = rdev;
1388 break;
1389 }
NeilBrown475b0322011-12-23 10:17:55 +11001390 if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) {
1391 atomic_inc(&rrdev->nr_pending);
1392 blocked_rdev = rrdev;
1393 break;
1394 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001395 if (rdev && (test_bit(Faulty, &rdev->flags)
1396 || test_bit(Unmerged, &rdev->flags)))
1397 rdev = NULL;
NeilBrown050b6612012-03-19 12:46:39 +11001398 if (rrdev && (test_bit(Faulty, &rrdev->flags)
1399 || test_bit(Unmerged, &rrdev->flags)))
NeilBrown475b0322011-12-23 10:17:55 +11001400 rrdev = NULL;
1401
NeilBrownd4432c22011-07-28 11:39:24 +10001402 r10_bio->devs[i].bio = NULL;
NeilBrown475b0322011-12-23 10:17:55 +11001403 r10_bio->devs[i].repl_bio = NULL;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001404
1405 if (!rdev && !rrdev) {
NeilBrown6cce3b22006-01-06 00:20:16 -08001406 set_bit(R10BIO_Degraded, &r10_bio->state);
NeilBrownd4432c22011-07-28 11:39:24 +10001407 continue;
NeilBrown6cce3b22006-01-06 00:20:16 -08001408 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001409 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) {
NeilBrownd4432c22011-07-28 11:39:24 +10001410 sector_t first_bad;
1411 sector_t dev_sector = r10_bio->devs[i].addr;
1412 int bad_sectors;
1413 int is_bad;
1414
1415 is_bad = is_badblock(rdev, dev_sector,
1416 max_sectors,
1417 &first_bad, &bad_sectors);
1418 if (is_bad < 0) {
1419 /* Mustn't write here until the bad block
1420 * is acknowledged
1421 */
1422 atomic_inc(&rdev->nr_pending);
1423 set_bit(BlockedBadBlocks, &rdev->flags);
1424 blocked_rdev = rdev;
1425 break;
1426 }
1427 if (is_bad && first_bad <= dev_sector) {
1428 /* Cannot write here at all */
1429 bad_sectors -= (dev_sector - first_bad);
1430 if (bad_sectors < max_sectors)
1431 /* Mustn't write more than bad_sectors
1432 * to other devices yet
1433 */
1434 max_sectors = bad_sectors;
1435 /* We don't set R10BIO_Degraded as that
1436 * only applies if the disk is missing,
1437 * so it might be re-added, and we want to
1438 * know to recover this chunk.
1439 * In this case the device is here, and the
1440 * fact that this chunk is not in-sync is
1441 * recorded in the bad block log.
1442 */
1443 continue;
1444 }
1445 if (is_bad) {
1446 int good_sectors = first_bad - dev_sector;
1447 if (good_sectors < max_sectors)
1448 max_sectors = good_sectors;
1449 }
1450 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001451 if (rdev) {
1452 r10_bio->devs[i].bio = bio;
1453 atomic_inc(&rdev->nr_pending);
1454 }
NeilBrown475b0322011-12-23 10:17:55 +11001455 if (rrdev) {
1456 r10_bio->devs[i].repl_bio = bio;
1457 atomic_inc(&rrdev->nr_pending);
1458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
1460 rcu_read_unlock();
1461
Dan Williams6bfe0b42008-04-30 00:52:32 -07001462 if (unlikely(blocked_rdev)) {
1463 /* Have to wait for this device to get unblocked, then retry */
1464 int j;
1465 int d;
1466
NeilBrown475b0322011-12-23 10:17:55 +11001467 for (j = 0; j < i; j++) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07001468 if (r10_bio->devs[j].bio) {
1469 d = r10_bio->devs[j].devnum;
1470 rdev_dec_pending(conf->mirrors[d].rdev, mddev);
1471 }
NeilBrown475b0322011-12-23 10:17:55 +11001472 if (r10_bio->devs[j].repl_bio) {
NeilBrown4ca40c22011-12-23 10:17:55 +11001473 struct md_rdev *rdev;
NeilBrown475b0322011-12-23 10:17:55 +11001474 d = r10_bio->devs[j].devnum;
NeilBrown4ca40c22011-12-23 10:17:55 +11001475 rdev = conf->mirrors[d].replacement;
1476 if (!rdev) {
1477 /* Race with remove_disk */
1478 smp_mb();
1479 rdev = conf->mirrors[d].rdev;
1480 }
1481 rdev_dec_pending(rdev, mddev);
NeilBrown475b0322011-12-23 10:17:55 +11001482 }
1483 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07001484 allow_barrier(conf);
1485 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1486 wait_barrier(conf);
1487 goto retry_write;
1488 }
1489
NeilBrownd4432c22011-07-28 11:39:24 +10001490 if (max_sectors < r10_bio->sectors) {
1491 /* We are splitting this into multiple parts, so
1492 * we need to prepare for allocating another r10_bio.
1493 */
1494 r10_bio->sectors = max_sectors;
1495 spin_lock_irq(&conf->device_lock);
1496 if (bio->bi_phys_segments == 0)
1497 bio->bi_phys_segments = 2;
1498 else
1499 bio->bi_phys_segments++;
1500 spin_unlock_irq(&conf->device_lock);
1501 }
1502 sectors_handled = r10_bio->sector + max_sectors - bio->bi_sector;
1503
NeilBrown4e780642010-10-19 12:54:01 +11001504 atomic_set(&r10_bio->remaining, 1);
NeilBrownd4432c22011-07-28 11:39:24 +10001505 bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0);
NeilBrown06d91a52005-06-21 17:17:12 -07001506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 for (i = 0; i < conf->copies; i++) {
1508 struct bio *mbio;
1509 int d = r10_bio->devs[i].devnum;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001510 if (r10_bio->devs[i].bio) {
1511 struct md_rdev *rdev = conf->mirrors[d].rdev;
1512 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1513 md_trim_bio(mbio, r10_bio->sector - bio->bi_sector,
1514 max_sectors);
1515 r10_bio->devs[i].bio = mbio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001517 mbio->bi_sector = (r10_bio->devs[i].addr+
1518 choose_data_offset(r10_bio,
1519 rdev));
1520 mbio->bi_bdev = rdev->bdev;
1521 mbio->bi_end_io = raid10_end_write_request;
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001522 mbio->bi_rw =
1523 WRITE | do_sync | do_fua | do_discard | do_same;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001524 mbio->bi_private = r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001526 atomic_inc(&r10_bio->remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001528 cb = blk_check_plugged(raid10_unplug, mddev,
1529 sizeof(*plug));
1530 if (cb)
1531 plug = container_of(cb, struct raid10_plug_cb,
1532 cb);
1533 else
1534 plug = NULL;
1535 spin_lock_irqsave(&conf->device_lock, flags);
1536 if (plug) {
1537 bio_list_add(&plug->pending, mbio);
1538 plug->pending_cnt++;
1539 } else {
1540 bio_list_add(&conf->pending_bio_list, mbio);
1541 conf->pending_count++;
1542 }
1543 spin_unlock_irqrestore(&conf->device_lock, flags);
1544 if (!plug)
1545 md_wakeup_thread(mddev->thread);
1546 }
NeilBrown57c67df2012-10-11 13:32:13 +11001547
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001548 if (r10_bio->devs[i].repl_bio) {
1549 struct md_rdev *rdev = conf->mirrors[d].replacement;
1550 if (rdev == NULL) {
1551 /* Replacement just got moved to main 'rdev' */
1552 smp_mb();
1553 rdev = conf->mirrors[d].rdev;
1554 }
1555 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1556 md_trim_bio(mbio, r10_bio->sector - bio->bi_sector,
1557 max_sectors);
1558 r10_bio->devs[i].repl_bio = mbio;
1559
1560 mbio->bi_sector = (r10_bio->devs[i].addr +
1561 choose_data_offset(
1562 r10_bio, rdev));
1563 mbio->bi_bdev = rdev->bdev;
1564 mbio->bi_end_io = raid10_end_write_request;
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001565 mbio->bi_rw =
1566 WRITE | do_sync | do_fua | do_discard | do_same;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001567 mbio->bi_private = r10_bio;
1568
1569 atomic_inc(&r10_bio->remaining);
1570 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown57c67df2012-10-11 13:32:13 +11001571 bio_list_add(&conf->pending_bio_list, mbio);
1572 conf->pending_count++;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001573 spin_unlock_irqrestore(&conf->device_lock, flags);
1574 if (!mddev_check_plugged(mddev))
1575 md_wakeup_thread(mddev->thread);
NeilBrown57c67df2012-10-11 13:32:13 +11001576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 }
1578
NeilBrown079fa162011-09-10 17:21:23 +10001579 /* Don't remove the bias on 'remaining' (one_write_done) until
1580 * after checking if we need to go around again.
1581 */
NeilBrowna35e63e2008-03-04 14:29:29 -08001582
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001583 if (sectors_handled < bio_sectors(bio)) {
NeilBrown079fa162011-09-10 17:21:23 +10001584 one_write_done(r10_bio);
NeilBrown5e570282011-07-28 11:39:25 +10001585 /* We need another r10_bio. It has already been counted
NeilBrownd4432c22011-07-28 11:39:24 +10001586 * in bio->bi_phys_segments.
1587 */
1588 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
1589
1590 r10_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001591 r10_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrownd4432c22011-07-28 11:39:24 +10001592
1593 r10_bio->mddev = mddev;
1594 r10_bio->sector = bio->bi_sector + sectors_handled;
1595 r10_bio->state = 0;
1596 goto retry_write;
1597 }
NeilBrown079fa162011-09-10 17:21:23 +10001598 one_write_done(r10_bio);
1599
1600 /* In case raid10d snuck in to freeze_array */
1601 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602}
1603
NeilBrownfd01b882011-10-11 16:47:53 +11001604static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605{
NeilBrowne879a872011-10-11 16:49:02 +11001606 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 int i;
1608
NeilBrown5cf00fc2012-05-21 09:28:20 +10001609 if (conf->geo.near_copies < conf->geo.raid_disks)
Andre Noll9d8f0362009-06-18 08:45:01 +10001610 seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2);
NeilBrown5cf00fc2012-05-21 09:28:20 +10001611 if (conf->geo.near_copies > 1)
1612 seq_printf(seq, " %d near-copies", conf->geo.near_copies);
1613 if (conf->geo.far_copies > 1) {
1614 if (conf->geo.far_offset)
1615 seq_printf(seq, " %d offset-copies", conf->geo.far_copies);
NeilBrownc93983b2006-06-26 00:27:41 -07001616 else
NeilBrown5cf00fc2012-05-21 09:28:20 +10001617 seq_printf(seq, " %d far-copies", conf->geo.far_copies);
NeilBrownc93983b2006-06-26 00:27:41 -07001618 }
NeilBrown5cf00fc2012-05-21 09:28:20 +10001619 seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks,
1620 conf->geo.raid_disks - mddev->degraded);
1621 for (i = 0; i < conf->geo.raid_disks; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 seq_printf(seq, "%s",
1623 conf->mirrors[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08001624 test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 seq_printf(seq, "]");
1626}
1627
NeilBrown700c7212011-07-27 11:00:36 +10001628/* check if there are enough drives for
1629 * every block to appear on atleast one.
1630 * Don't consider the device numbered 'ignore'
1631 * as we might be about to remove it.
1632 */
NeilBrown635f6412013-06-11 14:57:09 +10001633static int _enough(struct r10conf *conf, int previous, int ignore)
NeilBrown700c7212011-07-27 11:00:36 +10001634{
1635 int first = 0;
NeilBrown725d6e52013-06-11 15:08:03 +10001636 int has_enough = 0;
NeilBrown635f6412013-06-11 14:57:09 +10001637 int disks, ncopies;
1638 if (previous) {
1639 disks = conf->prev.raid_disks;
1640 ncopies = conf->prev.near_copies;
1641 } else {
1642 disks = conf->geo.raid_disks;
1643 ncopies = conf->geo.near_copies;
1644 }
NeilBrown700c7212011-07-27 11:00:36 +10001645
NeilBrown725d6e52013-06-11 15:08:03 +10001646 rcu_read_lock();
NeilBrown700c7212011-07-27 11:00:36 +10001647 do {
1648 int n = conf->copies;
1649 int cnt = 0;
NeilBrown80b48122012-09-27 12:35:21 +10001650 int this = first;
NeilBrown700c7212011-07-27 11:00:36 +10001651 while (n--) {
NeilBrown725d6e52013-06-11 15:08:03 +10001652 struct md_rdev *rdev;
1653 if (this != ignore &&
1654 (rdev = rcu_dereference(conf->mirrors[this].rdev)) &&
1655 test_bit(In_sync, &rdev->flags))
NeilBrown700c7212011-07-27 11:00:36 +10001656 cnt++;
NeilBrown635f6412013-06-11 14:57:09 +10001657 this = (this+1) % disks;
NeilBrown700c7212011-07-27 11:00:36 +10001658 }
1659 if (cnt == 0)
NeilBrown725d6e52013-06-11 15:08:03 +10001660 goto out;
NeilBrown635f6412013-06-11 14:57:09 +10001661 first = (first + ncopies) % disks;
NeilBrown700c7212011-07-27 11:00:36 +10001662 } while (first != 0);
NeilBrown725d6e52013-06-11 15:08:03 +10001663 has_enough = 1;
1664out:
1665 rcu_read_unlock();
1666 return has_enough;
NeilBrown700c7212011-07-27 11:00:36 +10001667}
1668
NeilBrownf8c9e742012-05-21 09:28:33 +10001669static int enough(struct r10conf *conf, int ignore)
1670{
NeilBrown635f6412013-06-11 14:57:09 +10001671 /* when calling 'enough', both 'prev' and 'geo' must
1672 * be stable.
1673 * This is ensured if ->reconfig_mutex or ->device_lock
1674 * is held.
1675 */
1676 return _enough(conf, 0, ignore) &&
1677 _enough(conf, 1, ignore);
NeilBrownf8c9e742012-05-21 09:28:33 +10001678}
1679
NeilBrownfd01b882011-10-11 16:47:53 +11001680static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
1682 char b[BDEVNAME_SIZE];
NeilBrowne879a872011-10-11 16:49:02 +11001683 struct r10conf *conf = mddev->private;
NeilBrown635f6412013-06-11 14:57:09 +10001684 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
1686 /*
1687 * If it is not operational, then we have already marked it as dead
1688 * else if it is the last working disks, ignore the error, let the
1689 * next level up know.
1690 * else mark the drive as failed
1691 */
NeilBrown635f6412013-06-11 14:57:09 +10001692 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownb2d444d2005-11-08 21:39:31 -08001693 if (test_bit(In_sync, &rdev->flags)
NeilBrown635f6412013-06-11 14:57:09 +10001694 && !enough(conf, rdev->raid_disk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 /*
1696 * Don't fail the drive, just return an IO error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 */
NeilBrownc04be0a2006-10-03 01:15:53 -07001698 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown635f6412013-06-11 14:57:09 +10001699 return;
1700 }
1701 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1702 mddev->degraded++;
1703 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 * if recovery is running, make sure it aborts.
1705 */
NeilBrowndfc70642008-05-23 13:04:39 -07001706 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 }
NeilBrownde393cd2011-07-28 11:31:48 +10001708 set_bit(Blocked, &rdev->flags);
NeilBrownb2d444d2005-11-08 21:39:31 -08001709 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001710 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown635f6412013-06-11 14:57:09 +10001711 spin_unlock_irqrestore(&conf->device_lock, flags);
Joe Perches067032b2011-01-14 09:14:33 +11001712 printk(KERN_ALERT
1713 "md/raid10:%s: Disk failure on %s, disabling device.\n"
1714 "md/raid10:%s: Operation continuing on %d devices.\n",
NeilBrown128595e2010-05-03 14:47:14 +10001715 mdname(mddev), bdevname(rdev->bdev, b),
NeilBrown5cf00fc2012-05-21 09:28:20 +10001716 mdname(mddev), conf->geo.raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717}
1718
NeilBrowne879a872011-10-11 16:49:02 +11001719static void print_conf(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720{
1721 int i;
Jonathan Brassowdc280d92012-07-31 10:03:52 +10001722 struct raid10_info *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
NeilBrown128595e2010-05-03 14:47:14 +10001724 printk(KERN_DEBUG "RAID10 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 if (!conf) {
NeilBrown128595e2010-05-03 14:47:14 +10001726 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 return;
1728 }
NeilBrown5cf00fc2012-05-21 09:28:20 +10001729 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded,
1730 conf->geo.raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
NeilBrown5cf00fc2012-05-21 09:28:20 +10001732 for (i = 0; i < conf->geo.raid_disks; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 char b[BDEVNAME_SIZE];
1734 tmp = conf->mirrors + i;
1735 if (tmp->rdev)
NeilBrown128595e2010-05-03 14:47:14 +10001736 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownb2d444d2005-11-08 21:39:31 -08001737 i, !test_bit(In_sync, &tmp->rdev->flags),
1738 !test_bit(Faulty, &tmp->rdev->flags),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 bdevname(tmp->rdev->bdev,b));
1740 }
1741}
1742
NeilBrowne879a872011-10-11 16:49:02 +11001743static void close_sync(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
NeilBrown0a27ec92006-01-06 00:20:13 -08001745 wait_barrier(conf);
1746 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
1748 mempool_destroy(conf->r10buf_pool);
1749 conf->r10buf_pool = NULL;
1750}
1751
NeilBrownfd01b882011-10-11 16:47:53 +11001752static int raid10_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
1754 int i;
NeilBrowne879a872011-10-11 16:49:02 +11001755 struct r10conf *conf = mddev->private;
Jonathan Brassowdc280d92012-07-31 10:03:52 +10001756 struct raid10_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10001757 int count = 0;
1758 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
1760 /*
1761 * Find all non-in_sync disks within the RAID10 configuration
1762 * and mark them in_sync
1763 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10001764 for (i = 0; i < conf->geo.raid_disks; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 tmp = conf->mirrors + i;
NeilBrown4ca40c22011-12-23 10:17:55 +11001766 if (tmp->replacement
1767 && tmp->replacement->recovery_offset == MaxSector
1768 && !test_bit(Faulty, &tmp->replacement->flags)
1769 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
1770 /* Replacement has just become active */
1771 if (!tmp->rdev
1772 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
1773 count++;
1774 if (tmp->rdev) {
1775 /* Replaced device not technically faulty,
1776 * but we need to be sure it gets removed
1777 * and never re-added.
1778 */
1779 set_bit(Faulty, &tmp->rdev->flags);
1780 sysfs_notify_dirent_safe(
1781 tmp->rdev->sysfs_state);
1782 }
1783 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
1784 } else if (tmp->rdev
1785 && !test_bit(Faulty, &tmp->rdev->flags)
1786 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001787 count++;
Jonathan Brassow2863b9e2012-10-11 13:38:58 +11001788 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 }
1790 }
NeilBrown6b965622010-08-18 11:56:59 +10001791 spin_lock_irqsave(&conf->device_lock, flags);
1792 mddev->degraded -= count;
1793 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
1795 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001796 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797}
1798
1799
NeilBrownfd01b882011-10-11 16:47:53 +11001800static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801{
NeilBrowne879a872011-10-11 16:49:02 +11001802 struct r10conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001803 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 int mirror;
Neil Brown6c2fce22008-06-28 08:31:31 +10001805 int first = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10001806 int last = conf->geo.raid_disks - 1;
NeilBrown050b6612012-03-19 12:46:39 +11001807 struct request_queue *q = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
1809 if (mddev->recovery_cp < MaxSector)
1810 /* only hot-add to in-sync arrays, as recovery is
1811 * very different from resync
1812 */
Neil Brown199050e2008-06-28 08:31:33 +10001813 return -EBUSY;
NeilBrown635f6412013-06-11 14:57:09 +10001814 if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1))
Neil Brown199050e2008-06-28 08:31:33 +10001815 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816
NeilBrowna53a6c82008-11-06 17:28:20 +11001817 if (rdev->raid_disk >= 0)
Neil Brown6c2fce22008-06-28 08:31:31 +10001818 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819
NeilBrown050b6612012-03-19 12:46:39 +11001820 if (q->merge_bvec_fn) {
1821 set_bit(Unmerged, &rdev->flags);
1822 mddev->merge_check_needed = 1;
1823 }
1824
Namhyung Kim2c4193d2011-07-18 17:38:43 +10001825 if (rdev->saved_raid_disk >= first &&
NeilBrown6cce3b22006-01-06 00:20:16 -08001826 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1827 mirror = rdev->saved_raid_disk;
1828 else
Neil Brown6c2fce22008-06-28 08:31:31 +10001829 mirror = first;
NeilBrown2bb77732011-07-27 11:00:36 +10001830 for ( ; mirror <= last ; mirror++) {
Jonathan Brassowdc280d92012-07-31 10:03:52 +10001831 struct raid10_info *p = &conf->mirrors[mirror];
NeilBrown2bb77732011-07-27 11:00:36 +10001832 if (p->recovery_disabled == mddev->recovery_disabled)
1833 continue;
NeilBrownb7044d42011-12-23 10:17:56 +11001834 if (p->rdev) {
1835 if (!test_bit(WantReplacement, &p->rdev->flags) ||
1836 p->replacement != NULL)
1837 continue;
1838 clear_bit(In_sync, &rdev->flags);
1839 set_bit(Replacement, &rdev->flags);
1840 rdev->raid_disk = mirror;
1841 err = 0;
Jonathan Brassow9092c022013-05-02 14:19:24 -05001842 if (mddev->gendisk)
1843 disk_stack_limits(mddev->gendisk, rdev->bdev,
1844 rdev->data_offset << 9);
NeilBrownb7044d42011-12-23 10:17:56 +11001845 conf->fullsync = 1;
1846 rcu_assign_pointer(p->replacement, rdev);
1847 break;
1848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
Jonathan Brassow9092c022013-05-02 14:19:24 -05001850 if (mddev->gendisk)
1851 disk_stack_limits(mddev->gendisk, rdev->bdev,
1852 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853
NeilBrown2bb77732011-07-27 11:00:36 +10001854 p->head_position = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11001855 p->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown2bb77732011-07-27 11:00:36 +10001856 rdev->raid_disk = mirror;
1857 err = 0;
1858 if (rdev->saved_raid_disk != mirror)
1859 conf->fullsync = 1;
1860 rcu_assign_pointer(p->rdev, rdev);
1861 break;
1862 }
NeilBrown050b6612012-03-19 12:46:39 +11001863 if (err == 0 && test_bit(Unmerged, &rdev->flags)) {
1864 /* Some requests might not have seen this new
1865 * merge_bvec_fn. We must wait for them to complete
1866 * before merging the device fully.
1867 * First we make sure any code which has tested
1868 * our function has submitted the request, then
1869 * we wait for all outstanding requests to complete.
1870 */
1871 synchronize_sched();
NeilBrowne2d59922013-06-12 11:01:22 +10001872 freeze_array(conf, 0);
1873 unfreeze_array(conf);
NeilBrown050b6612012-03-19 12:46:39 +11001874 clear_bit(Unmerged, &rdev->flags);
1875 }
Andre Nollac5e7112009-08-03 10:59:47 +10001876 md_integrity_add_rdev(rdev, mddev);
Jonathan Brassowed30be02012-10-31 11:42:30 +11001877 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Shaohua Li532a2a32012-10-11 13:30:52 +11001878 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1879
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001881 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882}
1883
NeilBrownb8321b62011-12-23 10:17:51 +11001884static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885{
NeilBrowne879a872011-10-11 16:49:02 +11001886 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001888 int number = rdev->raid_disk;
NeilBrownc8ab9032011-12-23 10:17:54 +11001889 struct md_rdev **rdevp;
Jonathan Brassowdc280d92012-07-31 10:03:52 +10001890 struct raid10_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891
1892 print_conf(conf);
NeilBrownc8ab9032011-12-23 10:17:54 +11001893 if (rdev == p->rdev)
1894 rdevp = &p->rdev;
1895 else if (rdev == p->replacement)
1896 rdevp = &p->replacement;
1897 else
1898 return 0;
1899
1900 if (test_bit(In_sync, &rdev->flags) ||
1901 atomic_read(&rdev->nr_pending)) {
1902 err = -EBUSY;
1903 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
NeilBrownc8ab9032011-12-23 10:17:54 +11001905 /* Only remove faulty devices if recovery
1906 * is not possible.
1907 */
1908 if (!test_bit(Faulty, &rdev->flags) &&
1909 mddev->recovery_disabled != p->recovery_disabled &&
NeilBrown4ca40c22011-12-23 10:17:55 +11001910 (!p->replacement || p->replacement == rdev) &&
NeilBrown63aced62012-05-22 13:55:33 +10001911 number < conf->geo.raid_disks &&
NeilBrownc8ab9032011-12-23 10:17:54 +11001912 enough(conf, -1)) {
1913 err = -EBUSY;
1914 goto abort;
1915 }
1916 *rdevp = NULL;
1917 synchronize_rcu();
1918 if (atomic_read(&rdev->nr_pending)) {
1919 /* lost the race, try later */
1920 err = -EBUSY;
1921 *rdevp = rdev;
1922 goto abort;
NeilBrown4ca40c22011-12-23 10:17:55 +11001923 } else if (p->replacement) {
1924 /* We must have just cleared 'rdev' */
1925 p->rdev = p->replacement;
1926 clear_bit(Replacement, &p->replacement->flags);
1927 smp_mb(); /* Make sure other CPUs may see both as identical
1928 * but will never see neither -- if they are careful.
1929 */
1930 p->replacement = NULL;
1931 clear_bit(WantReplacement, &rdev->flags);
1932 } else
1933 /* We might have just remove the Replacement as faulty
1934 * Clear the flag just in case
1935 */
1936 clear_bit(WantReplacement, &rdev->flags);
1937
NeilBrownc8ab9032011-12-23 10:17:54 +11001938 err = md_integrity_register(mddev);
1939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940abort:
1941
1942 print_conf(conf);
1943 return err;
1944}
1945
1946
NeilBrown6712ecf2007-09-27 12:47:43 +02001947static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001949 struct r10bio *r10_bio = bio->bi_private;
NeilBrowne879a872011-10-11 16:49:02 +11001950 struct r10conf *conf = r10_bio->mddev->private;
Namhyung Kim778ca012011-07-18 17:38:47 +10001951 int d;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
NeilBrown3ea7daa2012-05-22 13:53:47 +10001953 if (bio == r10_bio->master_bio) {
1954 /* this is a reshape read */
1955 d = r10_bio->read_slot; /* really the read dev */
1956 } else
1957 d = find_bio_disk(conf, r10_bio, bio, NULL, NULL);
NeilBrown0eb3ff12006-01-06 00:20:29 -08001958
1959 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1960 set_bit(R10BIO_Uptodate, &r10_bio->state);
NeilBrowne684e412011-07-28 11:39:25 +10001961 else
1962 /* The write handler will notice the lack of
1963 * R10BIO_Uptodate and record any errors etc
1964 */
NeilBrown4dbcdc72006-01-06 00:20:52 -08001965 atomic_add(r10_bio->sectors,
1966 &conf->mirrors[d].rdev->corrected_errors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
1968 /* for reconstruct, we always reschedule after a read.
1969 * for resync, only after all reads
1970 */
NeilBrown73d5c382009-02-25 13:18:47 +11001971 rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
1973 atomic_dec_and_test(&r10_bio->remaining)) {
1974 /* we have read all the blocks,
1975 * do the comparison in process context in raid10d
1976 */
1977 reschedule_retry(r10_bio);
1978 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979}
1980
NeilBrown9f2c9d12011-10-11 16:48:43 +11001981static void end_sync_request(struct r10bio *r10_bio)
NeilBrown5e570282011-07-28 11:39:25 +10001982{
NeilBrownfd01b882011-10-11 16:47:53 +11001983 struct mddev *mddev = r10_bio->mddev;
NeilBrown5e570282011-07-28 11:39:25 +10001984
1985 while (atomic_dec_and_test(&r10_bio->remaining)) {
1986 if (r10_bio->master_bio == NULL) {
1987 /* the primary of several recovery bios */
1988 sector_t s = r10_bio->sectors;
1989 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
1990 test_bit(R10BIO_WriteError, &r10_bio->state))
1991 reschedule_retry(r10_bio);
1992 else
1993 put_buf(r10_bio);
1994 md_done_sync(mddev, s, 1);
1995 break;
1996 } else {
NeilBrown9f2c9d12011-10-11 16:48:43 +11001997 struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio;
NeilBrown5e570282011-07-28 11:39:25 +10001998 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
1999 test_bit(R10BIO_WriteError, &r10_bio->state))
2000 reschedule_retry(r10_bio);
2001 else
2002 put_buf(r10_bio);
2003 r10_bio = r10_bio2;
2004 }
2005 }
2006}
2007
NeilBrown6712ecf2007-09-27 12:47:43 +02002008static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009{
2010 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +11002011 struct r10bio *r10_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11002012 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11002013 struct r10conf *conf = mddev->private;
Namhyung Kim778ca012011-07-18 17:38:47 +10002014 int d;
NeilBrown749c55e2011-07-28 11:39:24 +10002015 sector_t first_bad;
2016 int bad_sectors;
2017 int slot;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002018 int repl;
NeilBrown4ca40c22011-12-23 10:17:55 +11002019 struct md_rdev *rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
NeilBrown9ad1aef2011-12-23 10:17:55 +11002021 d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
2022 if (repl)
2023 rdev = conf->mirrors[d].replacement;
NeilBrown547414d2012-03-13 11:21:20 +11002024 else
NeilBrown9ad1aef2011-12-23 10:17:55 +11002025 rdev = conf->mirrors[d].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002027 if (!uptodate) {
NeilBrown9ad1aef2011-12-23 10:17:55 +11002028 if (repl)
2029 md_error(mddev, rdev);
2030 else {
2031 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002032 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2033 set_bit(MD_RECOVERY_NEEDED,
2034 &rdev->mddev->recovery);
NeilBrown9ad1aef2011-12-23 10:17:55 +11002035 set_bit(R10BIO_WriteError, &r10_bio->state);
2036 }
2037 } else if (is_badblock(rdev,
NeilBrown749c55e2011-07-28 11:39:24 +10002038 r10_bio->devs[slot].addr,
2039 r10_bio->sectors,
2040 &first_bad, &bad_sectors))
2041 set_bit(R10BIO_MadeGood, &r10_bio->state);
NeilBrowndfc70642008-05-23 13:04:39 -07002042
NeilBrown9ad1aef2011-12-23 10:17:55 +11002043 rdev_dec_pending(rdev, mddev);
NeilBrown5e570282011-07-28 11:39:25 +10002044
2045 end_sync_request(r10_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046}
2047
2048/*
2049 * Note: sync and recover and handled very differently for raid10
2050 * This code is for resync.
2051 * For resync, we read through virtual addresses and read all blocks.
2052 * If there is any error, we schedule a write. The lowest numbered
2053 * drive is authoritative.
2054 * However requests come for physical address, so we need to map.
2055 * For every physical address there are raid_disks/copies virtual addresses,
2056 * which is always are least one, but is not necessarly an integer.
2057 * This means that a physical address can span multiple chunks, so we may
2058 * have to submit multiple io requests for a single sync request.
2059 */
2060/*
2061 * We check if all blocks are in-sync and only write to blocks that
2062 * aren't in sync
2063 */
NeilBrown9f2c9d12011-10-11 16:48:43 +11002064static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065{
NeilBrowne879a872011-10-11 16:49:02 +11002066 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 int i, first;
2068 struct bio *tbio, *fbio;
majianpengf4380a92012-04-12 16:04:47 +10002069 int vcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071 atomic_set(&r10_bio->remaining, 1);
2072
2073 /* find the first device with a block */
2074 for (i=0; i<conf->copies; i++)
2075 if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags))
2076 break;
2077
2078 if (i == conf->copies)
2079 goto done;
2080
2081 first = i;
2082 fbio = r10_bio->devs[i].bio;
2083
majianpengf4380a92012-04-12 16:04:47 +10002084 vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 /* now find blocks with errors */
NeilBrown0eb3ff12006-01-06 00:20:29 -08002086 for (i=0 ; i < conf->copies ; i++) {
2087 int j, d;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 tbio = r10_bio->devs[i].bio;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002090
2091 if (tbio->bi_end_io != end_sync_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 continue;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002093 if (i == first)
2094 continue;
2095 if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) {
2096 /* We know that the bi_io_vec layout is the same for
2097 * both 'first' and 'i', so we just compare them.
2098 * All vec entries are PAGE_SIZE;
2099 */
NeilBrown7bb23c42013-07-16 16:50:47 +10002100 int sectors = r10_bio->sectors;
2101 for (j = 0; j < vcnt; j++) {
2102 int len = PAGE_SIZE;
2103 if (sectors < (len / 512))
2104 len = sectors * 512;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002105 if (memcmp(page_address(fbio->bi_io_vec[j].bv_page),
2106 page_address(tbio->bi_io_vec[j].bv_page),
NeilBrown7bb23c42013-07-16 16:50:47 +10002107 len))
NeilBrown0eb3ff12006-01-06 00:20:29 -08002108 break;
NeilBrown7bb23c42013-07-16 16:50:47 +10002109 sectors -= len/512;
2110 }
NeilBrown0eb3ff12006-01-06 00:20:29 -08002111 if (j == vcnt)
2112 continue;
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11002113 atomic64_add(r10_bio->sectors, &mddev->resync_mismatches);
NeilBrownf84ee362011-07-28 11:39:25 +10002114 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2115 /* Don't fix anything. */
2116 continue;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002117 }
NeilBrownf84ee362011-07-28 11:39:25 +10002118 /* Ok, we need to write this bio, either to correct an
2119 * inconsistency or to correct an unreadable block.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 * First we need to fixup bv_offset, bv_len and
2121 * bi_vecs, as the read request might have corrupted these
2122 */
Kent Overstreet8be185f2012-09-06 14:14:43 -07002123 bio_reset(tbio);
2124
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 tbio->bi_vcnt = vcnt;
2126 tbio->bi_size = r10_bio->sectors << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 tbio->bi_rw = WRITE;
2128 tbio->bi_private = r10_bio;
2129 tbio->bi_sector = r10_bio->devs[i].addr;
2130
2131 for (j=0; j < vcnt ; j++) {
2132 tbio->bi_io_vec[j].bv_offset = 0;
2133 tbio->bi_io_vec[j].bv_len = PAGE_SIZE;
2134
2135 memcpy(page_address(tbio->bi_io_vec[j].bv_page),
2136 page_address(fbio->bi_io_vec[j].bv_page),
2137 PAGE_SIZE);
2138 }
2139 tbio->bi_end_io = end_sync_write;
2140
2141 d = r10_bio->devs[i].devnum;
2142 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
2143 atomic_inc(&r10_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002144 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
2146 tbio->bi_sector += conf->mirrors[d].rdev->data_offset;
2147 tbio->bi_bdev = conf->mirrors[d].rdev->bdev;
2148 generic_make_request(tbio);
2149 }
2150
NeilBrown9ad1aef2011-12-23 10:17:55 +11002151 /* Now write out to any replacement devices
2152 * that are active
2153 */
2154 for (i = 0; i < conf->copies; i++) {
2155 int j, d;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002156
2157 tbio = r10_bio->devs[i].repl_bio;
2158 if (!tbio || !tbio->bi_end_io)
2159 continue;
2160 if (r10_bio->devs[i].bio->bi_end_io != end_sync_write
2161 && r10_bio->devs[i].bio != fbio)
2162 for (j = 0; j < vcnt; j++)
2163 memcpy(page_address(tbio->bi_io_vec[j].bv_page),
2164 page_address(fbio->bi_io_vec[j].bv_page),
2165 PAGE_SIZE);
2166 d = r10_bio->devs[i].devnum;
2167 atomic_inc(&r10_bio->remaining);
2168 md_sync_acct(conf->mirrors[d].replacement->bdev,
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002169 bio_sectors(tbio));
NeilBrown9ad1aef2011-12-23 10:17:55 +11002170 generic_make_request(tbio);
2171 }
2172
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173done:
2174 if (atomic_dec_and_test(&r10_bio->remaining)) {
2175 md_done_sync(mddev, r10_bio->sectors, 1);
2176 put_buf(r10_bio);
2177 }
2178}
2179
2180/*
2181 * Now for the recovery code.
2182 * Recovery happens across physical sectors.
2183 * We recover all non-is_sync drives by finding the virtual address of
2184 * each, and then choose a working drive that also has that virt address.
2185 * There is a separate r10_bio for each non-in_sync drive.
2186 * Only the first two slots are in use. The first for reading,
2187 * The second for writing.
2188 *
2189 */
NeilBrown9f2c9d12011-10-11 16:48:43 +11002190static void fix_recovery_read_error(struct r10bio *r10_bio)
NeilBrown5e570282011-07-28 11:39:25 +10002191{
2192 /* We got a read error during recovery.
2193 * We repeat the read in smaller page-sized sections.
2194 * If a read succeeds, write it to the new device or record
2195 * a bad block if we cannot.
2196 * If a read fails, record a bad block on both old and
2197 * new devices.
2198 */
NeilBrownfd01b882011-10-11 16:47:53 +11002199 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11002200 struct r10conf *conf = mddev->private;
NeilBrown5e570282011-07-28 11:39:25 +10002201 struct bio *bio = r10_bio->devs[0].bio;
2202 sector_t sect = 0;
2203 int sectors = r10_bio->sectors;
2204 int idx = 0;
2205 int dr = r10_bio->devs[0].devnum;
2206 int dw = r10_bio->devs[1].devnum;
2207
2208 while (sectors) {
2209 int s = sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11002210 struct md_rdev *rdev;
NeilBrown5e570282011-07-28 11:39:25 +10002211 sector_t addr;
2212 int ok;
2213
2214 if (s > (PAGE_SIZE>>9))
2215 s = PAGE_SIZE >> 9;
2216
2217 rdev = conf->mirrors[dr].rdev;
2218 addr = r10_bio->devs[0].addr + sect,
2219 ok = sync_page_io(rdev,
2220 addr,
2221 s << 9,
2222 bio->bi_io_vec[idx].bv_page,
2223 READ, false);
2224 if (ok) {
2225 rdev = conf->mirrors[dw].rdev;
2226 addr = r10_bio->devs[1].addr + sect;
2227 ok = sync_page_io(rdev,
2228 addr,
2229 s << 9,
2230 bio->bi_io_vec[idx].bv_page,
2231 WRITE, false);
NeilBrownb7044d42011-12-23 10:17:56 +11002232 if (!ok) {
NeilBrown5e570282011-07-28 11:39:25 +10002233 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002234 if (!test_and_set_bit(WantReplacement,
2235 &rdev->flags))
2236 set_bit(MD_RECOVERY_NEEDED,
2237 &rdev->mddev->recovery);
2238 }
NeilBrown5e570282011-07-28 11:39:25 +10002239 }
2240 if (!ok) {
2241 /* We don't worry if we cannot set a bad block -
2242 * it really is bad so there is no loss in not
2243 * recording it yet
2244 */
2245 rdev_set_badblocks(rdev, addr, s, 0);
2246
2247 if (rdev != conf->mirrors[dw].rdev) {
2248 /* need bad block on destination too */
NeilBrown3cb03002011-10-11 16:45:26 +11002249 struct md_rdev *rdev2 = conf->mirrors[dw].rdev;
NeilBrown5e570282011-07-28 11:39:25 +10002250 addr = r10_bio->devs[1].addr + sect;
2251 ok = rdev_set_badblocks(rdev2, addr, s, 0);
2252 if (!ok) {
2253 /* just abort the recovery */
2254 printk(KERN_NOTICE
2255 "md/raid10:%s: recovery aborted"
2256 " due to read error\n",
2257 mdname(mddev));
2258
2259 conf->mirrors[dw].recovery_disabled
2260 = mddev->recovery_disabled;
2261 set_bit(MD_RECOVERY_INTR,
2262 &mddev->recovery);
2263 break;
2264 }
2265 }
2266 }
2267
2268 sectors -= s;
2269 sect += s;
2270 idx++;
2271 }
2272}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
NeilBrown9f2c9d12011-10-11 16:48:43 +11002274static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275{
NeilBrowne879a872011-10-11 16:49:02 +11002276 struct r10conf *conf = mddev->private;
Namhyung Kimc65060a2011-07-18 17:38:49 +10002277 int d;
NeilBrown24afd802011-12-23 10:17:55 +11002278 struct bio *wbio, *wbio2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
NeilBrown5e570282011-07-28 11:39:25 +10002280 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) {
2281 fix_recovery_read_error(r10_bio);
2282 end_sync_request(r10_bio);
2283 return;
2284 }
2285
Namhyung Kimc65060a2011-07-18 17:38:49 +10002286 /*
2287 * share the pages with the first bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 * and submit the write request
2289 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 d = r10_bio->devs[1].devnum;
NeilBrown24afd802011-12-23 10:17:55 +11002291 wbio = r10_bio->devs[1].bio;
2292 wbio2 = r10_bio->devs[1].repl_bio;
2293 if (wbio->bi_end_io) {
2294 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002295 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio));
NeilBrown24afd802011-12-23 10:17:55 +11002296 generic_make_request(wbio);
2297 }
2298 if (wbio2 && wbio2->bi_end_io) {
2299 atomic_inc(&conf->mirrors[d].replacement->nr_pending);
2300 md_sync_acct(conf->mirrors[d].replacement->bdev,
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002301 bio_sectors(wbio2));
NeilBrown24afd802011-12-23 10:17:55 +11002302 generic_make_request(wbio2);
2303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304}
2305
2306
2307/*
Robert Becker1e509152009-12-14 12:49:58 +11002308 * Used by fix_read_error() to decay the per rdev read_errors.
2309 * We halve the read error count for every hour that has elapsed
2310 * since the last recorded read error.
2311 *
2312 */
NeilBrownfd01b882011-10-11 16:47:53 +11002313static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev)
Robert Becker1e509152009-12-14 12:49:58 +11002314{
2315 struct timespec cur_time_mon;
2316 unsigned long hours_since_last;
2317 unsigned int read_errors = atomic_read(&rdev->read_errors);
2318
2319 ktime_get_ts(&cur_time_mon);
2320
2321 if (rdev->last_read_error.tv_sec == 0 &&
2322 rdev->last_read_error.tv_nsec == 0) {
2323 /* first time we've seen a read error */
2324 rdev->last_read_error = cur_time_mon;
2325 return;
2326 }
2327
2328 hours_since_last = (cur_time_mon.tv_sec -
2329 rdev->last_read_error.tv_sec) / 3600;
2330
2331 rdev->last_read_error = cur_time_mon;
2332
2333 /*
2334 * if hours_since_last is > the number of bits in read_errors
2335 * just set read errors to 0. We do this to avoid
2336 * overflowing the shift of read_errors by hours_since_last.
2337 */
2338 if (hours_since_last >= 8 * sizeof(read_errors))
2339 atomic_set(&rdev->read_errors, 0);
2340 else
2341 atomic_set(&rdev->read_errors, read_errors >> hours_since_last);
2342}
2343
NeilBrown3cb03002011-10-11 16:45:26 +11002344static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrown58c54fc2011-07-28 11:39:25 +10002345 int sectors, struct page *page, int rw)
2346{
2347 sector_t first_bad;
2348 int bad_sectors;
2349
2350 if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors)
2351 && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags)))
2352 return -1;
2353 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
2354 /* success */
2355 return 1;
NeilBrownb7044d42011-12-23 10:17:56 +11002356 if (rw == WRITE) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002357 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002358 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2359 set_bit(MD_RECOVERY_NEEDED,
2360 &rdev->mddev->recovery);
2361 }
NeilBrown58c54fc2011-07-28 11:39:25 +10002362 /* need to record an error - either for the block or the device */
2363 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
2364 md_error(rdev->mddev, rdev);
2365 return 0;
2366}
2367
Robert Becker1e509152009-12-14 12:49:58 +11002368/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 * This is a kernel thread which:
2370 *
2371 * 1. Retries failed read operations on working mirrors.
2372 * 2. Updates the raid superblock when problems encounter.
NeilBrown6814d532006-10-03 01:15:45 -07002373 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 */
2375
NeilBrowne879a872011-10-11 16:49:02 +11002376static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio)
NeilBrown6814d532006-10-03 01:15:45 -07002377{
2378 int sect = 0; /* Offset from r10_bio->sector */
2379 int sectors = r10_bio->sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11002380 struct md_rdev*rdev;
Robert Becker1e509152009-12-14 12:49:58 +11002381 int max_read_errors = atomic_read(&mddev->max_corr_read_errors);
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002382 int d = r10_bio->devs[r10_bio->read_slot].devnum;
Robert Becker1e509152009-12-14 12:49:58 +11002383
NeilBrown7c4e06f2011-05-11 14:53:17 +10002384 /* still own a reference to this rdev, so it cannot
2385 * have been cleared recently.
2386 */
2387 rdev = conf->mirrors[d].rdev;
Robert Becker1e509152009-12-14 12:49:58 +11002388
NeilBrown7c4e06f2011-05-11 14:53:17 +10002389 if (test_bit(Faulty, &rdev->flags))
2390 /* drive has already been failed, just ignore any
2391 more fix_read_error() attempts */
2392 return;
2393
2394 check_decay_read_errors(mddev, rdev);
2395 atomic_inc(&rdev->read_errors);
2396 if (atomic_read(&rdev->read_errors) > max_read_errors) {
2397 char b[BDEVNAME_SIZE];
Robert Becker1e509152009-12-14 12:49:58 +11002398 bdevname(rdev->bdev, b);
2399
NeilBrown7c4e06f2011-05-11 14:53:17 +10002400 printk(KERN_NOTICE
2401 "md/raid10:%s: %s: Raid device exceeded "
2402 "read_error threshold [cur %d:max %d]\n",
2403 mdname(mddev), b,
2404 atomic_read(&rdev->read_errors), max_read_errors);
2405 printk(KERN_NOTICE
2406 "md/raid10:%s: %s: Failing raid device\n",
2407 mdname(mddev), b);
2408 md_error(mddev, conf->mirrors[d].rdev);
NeilBrownfae8cc52012-02-14 11:10:10 +11002409 r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED;
NeilBrown7c4e06f2011-05-11 14:53:17 +10002410 return;
Robert Becker1e509152009-12-14 12:49:58 +11002411 }
Robert Becker1e509152009-12-14 12:49:58 +11002412
NeilBrown6814d532006-10-03 01:15:45 -07002413 while(sectors) {
2414 int s = sectors;
2415 int sl = r10_bio->read_slot;
2416 int success = 0;
2417 int start;
2418
2419 if (s > (PAGE_SIZE>>9))
2420 s = PAGE_SIZE >> 9;
2421
2422 rcu_read_lock();
2423 do {
NeilBrown8dbed5c2011-07-28 11:39:24 +10002424 sector_t first_bad;
2425 int bad_sectors;
2426
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002427 d = r10_bio->devs[sl].devnum;
NeilBrown6814d532006-10-03 01:15:45 -07002428 rdev = rcu_dereference(conf->mirrors[d].rdev);
2429 if (rdev &&
NeilBrown050b6612012-03-19 12:46:39 +11002430 !test_bit(Unmerged, &rdev->flags) &&
NeilBrown8dbed5c2011-07-28 11:39:24 +10002431 test_bit(In_sync, &rdev->flags) &&
2432 is_badblock(rdev, r10_bio->devs[sl].addr + sect, s,
2433 &first_bad, &bad_sectors) == 0) {
NeilBrown6814d532006-10-03 01:15:45 -07002434 atomic_inc(&rdev->nr_pending);
2435 rcu_read_unlock();
NeilBrown2b193362010-10-27 15:16:40 +11002436 success = sync_page_io(rdev,
NeilBrown6814d532006-10-03 01:15:45 -07002437 r10_bio->devs[sl].addr +
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11002438 sect,
NeilBrown6814d532006-10-03 01:15:45 -07002439 s<<9,
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11002440 conf->tmppage, READ, false);
NeilBrown6814d532006-10-03 01:15:45 -07002441 rdev_dec_pending(rdev, mddev);
2442 rcu_read_lock();
2443 if (success)
2444 break;
2445 }
2446 sl++;
2447 if (sl == conf->copies)
2448 sl = 0;
2449 } while (!success && sl != r10_bio->read_slot);
2450 rcu_read_unlock();
2451
2452 if (!success) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002453 /* Cannot read from anywhere, just mark the block
2454 * as bad on the first device to discourage future
2455 * reads.
2456 */
NeilBrown6814d532006-10-03 01:15:45 -07002457 int dn = r10_bio->devs[r10_bio->read_slot].devnum;
NeilBrown58c54fc2011-07-28 11:39:25 +10002458 rdev = conf->mirrors[dn].rdev;
2459
2460 if (!rdev_set_badblocks(
2461 rdev,
2462 r10_bio->devs[r10_bio->read_slot].addr
2463 + sect,
NeilBrownfae8cc52012-02-14 11:10:10 +11002464 s, 0)) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002465 md_error(mddev, rdev);
NeilBrownfae8cc52012-02-14 11:10:10 +11002466 r10_bio->devs[r10_bio->read_slot].bio
2467 = IO_BLOCKED;
2468 }
NeilBrown6814d532006-10-03 01:15:45 -07002469 break;
2470 }
2471
2472 start = sl;
2473 /* write it back and re-read */
2474 rcu_read_lock();
2475 while (sl != r10_bio->read_slot) {
Robert Becker67b8dc42009-12-14 12:49:57 +11002476 char b[BDEVNAME_SIZE];
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002477
NeilBrown6814d532006-10-03 01:15:45 -07002478 if (sl==0)
2479 sl = conf->copies;
2480 sl--;
2481 d = r10_bio->devs[sl].devnum;
2482 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown1294b9c2011-07-28 11:39:23 +10002483 if (!rdev ||
NeilBrown050b6612012-03-19 12:46:39 +11002484 test_bit(Unmerged, &rdev->flags) ||
NeilBrown1294b9c2011-07-28 11:39:23 +10002485 !test_bit(In_sync, &rdev->flags))
2486 continue;
2487
2488 atomic_inc(&rdev->nr_pending);
2489 rcu_read_unlock();
NeilBrown58c54fc2011-07-28 11:39:25 +10002490 if (r10_sync_page_io(rdev,
2491 r10_bio->devs[sl].addr +
2492 sect,
NeilBrown055d3742012-07-03 15:55:33 +10002493 s, conf->tmppage, WRITE)
NeilBrown1294b9c2011-07-28 11:39:23 +10002494 == 0) {
2495 /* Well, this device is dead */
2496 printk(KERN_NOTICE
2497 "md/raid10:%s: read correction "
2498 "write failed"
2499 " (%d sectors at %llu on %s)\n",
2500 mdname(mddev), s,
2501 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002502 sect +
2503 choose_data_offset(r10_bio,
2504 rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002505 bdevname(rdev->bdev, b));
2506 printk(KERN_NOTICE "md/raid10:%s: %s: failing "
2507 "drive\n",
2508 mdname(mddev),
2509 bdevname(rdev->bdev, b));
NeilBrown6814d532006-10-03 01:15:45 -07002510 }
NeilBrown1294b9c2011-07-28 11:39:23 +10002511 rdev_dec_pending(rdev, mddev);
2512 rcu_read_lock();
NeilBrown6814d532006-10-03 01:15:45 -07002513 }
2514 sl = start;
2515 while (sl != r10_bio->read_slot) {
NeilBrown1294b9c2011-07-28 11:39:23 +10002516 char b[BDEVNAME_SIZE];
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002517
NeilBrown6814d532006-10-03 01:15:45 -07002518 if (sl==0)
2519 sl = conf->copies;
2520 sl--;
2521 d = r10_bio->devs[sl].devnum;
2522 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown1294b9c2011-07-28 11:39:23 +10002523 if (!rdev ||
2524 !test_bit(In_sync, &rdev->flags))
2525 continue;
Robert Becker67b8dc42009-12-14 12:49:57 +11002526
NeilBrown1294b9c2011-07-28 11:39:23 +10002527 atomic_inc(&rdev->nr_pending);
2528 rcu_read_unlock();
NeilBrown58c54fc2011-07-28 11:39:25 +10002529 switch (r10_sync_page_io(rdev,
2530 r10_bio->devs[sl].addr +
2531 sect,
NeilBrown055d3742012-07-03 15:55:33 +10002532 s, conf->tmppage,
NeilBrown58c54fc2011-07-28 11:39:25 +10002533 READ)) {
2534 case 0:
NeilBrown1294b9c2011-07-28 11:39:23 +10002535 /* Well, this device is dead */
2536 printk(KERN_NOTICE
2537 "md/raid10:%s: unable to read back "
2538 "corrected sectors"
2539 " (%d sectors at %llu on %s)\n",
2540 mdname(mddev), s,
2541 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002542 sect +
2543 choose_data_offset(r10_bio, rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002544 bdevname(rdev->bdev, b));
2545 printk(KERN_NOTICE "md/raid10:%s: %s: failing "
2546 "drive\n",
2547 mdname(mddev),
2548 bdevname(rdev->bdev, b));
NeilBrown58c54fc2011-07-28 11:39:25 +10002549 break;
2550 case 1:
NeilBrown1294b9c2011-07-28 11:39:23 +10002551 printk(KERN_INFO
2552 "md/raid10:%s: read error corrected"
2553 " (%d sectors at %llu on %s)\n",
2554 mdname(mddev), s,
2555 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002556 sect +
2557 choose_data_offset(r10_bio, rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002558 bdevname(rdev->bdev, b));
2559 atomic_add(s, &rdev->corrected_errors);
NeilBrown6814d532006-10-03 01:15:45 -07002560 }
NeilBrown1294b9c2011-07-28 11:39:23 +10002561
2562 rdev_dec_pending(rdev, mddev);
2563 rcu_read_lock();
NeilBrown6814d532006-10-03 01:15:45 -07002564 }
2565 rcu_read_unlock();
2566
2567 sectors -= s;
2568 sect += s;
2569 }
2570}
2571
NeilBrown9f2c9d12011-10-11 16:48:43 +11002572static int narrow_write_error(struct r10bio *r10_bio, int i)
NeilBrownbd870a12011-07-28 11:39:24 +10002573{
2574 struct bio *bio = r10_bio->master_bio;
NeilBrownfd01b882011-10-11 16:47:53 +11002575 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11002576 struct r10conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002577 struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev;
NeilBrownbd870a12011-07-28 11:39:24 +10002578 /* bio has the data to be written to slot 'i' where
2579 * we just recently had a write error.
2580 * We repeatedly clone the bio and trim down to one block,
2581 * then try the write. Where the write fails we record
2582 * a bad block.
2583 * It is conceivable that the bio doesn't exactly align with
2584 * blocks. We must handle this.
2585 *
2586 * We currently own a reference to the rdev.
2587 */
2588
2589 int block_sectors;
2590 sector_t sector;
2591 int sectors;
2592 int sect_to_write = r10_bio->sectors;
2593 int ok = 1;
2594
2595 if (rdev->badblocks.shift < 0)
2596 return 0;
2597
2598 block_sectors = 1 << rdev->badblocks.shift;
2599 sector = r10_bio->sector;
2600 sectors = ((r10_bio->sector + block_sectors)
2601 & ~(sector_t)(block_sectors - 1))
2602 - sector;
2603
2604 while (sect_to_write) {
2605 struct bio *wbio;
2606 if (sectors > sect_to_write)
2607 sectors = sect_to_write;
2608 /* Write at 'sector' for 'sectors' */
2609 wbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
2610 md_trim_bio(wbio, sector - bio->bi_sector, sectors);
2611 wbio->bi_sector = (r10_bio->devs[i].addr+
NeilBrownf8c9e742012-05-21 09:28:33 +10002612 choose_data_offset(r10_bio, rdev) +
NeilBrownbd870a12011-07-28 11:39:24 +10002613 (sector - r10_bio->sector));
2614 wbio->bi_bdev = rdev->bdev;
2615 if (submit_bio_wait(WRITE, wbio) == 0)
2616 /* Failure! */
2617 ok = rdev_set_badblocks(rdev, sector,
2618 sectors, 0)
2619 && ok;
2620
2621 bio_put(wbio);
2622 sect_to_write -= sectors;
2623 sector += sectors;
2624 sectors = block_sectors;
2625 }
2626 return ok;
2627}
2628
NeilBrown9f2c9d12011-10-11 16:48:43 +11002629static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio)
NeilBrown560f8e52011-07-28 11:39:23 +10002630{
2631 int slot = r10_bio->read_slot;
NeilBrown560f8e52011-07-28 11:39:23 +10002632 struct bio *bio;
NeilBrowne879a872011-10-11 16:49:02 +11002633 struct r10conf *conf = mddev->private;
NeilBrownabbf0982011-12-23 10:17:54 +11002634 struct md_rdev *rdev = r10_bio->devs[slot].rdev;
NeilBrown560f8e52011-07-28 11:39:23 +10002635 char b[BDEVNAME_SIZE];
2636 unsigned long do_sync;
NeilBrown856e08e2011-07-28 11:39:23 +10002637 int max_sectors;
NeilBrown560f8e52011-07-28 11:39:23 +10002638
2639 /* we got a read error. Maybe the drive is bad. Maybe just
2640 * the block and we can fix it.
2641 * We freeze all other IO, and try reading the block from
2642 * other devices. When we find one, we re-write
2643 * and check it that fixes the read error.
2644 * This is all done synchronously while the array is
2645 * frozen.
2646 */
NeilBrownfae8cc52012-02-14 11:10:10 +11002647 bio = r10_bio->devs[slot].bio;
2648 bdevname(bio->bi_bdev, b);
2649 bio_put(bio);
2650 r10_bio->devs[slot].bio = NULL;
2651
NeilBrown560f8e52011-07-28 11:39:23 +10002652 if (mddev->ro == 0) {
NeilBrowne2d59922013-06-12 11:01:22 +10002653 freeze_array(conf, 1);
NeilBrown560f8e52011-07-28 11:39:23 +10002654 fix_read_error(conf, mddev, r10_bio);
2655 unfreeze_array(conf);
NeilBrownfae8cc52012-02-14 11:10:10 +11002656 } else
2657 r10_bio->devs[slot].bio = IO_BLOCKED;
2658
NeilBrownabbf0982011-12-23 10:17:54 +11002659 rdev_dec_pending(rdev, mddev);
NeilBrown560f8e52011-07-28 11:39:23 +10002660
NeilBrown7399c312011-07-28 11:39:23 +10002661read_more:
NeilBrown96c3fd12011-12-23 10:17:54 +11002662 rdev = read_balance(conf, r10_bio, &max_sectors);
2663 if (rdev == NULL) {
NeilBrown560f8e52011-07-28 11:39:23 +10002664 printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O"
2665 " read error for block %llu\n",
NeilBrown7399c312011-07-28 11:39:23 +10002666 mdname(mddev), b,
NeilBrown560f8e52011-07-28 11:39:23 +10002667 (unsigned long long)r10_bio->sector);
2668 raid_end_bio_io(r10_bio);
NeilBrown560f8e52011-07-28 11:39:23 +10002669 return;
2670 }
2671
2672 do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC);
NeilBrown560f8e52011-07-28 11:39:23 +10002673 slot = r10_bio->read_slot;
NeilBrown560f8e52011-07-28 11:39:23 +10002674 printk_ratelimited(
2675 KERN_ERR
NeilBrown055d3742012-07-03 15:55:33 +10002676 "md/raid10:%s: %s: redirecting "
NeilBrown560f8e52011-07-28 11:39:23 +10002677 "sector %llu to another mirror\n",
2678 mdname(mddev),
2679 bdevname(rdev->bdev, b),
2680 (unsigned long long)r10_bio->sector);
2681 bio = bio_clone_mddev(r10_bio->master_bio,
2682 GFP_NOIO, mddev);
NeilBrown7399c312011-07-28 11:39:23 +10002683 md_trim_bio(bio,
2684 r10_bio->sector - bio->bi_sector,
2685 max_sectors);
NeilBrown560f8e52011-07-28 11:39:23 +10002686 r10_bio->devs[slot].bio = bio;
NeilBrownabbf0982011-12-23 10:17:54 +11002687 r10_bio->devs[slot].rdev = rdev;
NeilBrown560f8e52011-07-28 11:39:23 +10002688 bio->bi_sector = r10_bio->devs[slot].addr
NeilBrownf8c9e742012-05-21 09:28:33 +10002689 + choose_data_offset(r10_bio, rdev);
NeilBrown560f8e52011-07-28 11:39:23 +10002690 bio->bi_bdev = rdev->bdev;
2691 bio->bi_rw = READ | do_sync;
2692 bio->bi_private = r10_bio;
2693 bio->bi_end_io = raid10_end_read_request;
NeilBrown7399c312011-07-28 11:39:23 +10002694 if (max_sectors < r10_bio->sectors) {
2695 /* Drat - have to split this up more */
2696 struct bio *mbio = r10_bio->master_bio;
2697 int sectors_handled =
2698 r10_bio->sector + max_sectors
2699 - mbio->bi_sector;
2700 r10_bio->sectors = max_sectors;
2701 spin_lock_irq(&conf->device_lock);
2702 if (mbio->bi_phys_segments == 0)
2703 mbio->bi_phys_segments = 2;
2704 else
2705 mbio->bi_phys_segments++;
2706 spin_unlock_irq(&conf->device_lock);
2707 generic_make_request(bio);
NeilBrown7399c312011-07-28 11:39:23 +10002708
2709 r10_bio = mempool_alloc(conf->r10bio_pool,
2710 GFP_NOIO);
2711 r10_bio->master_bio = mbio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002712 r10_bio->sectors = bio_sectors(mbio) - sectors_handled;
NeilBrown7399c312011-07-28 11:39:23 +10002713 r10_bio->state = 0;
2714 set_bit(R10BIO_ReadError,
2715 &r10_bio->state);
2716 r10_bio->mddev = mddev;
2717 r10_bio->sector = mbio->bi_sector
2718 + sectors_handled;
2719
2720 goto read_more;
2721 } else
2722 generic_make_request(bio);
NeilBrown560f8e52011-07-28 11:39:23 +10002723}
2724
NeilBrowne879a872011-10-11 16:49:02 +11002725static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio)
NeilBrown749c55e2011-07-28 11:39:24 +10002726{
2727 /* Some sort of write request has finished and it
2728 * succeeded in writing where we thought there was a
2729 * bad block. So forget the bad block.
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002730 * Or possibly if failed and we need to record
2731 * a bad block.
NeilBrown749c55e2011-07-28 11:39:24 +10002732 */
2733 int m;
NeilBrown3cb03002011-10-11 16:45:26 +11002734 struct md_rdev *rdev;
NeilBrown749c55e2011-07-28 11:39:24 +10002735
2736 if (test_bit(R10BIO_IsSync, &r10_bio->state) ||
2737 test_bit(R10BIO_IsRecover, &r10_bio->state)) {
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002738 for (m = 0; m < conf->copies; m++) {
2739 int dev = r10_bio->devs[m].devnum;
2740 rdev = conf->mirrors[dev].rdev;
2741 if (r10_bio->devs[m].bio == NULL)
2742 continue;
2743 if (test_bit(BIO_UPTODATE,
NeilBrown749c55e2011-07-28 11:39:24 +10002744 &r10_bio->devs[m].bio->bi_flags)) {
NeilBrown749c55e2011-07-28 11:39:24 +10002745 rdev_clear_badblocks(
2746 rdev,
2747 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002748 r10_bio->sectors, 0);
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002749 } else {
2750 if (!rdev_set_badblocks(
2751 rdev,
2752 r10_bio->devs[m].addr,
2753 r10_bio->sectors, 0))
2754 md_error(conf->mddev, rdev);
NeilBrown749c55e2011-07-28 11:39:24 +10002755 }
NeilBrown9ad1aef2011-12-23 10:17:55 +11002756 rdev = conf->mirrors[dev].replacement;
2757 if (r10_bio->devs[m].repl_bio == NULL)
2758 continue;
2759 if (test_bit(BIO_UPTODATE,
2760 &r10_bio->devs[m].repl_bio->bi_flags)) {
2761 rdev_clear_badblocks(
2762 rdev,
2763 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002764 r10_bio->sectors, 0);
NeilBrown9ad1aef2011-12-23 10:17:55 +11002765 } else {
2766 if (!rdev_set_badblocks(
2767 rdev,
2768 r10_bio->devs[m].addr,
2769 r10_bio->sectors, 0))
2770 md_error(conf->mddev, rdev);
2771 }
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002772 }
NeilBrown749c55e2011-07-28 11:39:24 +10002773 put_buf(r10_bio);
2774 } else {
NeilBrownbd870a12011-07-28 11:39:24 +10002775 for (m = 0; m < conf->copies; m++) {
2776 int dev = r10_bio->devs[m].devnum;
2777 struct bio *bio = r10_bio->devs[m].bio;
2778 rdev = conf->mirrors[dev].rdev;
2779 if (bio == IO_MADE_GOOD) {
NeilBrown749c55e2011-07-28 11:39:24 +10002780 rdev_clear_badblocks(
2781 rdev,
2782 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002783 r10_bio->sectors, 0);
NeilBrown749c55e2011-07-28 11:39:24 +10002784 rdev_dec_pending(rdev, conf->mddev);
NeilBrownbd870a12011-07-28 11:39:24 +10002785 } else if (bio != NULL &&
2786 !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2787 if (!narrow_write_error(r10_bio, m)) {
2788 md_error(conf->mddev, rdev);
2789 set_bit(R10BIO_Degraded,
2790 &r10_bio->state);
2791 }
2792 rdev_dec_pending(rdev, conf->mddev);
NeilBrown749c55e2011-07-28 11:39:24 +10002793 }
NeilBrown475b0322011-12-23 10:17:55 +11002794 bio = r10_bio->devs[m].repl_bio;
2795 rdev = conf->mirrors[dev].replacement;
NeilBrown4ca40c22011-12-23 10:17:55 +11002796 if (rdev && bio == IO_MADE_GOOD) {
NeilBrown475b0322011-12-23 10:17:55 +11002797 rdev_clear_badblocks(
2798 rdev,
2799 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002800 r10_bio->sectors, 0);
NeilBrown475b0322011-12-23 10:17:55 +11002801 rdev_dec_pending(rdev, conf->mddev);
2802 }
NeilBrownbd870a12011-07-28 11:39:24 +10002803 }
2804 if (test_bit(R10BIO_WriteError,
2805 &r10_bio->state))
2806 close_write(r10_bio);
NeilBrown749c55e2011-07-28 11:39:24 +10002807 raid_end_bio_io(r10_bio);
2808 }
2809}
2810
Shaohua Li4ed87312012-10-11 13:34:00 +11002811static void raid10d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812{
Shaohua Li4ed87312012-10-11 13:34:00 +11002813 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002814 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 unsigned long flags;
NeilBrowne879a872011-10-11 16:49:02 +11002816 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002818 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819
2820 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002822 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002824
NeilBrown0021b7b2012-07-31 09:08:14 +02002825 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002826
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002828 if (list_empty(head)) {
NeilBrown6cce3b22006-01-06 00:20:16 -08002829 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002831 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002832 r10_bio = list_entry(head->prev, struct r10bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 list_del(head->prev);
NeilBrown4443ae12006-01-06 00:20:28 -08002834 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 spin_unlock_irqrestore(&conf->device_lock, flags);
2836
2837 mddev = r10_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002838 conf = mddev->private;
NeilBrownbd870a12011-07-28 11:39:24 +10002839 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
2840 test_bit(R10BIO_WriteError, &r10_bio->state))
NeilBrown749c55e2011-07-28 11:39:24 +10002841 handle_write_completed(conf, r10_bio);
NeilBrown3ea7daa2012-05-22 13:53:47 +10002842 else if (test_bit(R10BIO_IsReshape, &r10_bio->state))
2843 reshape_request_write(mddev, r10_bio);
NeilBrown749c55e2011-07-28 11:39:24 +10002844 else if (test_bit(R10BIO_IsSync, &r10_bio->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 sync_request_write(mddev, r10_bio);
Jens Axboe7eaceac2011-03-10 08:52:07 +01002846 else if (test_bit(R10BIO_IsRecover, &r10_bio->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 recovery_request_write(mddev, r10_bio);
NeilBrown856e08e2011-07-28 11:39:23 +10002848 else if (test_bit(R10BIO_ReadError, &r10_bio->state))
NeilBrown560f8e52011-07-28 11:39:23 +10002849 handle_read_error(mddev, r10_bio);
NeilBrown856e08e2011-07-28 11:39:23 +10002850 else {
2851 /* just a partial read to be scheduled from a
2852 * separate context
2853 */
2854 int slot = r10_bio->read_slot;
2855 generic_make_request(r10_bio->devs[slot].bio);
2856 }
NeilBrown4443ae12006-01-06 00:20:28 -08002857
NeilBrown1d9d5242009-10-16 15:55:32 +11002858 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002859 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2860 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002862 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863}
2864
2865
NeilBrowne879a872011-10-11 16:49:02 +11002866static int init_resync(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867{
2868 int buffs;
NeilBrown69335ef2011-12-23 10:17:54 +11002869 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
2871 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhennb6385482006-04-02 13:34:29 +02002872 BUG_ON(conf->r10buf_pool);
NeilBrown69335ef2011-12-23 10:17:54 +11002873 conf->have_replacement = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10002874 for (i = 0; i < conf->geo.raid_disks; i++)
NeilBrown69335ef2011-12-23 10:17:54 +11002875 if (conf->mirrors[i].replacement)
2876 conf->have_replacement = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf);
2878 if (!conf->r10buf_pool)
2879 return -ENOMEM;
2880 conf->next_resync = 0;
2881 return 0;
2882}
2883
2884/*
2885 * perform a "sync" on one "block"
2886 *
2887 * We need to make sure that no normal I/O request - particularly write
2888 * requests - conflict with active sync requests.
2889 *
2890 * This is achieved by tracking pending requests and a 'barrier' concept
2891 * that can be installed to exclude normal IO requests.
2892 *
2893 * Resync and recovery are handled very differently.
2894 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2895 *
2896 * For resync, we iterate over virtual addresses, read all copies,
2897 * and update if there are differences. If only one copy is live,
2898 * skip it.
2899 * For recovery, we iterate over physical addresses, read a good
2900 * value for each non-in_sync drive, and over-write.
2901 *
2902 * So, for recovery we may have several outstanding complex requests for a
2903 * given address, one for each out-of-sync device. We model this by allocating
2904 * a number of r10_bio structures, one for each out-of-sync device.
2905 * As we setup these structures, we collect all bio's together into a list
2906 * which we then process collectively to add pages, and then process again
2907 * to pass to generic_make_request.
2908 *
2909 * The r10_bio structures are linked using a borrowed master_bio pointer.
2910 * This link is counted in ->remaining. When the r10_bio that points to NULL
2911 * has its remaining count decremented to 0, the whole complex operation
2912 * is complete.
2913 *
2914 */
2915
NeilBrownfd01b882011-10-11 16:47:53 +11002916static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
NeilBrownab9d47e2011-05-11 14:54:41 +10002917 int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918{
NeilBrowne879a872011-10-11 16:49:02 +11002919 struct r10conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002920 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 struct bio *biolist = NULL, *bio;
2922 sector_t max_sector, nr_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 int i;
NeilBrown6cce3b22006-01-06 00:20:16 -08002924 int max_sync;
NeilBrown57dab0b2010-10-19 10:03:39 +11002925 sector_t sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 sector_t sectors_skipped = 0;
2927 int chunks_skipped = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10002928 sector_t chunk_mask = conf->geo.chunk_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929
2930 if (!conf->r10buf_pool)
2931 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002932 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002934 /*
2935 * Allow skipping a full rebuild for incremental assembly
2936 * of a clean array, like RAID1 does.
2937 */
2938 if (mddev->bitmap == NULL &&
2939 mddev->recovery_cp == MaxSector &&
NeilBrown13765122013-07-04 16:41:53 +10002940 mddev->reshape_position == MaxSector &&
2941 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002942 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown13765122013-07-04 16:41:53 +10002943 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002944 conf->fullsync == 0) {
2945 *skipped = 1;
NeilBrown13765122013-07-04 16:41:53 +10002946 return mddev->dev_sectors - sector_nr;
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002947 }
2948
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 skipped:
Andre Noll58c0fed2009-03-31 14:33:13 +11002950 max_sector = mddev->dev_sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10002951 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
2952 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 max_sector = mddev->resync_max_sectors;
2954 if (sector_nr >= max_sector) {
NeilBrown6cce3b22006-01-06 00:20:16 -08002955 /* If we aborted, we need to abort the
2956 * sync on the 'current' bitmap chucks (there can
2957 * be several when recovering multiple devices).
2958 * as we may have started syncing it but not finished.
2959 * We can find the current address in
2960 * mddev->curr_resync, but for recovery,
2961 * we need to convert that to several
2962 * virtual addresses.
2963 */
NeilBrown3ea7daa2012-05-22 13:53:47 +10002964 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
2965 end_reshape(conf);
2966 return 0;
2967 }
2968
NeilBrown6cce3b22006-01-06 00:20:16 -08002969 if (mddev->curr_resync < max_sector) { /* aborted */
2970 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2971 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2972 &sync_blocks, 1);
NeilBrown5cf00fc2012-05-21 09:28:20 +10002973 else for (i = 0; i < conf->geo.raid_disks; i++) {
NeilBrown6cce3b22006-01-06 00:20:16 -08002974 sector_t sect =
2975 raid10_find_virt(conf, mddev->curr_resync, i);
2976 bitmap_end_sync(mddev->bitmap, sect,
2977 &sync_blocks, 1);
2978 }
NeilBrown9ad1aef2011-12-23 10:17:55 +11002979 } else {
2980 /* completed sync */
2981 if ((!mddev->bitmap || conf->fullsync)
2982 && conf->have_replacement
2983 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2984 /* Completed a full sync so the replacements
2985 * are now fully recovered.
2986 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10002987 for (i = 0; i < conf->geo.raid_disks; i++)
NeilBrown9ad1aef2011-12-23 10:17:55 +11002988 if (conf->mirrors[i].replacement)
2989 conf->mirrors[i].replacement
2990 ->recovery_offset
2991 = MaxSector;
2992 }
NeilBrown6cce3b22006-01-06 00:20:16 -08002993 conf->fullsync = 0;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002994 }
NeilBrown6cce3b22006-01-06 00:20:16 -08002995 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 close_sync(conf);
NeilBrown57afd892005-06-21 17:17:13 -07002997 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 return sectors_skipped;
2999 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10003000
3001 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3002 return reshape_request(mddev, sector_nr, skipped);
3003
NeilBrown5cf00fc2012-05-21 09:28:20 +10003004 if (chunks_skipped >= conf->geo.raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 /* if there has been nothing to do on any drive,
3006 * then there is nothing to do at all..
3007 */
NeilBrown57afd892005-06-21 17:17:13 -07003008 *skipped = 1;
3009 return (max_sector - sector_nr) + sectors_skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 }
3011
NeilBrownc6207272008-02-06 01:39:52 -08003012 if (max_sector > mddev->resync_max)
3013 max_sector = mddev->resync_max; /* Don't do IO beyond here */
3014
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 /* make sure whole request will fit in a chunk - if chunks
3016 * are meaningful
3017 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003018 if (conf->geo.near_copies < conf->geo.raid_disks &&
3019 max_sector > (sector_nr | chunk_mask))
3020 max_sector = (sector_nr | chunk_mask) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 /*
3022 * If there is non-resync activity waiting for us then
3023 * put in a delay to throttle resync.
3024 */
NeilBrown0a27ec92006-01-06 00:20:13 -08003025 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 msleep_interruptible(1000);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027
3028 /* Again, very different code for resync and recovery.
3029 * Both must result in an r10bio with a list of bios that
3030 * have bi_end_io, bi_sector, bi_bdev set,
3031 * and bi_private set to the r10bio.
3032 * For recovery, we may actually create several r10bios
3033 * with 2 bios in each, that correspond to the bios in the main one.
3034 * In this case, the subordinate r10bios link back through a
3035 * borrowed master_bio pointer, and the counter in the master
3036 * includes a ref from each subordinate.
3037 */
3038 /* First, we decide what to do and set ->bi_end_io
3039 * To end_sync_read if we want to read, and
3040 * end_sync_write if we will want to write.
3041 */
3042
NeilBrown6cce3b22006-01-06 00:20:16 -08003043 max_sync = RESYNC_PAGES << (PAGE_SHIFT-9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3045 /* recovery... the complicated one */
NeilBrowne875ece2011-07-28 11:39:24 +10003046 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 r10_bio = NULL;
3048
NeilBrown5cf00fc2012-05-21 09:28:20 +10003049 for (i = 0 ; i < conf->geo.raid_disks; i++) {
NeilBrownab9d47e2011-05-11 14:54:41 +10003050 int still_degraded;
NeilBrown9f2c9d12011-10-11 16:48:43 +11003051 struct r10bio *rb2;
NeilBrownab9d47e2011-05-11 14:54:41 +10003052 sector_t sect;
3053 int must_sync;
NeilBrowne875ece2011-07-28 11:39:24 +10003054 int any_working;
Jonathan Brassowdc280d92012-07-31 10:03:52 +10003055 struct raid10_info *mirror = &conf->mirrors[i];
NeilBrownab9d47e2011-05-11 14:54:41 +10003056
NeilBrown24afd802011-12-23 10:17:55 +11003057 if ((mirror->rdev == NULL ||
3058 test_bit(In_sync, &mirror->rdev->flags))
3059 &&
3060 (mirror->replacement == NULL ||
3061 test_bit(Faulty,
3062 &mirror->replacement->flags)))
NeilBrownab9d47e2011-05-11 14:54:41 +10003063 continue;
3064
3065 still_degraded = 0;
3066 /* want to reconstruct this device */
3067 rb2 = r10_bio;
3068 sect = raid10_find_virt(conf, sector_nr, i);
NeilBrownfc448a12012-07-03 10:37:30 +10003069 if (sect >= mddev->resync_max_sectors) {
3070 /* last stripe is not complete - don't
3071 * try to recover this sector.
3072 */
3073 continue;
3074 }
NeilBrown24afd802011-12-23 10:17:55 +11003075 /* Unless we are doing a full sync, or a replacement
3076 * we only need to recover the block if it is set in
3077 * the bitmap
NeilBrownab9d47e2011-05-11 14:54:41 +10003078 */
3079 must_sync = bitmap_start_sync(mddev->bitmap, sect,
3080 &sync_blocks, 1);
3081 if (sync_blocks < max_sync)
3082 max_sync = sync_blocks;
3083 if (!must_sync &&
NeilBrown24afd802011-12-23 10:17:55 +11003084 mirror->replacement == NULL &&
NeilBrownab9d47e2011-05-11 14:54:41 +10003085 !conf->fullsync) {
3086 /* yep, skip the sync_blocks here, but don't assume
3087 * that there will never be anything to do here
NeilBrown6cce3b22006-01-06 00:20:16 -08003088 */
NeilBrownab9d47e2011-05-11 14:54:41 +10003089 chunks_skipped = -1;
3090 continue;
3091 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092
NeilBrownab9d47e2011-05-11 14:54:41 +10003093 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
3094 raise_barrier(conf, rb2 != NULL);
3095 atomic_set(&r10_bio->remaining, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
NeilBrownab9d47e2011-05-11 14:54:41 +10003097 r10_bio->master_bio = (struct bio*)rb2;
3098 if (rb2)
3099 atomic_inc(&rb2->remaining);
3100 r10_bio->mddev = mddev;
3101 set_bit(R10BIO_IsRecover, &r10_bio->state);
3102 r10_bio->sector = sect;
NeilBrown6cce3b22006-01-06 00:20:16 -08003103
NeilBrownab9d47e2011-05-11 14:54:41 +10003104 raid10_find_phys(conf, r10_bio);
NeilBrown18055562009-05-07 12:48:10 +10003105
NeilBrownab9d47e2011-05-11 14:54:41 +10003106 /* Need to check if the array will still be
3107 * degraded
3108 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003109 for (j = 0; j < conf->geo.raid_disks; j++)
NeilBrownab9d47e2011-05-11 14:54:41 +10003110 if (conf->mirrors[j].rdev == NULL ||
3111 test_bit(Faulty, &conf->mirrors[j].rdev->flags)) {
3112 still_degraded = 1;
NeilBrown87fc7672005-09-09 16:24:04 -07003113 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003115
3116 must_sync = bitmap_start_sync(mddev->bitmap, sect,
3117 &sync_blocks, still_degraded);
3118
NeilBrowne875ece2011-07-28 11:39:24 +10003119 any_working = 0;
NeilBrownab9d47e2011-05-11 14:54:41 +10003120 for (j=0; j<conf->copies;j++) {
NeilBrowne875ece2011-07-28 11:39:24 +10003121 int k;
NeilBrownab9d47e2011-05-11 14:54:41 +10003122 int d = r10_bio->devs[j].devnum;
NeilBrown5e570282011-07-28 11:39:25 +10003123 sector_t from_addr, to_addr;
NeilBrown3cb03002011-10-11 16:45:26 +11003124 struct md_rdev *rdev;
NeilBrown40c356c2011-07-28 11:39:24 +10003125 sector_t sector, first_bad;
3126 int bad_sectors;
NeilBrownab9d47e2011-05-11 14:54:41 +10003127 if (!conf->mirrors[d].rdev ||
3128 !test_bit(In_sync, &conf->mirrors[d].rdev->flags))
3129 continue;
3130 /* This is where we read from */
NeilBrowne875ece2011-07-28 11:39:24 +10003131 any_working = 1;
NeilBrown40c356c2011-07-28 11:39:24 +10003132 rdev = conf->mirrors[d].rdev;
3133 sector = r10_bio->devs[j].addr;
3134
3135 if (is_badblock(rdev, sector, max_sync,
3136 &first_bad, &bad_sectors)) {
3137 if (first_bad > sector)
3138 max_sync = first_bad - sector;
3139 else {
3140 bad_sectors -= (sector
3141 - first_bad);
3142 if (max_sync > bad_sectors)
3143 max_sync = bad_sectors;
3144 continue;
3145 }
3146 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003147 bio = r10_bio->devs[0].bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003148 bio_reset(bio);
NeilBrownab9d47e2011-05-11 14:54:41 +10003149 bio->bi_next = biolist;
3150 biolist = bio;
3151 bio->bi_private = r10_bio;
3152 bio->bi_end_io = end_sync_read;
3153 bio->bi_rw = READ;
NeilBrown5e570282011-07-28 11:39:25 +10003154 from_addr = r10_bio->devs[j].addr;
NeilBrown24afd802011-12-23 10:17:55 +11003155 bio->bi_sector = from_addr + rdev->data_offset;
3156 bio->bi_bdev = rdev->bdev;
3157 atomic_inc(&rdev->nr_pending);
3158 /* and we write to 'i' (if not in_sync) */
NeilBrownab9d47e2011-05-11 14:54:41 +10003159
3160 for (k=0; k<conf->copies; k++)
3161 if (r10_bio->devs[k].devnum == i)
3162 break;
3163 BUG_ON(k == conf->copies);
NeilBrown5e570282011-07-28 11:39:25 +10003164 to_addr = r10_bio->devs[k].addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003165 r10_bio->devs[0].devnum = d;
NeilBrown5e570282011-07-28 11:39:25 +10003166 r10_bio->devs[0].addr = from_addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003167 r10_bio->devs[1].devnum = i;
NeilBrown5e570282011-07-28 11:39:25 +10003168 r10_bio->devs[1].addr = to_addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003169
NeilBrown24afd802011-12-23 10:17:55 +11003170 rdev = mirror->rdev;
3171 if (!test_bit(In_sync, &rdev->flags)) {
3172 bio = r10_bio->devs[1].bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003173 bio_reset(bio);
NeilBrown24afd802011-12-23 10:17:55 +11003174 bio->bi_next = biolist;
3175 biolist = bio;
3176 bio->bi_private = r10_bio;
3177 bio->bi_end_io = end_sync_write;
3178 bio->bi_rw = WRITE;
3179 bio->bi_sector = to_addr
3180 + rdev->data_offset;
3181 bio->bi_bdev = rdev->bdev;
3182 atomic_inc(&r10_bio->remaining);
3183 } else
3184 r10_bio->devs[1].bio->bi_end_io = NULL;
3185
3186 /* and maybe write to replacement */
3187 bio = r10_bio->devs[1].repl_bio;
3188 if (bio)
3189 bio->bi_end_io = NULL;
3190 rdev = mirror->replacement;
3191 /* Note: if rdev != NULL, then bio
3192 * cannot be NULL as r10buf_pool_alloc will
3193 * have allocated it.
3194 * So the second test here is pointless.
3195 * But it keeps semantic-checkers happy, and
3196 * this comment keeps human reviewers
3197 * happy.
3198 */
3199 if (rdev == NULL || bio == NULL ||
3200 test_bit(Faulty, &rdev->flags))
3201 break;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003202 bio_reset(bio);
NeilBrown24afd802011-12-23 10:17:55 +11003203 bio->bi_next = biolist;
3204 biolist = bio;
3205 bio->bi_private = r10_bio;
3206 bio->bi_end_io = end_sync_write;
3207 bio->bi_rw = WRITE;
3208 bio->bi_sector = to_addr + rdev->data_offset;
3209 bio->bi_bdev = rdev->bdev;
3210 atomic_inc(&r10_bio->remaining);
NeilBrownab9d47e2011-05-11 14:54:41 +10003211 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003213 if (j == conf->copies) {
NeilBrowne875ece2011-07-28 11:39:24 +10003214 /* Cannot recover, so abort the recovery or
3215 * record a bad block */
NeilBrownab9d47e2011-05-11 14:54:41 +10003216 put_buf(r10_bio);
3217 if (rb2)
3218 atomic_dec(&rb2->remaining);
3219 r10_bio = rb2;
NeilBrowne875ece2011-07-28 11:39:24 +10003220 if (any_working) {
3221 /* problem is that there are bad blocks
3222 * on other device(s)
3223 */
3224 int k;
3225 for (k = 0; k < conf->copies; k++)
3226 if (r10_bio->devs[k].devnum == i)
3227 break;
NeilBrown24afd802011-12-23 10:17:55 +11003228 if (!test_bit(In_sync,
3229 &mirror->rdev->flags)
3230 && !rdev_set_badblocks(
3231 mirror->rdev,
3232 r10_bio->devs[k].addr,
3233 max_sync, 0))
3234 any_working = 0;
3235 if (mirror->replacement &&
3236 !rdev_set_badblocks(
3237 mirror->replacement,
NeilBrowne875ece2011-07-28 11:39:24 +10003238 r10_bio->devs[k].addr,
3239 max_sync, 0))
3240 any_working = 0;
3241 }
3242 if (!any_working) {
3243 if (!test_and_set_bit(MD_RECOVERY_INTR,
3244 &mddev->recovery))
3245 printk(KERN_INFO "md/raid10:%s: insufficient "
3246 "working devices for recovery.\n",
3247 mdname(mddev));
NeilBrown24afd802011-12-23 10:17:55 +11003248 mirror->recovery_disabled
NeilBrowne875ece2011-07-28 11:39:24 +10003249 = mddev->recovery_disabled;
3250 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003251 break;
3252 }
3253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 if (biolist == NULL) {
3255 while (r10_bio) {
NeilBrown9f2c9d12011-10-11 16:48:43 +11003256 struct r10bio *rb2 = r10_bio;
3257 r10_bio = (struct r10bio*) rb2->master_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 rb2->master_bio = NULL;
3259 put_buf(rb2);
3260 }
3261 goto giveup;
3262 }
3263 } else {
3264 /* resync. Schedule a read for every block at this virt offset */
3265 int count = 0;
NeilBrown6cce3b22006-01-06 00:20:16 -08003266
NeilBrown78200d42009-02-25 13:18:47 +11003267 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
3268
NeilBrown6cce3b22006-01-06 00:20:16 -08003269 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
3270 &sync_blocks, mddev->degraded) &&
NeilBrownab9d47e2011-05-11 14:54:41 +10003271 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED,
3272 &mddev->recovery)) {
NeilBrown6cce3b22006-01-06 00:20:16 -08003273 /* We can skip this block */
3274 *skipped = 1;
3275 return sync_blocks + sectors_skipped;
3276 }
3277 if (sync_blocks < max_sync)
3278 max_sync = sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
3280
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 r10_bio->mddev = mddev;
3282 atomic_set(&r10_bio->remaining, 0);
NeilBrown6cce3b22006-01-06 00:20:16 -08003283 raise_barrier(conf, 0);
3284 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285
3286 r10_bio->master_bio = NULL;
3287 r10_bio->sector = sector_nr;
3288 set_bit(R10BIO_IsSync, &r10_bio->state);
3289 raid10_find_phys(conf, r10_bio);
NeilBrown5cf00fc2012-05-21 09:28:20 +10003290 r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291
NeilBrown5cf00fc2012-05-21 09:28:20 +10003292 for (i = 0; i < conf->copies; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 int d = r10_bio->devs[i].devnum;
NeilBrown40c356c2011-07-28 11:39:24 +10003294 sector_t first_bad, sector;
3295 int bad_sectors;
3296
NeilBrown9ad1aef2011-12-23 10:17:55 +11003297 if (r10_bio->devs[i].repl_bio)
3298 r10_bio->devs[i].repl_bio->bi_end_io = NULL;
3299
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 bio = r10_bio->devs[i].bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003301 bio_reset(bio);
NeilBrownaf03b8e2007-06-16 10:16:06 -07003302 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 if (conf->mirrors[d].rdev == NULL ||
NeilBrownb2d444d2005-11-08 21:39:31 -08003304 test_bit(Faulty, &conf->mirrors[d].rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305 continue;
NeilBrown40c356c2011-07-28 11:39:24 +10003306 sector = r10_bio->devs[i].addr;
3307 if (is_badblock(conf->mirrors[d].rdev,
3308 sector, max_sync,
3309 &first_bad, &bad_sectors)) {
3310 if (first_bad > sector)
3311 max_sync = first_bad - sector;
3312 else {
3313 bad_sectors -= (sector - first_bad);
3314 if (max_sync > bad_sectors)
Dan Carpenter91502f02012-10-11 14:20:58 +11003315 max_sync = bad_sectors;
NeilBrown40c356c2011-07-28 11:39:24 +10003316 continue;
3317 }
3318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
3320 atomic_inc(&r10_bio->remaining);
3321 bio->bi_next = biolist;
3322 biolist = bio;
3323 bio->bi_private = r10_bio;
3324 bio->bi_end_io = end_sync_read;
NeilBrown802ba062006-12-13 00:34:13 -08003325 bio->bi_rw = READ;
NeilBrown40c356c2011-07-28 11:39:24 +10003326 bio->bi_sector = sector +
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 conf->mirrors[d].rdev->data_offset;
3328 bio->bi_bdev = conf->mirrors[d].rdev->bdev;
3329 count++;
NeilBrown9ad1aef2011-12-23 10:17:55 +11003330
3331 if (conf->mirrors[d].replacement == NULL ||
3332 test_bit(Faulty,
3333 &conf->mirrors[d].replacement->flags))
3334 continue;
3335
3336 /* Need to set up for writing to the replacement */
3337 bio = r10_bio->devs[i].repl_bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003338 bio_reset(bio);
NeilBrown9ad1aef2011-12-23 10:17:55 +11003339 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3340
3341 sector = r10_bio->devs[i].addr;
3342 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
3343 bio->bi_next = biolist;
3344 biolist = bio;
3345 bio->bi_private = r10_bio;
3346 bio->bi_end_io = end_sync_write;
3347 bio->bi_rw = WRITE;
3348 bio->bi_sector = sector +
3349 conf->mirrors[d].replacement->data_offset;
3350 bio->bi_bdev = conf->mirrors[d].replacement->bdev;
3351 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 }
3353
3354 if (count < 2) {
3355 for (i=0; i<conf->copies; i++) {
3356 int d = r10_bio->devs[i].devnum;
3357 if (r10_bio->devs[i].bio->bi_end_io)
NeilBrownab9d47e2011-05-11 14:54:41 +10003358 rdev_dec_pending(conf->mirrors[d].rdev,
3359 mddev);
NeilBrown9ad1aef2011-12-23 10:17:55 +11003360 if (r10_bio->devs[i].repl_bio &&
3361 r10_bio->devs[i].repl_bio->bi_end_io)
3362 rdev_dec_pending(
3363 conf->mirrors[d].replacement,
3364 mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 }
3366 put_buf(r10_bio);
3367 biolist = NULL;
3368 goto giveup;
3369 }
3370 }
3371
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 nr_sectors = 0;
NeilBrown6cce3b22006-01-06 00:20:16 -08003373 if (sector_nr + max_sync < max_sector)
3374 max_sector = sector_nr + max_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 do {
3376 struct page *page;
3377 int len = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378 if (sector_nr + (len>>9) > max_sector)
3379 len = (max_sector - sector_nr) << 9;
3380 if (len == 0)
3381 break;
3382 for (bio= biolist ; bio ; bio=bio->bi_next) {
NeilBrownab9d47e2011-05-11 14:54:41 +10003383 struct bio *bio2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003384 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
NeilBrownab9d47e2011-05-11 14:54:41 +10003385 if (bio_add_page(bio, page, len, 0))
3386 continue;
3387
3388 /* stop here */
3389 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
3390 for (bio2 = biolist;
3391 bio2 && bio2 != bio;
3392 bio2 = bio2->bi_next) {
3393 /* remove last page from this bio */
3394 bio2->bi_vcnt--;
3395 bio2->bi_size -= len;
3396 bio2->bi_flags &= ~(1<< BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003398 goto bio_full;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 }
3400 nr_sectors += len>>9;
3401 sector_nr += len>>9;
3402 } while (biolist->bi_vcnt < RESYNC_PAGES);
3403 bio_full:
3404 r10_bio->sectors = nr_sectors;
3405
3406 while (biolist) {
3407 bio = biolist;
3408 biolist = biolist->bi_next;
3409
3410 bio->bi_next = NULL;
3411 r10_bio = bio->bi_private;
3412 r10_bio->sectors = nr_sectors;
3413
3414 if (bio->bi_end_io == end_sync_read) {
3415 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrown7bb23c42013-07-16 16:50:47 +10003416 set_bit(BIO_UPTODATE, &bio->bi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 generic_make_request(bio);
3418 }
3419 }
3420
NeilBrown57afd892005-06-21 17:17:13 -07003421 if (sectors_skipped)
3422 /* pretend they weren't skipped, it makes
3423 * no important difference in this case
3424 */
3425 md_done_sync(mddev, sectors_skipped, 1);
3426
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 return sectors_skipped + nr_sectors;
3428 giveup:
3429 /* There is nowhere to write, so all non-sync
NeilBrowne875ece2011-07-28 11:39:24 +10003430 * drives must be failed or in resync, all drives
3431 * have a bad block, so try the next chunk...
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 */
NeilBrown09b40682009-02-25 13:18:47 +11003433 if (sector_nr + max_sync < max_sector)
3434 max_sector = sector_nr + max_sync;
3435
3436 sectors_skipped += (max_sector - sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437 chunks_skipped ++;
3438 sector_nr = max_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439 goto skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440}
3441
Dan Williams80c3a6c2009-03-17 18:10:40 -07003442static sector_t
NeilBrownfd01b882011-10-11 16:47:53 +11003443raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07003444{
3445 sector_t size;
NeilBrowne879a872011-10-11 16:49:02 +11003446 struct r10conf *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003447
3448 if (!raid_disks)
NeilBrown3ea7daa2012-05-22 13:53:47 +10003449 raid_disks = min(conf->geo.raid_disks,
3450 conf->prev.raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003451 if (!sectors)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003452 sectors = conf->dev_sectors;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003453
NeilBrown5cf00fc2012-05-21 09:28:20 +10003454 size = sectors >> conf->geo.chunk_shift;
3455 sector_div(size, conf->geo.far_copies);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003456 size = size * raid_disks;
NeilBrown5cf00fc2012-05-21 09:28:20 +10003457 sector_div(size, conf->geo.near_copies);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003458
NeilBrown5cf00fc2012-05-21 09:28:20 +10003459 return size << conf->geo.chunk_shift;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003460}
3461
NeilBrown6508fdb2012-05-17 10:08:45 +10003462static void calc_sectors(struct r10conf *conf, sector_t size)
3463{
3464 /* Calculate the number of sectors-per-device that will
3465 * actually be used, and set conf->dev_sectors and
3466 * conf->stride
3467 */
3468
NeilBrown5cf00fc2012-05-21 09:28:20 +10003469 size = size >> conf->geo.chunk_shift;
3470 sector_div(size, conf->geo.far_copies);
3471 size = size * conf->geo.raid_disks;
3472 sector_div(size, conf->geo.near_copies);
NeilBrown6508fdb2012-05-17 10:08:45 +10003473 /* 'size' is now the number of chunks in the array */
3474 /* calculate "used chunks per device" */
3475 size = size * conf->copies;
3476
3477 /* We need to round up when dividing by raid_disks to
3478 * get the stride size.
3479 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003480 size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks);
NeilBrown6508fdb2012-05-17 10:08:45 +10003481
NeilBrown5cf00fc2012-05-21 09:28:20 +10003482 conf->dev_sectors = size << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003483
NeilBrown5cf00fc2012-05-21 09:28:20 +10003484 if (conf->geo.far_offset)
3485 conf->geo.stride = 1 << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003486 else {
NeilBrown5cf00fc2012-05-21 09:28:20 +10003487 sector_div(size, conf->geo.far_copies);
3488 conf->geo.stride = size << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003489 }
3490}
Trela, Maciejdab8b292010-03-08 16:02:45 +11003491
NeilBrowndeb200d2012-05-21 09:28:33 +10003492enum geo_type {geo_new, geo_old, geo_start};
3493static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new)
3494{
3495 int nc, fc, fo;
3496 int layout, chunk, disks;
3497 switch (new) {
3498 case geo_old:
3499 layout = mddev->layout;
3500 chunk = mddev->chunk_sectors;
3501 disks = mddev->raid_disks - mddev->delta_disks;
3502 break;
3503 case geo_new:
3504 layout = mddev->new_layout;
3505 chunk = mddev->new_chunk_sectors;
3506 disks = mddev->raid_disks;
3507 break;
3508 default: /* avoid 'may be unused' warnings */
3509 case geo_start: /* new when starting reshape - raid_disks not
3510 * updated yet. */
3511 layout = mddev->new_layout;
3512 chunk = mddev->new_chunk_sectors;
3513 disks = mddev->raid_disks + mddev->delta_disks;
3514 break;
3515 }
Jonathan Brassow475901a2013-02-21 13:28:10 +11003516 if (layout >> 18)
NeilBrowndeb200d2012-05-21 09:28:33 +10003517 return -1;
3518 if (chunk < (PAGE_SIZE >> 9) ||
3519 !is_power_of_2(chunk))
3520 return -2;
3521 nc = layout & 255;
3522 fc = (layout >> 8) & 255;
3523 fo = layout & (1<<16);
3524 geo->raid_disks = disks;
3525 geo->near_copies = nc;
3526 geo->far_copies = fc;
3527 geo->far_offset = fo;
Jonathan Brassow475901a2013-02-21 13:28:10 +11003528 geo->far_set_size = (layout & (1<<17)) ? disks / fc : disks;
NeilBrowndeb200d2012-05-21 09:28:33 +10003529 geo->chunk_mask = chunk - 1;
3530 geo->chunk_shift = ffz(~chunk);
3531 return nc*fc;
3532}
3533
NeilBrowne879a872011-10-11 16:49:02 +11003534static struct r10conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535{
NeilBrowne879a872011-10-11 16:49:02 +11003536 struct r10conf *conf = NULL;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003537 int err = -EINVAL;
NeilBrowndeb200d2012-05-21 09:28:33 +10003538 struct geom geo;
3539 int copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540
NeilBrowndeb200d2012-05-21 09:28:33 +10003541 copies = setup_geo(&geo, mddev, geo_new);
3542
3543 if (copies == -2) {
NeilBrown128595e2010-05-03 14:47:14 +10003544 printk(KERN_ERR "md/raid10:%s: chunk size must be "
3545 "at least PAGE_SIZE(%ld) and be a power of 2.\n",
3546 mdname(mddev), PAGE_SIZE);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003547 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 }
NeilBrown2604b702006-01-06 00:20:36 -08003549
NeilBrowndeb200d2012-05-21 09:28:33 +10003550 if (copies < 2 || copies > mddev->raid_disks) {
NeilBrown128595e2010-05-03 14:47:14 +10003551 printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n",
Maciej Trelaf73ea872010-06-16 11:46:29 +01003552 mdname(mddev), mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553 goto out;
3554 }
Trela, Maciejdab8b292010-03-08 16:02:45 +11003555
3556 err = -ENOMEM;
NeilBrowne879a872011-10-11 16:49:02 +11003557 conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003558 if (!conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559 goto out;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003560
NeilBrown3ea7daa2012-05-22 13:53:47 +10003561 /* FIXME calc properly */
Jonathan Brassowdc280d92012-07-31 10:03:52 +10003562 conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks +
NeilBrown78eaa0d2013-07-02 15:58:05 +10003563 max(0,-mddev->delta_disks)),
Trela, Maciejdab8b292010-03-08 16:02:45 +11003564 GFP_KERNEL);
3565 if (!conf->mirrors)
3566 goto out;
NeilBrown4443ae12006-01-06 00:20:28 -08003567
3568 conf->tmppage = alloc_page(GFP_KERNEL);
3569 if (!conf->tmppage)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003570 goto out;
3571
NeilBrowndeb200d2012-05-21 09:28:33 +10003572 conf->geo = geo;
3573 conf->copies = copies;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003574 conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc,
3575 r10bio_pool_free, conf);
3576 if (!conf->r10bio_pool)
3577 goto out;
3578
NeilBrown6508fdb2012-05-17 10:08:45 +10003579 calc_sectors(conf, mddev->dev_sectors);
NeilBrown3ea7daa2012-05-22 13:53:47 +10003580 if (mddev->reshape_position == MaxSector) {
3581 conf->prev = conf->geo;
3582 conf->reshape_progress = MaxSector;
3583 } else {
3584 if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) {
3585 err = -EINVAL;
3586 goto out;
3587 }
3588 conf->reshape_progress = mddev->reshape_position;
3589 if (conf->prev.far_offset)
3590 conf->prev.stride = 1 << conf->prev.chunk_shift;
3591 else
3592 /* far_copies must be 1 */
3593 conf->prev.stride = conf->dev_sectors;
3594 }
Neil Browne7e72bf2008-05-14 16:05:54 -07003595 spin_lock_init(&conf->device_lock);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003596 INIT_LIST_HEAD(&conf->retry_list);
3597
3598 spin_lock_init(&conf->resync_lock);
3599 init_waitqueue_head(&conf->wait_barrier);
3600
NeilBrown02326052012-07-03 15:56:52 +10003601 conf->thread = md_register_thread(raid10d, mddev, "raid10");
Trela, Maciejdab8b292010-03-08 16:02:45 +11003602 if (!conf->thread)
3603 goto out;
3604
Trela, Maciejdab8b292010-03-08 16:02:45 +11003605 conf->mddev = mddev;
3606 return conf;
3607
3608 out:
NeilBrown3ea7daa2012-05-22 13:53:47 +10003609 if (err == -ENOMEM)
3610 printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n",
3611 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003612 if (conf) {
3613 if (conf->r10bio_pool)
3614 mempool_destroy(conf->r10bio_pool);
3615 kfree(conf->mirrors);
3616 safe_put_page(conf->tmppage);
3617 kfree(conf);
3618 }
3619 return ERR_PTR(err);
3620}
3621
NeilBrownfd01b882011-10-11 16:47:53 +11003622static int run(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003623{
NeilBrowne879a872011-10-11 16:49:02 +11003624 struct r10conf *conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003625 int i, disk_idx, chunk_size;
Jonathan Brassowdc280d92012-07-31 10:03:52 +10003626 struct raid10_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11003627 struct md_rdev *rdev;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003628 sector_t size;
NeilBrown3ea7daa2012-05-22 13:53:47 +10003629 sector_t min_offset_diff = 0;
3630 int first = 1;
Shaohua Li532a2a32012-10-11 13:30:52 +11003631 bool discard_supported = false;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003632
3633 if (mddev->private == NULL) {
3634 conf = setup_conf(mddev);
3635 if (IS_ERR(conf))
3636 return PTR_ERR(conf);
3637 mddev->private = conf;
3638 }
3639 conf = mddev->private;
3640 if (!conf)
3641 goto out;
3642
Trela, Maciejdab8b292010-03-08 16:02:45 +11003643 mddev->thread = conf->thread;
3644 conf->thread = NULL;
3645
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10003646 chunk_size = mddev->chunk_sectors << 9;
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003647 if (mddev->queue) {
Shaohua Li532a2a32012-10-11 13:30:52 +11003648 blk_queue_max_discard_sectors(mddev->queue,
3649 mddev->chunk_sectors);
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07003650 blk_queue_max_write_same_sectors(mddev->queue, 0);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003651 blk_queue_io_min(mddev->queue, chunk_size);
3652 if (conf->geo.raid_disks % conf->geo.near_copies)
3653 blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks);
3654 else
3655 blk_queue_io_opt(mddev->queue, chunk_size *
3656 (conf->geo.raid_disks / conf->geo.near_copies));
3657 }
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10003658
NeilBrowndafb20f2012-03-19 12:46:39 +11003659 rdev_for_each(rdev, mddev) {
NeilBrown3ea7daa2012-05-22 13:53:47 +10003660 long long diff;
NeilBrownaba336b2012-05-31 15:39:11 +10003661 struct request_queue *q;
NeilBrown34b343c2011-07-28 11:31:47 +10003662
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 disk_idx = rdev->raid_disk;
NeilBrownf8c9e742012-05-21 09:28:33 +10003664 if (disk_idx < 0)
3665 continue;
3666 if (disk_idx >= conf->geo.raid_disks &&
3667 disk_idx >= conf->prev.raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 continue;
3669 disk = conf->mirrors + disk_idx;
3670
NeilBrown56a25592011-12-23 10:17:55 +11003671 if (test_bit(Replacement, &rdev->flags)) {
3672 if (disk->replacement)
3673 goto out_free_conf;
3674 disk->replacement = rdev;
3675 } else {
3676 if (disk->rdev)
3677 goto out_free_conf;
3678 disk->rdev = rdev;
3679 }
NeilBrownaba336b2012-05-31 15:39:11 +10003680 q = bdev_get_queue(rdev->bdev);
3681 if (q->merge_bvec_fn)
3682 mddev->merge_check_needed = 1;
NeilBrown3ea7daa2012-05-22 13:53:47 +10003683 diff = (rdev->new_data_offset - rdev->data_offset);
3684 if (!mddev->reshape_backwards)
3685 diff = -diff;
3686 if (diff < 0)
3687 diff = 0;
3688 if (first || diff < min_offset_diff)
3689 min_offset_diff = diff;
NeilBrown56a25592011-12-23 10:17:55 +11003690
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003691 if (mddev->gendisk)
3692 disk_stack_limits(mddev->gendisk, rdev->bdev,
3693 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694
3695 disk->head_position = 0;
Shaohua Li532a2a32012-10-11 13:30:52 +11003696
3697 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
3698 discard_supported = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10003700
Jonathan Brassowed30be02012-10-31 11:42:30 +11003701 if (mddev->queue) {
3702 if (discard_supported)
3703 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
3704 mddev->queue);
3705 else
3706 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
3707 mddev->queue);
3708 }
NeilBrown6d508242005-09-09 16:24:03 -07003709 /* need to check that every block has at least one working mirror */
NeilBrown700c7212011-07-27 11:00:36 +10003710 if (!enough(conf, -1)) {
NeilBrown128595e2010-05-03 14:47:14 +10003711 printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n",
NeilBrown6d508242005-09-09 16:24:03 -07003712 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 goto out_free_conf;
3714 }
3715
NeilBrown3ea7daa2012-05-22 13:53:47 +10003716 if (conf->reshape_progress != MaxSector) {
3717 /* must ensure that shape change is supported */
3718 if (conf->geo.far_copies != 1 &&
3719 conf->geo.far_offset == 0)
3720 goto out_free_conf;
3721 if (conf->prev.far_copies != 1 &&
NeilBrown78eaa0d2013-07-02 15:58:05 +10003722 conf->prev.far_offset == 0)
NeilBrown3ea7daa2012-05-22 13:53:47 +10003723 goto out_free_conf;
3724 }
3725
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726 mddev->degraded = 0;
NeilBrownf8c9e742012-05-21 09:28:33 +10003727 for (i = 0;
3728 i < conf->geo.raid_disks
3729 || i < conf->prev.raid_disks;
3730 i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731
3732 disk = conf->mirrors + i;
3733
NeilBrown56a25592011-12-23 10:17:55 +11003734 if (!disk->rdev && disk->replacement) {
3735 /* The replacement is all we have - use it */
3736 disk->rdev = disk->replacement;
3737 disk->replacement = NULL;
3738 clear_bit(Replacement, &disk->rdev->flags);
3739 }
3740
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003741 if (!disk->rdev ||
NeilBrown2e333e82006-10-21 10:24:07 -07003742 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 disk->head_position = 0;
3744 mddev->degraded++;
Neil Brown8c2e8702008-06-28 08:30:52 +10003745 if (disk->rdev)
3746 conf->fullsync = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 }
NeilBrownd890fa22011-10-26 11:54:39 +11003748 disk->recovery_disabled = mddev->recovery_disabled - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 }
3750
Andre Noll8c6ac862009-06-18 08:48:06 +10003751 if (mddev->recovery_cp != MaxSector)
NeilBrown128595e2010-05-03 14:47:14 +10003752 printk(KERN_NOTICE "md/raid10:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10003753 " -- starting background reconstruction\n",
3754 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 printk(KERN_INFO
NeilBrown128595e2010-05-03 14:47:14 +10003756 "md/raid10:%s: active with %d out of %d devices\n",
NeilBrown5cf00fc2012-05-21 09:28:20 +10003757 mdname(mddev), conf->geo.raid_disks - mddev->degraded,
3758 conf->geo.raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 /*
3760 * Ok, everything is just fine now
3761 */
Trela, Maciejdab8b292010-03-08 16:02:45 +11003762 mddev->dev_sectors = conf->dev_sectors;
3763 size = raid10_size(mddev, 0, 0);
3764 md_set_array_sectors(mddev, size);
3765 mddev->resync_max_sectors = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003767 if (mddev->queue) {
NeilBrown5cf00fc2012-05-21 09:28:20 +10003768 int stripe = conf->geo.raid_disks *
Andre Noll9d8f0362009-06-18 08:45:01 +10003769 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003770 mddev->queue->backing_dev_info.congested_fn = raid10_congested;
3771 mddev->queue->backing_dev_info.congested_data = mddev;
3772
3773 /* Calculate max read-ahead size.
3774 * We need to readahead at least twice a whole stripe....
3775 * maybe...
3776 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003777 stripe /= conf->geo.near_copies;
NeilBrown3ea7daa2012-05-22 13:53:47 +10003778 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
3779 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003780 blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 }
3782
Martin K. Petersena91a2782011-03-17 11:11:05 +01003783
3784 if (md_integrity_register(mddev))
3785 goto out_free_conf;
3786
NeilBrown3ea7daa2012-05-22 13:53:47 +10003787 if (conf->reshape_progress != MaxSector) {
3788 unsigned long before_length, after_length;
3789
3790 before_length = ((1 << conf->prev.chunk_shift) *
3791 conf->prev.far_copies);
3792 after_length = ((1 << conf->geo.chunk_shift) *
3793 conf->geo.far_copies);
3794
3795 if (max(before_length, after_length) > min_offset_diff) {
3796 /* This cannot work */
3797 printk("md/raid10: offset difference not enough to continue reshape\n");
3798 goto out_free_conf;
3799 }
3800 conf->offset_diff = min_offset_diff;
3801
3802 conf->reshape_safe = conf->reshape_progress;
3803 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3804 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3805 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
3806 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3807 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
3808 "reshape");
3809 }
3810
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 return 0;
3812
3813out_free_conf:
NeilBrown01f96c02011-09-21 15:30:20 +10003814 md_unregister_thread(&mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 if (conf->r10bio_pool)
3816 mempool_destroy(conf->r10bio_pool);
NeilBrown1345b1d2006-01-06 00:20:40 -08003817 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003818 kfree(conf->mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 kfree(conf);
3820 mddev->private = NULL;
3821out:
3822 return -EIO;
3823}
3824
NeilBrownfd01b882011-10-11 16:47:53 +11003825static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826{
NeilBrowne879a872011-10-11 16:49:02 +11003827 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828
NeilBrown409c57f2009-03-31 14:39:39 +11003829 raise_barrier(conf, 0);
3830 lower_barrier(conf);
3831
NeilBrown01f96c02011-09-21 15:30:20 +10003832 md_unregister_thread(&mddev->thread);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003833 if (mddev->queue)
3834 /* the unplug fn references 'conf'*/
3835 blk_sync_queue(mddev->queue);
3836
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837 if (conf->r10bio_pool)
3838 mempool_destroy(conf->r10bio_pool);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10003839 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003840 kfree(conf->mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 kfree(conf);
3842 mddev->private = NULL;
3843 return 0;
3844}
3845
NeilBrownfd01b882011-10-11 16:47:53 +11003846static void raid10_quiesce(struct mddev *mddev, int state)
NeilBrown6cce3b22006-01-06 00:20:16 -08003847{
NeilBrowne879a872011-10-11 16:49:02 +11003848 struct r10conf *conf = mddev->private;
NeilBrown6cce3b22006-01-06 00:20:16 -08003849
3850 switch(state) {
3851 case 1:
3852 raise_barrier(conf, 0);
3853 break;
3854 case 0:
3855 lower_barrier(conf);
3856 break;
3857 }
NeilBrown6cce3b22006-01-06 00:20:16 -08003858}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859
NeilBrown006a09a2012-03-19 12:46:40 +11003860static int raid10_resize(struct mddev *mddev, sector_t sectors)
3861{
3862 /* Resize of 'far' arrays is not supported.
3863 * For 'near' and 'offset' arrays we can set the
3864 * number of sectors used to be an appropriate multiple
3865 * of the chunk size.
3866 * For 'offset', this is far_copies*chunksize.
3867 * For 'near' the multiplier is the LCM of
3868 * near_copies and raid_disks.
3869 * So if far_copies > 1 && !far_offset, fail.
3870 * Else find LCM(raid_disks, near_copy)*far_copies and
3871 * multiply by chunk_size. Then round to this number.
3872 * This is mostly done by raid10_size()
3873 */
3874 struct r10conf *conf = mddev->private;
3875 sector_t oldsize, size;
3876
NeilBrownf8c9e742012-05-21 09:28:33 +10003877 if (mddev->reshape_position != MaxSector)
3878 return -EBUSY;
3879
NeilBrown5cf00fc2012-05-21 09:28:20 +10003880 if (conf->geo.far_copies > 1 && !conf->geo.far_offset)
NeilBrown006a09a2012-03-19 12:46:40 +11003881 return -EINVAL;
3882
3883 oldsize = raid10_size(mddev, 0, 0);
3884 size = raid10_size(mddev, sectors, 0);
NeilBrowna4a61252012-05-22 13:55:27 +10003885 if (mddev->external_size &&
3886 mddev->array_sectors > size)
NeilBrown006a09a2012-03-19 12:46:40 +11003887 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10003888 if (mddev->bitmap) {
3889 int ret = bitmap_resize(mddev->bitmap, size, 0, 0);
3890 if (ret)
3891 return ret;
3892 }
3893 md_set_array_sectors(mddev, size);
NeilBrown006a09a2012-03-19 12:46:40 +11003894 set_capacity(mddev->gendisk, mddev->array_sectors);
3895 revalidate_disk(mddev->gendisk);
3896 if (sectors > mddev->dev_sectors &&
3897 mddev->recovery_cp > oldsize) {
3898 mddev->recovery_cp = oldsize;
3899 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3900 }
NeilBrown6508fdb2012-05-17 10:08:45 +10003901 calc_sectors(conf, sectors);
3902 mddev->dev_sectors = conf->dev_sectors;
NeilBrown006a09a2012-03-19 12:46:40 +11003903 mddev->resync_max_sectors = size;
3904 return 0;
3905}
3906
NeilBrownfd01b882011-10-11 16:47:53 +11003907static void *raid10_takeover_raid0(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003908{
NeilBrown3cb03002011-10-11 16:45:26 +11003909 struct md_rdev *rdev;
NeilBrowne879a872011-10-11 16:49:02 +11003910 struct r10conf *conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003911
3912 if (mddev->degraded > 0) {
NeilBrown128595e2010-05-03 14:47:14 +10003913 printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n",
3914 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003915 return ERR_PTR(-EINVAL);
3916 }
3917
Trela, Maciejdab8b292010-03-08 16:02:45 +11003918 /* Set new parameters */
3919 mddev->new_level = 10;
3920 /* new layout: far_copies = 1, near_copies = 2 */
3921 mddev->new_layout = (1<<8) + 2;
3922 mddev->new_chunk_sectors = mddev->chunk_sectors;
3923 mddev->delta_disks = mddev->raid_disks;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003924 mddev->raid_disks *= 2;
3925 /* make sure it will be not marked as dirty */
3926 mddev->recovery_cp = MaxSector;
3927
3928 conf = setup_conf(mddev);
Krzysztof Wojcik02214dc2011-02-04 14:18:26 +01003929 if (!IS_ERR(conf)) {
NeilBrowndafb20f2012-03-19 12:46:39 +11003930 rdev_for_each(rdev, mddev)
NeilBrowne93f68a2010-06-15 09:36:03 +01003931 if (rdev->raid_disk >= 0)
3932 rdev->new_raid_disk = rdev->raid_disk * 2;
Krzysztof Wojcik02214dc2011-02-04 14:18:26 +01003933 conf->barrier = 1;
3934 }
3935
Trela, Maciejdab8b292010-03-08 16:02:45 +11003936 return conf;
3937}
3938
NeilBrownfd01b882011-10-11 16:47:53 +11003939static void *raid10_takeover(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003940{
NeilBrowne373ab12011-10-11 16:48:59 +11003941 struct r0conf *raid0_conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003942
3943 /* raid10 can take over:
3944 * raid0 - providing it has only two drives
3945 */
3946 if (mddev->level == 0) {
3947 /* for raid0 takeover only one zone is supported */
NeilBrowne373ab12011-10-11 16:48:59 +11003948 raid0_conf = mddev->private;
3949 if (raid0_conf->nr_strip_zones > 1) {
NeilBrown128595e2010-05-03 14:47:14 +10003950 printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0"
3951 " with more than one zone.\n",
3952 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003953 return ERR_PTR(-EINVAL);
3954 }
3955 return raid10_takeover_raid0(mddev);
3956 }
3957 return ERR_PTR(-EINVAL);
3958}
3959
NeilBrown3ea7daa2012-05-22 13:53:47 +10003960static int raid10_check_reshape(struct mddev *mddev)
3961{
3962 /* Called when there is a request to change
3963 * - layout (to ->new_layout)
3964 * - chunk size (to ->new_chunk_sectors)
3965 * - raid_disks (by delta_disks)
3966 * or when trying to restart a reshape that was ongoing.
3967 *
3968 * We need to validate the request and possibly allocate
3969 * space if that might be an issue later.
3970 *
3971 * Currently we reject any reshape of a 'far' mode array,
3972 * allow chunk size to change if new is generally acceptable,
3973 * allow raid_disks to increase, and allow
3974 * a switch between 'near' mode and 'offset' mode.
3975 */
3976 struct r10conf *conf = mddev->private;
3977 struct geom geo;
3978
3979 if (conf->geo.far_copies != 1 && !conf->geo.far_offset)
3980 return -EINVAL;
3981
3982 if (setup_geo(&geo, mddev, geo_start) != conf->copies)
3983 /* mustn't change number of copies */
3984 return -EINVAL;
3985 if (geo.far_copies > 1 && !geo.far_offset)
3986 /* Cannot switch to 'far' mode */
3987 return -EINVAL;
3988
3989 if (mddev->array_sectors & geo.chunk_mask)
3990 /* not factor of array size */
3991 return -EINVAL;
3992
NeilBrown3ea7daa2012-05-22 13:53:47 +10003993 if (!enough(conf, -1))
3994 return -EINVAL;
3995
3996 kfree(conf->mirrors_new);
3997 conf->mirrors_new = NULL;
3998 if (mddev->delta_disks > 0) {
3999 /* allocate new 'mirrors' list */
4000 conf->mirrors_new = kzalloc(
Jonathan Brassowdc280d92012-07-31 10:03:52 +10004001 sizeof(struct raid10_info)
NeilBrown3ea7daa2012-05-22 13:53:47 +10004002 *(mddev->raid_disks +
4003 mddev->delta_disks),
4004 GFP_KERNEL);
4005 if (!conf->mirrors_new)
4006 return -ENOMEM;
4007 }
4008 return 0;
4009}
4010
4011/*
4012 * Need to check if array has failed when deciding whether to:
4013 * - start an array
4014 * - remove non-faulty devices
4015 * - add a spare
4016 * - allow a reshape
4017 * This determination is simple when no reshape is happening.
4018 * However if there is a reshape, we need to carefully check
4019 * both the before and after sections.
4020 * This is because some failed devices may only affect one
4021 * of the two sections, and some non-in_sync devices may
4022 * be insync in the section most affected by failed devices.
4023 */
4024static int calc_degraded(struct r10conf *conf)
4025{
4026 int degraded, degraded2;
4027 int i;
4028
4029 rcu_read_lock();
4030 degraded = 0;
4031 /* 'prev' section first */
4032 for (i = 0; i < conf->prev.raid_disks; i++) {
4033 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
4034 if (!rdev || test_bit(Faulty, &rdev->flags))
4035 degraded++;
4036 else if (!test_bit(In_sync, &rdev->flags))
4037 /* When we can reduce the number of devices in
4038 * an array, this might not contribute to
4039 * 'degraded'. It does now.
4040 */
4041 degraded++;
4042 }
4043 rcu_read_unlock();
4044 if (conf->geo.raid_disks == conf->prev.raid_disks)
4045 return degraded;
4046 rcu_read_lock();
4047 degraded2 = 0;
4048 for (i = 0; i < conf->geo.raid_disks; i++) {
4049 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
4050 if (!rdev || test_bit(Faulty, &rdev->flags))
4051 degraded2++;
4052 else if (!test_bit(In_sync, &rdev->flags)) {
4053 /* If reshape is increasing the number of devices,
4054 * this section has already been recovered, so
4055 * it doesn't contribute to degraded.
4056 * else it does.
4057 */
4058 if (conf->geo.raid_disks <= conf->prev.raid_disks)
4059 degraded2++;
4060 }
4061 }
4062 rcu_read_unlock();
4063 if (degraded2 > degraded)
4064 return degraded2;
4065 return degraded;
4066}
4067
4068static int raid10_start_reshape(struct mddev *mddev)
4069{
4070 /* A 'reshape' has been requested. This commits
4071 * the various 'new' fields and sets MD_RECOVER_RESHAPE
4072 * This also checks if there are enough spares and adds them
4073 * to the array.
4074 * We currently require enough spares to make the final
4075 * array non-degraded. We also require that the difference
4076 * between old and new data_offset - on each device - is
4077 * enough that we never risk over-writing.
4078 */
4079
4080 unsigned long before_length, after_length;
4081 sector_t min_offset_diff = 0;
4082 int first = 1;
4083 struct geom new;
4084 struct r10conf *conf = mddev->private;
4085 struct md_rdev *rdev;
4086 int spares = 0;
NeilBrownbb63a702012-05-22 13:55:28 +10004087 int ret;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004088
4089 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4090 return -EBUSY;
4091
4092 if (setup_geo(&new, mddev, geo_start) != conf->copies)
4093 return -EINVAL;
4094
4095 before_length = ((1 << conf->prev.chunk_shift) *
4096 conf->prev.far_copies);
4097 after_length = ((1 << conf->geo.chunk_shift) *
4098 conf->geo.far_copies);
4099
4100 rdev_for_each(rdev, mddev) {
4101 if (!test_bit(In_sync, &rdev->flags)
4102 && !test_bit(Faulty, &rdev->flags))
4103 spares++;
4104 if (rdev->raid_disk >= 0) {
4105 long long diff = (rdev->new_data_offset
4106 - rdev->data_offset);
4107 if (!mddev->reshape_backwards)
4108 diff = -diff;
4109 if (diff < 0)
4110 diff = 0;
4111 if (first || diff < min_offset_diff)
4112 min_offset_diff = diff;
4113 }
4114 }
4115
4116 if (max(before_length, after_length) > min_offset_diff)
4117 return -EINVAL;
4118
4119 if (spares < mddev->delta_disks)
4120 return -EINVAL;
4121
4122 conf->offset_diff = min_offset_diff;
4123 spin_lock_irq(&conf->device_lock);
4124 if (conf->mirrors_new) {
4125 memcpy(conf->mirrors_new, conf->mirrors,
Jonathan Brassowdc280d92012-07-31 10:03:52 +10004126 sizeof(struct raid10_info)*conf->prev.raid_disks);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004127 smp_mb();
4128 kfree(conf->mirrors_old); /* FIXME and elsewhere */
4129 conf->mirrors_old = conf->mirrors;
4130 conf->mirrors = conf->mirrors_new;
4131 conf->mirrors_new = NULL;
4132 }
4133 setup_geo(&conf->geo, mddev, geo_start);
4134 smp_mb();
4135 if (mddev->reshape_backwards) {
4136 sector_t size = raid10_size(mddev, 0, 0);
4137 if (size < mddev->array_sectors) {
4138 spin_unlock_irq(&conf->device_lock);
4139 printk(KERN_ERR "md/raid10:%s: array size must be reduce before number of disks\n",
4140 mdname(mddev));
4141 return -EINVAL;
4142 }
4143 mddev->resync_max_sectors = size;
4144 conf->reshape_progress = size;
4145 } else
4146 conf->reshape_progress = 0;
4147 spin_unlock_irq(&conf->device_lock);
4148
NeilBrownbb63a702012-05-22 13:55:28 +10004149 if (mddev->delta_disks && mddev->bitmap) {
4150 ret = bitmap_resize(mddev->bitmap,
4151 raid10_size(mddev, 0,
4152 conf->geo.raid_disks),
4153 0, 0);
4154 if (ret)
4155 goto abort;
4156 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004157 if (mddev->delta_disks > 0) {
4158 rdev_for_each(rdev, mddev)
4159 if (rdev->raid_disk < 0 &&
4160 !test_bit(Faulty, &rdev->flags)) {
4161 if (raid10_add_disk(mddev, rdev) == 0) {
4162 if (rdev->raid_disk >=
4163 conf->prev.raid_disks)
4164 set_bit(In_sync, &rdev->flags);
4165 else
4166 rdev->recovery_offset = 0;
4167
4168 if (sysfs_link_rdev(mddev, rdev))
4169 /* Failure here is OK */;
4170 }
4171 } else if (rdev->raid_disk >= conf->prev.raid_disks
4172 && !test_bit(Faulty, &rdev->flags)) {
4173 /* This is a spare that was manually added */
4174 set_bit(In_sync, &rdev->flags);
4175 }
4176 }
4177 /* When a reshape changes the number of devices,
4178 * ->degraded is measured against the larger of the
4179 * pre and post numbers.
4180 */
4181 spin_lock_irq(&conf->device_lock);
4182 mddev->degraded = calc_degraded(conf);
4183 spin_unlock_irq(&conf->device_lock);
4184 mddev->raid_disks = conf->geo.raid_disks;
4185 mddev->reshape_position = conf->reshape_progress;
4186 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4187
4188 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4189 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4190 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4191 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4192
4193 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
4194 "reshape");
4195 if (!mddev->sync_thread) {
NeilBrownbb63a702012-05-22 13:55:28 +10004196 ret = -EAGAIN;
4197 goto abort;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004198 }
4199 conf->reshape_checkpoint = jiffies;
4200 md_wakeup_thread(mddev->sync_thread);
4201 md_new_event(mddev);
4202 return 0;
NeilBrownbb63a702012-05-22 13:55:28 +10004203
4204abort:
4205 mddev->recovery = 0;
4206 spin_lock_irq(&conf->device_lock);
4207 conf->geo = conf->prev;
4208 mddev->raid_disks = conf->geo.raid_disks;
4209 rdev_for_each(rdev, mddev)
4210 rdev->new_data_offset = rdev->data_offset;
4211 smp_wmb();
4212 conf->reshape_progress = MaxSector;
4213 mddev->reshape_position = MaxSector;
4214 spin_unlock_irq(&conf->device_lock);
4215 return ret;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004216}
4217
4218/* Calculate the last device-address that could contain
4219 * any block from the chunk that includes the array-address 's'
4220 * and report the next address.
4221 * i.e. the address returned will be chunk-aligned and after
4222 * any data that is in the chunk containing 's'.
4223 */
4224static sector_t last_dev_address(sector_t s, struct geom *geo)
4225{
4226 s = (s | geo->chunk_mask) + 1;
4227 s >>= geo->chunk_shift;
4228 s *= geo->near_copies;
4229 s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks);
4230 s *= geo->far_copies;
4231 s <<= geo->chunk_shift;
4232 return s;
4233}
4234
4235/* Calculate the first device-address that could contain
4236 * any block from the chunk that includes the array-address 's'.
4237 * This too will be the start of a chunk
4238 */
4239static sector_t first_dev_address(sector_t s, struct geom *geo)
4240{
4241 s >>= geo->chunk_shift;
4242 s *= geo->near_copies;
4243 sector_div(s, geo->raid_disks);
4244 s *= geo->far_copies;
4245 s <<= geo->chunk_shift;
4246 return s;
4247}
4248
4249static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
4250 int *skipped)
4251{
4252 /* We simply copy at most one chunk (smallest of old and new)
4253 * at a time, possibly less if that exceeds RESYNC_PAGES,
4254 * or we hit a bad block or something.
4255 * This might mean we pause for normal IO in the middle of
4256 * a chunk, but that is not a problem was mddev->reshape_position
4257 * can record any location.
4258 *
4259 * If we will want to write to a location that isn't
4260 * yet recorded as 'safe' (i.e. in metadata on disk) then
4261 * we need to flush all reshape requests and update the metadata.
4262 *
4263 * When reshaping forwards (e.g. to more devices), we interpret
4264 * 'safe' as the earliest block which might not have been copied
4265 * down yet. We divide this by previous stripe size and multiply
4266 * by previous stripe length to get lowest device offset that we
4267 * cannot write to yet.
4268 * We interpret 'sector_nr' as an address that we want to write to.
4269 * From this we use last_device_address() to find where we might
4270 * write to, and first_device_address on the 'safe' position.
4271 * If this 'next' write position is after the 'safe' position,
4272 * we must update the metadata to increase the 'safe' position.
4273 *
4274 * When reshaping backwards, we round in the opposite direction
4275 * and perform the reverse test: next write position must not be
4276 * less than current safe position.
4277 *
4278 * In all this the minimum difference in data offsets
4279 * (conf->offset_diff - always positive) allows a bit of slack,
4280 * so next can be after 'safe', but not by more than offset_disk
4281 *
4282 * We need to prepare all the bios here before we start any IO
4283 * to ensure the size we choose is acceptable to all devices.
4284 * The means one for each copy for write-out and an extra one for
4285 * read-in.
4286 * We store the read-in bio in ->master_bio and the others in
4287 * ->devs[x].bio and ->devs[x].repl_bio.
4288 */
4289 struct r10conf *conf = mddev->private;
4290 struct r10bio *r10_bio;
4291 sector_t next, safe, last;
4292 int max_sectors;
4293 int nr_sectors;
4294 int s;
4295 struct md_rdev *rdev;
4296 int need_flush = 0;
4297 struct bio *blist;
4298 struct bio *bio, *read_bio;
4299 int sectors_done = 0;
4300
4301 if (sector_nr == 0) {
4302 /* If restarting in the middle, skip the initial sectors */
4303 if (mddev->reshape_backwards &&
4304 conf->reshape_progress < raid10_size(mddev, 0, 0)) {
4305 sector_nr = (raid10_size(mddev, 0, 0)
4306 - conf->reshape_progress);
4307 } else if (!mddev->reshape_backwards &&
4308 conf->reshape_progress > 0)
4309 sector_nr = conf->reshape_progress;
4310 if (sector_nr) {
4311 mddev->curr_resync_completed = sector_nr;
4312 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
4313 *skipped = 1;
4314 return sector_nr;
4315 }
4316 }
4317
4318 /* We don't use sector_nr to track where we are up to
4319 * as that doesn't work well for ->reshape_backwards.
4320 * So just use ->reshape_progress.
4321 */
4322 if (mddev->reshape_backwards) {
4323 /* 'next' is the earliest device address that we might
4324 * write to for this chunk in the new layout
4325 */
4326 next = first_dev_address(conf->reshape_progress - 1,
4327 &conf->geo);
4328
4329 /* 'safe' is the last device address that we might read from
4330 * in the old layout after a restart
4331 */
4332 safe = last_dev_address(conf->reshape_safe - 1,
4333 &conf->prev);
4334
4335 if (next + conf->offset_diff < safe)
4336 need_flush = 1;
4337
4338 last = conf->reshape_progress - 1;
4339 sector_nr = last & ~(sector_t)(conf->geo.chunk_mask
4340 & conf->prev.chunk_mask);
4341 if (sector_nr + RESYNC_BLOCK_SIZE/512 < last)
4342 sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512;
4343 } else {
4344 /* 'next' is after the last device address that we
4345 * might write to for this chunk in the new layout
4346 */
4347 next = last_dev_address(conf->reshape_progress, &conf->geo);
4348
4349 /* 'safe' is the earliest device address that we might
4350 * read from in the old layout after a restart
4351 */
4352 safe = first_dev_address(conf->reshape_safe, &conf->prev);
4353
4354 /* Need to update metadata if 'next' might be beyond 'safe'
4355 * as that would possibly corrupt data
4356 */
4357 if (next > safe + conf->offset_diff)
4358 need_flush = 1;
4359
4360 sector_nr = conf->reshape_progress;
4361 last = sector_nr | (conf->geo.chunk_mask
4362 & conf->prev.chunk_mask);
4363
4364 if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last)
4365 last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1;
4366 }
4367
4368 if (need_flush ||
4369 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
4370 /* Need to update reshape_position in metadata */
4371 wait_barrier(conf);
4372 mddev->reshape_position = conf->reshape_progress;
4373 if (mddev->reshape_backwards)
4374 mddev->curr_resync_completed = raid10_size(mddev, 0, 0)
4375 - conf->reshape_progress;
4376 else
4377 mddev->curr_resync_completed = conf->reshape_progress;
4378 conf->reshape_checkpoint = jiffies;
4379 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4380 md_wakeup_thread(mddev->thread);
4381 wait_event(mddev->sb_wait, mddev->flags == 0 ||
4382 kthread_should_stop());
4383 conf->reshape_safe = mddev->reshape_position;
4384 allow_barrier(conf);
4385 }
4386
4387read_more:
4388 /* Now schedule reads for blocks from sector_nr to last */
4389 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
4390 raise_barrier(conf, sectors_done != 0);
4391 atomic_set(&r10_bio->remaining, 0);
4392 r10_bio->mddev = mddev;
4393 r10_bio->sector = sector_nr;
4394 set_bit(R10BIO_IsReshape, &r10_bio->state);
4395 r10_bio->sectors = last - sector_nr + 1;
4396 rdev = read_balance(conf, r10_bio, &max_sectors);
4397 BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state));
4398
4399 if (!rdev) {
4400 /* Cannot read from here, so need to record bad blocks
4401 * on all the target devices.
4402 */
4403 // FIXME
4404 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4405 return sectors_done;
4406 }
4407
4408 read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev);
4409
4410 read_bio->bi_bdev = rdev->bdev;
4411 read_bio->bi_sector = (r10_bio->devs[r10_bio->read_slot].addr
4412 + rdev->data_offset);
4413 read_bio->bi_private = r10_bio;
4414 read_bio->bi_end_io = end_sync_read;
4415 read_bio->bi_rw = READ;
4416 read_bio->bi_flags &= ~(BIO_POOL_MASK - 1);
4417 read_bio->bi_flags |= 1 << BIO_UPTODATE;
4418 read_bio->bi_vcnt = 0;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004419 read_bio->bi_size = 0;
4420 r10_bio->master_bio = read_bio;
4421 r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum;
4422
4423 /* Now find the locations in the new layout */
4424 __raid10_find_phys(&conf->geo, r10_bio);
4425
4426 blist = read_bio;
4427 read_bio->bi_next = NULL;
4428
4429 for (s = 0; s < conf->copies*2; s++) {
4430 struct bio *b;
4431 int d = r10_bio->devs[s/2].devnum;
4432 struct md_rdev *rdev2;
4433 if (s&1) {
4434 rdev2 = conf->mirrors[d].replacement;
4435 b = r10_bio->devs[s/2].repl_bio;
4436 } else {
4437 rdev2 = conf->mirrors[d].rdev;
4438 b = r10_bio->devs[s/2].bio;
4439 }
4440 if (!rdev2 || test_bit(Faulty, &rdev2->flags))
4441 continue;
Kent Overstreet8be185f2012-09-06 14:14:43 -07004442
4443 bio_reset(b);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004444 b->bi_bdev = rdev2->bdev;
4445 b->bi_sector = r10_bio->devs[s/2].addr + rdev2->new_data_offset;
4446 b->bi_private = r10_bio;
4447 b->bi_end_io = end_reshape_write;
4448 b->bi_rw = WRITE;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004449 b->bi_next = blist;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004450 blist = b;
4451 }
4452
4453 /* Now add as many pages as possible to all of these bios. */
4454
4455 nr_sectors = 0;
4456 for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) {
4457 struct page *page = r10_bio->devs[0].bio->bi_io_vec[s/(PAGE_SIZE>>9)].bv_page;
4458 int len = (max_sectors - s) << 9;
4459 if (len > PAGE_SIZE)
4460 len = PAGE_SIZE;
4461 for (bio = blist; bio ; bio = bio->bi_next) {
4462 struct bio *bio2;
4463 if (bio_add_page(bio, page, len, 0))
4464 continue;
4465
4466 /* Didn't fit, must stop */
4467 for (bio2 = blist;
4468 bio2 && bio2 != bio;
4469 bio2 = bio2->bi_next) {
4470 /* Remove last page from this bio */
4471 bio2->bi_vcnt--;
4472 bio2->bi_size -= len;
4473 bio2->bi_flags &= ~(1<<BIO_SEG_VALID);
4474 }
4475 goto bio_full;
4476 }
4477 sector_nr += len >> 9;
4478 nr_sectors += len >> 9;
4479 }
4480bio_full:
4481 r10_bio->sectors = nr_sectors;
4482
4483 /* Now submit the read */
4484 md_sync_acct(read_bio->bi_bdev, r10_bio->sectors);
4485 atomic_inc(&r10_bio->remaining);
4486 read_bio->bi_next = NULL;
4487 generic_make_request(read_bio);
4488 sector_nr += nr_sectors;
4489 sectors_done += nr_sectors;
4490 if (sector_nr <= last)
4491 goto read_more;
4492
4493 /* Now that we have done the whole section we can
4494 * update reshape_progress
4495 */
4496 if (mddev->reshape_backwards)
4497 conf->reshape_progress -= sectors_done;
4498 else
4499 conf->reshape_progress += sectors_done;
4500
4501 return sectors_done;
4502}
4503
4504static void end_reshape_request(struct r10bio *r10_bio);
4505static int handle_reshape_read_error(struct mddev *mddev,
4506 struct r10bio *r10_bio);
4507static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio)
4508{
4509 /* Reshape read completed. Hopefully we have a block
4510 * to write out.
4511 * If we got a read error then we do sync 1-page reads from
4512 * elsewhere until we find the data - or give up.
4513 */
4514 struct r10conf *conf = mddev->private;
4515 int s;
4516
4517 if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
4518 if (handle_reshape_read_error(mddev, r10_bio) < 0) {
4519 /* Reshape has been aborted */
4520 md_done_sync(mddev, r10_bio->sectors, 0);
4521 return;
4522 }
4523
4524 /* We definitely have the data in the pages, schedule the
4525 * writes.
4526 */
4527 atomic_set(&r10_bio->remaining, 1);
4528 for (s = 0; s < conf->copies*2; s++) {
4529 struct bio *b;
4530 int d = r10_bio->devs[s/2].devnum;
4531 struct md_rdev *rdev;
4532 if (s&1) {
4533 rdev = conf->mirrors[d].replacement;
4534 b = r10_bio->devs[s/2].repl_bio;
4535 } else {
4536 rdev = conf->mirrors[d].rdev;
4537 b = r10_bio->devs[s/2].bio;
4538 }
4539 if (!rdev || test_bit(Faulty, &rdev->flags))
4540 continue;
4541 atomic_inc(&rdev->nr_pending);
4542 md_sync_acct(b->bi_bdev, r10_bio->sectors);
4543 atomic_inc(&r10_bio->remaining);
4544 b->bi_next = NULL;
4545 generic_make_request(b);
4546 }
4547 end_reshape_request(r10_bio);
4548}
4549
4550static void end_reshape(struct r10conf *conf)
4551{
4552 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery))
4553 return;
4554
4555 spin_lock_irq(&conf->device_lock);
4556 conf->prev = conf->geo;
4557 md_finish_reshape(conf->mddev);
4558 smp_wmb();
4559 conf->reshape_progress = MaxSector;
4560 spin_unlock_irq(&conf->device_lock);
4561
4562 /* read-ahead size must cover two whole stripes, which is
4563 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4564 */
4565 if (conf->mddev->queue) {
4566 int stripe = conf->geo.raid_disks *
4567 ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE);
4568 stripe /= conf->geo.near_copies;
4569 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
4570 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
4571 }
4572 conf->fullsync = 0;
4573}
4574
4575
4576static int handle_reshape_read_error(struct mddev *mddev,
4577 struct r10bio *r10_bio)
4578{
4579 /* Use sync reads to get the blocks from somewhere else */
4580 int sectors = r10_bio->sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004581 struct r10conf *conf = mddev->private;
NeilBrowne0ee7782012-08-18 09:51:42 +10004582 struct {
4583 struct r10bio r10_bio;
4584 struct r10dev devs[conf->copies];
4585 } on_stack;
4586 struct r10bio *r10b = &on_stack.r10_bio;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004587 int slot = 0;
4588 int idx = 0;
4589 struct bio_vec *bvec = r10_bio->master_bio->bi_io_vec;
4590
NeilBrowne0ee7782012-08-18 09:51:42 +10004591 r10b->sector = r10_bio->sector;
4592 __raid10_find_phys(&conf->prev, r10b);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004593
4594 while (sectors) {
4595 int s = sectors;
4596 int success = 0;
4597 int first_slot = slot;
4598
4599 if (s > (PAGE_SIZE >> 9))
4600 s = PAGE_SIZE >> 9;
4601
4602 while (!success) {
NeilBrowne0ee7782012-08-18 09:51:42 +10004603 int d = r10b->devs[slot].devnum;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004604 struct md_rdev *rdev = conf->mirrors[d].rdev;
4605 sector_t addr;
4606 if (rdev == NULL ||
4607 test_bit(Faulty, &rdev->flags) ||
4608 !test_bit(In_sync, &rdev->flags))
4609 goto failed;
4610
NeilBrowne0ee7782012-08-18 09:51:42 +10004611 addr = r10b->devs[slot].addr + idx * PAGE_SIZE;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004612 success = sync_page_io(rdev,
4613 addr,
4614 s << 9,
4615 bvec[idx].bv_page,
4616 READ, false);
4617 if (success)
4618 break;
4619 failed:
4620 slot++;
4621 if (slot >= conf->copies)
4622 slot = 0;
4623 if (slot == first_slot)
4624 break;
4625 }
4626 if (!success) {
4627 /* couldn't read this block, must give up */
4628 set_bit(MD_RECOVERY_INTR,
4629 &mddev->recovery);
4630 return -EIO;
4631 }
4632 sectors -= s;
4633 idx++;
4634 }
4635 return 0;
4636}
4637
4638static void end_reshape_write(struct bio *bio, int error)
4639{
4640 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
4641 struct r10bio *r10_bio = bio->bi_private;
4642 struct mddev *mddev = r10_bio->mddev;
4643 struct r10conf *conf = mddev->private;
4644 int d;
4645 int slot;
4646 int repl;
4647 struct md_rdev *rdev = NULL;
4648
4649 d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
4650 if (repl)
4651 rdev = conf->mirrors[d].replacement;
4652 if (!rdev) {
4653 smp_mb();
4654 rdev = conf->mirrors[d].rdev;
4655 }
4656
4657 if (!uptodate) {
4658 /* FIXME should record badblock */
4659 md_error(mddev, rdev);
4660 }
4661
4662 rdev_dec_pending(rdev, mddev);
4663 end_reshape_request(r10_bio);
4664}
4665
4666static void end_reshape_request(struct r10bio *r10_bio)
4667{
4668 if (!atomic_dec_and_test(&r10_bio->remaining))
4669 return;
4670 md_done_sync(r10_bio->mddev, r10_bio->sectors, 1);
4671 bio_put(r10_bio->master_bio);
4672 put_buf(r10_bio);
4673}
4674
4675static void raid10_finish_reshape(struct mddev *mddev)
4676{
4677 struct r10conf *conf = mddev->private;
4678
4679 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4680 return;
4681
4682 if (mddev->delta_disks > 0) {
4683 sector_t size = raid10_size(mddev, 0, 0);
4684 md_set_array_sectors(mddev, size);
4685 if (mddev->recovery_cp > mddev->resync_max_sectors) {
4686 mddev->recovery_cp = mddev->resync_max_sectors;
4687 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4688 }
4689 mddev->resync_max_sectors = size;
4690 set_capacity(mddev->gendisk, mddev->array_sectors);
4691 revalidate_disk(mddev->gendisk);
NeilBrown63aced62012-05-22 13:55:33 +10004692 } else {
4693 int d;
4694 for (d = conf->geo.raid_disks ;
4695 d < conf->geo.raid_disks - mddev->delta_disks;
4696 d++) {
4697 struct md_rdev *rdev = conf->mirrors[d].rdev;
4698 if (rdev)
4699 clear_bit(In_sync, &rdev->flags);
4700 rdev = conf->mirrors[d].replacement;
4701 if (rdev)
4702 clear_bit(In_sync, &rdev->flags);
4703 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004704 }
4705 mddev->layout = mddev->new_layout;
4706 mddev->chunk_sectors = 1 << conf->geo.chunk_shift;
4707 mddev->reshape_position = MaxSector;
4708 mddev->delta_disks = 0;
4709 mddev->reshape_backwards = 0;
4710}
4711
NeilBrown84fc4b52011-10-11 16:49:58 +11004712static struct md_personality raid10_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07004713{
4714 .name = "raid10",
NeilBrown2604b702006-01-06 00:20:36 -08004715 .level = 10,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716 .owner = THIS_MODULE,
4717 .make_request = make_request,
4718 .run = run,
4719 .stop = stop,
4720 .status = status,
4721 .error_handler = error,
4722 .hot_add_disk = raid10_add_disk,
4723 .hot_remove_disk= raid10_remove_disk,
4724 .spare_active = raid10_spare_active,
4725 .sync_request = sync_request,
NeilBrown6cce3b22006-01-06 00:20:16 -08004726 .quiesce = raid10_quiesce,
Dan Williams80c3a6c2009-03-17 18:10:40 -07004727 .size = raid10_size,
NeilBrown006a09a2012-03-19 12:46:40 +11004728 .resize = raid10_resize,
Trela, Maciejdab8b292010-03-08 16:02:45 +11004729 .takeover = raid10_takeover,
NeilBrown3ea7daa2012-05-22 13:53:47 +10004730 .check_reshape = raid10_check_reshape,
4731 .start_reshape = raid10_start_reshape,
4732 .finish_reshape = raid10_finish_reshape,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004733};
4734
4735static int __init raid_init(void)
4736{
NeilBrown2604b702006-01-06 00:20:36 -08004737 return register_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004738}
4739
4740static void raid_exit(void)
4741{
NeilBrown2604b702006-01-06 00:20:36 -08004742 unregister_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004743}
4744
4745module_init(raid_init);
4746module_exit(raid_exit);
4747MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11004748MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004749MODULE_ALIAS("md-personality-9"); /* RAID10 */
NeilBrownd9d166c2006-01-06 00:20:51 -08004750MODULE_ALIAS("md-raid10");
NeilBrown2604b702006-01-06 00:20:36 -08004751MODULE_ALIAS("md-level-10");
NeilBrown34db0cd2011-10-11 16:50:01 +11004752
4753module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);