blob: 6703751d87d7fd7314149cfd8ba5bef314927235 [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 */
NeilBrown6cce3b232006-01-06 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
NeilBrown6cce3b232006-01-06 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{
NeilBrown6cce3b232006-01-06 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
NeilBrown6cce3b232006-01-06 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
Kent Overstreet20d01892013-11-23 18:21:01 -08001155static 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;
Jens Axboea3623572005-11-01 09:26:16 +01001161 const int rw = bio_data_dir(bio);
NeilBrown2c7d46e2010-08-18 16:16:05 +10001162 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
Tejun Heoe9c74692010-09-03 11:56:18 +02001163 const unsigned long do_fua = (bio->bi_rw & REQ_FUA);
Shaohua Li532a2a32012-10-11 13:30:52 +11001164 const unsigned long do_discard = (bio->bi_rw
1165 & (REQ_DISCARD | REQ_SECURE));
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001166 const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME);
NeilBrown6cce3b232006-01-06 00:20:16 -08001167 unsigned long flags;
NeilBrown3cb03002011-10-11 16:45:26 +11001168 struct md_rdev *blocked_rdev;
NeilBrown57c67df2012-10-11 13:32:13 +11001169 struct blk_plug_cb *cb;
1170 struct raid10_plug_cb *plug = NULL;
NeilBrownd4432c22011-07-28 11:39:24 +10001171 int sectors_handled;
1172 int max_sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10001173 int sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
NeilBrowncc13b1d2014-05-05 13:34:37 +10001175 /*
1176 * Register the new request and wait if the reconstruction
1177 * thread has put up a bar for new requests.
1178 * Continue immediately if no resync is active currently.
1179 */
1180 wait_barrier(conf);
1181
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001182 sectors = bio_sectors(bio);
NeilBrown3ea7daa2012-05-22 13:53:47 +10001183 while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
Kent Overstreet4f024f32013-10-11 15:44:27 -07001184 bio->bi_iter.bi_sector < conf->reshape_progress &&
1185 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) {
NeilBrown3ea7daa2012-05-22 13:53:47 +10001186 /* IO spans the reshape position. Need to wait for
1187 * reshape to pass
1188 */
1189 allow_barrier(conf);
1190 wait_event(conf->wait_barrier,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001191 conf->reshape_progress <= bio->bi_iter.bi_sector ||
1192 conf->reshape_progress >= bio->bi_iter.bi_sector +
1193 sectors);
NeilBrown3ea7daa2012-05-22 13:53:47 +10001194 wait_barrier(conf);
1195 }
1196 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
1197 bio_data_dir(bio) == WRITE &&
1198 (mddev->reshape_backwards
Kent Overstreet4f024f32013-10-11 15:44:27 -07001199 ? (bio->bi_iter.bi_sector < conf->reshape_safe &&
1200 bio->bi_iter.bi_sector + sectors > conf->reshape_progress)
1201 : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe &&
1202 bio->bi_iter.bi_sector < conf->reshape_progress))) {
NeilBrown3ea7daa2012-05-22 13:53:47 +10001203 /* Need to update reshape_position in metadata */
1204 mddev->reshape_position = conf->reshape_progress;
1205 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1206 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1207 md_wakeup_thread(mddev->thread);
1208 wait_event(mddev->sb_wait,
1209 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1210
1211 conf->reshape_safe = mddev->reshape_position;
1212 }
1213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
1215
1216 r10_bio->master_bio = bio;
NeilBrown3ea7daa2012-05-22 13:53:47 +10001217 r10_bio->sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
1219 r10_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001220 r10_bio->sector = bio->bi_iter.bi_sector;
NeilBrown6cce3b232006-01-06 00:20:16 -08001221 r10_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
NeilBrown856e08e2011-07-28 11:39:23 +10001223 /* We might need to issue multiple reads to different
1224 * devices if there are bad blocks around, so we keep
1225 * track of the number of reads in bio->bi_phys_segments.
1226 * If this is 0, there is only one r10_bio and no locking
1227 * will be needed when the request completes. If it is
1228 * non-zero, then it is the number of not-completed requests.
1229 */
1230 bio->bi_phys_segments = 0;
1231 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1232
Jens Axboea3623572005-11-01 09:26:16 +01001233 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 /*
1235 * read balancing logic:
1236 */
NeilBrown96c3fd12011-12-23 10:17:54 +11001237 struct md_rdev *rdev;
NeilBrown856e08e2011-07-28 11:39:23 +10001238 int slot;
1239
1240read_again:
NeilBrown96c3fd12011-12-23 10:17:54 +11001241 rdev = read_balance(conf, r10_bio, &max_sectors);
1242 if (!rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 raid_end_bio_io(r10_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001244 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 }
NeilBrown96c3fd12011-12-23 10:17:54 +11001246 slot = r10_bio->read_slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
NeilBrowna167f662010-10-26 18:31:13 +11001248 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001249 bio_trim(read_bio, r10_bio->sector - bio->bi_iter.bi_sector,
Kent Overstreet6678d832013-08-07 11:14:32 -07001250 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
1252 r10_bio->devs[slot].bio = read_bio;
NeilBrownabbf0982011-12-23 10:17:54 +11001253 r10_bio->devs[slot].rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Kent Overstreet4f024f32013-10-11 15:44:27 -07001255 read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr +
NeilBrownf8c9e742012-05-21 09:28:33 +10001256 choose_data_offset(r10_bio, rdev);
NeilBrown96c3fd12011-12-23 10:17:54 +11001257 read_bio->bi_bdev = rdev->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 read_bio->bi_end_io = raid10_end_read_request;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001259 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 read_bio->bi_private = r10_bio;
1261
NeilBrown856e08e2011-07-28 11:39:23 +10001262 if (max_sectors < r10_bio->sectors) {
1263 /* Could not read all from this device, so we will
1264 * need another r10_bio.
1265 */
NeilBrownb50c2592014-01-14 10:38:09 +11001266 sectors_handled = (r10_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07001267 - bio->bi_iter.bi_sector);
NeilBrown856e08e2011-07-28 11:39:23 +10001268 r10_bio->sectors = max_sectors;
1269 spin_lock_irq(&conf->device_lock);
1270 if (bio->bi_phys_segments == 0)
1271 bio->bi_phys_segments = 2;
1272 else
1273 bio->bi_phys_segments++;
NeilBrownb50c2592014-01-14 10:38:09 +11001274 spin_unlock_irq(&conf->device_lock);
NeilBrown856e08e2011-07-28 11:39:23 +10001275 /* Cannot call generic_make_request directly
1276 * as that will be queued in __generic_make_request
1277 * and subsequent mempool_alloc might block
1278 * waiting for it. so hand bio over to raid10d.
1279 */
1280 reschedule_retry(r10_bio);
1281
1282 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
1283
1284 r10_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001285 r10_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrown856e08e2011-07-28 11:39:23 +10001286 r10_bio->state = 0;
1287 r10_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001288 r10_bio->sector = bio->bi_iter.bi_sector +
1289 sectors_handled;
NeilBrown856e08e2011-07-28 11:39:23 +10001290 goto read_again;
1291 } else
1292 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001293 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
1295
1296 /*
1297 * WRITE:
1298 */
NeilBrown34db0cd2011-10-11 16:50:01 +11001299 if (conf->pending_count >= max_queued_requests) {
1300 md_wakeup_thread(mddev->thread);
1301 wait_event(conf->wait_barrier,
1302 conf->pending_count < max_queued_requests);
1303 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07001304 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 * inc refcount on their rdev. Record them by setting
1306 * bios[x] to bio
NeilBrownd4432c22011-07-28 11:39:24 +10001307 * If there are known/acknowledged bad blocks on any device
1308 * on which we have seen a write error, we want to avoid
1309 * writing to those blocks. This potentially requires several
1310 * writes to write around the bad blocks. Each set of writes
1311 * gets its own r10_bio with a set of bios attached. The number
1312 * of r10_bios is recored in bio->bi_phys_segments just as with
1313 * the read case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001315
NeilBrown69335ef2011-12-23 10:17:54 +11001316 r10_bio->read_slot = -1; /* make sure repl_bio gets freed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 raid10_find_phys(conf, r10_bio);
NeilBrownd4432c22011-07-28 11:39:24 +10001318retry_write:
Harvey Harrisoncb6969e2008-05-06 20:42:32 -07001319 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 rcu_read_lock();
NeilBrownd4432c22011-07-28 11:39:24 +10001321 max_sectors = r10_bio->sectors;
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 for (i = 0; i < conf->copies; i++) {
1324 int d = r10_bio->devs[i].devnum;
NeilBrown3cb03002011-10-11 16:45:26 +11001325 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown475b0322011-12-23 10:17:55 +11001326 struct md_rdev *rrdev = rcu_dereference(
1327 conf->mirrors[d].replacement);
NeilBrown4ca40c22011-12-23 10:17:55 +11001328 if (rdev == rrdev)
1329 rrdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001330 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1331 atomic_inc(&rdev->nr_pending);
1332 blocked_rdev = rdev;
1333 break;
1334 }
NeilBrown475b0322011-12-23 10:17:55 +11001335 if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) {
1336 atomic_inc(&rrdev->nr_pending);
1337 blocked_rdev = rrdev;
1338 break;
1339 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001340 if (rdev && (test_bit(Faulty, &rdev->flags)
1341 || test_bit(Unmerged, &rdev->flags)))
1342 rdev = NULL;
NeilBrown050b6612012-03-19 12:46:39 +11001343 if (rrdev && (test_bit(Faulty, &rrdev->flags)
1344 || test_bit(Unmerged, &rrdev->flags)))
NeilBrown475b0322011-12-23 10:17:55 +11001345 rrdev = NULL;
1346
NeilBrownd4432c22011-07-28 11:39:24 +10001347 r10_bio->devs[i].bio = NULL;
NeilBrown475b0322011-12-23 10:17:55 +11001348 r10_bio->devs[i].repl_bio = NULL;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001349
1350 if (!rdev && !rrdev) {
NeilBrown6cce3b232006-01-06 00:20:16 -08001351 set_bit(R10BIO_Degraded, &r10_bio->state);
NeilBrownd4432c22011-07-28 11:39:24 +10001352 continue;
NeilBrown6cce3b232006-01-06 00:20:16 -08001353 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001354 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) {
NeilBrownd4432c22011-07-28 11:39:24 +10001355 sector_t first_bad;
1356 sector_t dev_sector = r10_bio->devs[i].addr;
1357 int bad_sectors;
1358 int is_bad;
1359
1360 is_bad = is_badblock(rdev, dev_sector,
1361 max_sectors,
1362 &first_bad, &bad_sectors);
1363 if (is_bad < 0) {
1364 /* Mustn't write here until the bad block
1365 * is acknowledged
1366 */
1367 atomic_inc(&rdev->nr_pending);
1368 set_bit(BlockedBadBlocks, &rdev->flags);
1369 blocked_rdev = rdev;
1370 break;
1371 }
1372 if (is_bad && first_bad <= dev_sector) {
1373 /* Cannot write here at all */
1374 bad_sectors -= (dev_sector - first_bad);
1375 if (bad_sectors < max_sectors)
1376 /* Mustn't write more than bad_sectors
1377 * to other devices yet
1378 */
1379 max_sectors = bad_sectors;
1380 /* We don't set R10BIO_Degraded as that
1381 * only applies if the disk is missing,
1382 * so it might be re-added, and we want to
1383 * know to recover this chunk.
1384 * In this case the device is here, and the
1385 * fact that this chunk is not in-sync is
1386 * recorded in the bad block log.
1387 */
1388 continue;
1389 }
1390 if (is_bad) {
1391 int good_sectors = first_bad - dev_sector;
1392 if (good_sectors < max_sectors)
1393 max_sectors = good_sectors;
1394 }
1395 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001396 if (rdev) {
1397 r10_bio->devs[i].bio = bio;
1398 atomic_inc(&rdev->nr_pending);
1399 }
NeilBrown475b0322011-12-23 10:17:55 +11001400 if (rrdev) {
1401 r10_bio->devs[i].repl_bio = bio;
1402 atomic_inc(&rrdev->nr_pending);
1403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 }
1405 rcu_read_unlock();
1406
Dan Williams6bfe0b42008-04-30 00:52:32 -07001407 if (unlikely(blocked_rdev)) {
1408 /* Have to wait for this device to get unblocked, then retry */
1409 int j;
1410 int d;
1411
NeilBrown475b0322011-12-23 10:17:55 +11001412 for (j = 0; j < i; j++) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07001413 if (r10_bio->devs[j].bio) {
1414 d = r10_bio->devs[j].devnum;
1415 rdev_dec_pending(conf->mirrors[d].rdev, mddev);
1416 }
NeilBrown475b0322011-12-23 10:17:55 +11001417 if (r10_bio->devs[j].repl_bio) {
NeilBrown4ca40c22011-12-23 10:17:55 +11001418 struct md_rdev *rdev;
NeilBrown475b0322011-12-23 10:17:55 +11001419 d = r10_bio->devs[j].devnum;
NeilBrown4ca40c22011-12-23 10:17:55 +11001420 rdev = conf->mirrors[d].replacement;
1421 if (!rdev) {
1422 /* Race with remove_disk */
1423 smp_mb();
1424 rdev = conf->mirrors[d].rdev;
1425 }
1426 rdev_dec_pending(rdev, mddev);
NeilBrown475b0322011-12-23 10:17:55 +11001427 }
1428 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07001429 allow_barrier(conf);
1430 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1431 wait_barrier(conf);
1432 goto retry_write;
1433 }
1434
NeilBrownd4432c22011-07-28 11:39:24 +10001435 if (max_sectors < r10_bio->sectors) {
1436 /* We are splitting this into multiple parts, so
1437 * we need to prepare for allocating another r10_bio.
1438 */
1439 r10_bio->sectors = max_sectors;
1440 spin_lock_irq(&conf->device_lock);
1441 if (bio->bi_phys_segments == 0)
1442 bio->bi_phys_segments = 2;
1443 else
1444 bio->bi_phys_segments++;
1445 spin_unlock_irq(&conf->device_lock);
1446 }
Kent Overstreet4f024f32013-10-11 15:44:27 -07001447 sectors_handled = r10_bio->sector + max_sectors -
1448 bio->bi_iter.bi_sector;
NeilBrownd4432c22011-07-28 11:39:24 +10001449
NeilBrown4e780642010-10-19 12:54:01 +11001450 atomic_set(&r10_bio->remaining, 1);
NeilBrownd4432c22011-07-28 11:39:24 +10001451 bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0);
NeilBrown06d91a52005-06-21 17:17:12 -07001452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 for (i = 0; i < conf->copies; i++) {
1454 struct bio *mbio;
1455 int d = r10_bio->devs[i].devnum;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001456 if (r10_bio->devs[i].bio) {
1457 struct md_rdev *rdev = conf->mirrors[d].rdev;
1458 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001459 bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector,
Kent Overstreet6678d832013-08-07 11:14:32 -07001460 max_sectors);
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001461 r10_bio->devs[i].bio = mbio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Kent Overstreet4f024f32013-10-11 15:44:27 -07001463 mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr+
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001464 choose_data_offset(r10_bio,
1465 rdev));
1466 mbio->bi_bdev = rdev->bdev;
1467 mbio->bi_end_io = raid10_end_write_request;
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001468 mbio->bi_rw =
1469 WRITE | do_sync | do_fua | do_discard | do_same;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001470 mbio->bi_private = r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001472 atomic_inc(&r10_bio->remaining);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001474 cb = blk_check_plugged(raid10_unplug, mddev,
1475 sizeof(*plug));
1476 if (cb)
1477 plug = container_of(cb, struct raid10_plug_cb,
1478 cb);
1479 else
1480 plug = NULL;
1481 spin_lock_irqsave(&conf->device_lock, flags);
1482 if (plug) {
1483 bio_list_add(&plug->pending, mbio);
1484 plug->pending_cnt++;
1485 } else {
1486 bio_list_add(&conf->pending_bio_list, mbio);
1487 conf->pending_count++;
1488 }
1489 spin_unlock_irqrestore(&conf->device_lock, flags);
1490 if (!plug)
1491 md_wakeup_thread(mddev->thread);
1492 }
NeilBrown57c67df2012-10-11 13:32:13 +11001493
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001494 if (r10_bio->devs[i].repl_bio) {
1495 struct md_rdev *rdev = conf->mirrors[d].replacement;
1496 if (rdev == NULL) {
1497 /* Replacement just got moved to main 'rdev' */
1498 smp_mb();
1499 rdev = conf->mirrors[d].rdev;
1500 }
1501 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001502 bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector,
Kent Overstreet6678d832013-08-07 11:14:32 -07001503 max_sectors);
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001504 r10_bio->devs[i].repl_bio = mbio;
1505
Kent Overstreet4f024f32013-10-11 15:44:27 -07001506 mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr +
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001507 choose_data_offset(
1508 r10_bio, rdev));
1509 mbio->bi_bdev = rdev->bdev;
1510 mbio->bi_end_io = raid10_end_write_request;
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001511 mbio->bi_rw =
1512 WRITE | do_sync | do_fua | do_discard | do_same;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001513 mbio->bi_private = r10_bio;
1514
1515 atomic_inc(&r10_bio->remaining);
1516 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown57c67df2012-10-11 13:32:13 +11001517 bio_list_add(&conf->pending_bio_list, mbio);
1518 conf->pending_count++;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001519 spin_unlock_irqrestore(&conf->device_lock, flags);
1520 if (!mddev_check_plugged(mddev))
1521 md_wakeup_thread(mddev->thread);
NeilBrown57c67df2012-10-11 13:32:13 +11001522 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 }
1524
NeilBrown079fa162011-09-10 17:21:23 +10001525 /* Don't remove the bias on 'remaining' (one_write_done) until
1526 * after checking if we need to go around again.
1527 */
NeilBrowna35e63e2008-03-04 14:29:29 -08001528
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001529 if (sectors_handled < bio_sectors(bio)) {
NeilBrown079fa162011-09-10 17:21:23 +10001530 one_write_done(r10_bio);
NeilBrown5e570282011-07-28 11:39:25 +10001531 /* We need another r10_bio. It has already been counted
NeilBrownd4432c22011-07-28 11:39:24 +10001532 * in bio->bi_phys_segments.
1533 */
1534 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
1535
1536 r10_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001537 r10_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrownd4432c22011-07-28 11:39:24 +10001538
1539 r10_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001540 r10_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
NeilBrownd4432c22011-07-28 11:39:24 +10001541 r10_bio->state = 0;
1542 goto retry_write;
1543 }
NeilBrown079fa162011-09-10 17:21:23 +10001544 one_write_done(r10_bio);
Kent Overstreet20d01892013-11-23 18:21:01 -08001545}
1546
1547static void make_request(struct mddev *mddev, struct bio *bio)
1548{
1549 struct r10conf *conf = mddev->private;
1550 sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask);
1551 int chunk_sects = chunk_mask + 1;
1552
1553 struct bio *split;
1554
1555 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
1556 md_flush_request(mddev, bio);
1557 return;
1558 }
1559
1560 md_write_start(mddev, bio);
1561
Kent Overstreet20d01892013-11-23 18:21:01 -08001562
1563 do {
1564
1565 /*
1566 * If this request crosses a chunk boundary, we need to split
1567 * it.
1568 */
1569 if (unlikely((bio->bi_iter.bi_sector & chunk_mask) +
1570 bio_sectors(bio) > chunk_sects
1571 && (conf->geo.near_copies < conf->geo.raid_disks
1572 || conf->prev.near_copies <
1573 conf->prev.raid_disks))) {
1574 split = bio_split(bio, chunk_sects -
1575 (bio->bi_iter.bi_sector &
1576 (chunk_sects - 1)),
1577 GFP_NOIO, fs_bio_set);
1578 bio_chain(split, bio);
1579 } else {
1580 split = bio;
1581 }
1582
1583 __make_request(mddev, split);
1584 } while (split != bio);
NeilBrown079fa162011-09-10 17:21:23 +10001585
1586 /* In case raid10d snuck in to freeze_array */
1587 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588}
1589
NeilBrownfd01b882011-10-11 16:47:53 +11001590static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591{
NeilBrowne879a872011-10-11 16:49:02 +11001592 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 int i;
1594
NeilBrown5cf00fc2012-05-21 09:28:20 +10001595 if (conf->geo.near_copies < conf->geo.raid_disks)
Andre Noll9d8f0362009-06-18 08:45:01 +10001596 seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2);
NeilBrown5cf00fc2012-05-21 09:28:20 +10001597 if (conf->geo.near_copies > 1)
1598 seq_printf(seq, " %d near-copies", conf->geo.near_copies);
1599 if (conf->geo.far_copies > 1) {
1600 if (conf->geo.far_offset)
1601 seq_printf(seq, " %d offset-copies", conf->geo.far_copies);
NeilBrownc93983b2006-06-26 00:27:41 -07001602 else
NeilBrown5cf00fc2012-05-21 09:28:20 +10001603 seq_printf(seq, " %d far-copies", conf->geo.far_copies);
NeilBrownc93983b2006-06-26 00:27:41 -07001604 }
NeilBrown5cf00fc2012-05-21 09:28:20 +10001605 seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks,
1606 conf->geo.raid_disks - mddev->degraded);
1607 for (i = 0; i < conf->geo.raid_disks; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 seq_printf(seq, "%s",
1609 conf->mirrors[i].rdev &&
NeilBrownb2d444d2005-11-08 21:39:31 -08001610 test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 seq_printf(seq, "]");
1612}
1613
NeilBrown700c7212011-07-27 11:00:36 +10001614/* check if there are enough drives for
1615 * every block to appear on atleast one.
1616 * Don't consider the device numbered 'ignore'
1617 * as we might be about to remove it.
1618 */
NeilBrown635f6412013-06-11 14:57:09 +10001619static int _enough(struct r10conf *conf, int previous, int ignore)
NeilBrown700c7212011-07-27 11:00:36 +10001620{
1621 int first = 0;
NeilBrown725d6e52013-06-11 15:08:03 +10001622 int has_enough = 0;
NeilBrown635f6412013-06-11 14:57:09 +10001623 int disks, ncopies;
1624 if (previous) {
1625 disks = conf->prev.raid_disks;
1626 ncopies = conf->prev.near_copies;
1627 } else {
1628 disks = conf->geo.raid_disks;
1629 ncopies = conf->geo.near_copies;
1630 }
NeilBrown700c7212011-07-27 11:00:36 +10001631
NeilBrown725d6e52013-06-11 15:08:03 +10001632 rcu_read_lock();
NeilBrown700c7212011-07-27 11:00:36 +10001633 do {
1634 int n = conf->copies;
1635 int cnt = 0;
NeilBrown80b48122012-09-27 12:35:21 +10001636 int this = first;
NeilBrown700c7212011-07-27 11:00:36 +10001637 while (n--) {
NeilBrown725d6e52013-06-11 15:08:03 +10001638 struct md_rdev *rdev;
1639 if (this != ignore &&
1640 (rdev = rcu_dereference(conf->mirrors[this].rdev)) &&
1641 test_bit(In_sync, &rdev->flags))
NeilBrown700c7212011-07-27 11:00:36 +10001642 cnt++;
NeilBrown635f6412013-06-11 14:57:09 +10001643 this = (this+1) % disks;
NeilBrown700c7212011-07-27 11:00:36 +10001644 }
1645 if (cnt == 0)
NeilBrown725d6e52013-06-11 15:08:03 +10001646 goto out;
NeilBrown635f6412013-06-11 14:57:09 +10001647 first = (first + ncopies) % disks;
NeilBrown700c7212011-07-27 11:00:36 +10001648 } while (first != 0);
NeilBrown725d6e52013-06-11 15:08:03 +10001649 has_enough = 1;
1650out:
1651 rcu_read_unlock();
1652 return has_enough;
NeilBrown700c7212011-07-27 11:00:36 +10001653}
1654
NeilBrownf8c9e742012-05-21 09:28:33 +10001655static int enough(struct r10conf *conf, int ignore)
1656{
NeilBrown635f6412013-06-11 14:57:09 +10001657 /* when calling 'enough', both 'prev' and 'geo' must
1658 * be stable.
1659 * This is ensured if ->reconfig_mutex or ->device_lock
1660 * is held.
1661 */
1662 return _enough(conf, 0, ignore) &&
1663 _enough(conf, 1, ignore);
NeilBrownf8c9e742012-05-21 09:28:33 +10001664}
1665
NeilBrownfd01b882011-10-11 16:47:53 +11001666static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667{
1668 char b[BDEVNAME_SIZE];
NeilBrowne879a872011-10-11 16:49:02 +11001669 struct r10conf *conf = mddev->private;
NeilBrown635f6412013-06-11 14:57:09 +10001670 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
1672 /*
1673 * If it is not operational, then we have already marked it as dead
1674 * else if it is the last working disks, ignore the error, let the
1675 * next level up know.
1676 * else mark the drive as failed
1677 */
NeilBrown635f6412013-06-11 14:57:09 +10001678 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownb2d444d2005-11-08 21:39:31 -08001679 if (test_bit(In_sync, &rdev->flags)
NeilBrown635f6412013-06-11 14:57:09 +10001680 && !enough(conf, rdev->raid_disk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 /*
1682 * Don't fail the drive, just return an IO error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 */
NeilBrownc04be0a2006-10-03 01:15:53 -07001684 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown635f6412013-06-11 14:57:09 +10001685 return;
1686 }
NeilBrown2446dba2014-07-31 10:16:29 +10001687 if (test_and_clear_bit(In_sync, &rdev->flags))
NeilBrown635f6412013-06-11 14:57:09 +10001688 mddev->degraded++;
NeilBrown2446dba2014-07-31 10:16:29 +10001689 /*
1690 * If recovery is running, make sure it aborts.
1691 */
1692 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrownde393cd2011-07-28 11:31:48 +10001693 set_bit(Blocked, &rdev->flags);
NeilBrownb2d444d2005-11-08 21:39:31 -08001694 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001695 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrown635f6412013-06-11 14:57:09 +10001696 spin_unlock_irqrestore(&conf->device_lock, flags);
Joe Perches067032b2011-01-14 09:14:33 +11001697 printk(KERN_ALERT
1698 "md/raid10:%s: Disk failure on %s, disabling device.\n"
1699 "md/raid10:%s: Operation continuing on %d devices.\n",
NeilBrown128595e2010-05-03 14:47:14 +10001700 mdname(mddev), bdevname(rdev->bdev, b),
NeilBrown5cf00fc2012-05-21 09:28:20 +10001701 mdname(mddev), conf->geo.raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702}
1703
NeilBrowne879a872011-10-11 16:49:02 +11001704static void print_conf(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
1706 int i;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10001707 struct raid10_info *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
NeilBrown128595e2010-05-03 14:47:14 +10001709 printk(KERN_DEBUG "RAID10 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 if (!conf) {
NeilBrown128595e2010-05-03 14:47:14 +10001711 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 return;
1713 }
NeilBrown5cf00fc2012-05-21 09:28:20 +10001714 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded,
1715 conf->geo.raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716
NeilBrown5cf00fc2012-05-21 09:28:20 +10001717 for (i = 0; i < conf->geo.raid_disks; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 char b[BDEVNAME_SIZE];
1719 tmp = conf->mirrors + i;
1720 if (tmp->rdev)
NeilBrown128595e2010-05-03 14:47:14 +10001721 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownb2d444d2005-11-08 21:39:31 -08001722 i, !test_bit(In_sync, &tmp->rdev->flags),
1723 !test_bit(Faulty, &tmp->rdev->flags),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 bdevname(tmp->rdev->bdev,b));
1725 }
1726}
1727
NeilBrowne879a872011-10-11 16:49:02 +11001728static void close_sync(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729{
NeilBrown0a27ec92006-01-06 00:20:13 -08001730 wait_barrier(conf);
1731 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 mempool_destroy(conf->r10buf_pool);
1734 conf->r10buf_pool = NULL;
1735}
1736
NeilBrownfd01b882011-10-11 16:47:53 +11001737static int raid10_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738{
1739 int i;
NeilBrowne879a872011-10-11 16:49:02 +11001740 struct r10conf *conf = mddev->private;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10001741 struct raid10_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10001742 int count = 0;
1743 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
1745 /*
1746 * Find all non-in_sync disks within the RAID10 configuration
1747 * and mark them in_sync
1748 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10001749 for (i = 0; i < conf->geo.raid_disks; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 tmp = conf->mirrors + i;
NeilBrown4ca40c22011-12-23 10:17:55 +11001751 if (tmp->replacement
1752 && tmp->replacement->recovery_offset == MaxSector
1753 && !test_bit(Faulty, &tmp->replacement->flags)
1754 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
1755 /* Replacement has just become active */
1756 if (!tmp->rdev
1757 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
1758 count++;
1759 if (tmp->rdev) {
1760 /* Replaced device not technically faulty,
1761 * but we need to be sure it gets removed
1762 * and never re-added.
1763 */
1764 set_bit(Faulty, &tmp->rdev->flags);
1765 sysfs_notify_dirent_safe(
1766 tmp->rdev->sysfs_state);
1767 }
1768 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
1769 } else if (tmp->rdev
Lukasz Dorau61e49472013-10-24 12:55:17 +11001770 && tmp->rdev->recovery_offset == MaxSector
NeilBrown4ca40c22011-12-23 10:17:55 +11001771 && !test_bit(Faulty, &tmp->rdev->flags)
1772 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001773 count++;
Jonathan Brassow2863b9e2012-10-11 13:38:58 +11001774 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 }
1776 }
NeilBrown6b965622010-08-18 11:56:59 +10001777 spin_lock_irqsave(&conf->device_lock, flags);
1778 mddev->degraded -= count;
1779 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
1781 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001782 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783}
1784
1785
NeilBrownfd01b882011-10-11 16:47:53 +11001786static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787{
NeilBrowne879a872011-10-11 16:49:02 +11001788 struct r10conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001789 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 int mirror;
Neil Brown6c2fce22008-06-28 08:31:31 +10001791 int first = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10001792 int last = conf->geo.raid_disks - 1;
NeilBrown050b6612012-03-19 12:46:39 +11001793 struct request_queue *q = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
1795 if (mddev->recovery_cp < MaxSector)
1796 /* only hot-add to in-sync arrays, as recovery is
1797 * very different from resync
1798 */
Neil Brown199050e2008-06-28 08:31:33 +10001799 return -EBUSY;
NeilBrown635f6412013-06-11 14:57:09 +10001800 if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1))
Neil Brown199050e2008-06-28 08:31:33 +10001801 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
NeilBrowna53a6c82008-11-06 17:28:20 +11001803 if (rdev->raid_disk >= 0)
Neil Brown6c2fce22008-06-28 08:31:31 +10001804 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
NeilBrown050b6612012-03-19 12:46:39 +11001806 if (q->merge_bvec_fn) {
1807 set_bit(Unmerged, &rdev->flags);
1808 mddev->merge_check_needed = 1;
1809 }
1810
Namhyung Kim2c4193d2011-07-18 17:38:43 +10001811 if (rdev->saved_raid_disk >= first &&
NeilBrown6cce3b232006-01-06 00:20:16 -08001812 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1813 mirror = rdev->saved_raid_disk;
1814 else
Neil Brown6c2fce22008-06-28 08:31:31 +10001815 mirror = first;
NeilBrown2bb77732011-07-27 11:00:36 +10001816 for ( ; mirror <= last ; mirror++) {
Jonathan Brassowdc280d982012-07-31 10:03:52 +10001817 struct raid10_info *p = &conf->mirrors[mirror];
NeilBrown2bb77732011-07-27 11:00:36 +10001818 if (p->recovery_disabled == mddev->recovery_disabled)
1819 continue;
NeilBrownb7044d42011-12-23 10:17:56 +11001820 if (p->rdev) {
1821 if (!test_bit(WantReplacement, &p->rdev->flags) ||
1822 p->replacement != NULL)
1823 continue;
1824 clear_bit(In_sync, &rdev->flags);
1825 set_bit(Replacement, &rdev->flags);
1826 rdev->raid_disk = mirror;
1827 err = 0;
Jonathan Brassow9092c022013-05-02 14:19:24 -05001828 if (mddev->gendisk)
1829 disk_stack_limits(mddev->gendisk, rdev->bdev,
1830 rdev->data_offset << 9);
NeilBrownb7044d42011-12-23 10:17:56 +11001831 conf->fullsync = 1;
1832 rcu_assign_pointer(p->replacement, rdev);
1833 break;
1834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Jonathan Brassow9092c022013-05-02 14:19:24 -05001836 if (mddev->gendisk)
1837 disk_stack_limits(mddev->gendisk, rdev->bdev,
1838 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
NeilBrown2bb77732011-07-27 11:00:36 +10001840 p->head_position = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11001841 p->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown2bb77732011-07-27 11:00:36 +10001842 rdev->raid_disk = mirror;
1843 err = 0;
1844 if (rdev->saved_raid_disk != mirror)
1845 conf->fullsync = 1;
1846 rcu_assign_pointer(p->rdev, rdev);
1847 break;
1848 }
NeilBrown050b6612012-03-19 12:46:39 +11001849 if (err == 0 && test_bit(Unmerged, &rdev->flags)) {
1850 /* Some requests might not have seen this new
1851 * merge_bvec_fn. We must wait for them to complete
1852 * before merging the device fully.
1853 * First we make sure any code which has tested
1854 * our function has submitted the request, then
1855 * we wait for all outstanding requests to complete.
1856 */
1857 synchronize_sched();
NeilBrowne2d59922013-06-12 11:01:22 +10001858 freeze_array(conf, 0);
1859 unfreeze_array(conf);
NeilBrown050b6612012-03-19 12:46:39 +11001860 clear_bit(Unmerged, &rdev->flags);
1861 }
Andre Nollac5e7112009-08-03 10:59:47 +10001862 md_integrity_add_rdev(rdev, mddev);
Jonathan Brassowed30be02012-10-31 11:42:30 +11001863 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Shaohua Li532a2a32012-10-11 13:30:52 +11001864 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001867 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868}
1869
NeilBrownb8321b62011-12-23 10:17:51 +11001870static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
NeilBrowne879a872011-10-11 16:49:02 +11001872 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001874 int number = rdev->raid_disk;
NeilBrownc8ab9032011-12-23 10:17:54 +11001875 struct md_rdev **rdevp;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10001876 struct raid10_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
1878 print_conf(conf);
NeilBrownc8ab9032011-12-23 10:17:54 +11001879 if (rdev == p->rdev)
1880 rdevp = &p->rdev;
1881 else if (rdev == p->replacement)
1882 rdevp = &p->replacement;
1883 else
1884 return 0;
1885
1886 if (test_bit(In_sync, &rdev->flags) ||
1887 atomic_read(&rdev->nr_pending)) {
1888 err = -EBUSY;
1889 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 }
NeilBrownc8ab9032011-12-23 10:17:54 +11001891 /* Only remove faulty devices if recovery
1892 * is not possible.
1893 */
1894 if (!test_bit(Faulty, &rdev->flags) &&
1895 mddev->recovery_disabled != p->recovery_disabled &&
NeilBrown4ca40c22011-12-23 10:17:55 +11001896 (!p->replacement || p->replacement == rdev) &&
NeilBrown63aced62012-05-22 13:55:33 +10001897 number < conf->geo.raid_disks &&
NeilBrownc8ab9032011-12-23 10:17:54 +11001898 enough(conf, -1)) {
1899 err = -EBUSY;
1900 goto abort;
1901 }
1902 *rdevp = NULL;
1903 synchronize_rcu();
1904 if (atomic_read(&rdev->nr_pending)) {
1905 /* lost the race, try later */
1906 err = -EBUSY;
1907 *rdevp = rdev;
1908 goto abort;
NeilBrown4ca40c22011-12-23 10:17:55 +11001909 } else if (p->replacement) {
1910 /* We must have just cleared 'rdev' */
1911 p->rdev = p->replacement;
1912 clear_bit(Replacement, &p->replacement->flags);
1913 smp_mb(); /* Make sure other CPUs may see both as identical
1914 * but will never see neither -- if they are careful.
1915 */
1916 p->replacement = NULL;
1917 clear_bit(WantReplacement, &rdev->flags);
1918 } else
1919 /* We might have just remove the Replacement as faulty
1920 * Clear the flag just in case
1921 */
1922 clear_bit(WantReplacement, &rdev->flags);
1923
NeilBrownc8ab9032011-12-23 10:17:54 +11001924 err = md_integrity_register(mddev);
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926abort:
1927
1928 print_conf(conf);
1929 return err;
1930}
1931
1932
NeilBrown6712ecf2007-09-27 12:47:43 +02001933static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001935 struct r10bio *r10_bio = bio->bi_private;
NeilBrowne879a872011-10-11 16:49:02 +11001936 struct r10conf *conf = r10_bio->mddev->private;
Namhyung Kim778ca012011-07-18 17:38:47 +10001937 int d;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
NeilBrown3ea7daa2012-05-22 13:53:47 +10001939 if (bio == r10_bio->master_bio) {
1940 /* this is a reshape read */
1941 d = r10_bio->read_slot; /* really the read dev */
1942 } else
1943 d = find_bio_disk(conf, r10_bio, bio, NULL, NULL);
NeilBrown0eb3ff12006-01-06 00:20:29 -08001944
1945 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1946 set_bit(R10BIO_Uptodate, &r10_bio->state);
NeilBrowne684e412011-07-28 11:39:25 +10001947 else
1948 /* The write handler will notice the lack of
1949 * R10BIO_Uptodate and record any errors etc
1950 */
NeilBrown4dbcdc72006-01-06 00:20:52 -08001951 atomic_add(r10_bio->sectors,
1952 &conf->mirrors[d].rdev->corrected_errors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 /* for reconstruct, we always reschedule after a read.
1955 * for resync, only after all reads
1956 */
NeilBrown73d5c382009-02-25 13:18:47 +11001957 rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
1959 atomic_dec_and_test(&r10_bio->remaining)) {
1960 /* we have read all the blocks,
1961 * do the comparison in process context in raid10d
1962 */
1963 reschedule_retry(r10_bio);
1964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965}
1966
NeilBrown9f2c9d12011-10-11 16:48:43 +11001967static void end_sync_request(struct r10bio *r10_bio)
NeilBrown5e570282011-07-28 11:39:25 +10001968{
NeilBrownfd01b882011-10-11 16:47:53 +11001969 struct mddev *mddev = r10_bio->mddev;
NeilBrown5e570282011-07-28 11:39:25 +10001970
1971 while (atomic_dec_and_test(&r10_bio->remaining)) {
1972 if (r10_bio->master_bio == NULL) {
1973 /* the primary of several recovery bios */
1974 sector_t s = r10_bio->sectors;
1975 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
1976 test_bit(R10BIO_WriteError, &r10_bio->state))
1977 reschedule_retry(r10_bio);
1978 else
1979 put_buf(r10_bio);
1980 md_done_sync(mddev, s, 1);
1981 break;
1982 } else {
NeilBrown9f2c9d12011-10-11 16:48:43 +11001983 struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio;
NeilBrown5e570282011-07-28 11:39:25 +10001984 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
1985 test_bit(R10BIO_WriteError, &r10_bio->state))
1986 reschedule_retry(r10_bio);
1987 else
1988 put_buf(r10_bio);
1989 r10_bio = r10_bio2;
1990 }
1991 }
1992}
1993
NeilBrown6712ecf2007-09-27 12:47:43 +02001994static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995{
1996 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +11001997 struct r10bio *r10_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001998 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11001999 struct r10conf *conf = mddev->private;
Namhyung Kim778ca012011-07-18 17:38:47 +10002000 int d;
NeilBrown749c55e2011-07-28 11:39:24 +10002001 sector_t first_bad;
2002 int bad_sectors;
2003 int slot;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002004 int repl;
NeilBrown4ca40c22011-12-23 10:17:55 +11002005 struct md_rdev *rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
NeilBrown9ad1aef2011-12-23 10:17:55 +11002007 d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
2008 if (repl)
2009 rdev = conf->mirrors[d].replacement;
NeilBrown547414d2012-03-13 11:21:20 +11002010 else
NeilBrown9ad1aef2011-12-23 10:17:55 +11002011 rdev = conf->mirrors[d].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002013 if (!uptodate) {
NeilBrown9ad1aef2011-12-23 10:17:55 +11002014 if (repl)
2015 md_error(mddev, rdev);
2016 else {
2017 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002018 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2019 set_bit(MD_RECOVERY_NEEDED,
2020 &rdev->mddev->recovery);
NeilBrown9ad1aef2011-12-23 10:17:55 +11002021 set_bit(R10BIO_WriteError, &r10_bio->state);
2022 }
2023 } else if (is_badblock(rdev,
NeilBrown749c55e2011-07-28 11:39:24 +10002024 r10_bio->devs[slot].addr,
2025 r10_bio->sectors,
2026 &first_bad, &bad_sectors))
2027 set_bit(R10BIO_MadeGood, &r10_bio->state);
NeilBrowndfc70642008-05-23 13:04:39 -07002028
NeilBrown9ad1aef2011-12-23 10:17:55 +11002029 rdev_dec_pending(rdev, mddev);
NeilBrown5e570282011-07-28 11:39:25 +10002030
2031 end_sync_request(r10_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032}
2033
2034/*
2035 * Note: sync and recover and handled very differently for raid10
2036 * This code is for resync.
2037 * For resync, we read through virtual addresses and read all blocks.
2038 * If there is any error, we schedule a write. The lowest numbered
2039 * drive is authoritative.
2040 * However requests come for physical address, so we need to map.
2041 * For every physical address there are raid_disks/copies virtual addresses,
2042 * which is always are least one, but is not necessarly an integer.
2043 * This means that a physical address can span multiple chunks, so we may
2044 * have to submit multiple io requests for a single sync request.
2045 */
2046/*
2047 * We check if all blocks are in-sync and only write to blocks that
2048 * aren't in sync
2049 */
NeilBrown9f2c9d12011-10-11 16:48:43 +11002050static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051{
NeilBrowne879a872011-10-11 16:49:02 +11002052 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 int i, first;
2054 struct bio *tbio, *fbio;
majianpengf4380a92012-04-12 16:04:47 +10002055 int vcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
2057 atomic_set(&r10_bio->remaining, 1);
2058
2059 /* find the first device with a block */
2060 for (i=0; i<conf->copies; i++)
2061 if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags))
2062 break;
2063
2064 if (i == conf->copies)
2065 goto done;
2066
2067 first = i;
2068 fbio = r10_bio->devs[i].bio;
2069
majianpengf4380a92012-04-12 16:04:47 +10002070 vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 /* now find blocks with errors */
NeilBrown0eb3ff12006-01-06 00:20:29 -08002072 for (i=0 ; i < conf->copies ; i++) {
2073 int j, d;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 tbio = r10_bio->devs[i].bio;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002076
2077 if (tbio->bi_end_io != end_sync_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 continue;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002079 if (i == first)
2080 continue;
2081 if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) {
2082 /* We know that the bi_io_vec layout is the same for
2083 * both 'first' and 'i', so we just compare them.
2084 * All vec entries are PAGE_SIZE;
2085 */
NeilBrown7bb23c42013-07-16 16:50:47 +10002086 int sectors = r10_bio->sectors;
2087 for (j = 0; j < vcnt; j++) {
2088 int len = PAGE_SIZE;
2089 if (sectors < (len / 512))
2090 len = sectors * 512;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002091 if (memcmp(page_address(fbio->bi_io_vec[j].bv_page),
2092 page_address(tbio->bi_io_vec[j].bv_page),
NeilBrown7bb23c42013-07-16 16:50:47 +10002093 len))
NeilBrown0eb3ff12006-01-06 00:20:29 -08002094 break;
NeilBrown7bb23c42013-07-16 16:50:47 +10002095 sectors -= len/512;
2096 }
NeilBrown0eb3ff12006-01-06 00:20:29 -08002097 if (j == vcnt)
2098 continue;
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11002099 atomic64_add(r10_bio->sectors, &mddev->resync_mismatches);
NeilBrownf84ee362011-07-28 11:39:25 +10002100 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2101 /* Don't fix anything. */
2102 continue;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002103 }
NeilBrownf84ee362011-07-28 11:39:25 +10002104 /* Ok, we need to write this bio, either to correct an
2105 * inconsistency or to correct an unreadable block.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 * First we need to fixup bv_offset, bv_len and
2107 * bi_vecs, as the read request might have corrupted these
2108 */
Kent Overstreet8be185f2012-09-06 14:14:43 -07002109 bio_reset(tbio);
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 tbio->bi_vcnt = vcnt;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002112 tbio->bi_iter.bi_size = r10_bio->sectors << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 tbio->bi_rw = WRITE;
2114 tbio->bi_private = r10_bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002115 tbio->bi_iter.bi_sector = r10_bio->devs[i].addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117 for (j=0; j < vcnt ; j++) {
2118 tbio->bi_io_vec[j].bv_offset = 0;
2119 tbio->bi_io_vec[j].bv_len = PAGE_SIZE;
2120
2121 memcpy(page_address(tbio->bi_io_vec[j].bv_page),
2122 page_address(fbio->bi_io_vec[j].bv_page),
2123 PAGE_SIZE);
2124 }
2125 tbio->bi_end_io = end_sync_write;
2126
2127 d = r10_bio->devs[i].devnum;
2128 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
2129 atomic_inc(&r10_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002130 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Kent Overstreet4f024f32013-10-11 15:44:27 -07002132 tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 tbio->bi_bdev = conf->mirrors[d].rdev->bdev;
2134 generic_make_request(tbio);
2135 }
2136
NeilBrown9ad1aef2011-12-23 10:17:55 +11002137 /* Now write out to any replacement devices
2138 * that are active
2139 */
2140 for (i = 0; i < conf->copies; i++) {
2141 int j, d;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002142
2143 tbio = r10_bio->devs[i].repl_bio;
2144 if (!tbio || !tbio->bi_end_io)
2145 continue;
2146 if (r10_bio->devs[i].bio->bi_end_io != end_sync_write
2147 && r10_bio->devs[i].bio != fbio)
2148 for (j = 0; j < vcnt; j++)
2149 memcpy(page_address(tbio->bi_io_vec[j].bv_page),
2150 page_address(fbio->bi_io_vec[j].bv_page),
2151 PAGE_SIZE);
2152 d = r10_bio->devs[i].devnum;
2153 atomic_inc(&r10_bio->remaining);
2154 md_sync_acct(conf->mirrors[d].replacement->bdev,
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002155 bio_sectors(tbio));
NeilBrown9ad1aef2011-12-23 10:17:55 +11002156 generic_make_request(tbio);
2157 }
2158
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159done:
2160 if (atomic_dec_and_test(&r10_bio->remaining)) {
2161 md_done_sync(mddev, r10_bio->sectors, 1);
2162 put_buf(r10_bio);
2163 }
2164}
2165
2166/*
2167 * Now for the recovery code.
2168 * Recovery happens across physical sectors.
2169 * We recover all non-is_sync drives by finding the virtual address of
2170 * each, and then choose a working drive that also has that virt address.
2171 * There is a separate r10_bio for each non-in_sync drive.
2172 * Only the first two slots are in use. The first for reading,
2173 * The second for writing.
2174 *
2175 */
NeilBrown9f2c9d12011-10-11 16:48:43 +11002176static void fix_recovery_read_error(struct r10bio *r10_bio)
NeilBrown5e570282011-07-28 11:39:25 +10002177{
2178 /* We got a read error during recovery.
2179 * We repeat the read in smaller page-sized sections.
2180 * If a read succeeds, write it to the new device or record
2181 * a bad block if we cannot.
2182 * If a read fails, record a bad block on both old and
2183 * new devices.
2184 */
NeilBrownfd01b882011-10-11 16:47:53 +11002185 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11002186 struct r10conf *conf = mddev->private;
NeilBrown5e570282011-07-28 11:39:25 +10002187 struct bio *bio = r10_bio->devs[0].bio;
2188 sector_t sect = 0;
2189 int sectors = r10_bio->sectors;
2190 int idx = 0;
2191 int dr = r10_bio->devs[0].devnum;
2192 int dw = r10_bio->devs[1].devnum;
2193
2194 while (sectors) {
2195 int s = sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11002196 struct md_rdev *rdev;
NeilBrown5e570282011-07-28 11:39:25 +10002197 sector_t addr;
2198 int ok;
2199
2200 if (s > (PAGE_SIZE>>9))
2201 s = PAGE_SIZE >> 9;
2202
2203 rdev = conf->mirrors[dr].rdev;
2204 addr = r10_bio->devs[0].addr + sect,
2205 ok = sync_page_io(rdev,
2206 addr,
2207 s << 9,
2208 bio->bi_io_vec[idx].bv_page,
2209 READ, false);
2210 if (ok) {
2211 rdev = conf->mirrors[dw].rdev;
2212 addr = r10_bio->devs[1].addr + sect;
2213 ok = sync_page_io(rdev,
2214 addr,
2215 s << 9,
2216 bio->bi_io_vec[idx].bv_page,
2217 WRITE, false);
NeilBrownb7044d42011-12-23 10:17:56 +11002218 if (!ok) {
NeilBrown5e570282011-07-28 11:39:25 +10002219 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002220 if (!test_and_set_bit(WantReplacement,
2221 &rdev->flags))
2222 set_bit(MD_RECOVERY_NEEDED,
2223 &rdev->mddev->recovery);
2224 }
NeilBrown5e570282011-07-28 11:39:25 +10002225 }
2226 if (!ok) {
2227 /* We don't worry if we cannot set a bad block -
2228 * it really is bad so there is no loss in not
2229 * recording it yet
2230 */
2231 rdev_set_badblocks(rdev, addr, s, 0);
2232
2233 if (rdev != conf->mirrors[dw].rdev) {
2234 /* need bad block on destination too */
NeilBrown3cb03002011-10-11 16:45:26 +11002235 struct md_rdev *rdev2 = conf->mirrors[dw].rdev;
NeilBrown5e570282011-07-28 11:39:25 +10002236 addr = r10_bio->devs[1].addr + sect;
2237 ok = rdev_set_badblocks(rdev2, addr, s, 0);
2238 if (!ok) {
2239 /* just abort the recovery */
2240 printk(KERN_NOTICE
2241 "md/raid10:%s: recovery aborted"
2242 " due to read error\n",
2243 mdname(mddev));
2244
2245 conf->mirrors[dw].recovery_disabled
2246 = mddev->recovery_disabled;
2247 set_bit(MD_RECOVERY_INTR,
2248 &mddev->recovery);
2249 break;
2250 }
2251 }
2252 }
2253
2254 sectors -= s;
2255 sect += s;
2256 idx++;
2257 }
2258}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259
NeilBrown9f2c9d12011-10-11 16:48:43 +11002260static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261{
NeilBrowne879a872011-10-11 16:49:02 +11002262 struct r10conf *conf = mddev->private;
Namhyung Kimc65060a2011-07-18 17:38:49 +10002263 int d;
NeilBrown24afd802011-12-23 10:17:55 +11002264 struct bio *wbio, *wbio2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
NeilBrown5e570282011-07-28 11:39:25 +10002266 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) {
2267 fix_recovery_read_error(r10_bio);
2268 end_sync_request(r10_bio);
2269 return;
2270 }
2271
Namhyung Kimc65060a2011-07-18 17:38:49 +10002272 /*
2273 * share the pages with the first bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 * and submit the write request
2275 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 d = r10_bio->devs[1].devnum;
NeilBrown24afd802011-12-23 10:17:55 +11002277 wbio = r10_bio->devs[1].bio;
2278 wbio2 = r10_bio->devs[1].repl_bio;
NeilBrown0eb25bb2013-07-24 15:37:42 +10002279 /* Need to test wbio2->bi_end_io before we call
2280 * generic_make_request as if the former is NULL,
2281 * the latter is free to free wbio2.
2282 */
2283 if (wbio2 && !wbio2->bi_end_io)
2284 wbio2 = NULL;
NeilBrown24afd802011-12-23 10:17:55 +11002285 if (wbio->bi_end_io) {
2286 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002287 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio));
NeilBrown24afd802011-12-23 10:17:55 +11002288 generic_make_request(wbio);
2289 }
NeilBrown0eb25bb2013-07-24 15:37:42 +10002290 if (wbio2) {
NeilBrown24afd802011-12-23 10:17:55 +11002291 atomic_inc(&conf->mirrors[d].replacement->nr_pending);
2292 md_sync_acct(conf->mirrors[d].replacement->bdev,
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002293 bio_sectors(wbio2));
NeilBrown24afd802011-12-23 10:17:55 +11002294 generic_make_request(wbio2);
2295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296}
2297
2298
2299/*
Robert Becker1e509152009-12-14 12:49:58 +11002300 * Used by fix_read_error() to decay the per rdev read_errors.
2301 * We halve the read error count for every hour that has elapsed
2302 * since the last recorded read error.
2303 *
2304 */
NeilBrownfd01b882011-10-11 16:47:53 +11002305static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev)
Robert Becker1e509152009-12-14 12:49:58 +11002306{
2307 struct timespec cur_time_mon;
2308 unsigned long hours_since_last;
2309 unsigned int read_errors = atomic_read(&rdev->read_errors);
2310
2311 ktime_get_ts(&cur_time_mon);
2312
2313 if (rdev->last_read_error.tv_sec == 0 &&
2314 rdev->last_read_error.tv_nsec == 0) {
2315 /* first time we've seen a read error */
2316 rdev->last_read_error = cur_time_mon;
2317 return;
2318 }
2319
2320 hours_since_last = (cur_time_mon.tv_sec -
2321 rdev->last_read_error.tv_sec) / 3600;
2322
2323 rdev->last_read_error = cur_time_mon;
2324
2325 /*
2326 * if hours_since_last is > the number of bits in read_errors
2327 * just set read errors to 0. We do this to avoid
2328 * overflowing the shift of read_errors by hours_since_last.
2329 */
2330 if (hours_since_last >= 8 * sizeof(read_errors))
2331 atomic_set(&rdev->read_errors, 0);
2332 else
2333 atomic_set(&rdev->read_errors, read_errors >> hours_since_last);
2334}
2335
NeilBrown3cb03002011-10-11 16:45:26 +11002336static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrown58c54fc2011-07-28 11:39:25 +10002337 int sectors, struct page *page, int rw)
2338{
2339 sector_t first_bad;
2340 int bad_sectors;
2341
2342 if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors)
2343 && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags)))
2344 return -1;
2345 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
2346 /* success */
2347 return 1;
NeilBrownb7044d42011-12-23 10:17:56 +11002348 if (rw == WRITE) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002349 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002350 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2351 set_bit(MD_RECOVERY_NEEDED,
2352 &rdev->mddev->recovery);
2353 }
NeilBrown58c54fc2011-07-28 11:39:25 +10002354 /* need to record an error - either for the block or the device */
2355 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
2356 md_error(rdev->mddev, rdev);
2357 return 0;
2358}
2359
Robert Becker1e509152009-12-14 12:49:58 +11002360/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 * This is a kernel thread which:
2362 *
2363 * 1. Retries failed read operations on working mirrors.
2364 * 2. Updates the raid superblock when problems encounter.
NeilBrown6814d532006-10-03 01:15:45 -07002365 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 */
2367
NeilBrowne879a872011-10-11 16:49:02 +11002368static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio)
NeilBrown6814d532006-10-03 01:15:45 -07002369{
2370 int sect = 0; /* Offset from r10_bio->sector */
2371 int sectors = r10_bio->sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11002372 struct md_rdev*rdev;
Robert Becker1e509152009-12-14 12:49:58 +11002373 int max_read_errors = atomic_read(&mddev->max_corr_read_errors);
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002374 int d = r10_bio->devs[r10_bio->read_slot].devnum;
Robert Becker1e509152009-12-14 12:49:58 +11002375
NeilBrown7c4e06f2011-05-11 14:53:17 +10002376 /* still own a reference to this rdev, so it cannot
2377 * have been cleared recently.
2378 */
2379 rdev = conf->mirrors[d].rdev;
Robert Becker1e509152009-12-14 12:49:58 +11002380
NeilBrown7c4e06f2011-05-11 14:53:17 +10002381 if (test_bit(Faulty, &rdev->flags))
2382 /* drive has already been failed, just ignore any
2383 more fix_read_error() attempts */
2384 return;
2385
2386 check_decay_read_errors(mddev, rdev);
2387 atomic_inc(&rdev->read_errors);
2388 if (atomic_read(&rdev->read_errors) > max_read_errors) {
2389 char b[BDEVNAME_SIZE];
Robert Becker1e509152009-12-14 12:49:58 +11002390 bdevname(rdev->bdev, b);
2391
NeilBrown7c4e06f2011-05-11 14:53:17 +10002392 printk(KERN_NOTICE
2393 "md/raid10:%s: %s: Raid device exceeded "
2394 "read_error threshold [cur %d:max %d]\n",
2395 mdname(mddev), b,
2396 atomic_read(&rdev->read_errors), max_read_errors);
2397 printk(KERN_NOTICE
2398 "md/raid10:%s: %s: Failing raid device\n",
2399 mdname(mddev), b);
2400 md_error(mddev, conf->mirrors[d].rdev);
NeilBrownfae8cc52012-02-14 11:10:10 +11002401 r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED;
NeilBrown7c4e06f2011-05-11 14:53:17 +10002402 return;
Robert Becker1e509152009-12-14 12:49:58 +11002403 }
Robert Becker1e509152009-12-14 12:49:58 +11002404
NeilBrown6814d532006-10-03 01:15:45 -07002405 while(sectors) {
2406 int s = sectors;
2407 int sl = r10_bio->read_slot;
2408 int success = 0;
2409 int start;
2410
2411 if (s > (PAGE_SIZE>>9))
2412 s = PAGE_SIZE >> 9;
2413
2414 rcu_read_lock();
2415 do {
NeilBrown8dbed5c2011-07-28 11:39:24 +10002416 sector_t first_bad;
2417 int bad_sectors;
2418
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002419 d = r10_bio->devs[sl].devnum;
NeilBrown6814d532006-10-03 01:15:45 -07002420 rdev = rcu_dereference(conf->mirrors[d].rdev);
2421 if (rdev &&
NeilBrown050b6612012-03-19 12:46:39 +11002422 !test_bit(Unmerged, &rdev->flags) &&
NeilBrown8dbed5c2011-07-28 11:39:24 +10002423 test_bit(In_sync, &rdev->flags) &&
2424 is_badblock(rdev, r10_bio->devs[sl].addr + sect, s,
2425 &first_bad, &bad_sectors) == 0) {
NeilBrown6814d532006-10-03 01:15:45 -07002426 atomic_inc(&rdev->nr_pending);
2427 rcu_read_unlock();
NeilBrown2b193362010-10-27 15:16:40 +11002428 success = sync_page_io(rdev,
NeilBrown6814d532006-10-03 01:15:45 -07002429 r10_bio->devs[sl].addr +
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11002430 sect,
NeilBrown6814d532006-10-03 01:15:45 -07002431 s<<9,
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11002432 conf->tmppage, READ, false);
NeilBrown6814d532006-10-03 01:15:45 -07002433 rdev_dec_pending(rdev, mddev);
2434 rcu_read_lock();
2435 if (success)
2436 break;
2437 }
2438 sl++;
2439 if (sl == conf->copies)
2440 sl = 0;
2441 } while (!success && sl != r10_bio->read_slot);
2442 rcu_read_unlock();
2443
2444 if (!success) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002445 /* Cannot read from anywhere, just mark the block
2446 * as bad on the first device to discourage future
2447 * reads.
2448 */
NeilBrown6814d532006-10-03 01:15:45 -07002449 int dn = r10_bio->devs[r10_bio->read_slot].devnum;
NeilBrown58c54fc2011-07-28 11:39:25 +10002450 rdev = conf->mirrors[dn].rdev;
2451
2452 if (!rdev_set_badblocks(
2453 rdev,
2454 r10_bio->devs[r10_bio->read_slot].addr
2455 + sect,
NeilBrownfae8cc52012-02-14 11:10:10 +11002456 s, 0)) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002457 md_error(mddev, rdev);
NeilBrownfae8cc52012-02-14 11:10:10 +11002458 r10_bio->devs[r10_bio->read_slot].bio
2459 = IO_BLOCKED;
2460 }
NeilBrown6814d532006-10-03 01:15:45 -07002461 break;
2462 }
2463
2464 start = sl;
2465 /* write it back and re-read */
2466 rcu_read_lock();
2467 while (sl != r10_bio->read_slot) {
Robert Becker67b8dc42009-12-14 12:49:57 +11002468 char b[BDEVNAME_SIZE];
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002469
NeilBrown6814d532006-10-03 01:15:45 -07002470 if (sl==0)
2471 sl = conf->copies;
2472 sl--;
2473 d = r10_bio->devs[sl].devnum;
2474 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown1294b9c2011-07-28 11:39:23 +10002475 if (!rdev ||
NeilBrown050b6612012-03-19 12:46:39 +11002476 test_bit(Unmerged, &rdev->flags) ||
NeilBrown1294b9c2011-07-28 11:39:23 +10002477 !test_bit(In_sync, &rdev->flags))
2478 continue;
2479
2480 atomic_inc(&rdev->nr_pending);
2481 rcu_read_unlock();
NeilBrown58c54fc2011-07-28 11:39:25 +10002482 if (r10_sync_page_io(rdev,
2483 r10_bio->devs[sl].addr +
2484 sect,
NeilBrown055d3742012-07-03 15:55:33 +10002485 s, conf->tmppage, WRITE)
NeilBrown1294b9c2011-07-28 11:39:23 +10002486 == 0) {
2487 /* Well, this device is dead */
2488 printk(KERN_NOTICE
2489 "md/raid10:%s: read correction "
2490 "write failed"
2491 " (%d sectors at %llu on %s)\n",
2492 mdname(mddev), s,
2493 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002494 sect +
2495 choose_data_offset(r10_bio,
2496 rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002497 bdevname(rdev->bdev, b));
2498 printk(KERN_NOTICE "md/raid10:%s: %s: failing "
2499 "drive\n",
2500 mdname(mddev),
2501 bdevname(rdev->bdev, b));
NeilBrown6814d532006-10-03 01:15:45 -07002502 }
NeilBrown1294b9c2011-07-28 11:39:23 +10002503 rdev_dec_pending(rdev, mddev);
2504 rcu_read_lock();
NeilBrown6814d532006-10-03 01:15:45 -07002505 }
2506 sl = start;
2507 while (sl != r10_bio->read_slot) {
NeilBrown1294b9c2011-07-28 11:39:23 +10002508 char b[BDEVNAME_SIZE];
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002509
NeilBrown6814d532006-10-03 01:15:45 -07002510 if (sl==0)
2511 sl = conf->copies;
2512 sl--;
2513 d = r10_bio->devs[sl].devnum;
2514 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown1294b9c2011-07-28 11:39:23 +10002515 if (!rdev ||
2516 !test_bit(In_sync, &rdev->flags))
2517 continue;
Robert Becker67b8dc42009-12-14 12:49:57 +11002518
NeilBrown1294b9c2011-07-28 11:39:23 +10002519 atomic_inc(&rdev->nr_pending);
2520 rcu_read_unlock();
NeilBrown58c54fc2011-07-28 11:39:25 +10002521 switch (r10_sync_page_io(rdev,
2522 r10_bio->devs[sl].addr +
2523 sect,
NeilBrown055d3742012-07-03 15:55:33 +10002524 s, conf->tmppage,
NeilBrown58c54fc2011-07-28 11:39:25 +10002525 READ)) {
2526 case 0:
NeilBrown1294b9c2011-07-28 11:39:23 +10002527 /* Well, this device is dead */
2528 printk(KERN_NOTICE
2529 "md/raid10:%s: unable to read back "
2530 "corrected sectors"
2531 " (%d sectors at %llu on %s)\n",
2532 mdname(mddev), s,
2533 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002534 sect +
2535 choose_data_offset(r10_bio, rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002536 bdevname(rdev->bdev, b));
2537 printk(KERN_NOTICE "md/raid10:%s: %s: failing "
2538 "drive\n",
2539 mdname(mddev),
2540 bdevname(rdev->bdev, b));
NeilBrown58c54fc2011-07-28 11:39:25 +10002541 break;
2542 case 1:
NeilBrown1294b9c2011-07-28 11:39:23 +10002543 printk(KERN_INFO
2544 "md/raid10:%s: read error corrected"
2545 " (%d sectors at %llu on %s)\n",
2546 mdname(mddev), s,
2547 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002548 sect +
2549 choose_data_offset(r10_bio, rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002550 bdevname(rdev->bdev, b));
2551 atomic_add(s, &rdev->corrected_errors);
NeilBrown6814d532006-10-03 01:15:45 -07002552 }
NeilBrown1294b9c2011-07-28 11:39:23 +10002553
2554 rdev_dec_pending(rdev, mddev);
2555 rcu_read_lock();
NeilBrown6814d532006-10-03 01:15:45 -07002556 }
2557 rcu_read_unlock();
2558
2559 sectors -= s;
2560 sect += s;
2561 }
2562}
2563
NeilBrown9f2c9d12011-10-11 16:48:43 +11002564static int narrow_write_error(struct r10bio *r10_bio, int i)
NeilBrownbd870a12011-07-28 11:39:24 +10002565{
2566 struct bio *bio = r10_bio->master_bio;
NeilBrownfd01b882011-10-11 16:47:53 +11002567 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11002568 struct r10conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002569 struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev;
NeilBrownbd870a12011-07-28 11:39:24 +10002570 /* bio has the data to be written to slot 'i' where
2571 * we just recently had a write error.
2572 * We repeatedly clone the bio and trim down to one block,
2573 * then try the write. Where the write fails we record
2574 * a bad block.
2575 * It is conceivable that the bio doesn't exactly align with
2576 * blocks. We must handle this.
2577 *
2578 * We currently own a reference to the rdev.
2579 */
2580
2581 int block_sectors;
2582 sector_t sector;
2583 int sectors;
2584 int sect_to_write = r10_bio->sectors;
2585 int ok = 1;
2586
2587 if (rdev->badblocks.shift < 0)
2588 return 0;
2589
2590 block_sectors = 1 << rdev->badblocks.shift;
2591 sector = r10_bio->sector;
2592 sectors = ((r10_bio->sector + block_sectors)
2593 & ~(sector_t)(block_sectors - 1))
2594 - sector;
2595
2596 while (sect_to_write) {
2597 struct bio *wbio;
2598 if (sectors > sect_to_write)
2599 sectors = sect_to_write;
2600 /* Write at 'sector' for 'sectors' */
2601 wbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002602 bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors);
2603 wbio->bi_iter.bi_sector = (r10_bio->devs[i].addr+
NeilBrownf8c9e742012-05-21 09:28:33 +10002604 choose_data_offset(r10_bio, rdev) +
NeilBrownbd870a12011-07-28 11:39:24 +10002605 (sector - r10_bio->sector));
2606 wbio->bi_bdev = rdev->bdev;
2607 if (submit_bio_wait(WRITE, wbio) == 0)
2608 /* Failure! */
2609 ok = rdev_set_badblocks(rdev, sector,
2610 sectors, 0)
2611 && ok;
2612
2613 bio_put(wbio);
2614 sect_to_write -= sectors;
2615 sector += sectors;
2616 sectors = block_sectors;
2617 }
2618 return ok;
2619}
2620
NeilBrown9f2c9d12011-10-11 16:48:43 +11002621static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio)
NeilBrown560f8e52011-07-28 11:39:23 +10002622{
2623 int slot = r10_bio->read_slot;
NeilBrown560f8e52011-07-28 11:39:23 +10002624 struct bio *bio;
NeilBrowne879a872011-10-11 16:49:02 +11002625 struct r10conf *conf = mddev->private;
NeilBrownabbf0982011-12-23 10:17:54 +11002626 struct md_rdev *rdev = r10_bio->devs[slot].rdev;
NeilBrown560f8e52011-07-28 11:39:23 +10002627 char b[BDEVNAME_SIZE];
2628 unsigned long do_sync;
NeilBrown856e08e2011-07-28 11:39:23 +10002629 int max_sectors;
NeilBrown560f8e52011-07-28 11:39:23 +10002630
2631 /* we got a read error. Maybe the drive is bad. Maybe just
2632 * the block and we can fix it.
2633 * We freeze all other IO, and try reading the block from
2634 * other devices. When we find one, we re-write
2635 * and check it that fixes the read error.
2636 * This is all done synchronously while the array is
2637 * frozen.
2638 */
NeilBrownfae8cc52012-02-14 11:10:10 +11002639 bio = r10_bio->devs[slot].bio;
2640 bdevname(bio->bi_bdev, b);
2641 bio_put(bio);
2642 r10_bio->devs[slot].bio = NULL;
2643
NeilBrown560f8e52011-07-28 11:39:23 +10002644 if (mddev->ro == 0) {
NeilBrowne2d59922013-06-12 11:01:22 +10002645 freeze_array(conf, 1);
NeilBrown560f8e52011-07-28 11:39:23 +10002646 fix_read_error(conf, mddev, r10_bio);
2647 unfreeze_array(conf);
NeilBrownfae8cc52012-02-14 11:10:10 +11002648 } else
2649 r10_bio->devs[slot].bio = IO_BLOCKED;
2650
NeilBrownabbf0982011-12-23 10:17:54 +11002651 rdev_dec_pending(rdev, mddev);
NeilBrown560f8e52011-07-28 11:39:23 +10002652
NeilBrown7399c312011-07-28 11:39:23 +10002653read_more:
NeilBrown96c3fd12011-12-23 10:17:54 +11002654 rdev = read_balance(conf, r10_bio, &max_sectors);
2655 if (rdev == NULL) {
NeilBrown560f8e52011-07-28 11:39:23 +10002656 printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O"
2657 " read error for block %llu\n",
NeilBrown7399c312011-07-28 11:39:23 +10002658 mdname(mddev), b,
NeilBrown560f8e52011-07-28 11:39:23 +10002659 (unsigned long long)r10_bio->sector);
2660 raid_end_bio_io(r10_bio);
NeilBrown560f8e52011-07-28 11:39:23 +10002661 return;
2662 }
2663
2664 do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC);
NeilBrown560f8e52011-07-28 11:39:23 +10002665 slot = r10_bio->read_slot;
NeilBrown560f8e52011-07-28 11:39:23 +10002666 printk_ratelimited(
2667 KERN_ERR
NeilBrown055d3742012-07-03 15:55:33 +10002668 "md/raid10:%s: %s: redirecting "
NeilBrown560f8e52011-07-28 11:39:23 +10002669 "sector %llu to another mirror\n",
2670 mdname(mddev),
2671 bdevname(rdev->bdev, b),
2672 (unsigned long long)r10_bio->sector);
2673 bio = bio_clone_mddev(r10_bio->master_bio,
2674 GFP_NOIO, mddev);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002675 bio_trim(bio, r10_bio->sector - bio->bi_iter.bi_sector, max_sectors);
NeilBrown560f8e52011-07-28 11:39:23 +10002676 r10_bio->devs[slot].bio = bio;
NeilBrownabbf0982011-12-23 10:17:54 +11002677 r10_bio->devs[slot].rdev = rdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002678 bio->bi_iter.bi_sector = r10_bio->devs[slot].addr
NeilBrownf8c9e742012-05-21 09:28:33 +10002679 + choose_data_offset(r10_bio, rdev);
NeilBrown560f8e52011-07-28 11:39:23 +10002680 bio->bi_bdev = rdev->bdev;
2681 bio->bi_rw = READ | do_sync;
2682 bio->bi_private = r10_bio;
2683 bio->bi_end_io = raid10_end_read_request;
NeilBrown7399c312011-07-28 11:39:23 +10002684 if (max_sectors < r10_bio->sectors) {
2685 /* Drat - have to split this up more */
2686 struct bio *mbio = r10_bio->master_bio;
2687 int sectors_handled =
2688 r10_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07002689 - mbio->bi_iter.bi_sector;
NeilBrown7399c312011-07-28 11:39:23 +10002690 r10_bio->sectors = max_sectors;
2691 spin_lock_irq(&conf->device_lock);
2692 if (mbio->bi_phys_segments == 0)
2693 mbio->bi_phys_segments = 2;
2694 else
2695 mbio->bi_phys_segments++;
2696 spin_unlock_irq(&conf->device_lock);
2697 generic_make_request(bio);
NeilBrown7399c312011-07-28 11:39:23 +10002698
2699 r10_bio = mempool_alloc(conf->r10bio_pool,
2700 GFP_NOIO);
2701 r10_bio->master_bio = mbio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002702 r10_bio->sectors = bio_sectors(mbio) - sectors_handled;
NeilBrown7399c312011-07-28 11:39:23 +10002703 r10_bio->state = 0;
2704 set_bit(R10BIO_ReadError,
2705 &r10_bio->state);
2706 r10_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002707 r10_bio->sector = mbio->bi_iter.bi_sector
NeilBrown7399c312011-07-28 11:39:23 +10002708 + sectors_handled;
2709
2710 goto read_more;
2711 } else
2712 generic_make_request(bio);
NeilBrown560f8e52011-07-28 11:39:23 +10002713}
2714
NeilBrowne879a872011-10-11 16:49:02 +11002715static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio)
NeilBrown749c55e2011-07-28 11:39:24 +10002716{
2717 /* Some sort of write request has finished and it
2718 * succeeded in writing where we thought there was a
2719 * bad block. So forget the bad block.
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002720 * Or possibly if failed and we need to record
2721 * a bad block.
NeilBrown749c55e2011-07-28 11:39:24 +10002722 */
2723 int m;
NeilBrown3cb03002011-10-11 16:45:26 +11002724 struct md_rdev *rdev;
NeilBrown749c55e2011-07-28 11:39:24 +10002725
2726 if (test_bit(R10BIO_IsSync, &r10_bio->state) ||
2727 test_bit(R10BIO_IsRecover, &r10_bio->state)) {
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002728 for (m = 0; m < conf->copies; m++) {
2729 int dev = r10_bio->devs[m].devnum;
2730 rdev = conf->mirrors[dev].rdev;
2731 if (r10_bio->devs[m].bio == NULL)
2732 continue;
2733 if (test_bit(BIO_UPTODATE,
NeilBrown749c55e2011-07-28 11:39:24 +10002734 &r10_bio->devs[m].bio->bi_flags)) {
NeilBrown749c55e2011-07-28 11:39:24 +10002735 rdev_clear_badblocks(
2736 rdev,
2737 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002738 r10_bio->sectors, 0);
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002739 } else {
2740 if (!rdev_set_badblocks(
2741 rdev,
2742 r10_bio->devs[m].addr,
2743 r10_bio->sectors, 0))
2744 md_error(conf->mddev, rdev);
NeilBrown749c55e2011-07-28 11:39:24 +10002745 }
NeilBrown9ad1aef2011-12-23 10:17:55 +11002746 rdev = conf->mirrors[dev].replacement;
2747 if (r10_bio->devs[m].repl_bio == NULL)
2748 continue;
2749 if (test_bit(BIO_UPTODATE,
2750 &r10_bio->devs[m].repl_bio->bi_flags)) {
2751 rdev_clear_badblocks(
2752 rdev,
2753 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002754 r10_bio->sectors, 0);
NeilBrown9ad1aef2011-12-23 10:17:55 +11002755 } else {
2756 if (!rdev_set_badblocks(
2757 rdev,
2758 r10_bio->devs[m].addr,
2759 r10_bio->sectors, 0))
2760 md_error(conf->mddev, rdev);
2761 }
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002762 }
NeilBrown749c55e2011-07-28 11:39:24 +10002763 put_buf(r10_bio);
2764 } else {
NeilBrownbd870a12011-07-28 11:39:24 +10002765 for (m = 0; m < conf->copies; m++) {
2766 int dev = r10_bio->devs[m].devnum;
2767 struct bio *bio = r10_bio->devs[m].bio;
2768 rdev = conf->mirrors[dev].rdev;
2769 if (bio == IO_MADE_GOOD) {
NeilBrown749c55e2011-07-28 11:39:24 +10002770 rdev_clear_badblocks(
2771 rdev,
2772 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002773 r10_bio->sectors, 0);
NeilBrown749c55e2011-07-28 11:39:24 +10002774 rdev_dec_pending(rdev, conf->mddev);
NeilBrownbd870a12011-07-28 11:39:24 +10002775 } else if (bio != NULL &&
2776 !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
2777 if (!narrow_write_error(r10_bio, m)) {
2778 md_error(conf->mddev, rdev);
2779 set_bit(R10BIO_Degraded,
2780 &r10_bio->state);
2781 }
2782 rdev_dec_pending(rdev, conf->mddev);
NeilBrown749c55e2011-07-28 11:39:24 +10002783 }
NeilBrown475b0322011-12-23 10:17:55 +11002784 bio = r10_bio->devs[m].repl_bio;
2785 rdev = conf->mirrors[dev].replacement;
NeilBrown4ca40c22011-12-23 10:17:55 +11002786 if (rdev && bio == IO_MADE_GOOD) {
NeilBrown475b0322011-12-23 10:17:55 +11002787 rdev_clear_badblocks(
2788 rdev,
2789 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002790 r10_bio->sectors, 0);
NeilBrown475b0322011-12-23 10:17:55 +11002791 rdev_dec_pending(rdev, conf->mddev);
2792 }
NeilBrownbd870a12011-07-28 11:39:24 +10002793 }
2794 if (test_bit(R10BIO_WriteError,
2795 &r10_bio->state))
2796 close_write(r10_bio);
NeilBrown749c55e2011-07-28 11:39:24 +10002797 raid_end_bio_io(r10_bio);
2798 }
2799}
2800
Shaohua Li4ed87312012-10-11 13:34:00 +11002801static void raid10d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802{
Shaohua Li4ed87312012-10-11 13:34:00 +11002803 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002804 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 unsigned long flags;
NeilBrowne879a872011-10-11 16:49:02 +11002806 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002808 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
2810 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002812 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002814
NeilBrown0021b7b2012-07-31 09:08:14 +02002815 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002816
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002818 if (list_empty(head)) {
NeilBrown6cce3b232006-01-06 00:20:16 -08002819 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002821 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002822 r10_bio = list_entry(head->prev, struct r10bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 list_del(head->prev);
NeilBrown4443ae12006-01-06 00:20:28 -08002824 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 spin_unlock_irqrestore(&conf->device_lock, flags);
2826
2827 mddev = r10_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002828 conf = mddev->private;
NeilBrownbd870a12011-07-28 11:39:24 +10002829 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
2830 test_bit(R10BIO_WriteError, &r10_bio->state))
NeilBrown749c55e2011-07-28 11:39:24 +10002831 handle_write_completed(conf, r10_bio);
NeilBrown3ea7daa2012-05-22 13:53:47 +10002832 else if (test_bit(R10BIO_IsReshape, &r10_bio->state))
2833 reshape_request_write(mddev, r10_bio);
NeilBrown749c55e2011-07-28 11:39:24 +10002834 else if (test_bit(R10BIO_IsSync, &r10_bio->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 sync_request_write(mddev, r10_bio);
Jens Axboe7eaceac2011-03-10 08:52:07 +01002836 else if (test_bit(R10BIO_IsRecover, &r10_bio->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 recovery_request_write(mddev, r10_bio);
NeilBrown856e08e2011-07-28 11:39:23 +10002838 else if (test_bit(R10BIO_ReadError, &r10_bio->state))
NeilBrown560f8e52011-07-28 11:39:23 +10002839 handle_read_error(mddev, r10_bio);
NeilBrown856e08e2011-07-28 11:39:23 +10002840 else {
2841 /* just a partial read to be scheduled from a
2842 * separate context
2843 */
2844 int slot = r10_bio->read_slot;
2845 generic_make_request(r10_bio->devs[slot].bio);
2846 }
NeilBrown4443ae12006-01-06 00:20:28 -08002847
NeilBrown1d9d5242009-10-16 15:55:32 +11002848 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002849 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2850 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002852 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853}
2854
2855
NeilBrowne879a872011-10-11 16:49:02 +11002856static int init_resync(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857{
2858 int buffs;
NeilBrown69335ef2011-12-23 10:17:54 +11002859 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860
2861 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhennb6385482006-04-02 13:34:29 +02002862 BUG_ON(conf->r10buf_pool);
NeilBrown69335ef2011-12-23 10:17:54 +11002863 conf->have_replacement = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10002864 for (i = 0; i < conf->geo.raid_disks; i++)
NeilBrown69335ef2011-12-23 10:17:54 +11002865 if (conf->mirrors[i].replacement)
2866 conf->have_replacement = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf);
2868 if (!conf->r10buf_pool)
2869 return -ENOMEM;
2870 conf->next_resync = 0;
2871 return 0;
2872}
2873
2874/*
2875 * perform a "sync" on one "block"
2876 *
2877 * We need to make sure that no normal I/O request - particularly write
2878 * requests - conflict with active sync requests.
2879 *
2880 * This is achieved by tracking pending requests and a 'barrier' concept
2881 * that can be installed to exclude normal IO requests.
2882 *
2883 * Resync and recovery are handled very differently.
2884 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2885 *
2886 * For resync, we iterate over virtual addresses, read all copies,
2887 * and update if there are differences. If only one copy is live,
2888 * skip it.
2889 * For recovery, we iterate over physical addresses, read a good
2890 * value for each non-in_sync drive, and over-write.
2891 *
2892 * So, for recovery we may have several outstanding complex requests for a
2893 * given address, one for each out-of-sync device. We model this by allocating
2894 * a number of r10_bio structures, one for each out-of-sync device.
2895 * As we setup these structures, we collect all bio's together into a list
2896 * which we then process collectively to add pages, and then process again
2897 * to pass to generic_make_request.
2898 *
2899 * The r10_bio structures are linked using a borrowed master_bio pointer.
2900 * This link is counted in ->remaining. When the r10_bio that points to NULL
2901 * has its remaining count decremented to 0, the whole complex operation
2902 * is complete.
2903 *
2904 */
2905
NeilBrownfd01b882011-10-11 16:47:53 +11002906static sector_t sync_request(struct mddev *mddev, sector_t sector_nr,
NeilBrownab9d47e2011-05-11 14:54:41 +10002907 int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908{
NeilBrowne879a872011-10-11 16:49:02 +11002909 struct r10conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002910 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 struct bio *biolist = NULL, *bio;
2912 sector_t max_sector, nr_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 int i;
NeilBrown6cce3b232006-01-06 00:20:16 -08002914 int max_sync;
NeilBrown57dab0b2010-10-19 10:03:39 +11002915 sector_t sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 sector_t sectors_skipped = 0;
2917 int chunks_skipped = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10002918 sector_t chunk_mask = conf->geo.chunk_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
2920 if (!conf->r10buf_pool)
2921 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002922 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002924 /*
2925 * Allow skipping a full rebuild for incremental assembly
2926 * of a clean array, like RAID1 does.
2927 */
2928 if (mddev->bitmap == NULL &&
2929 mddev->recovery_cp == MaxSector &&
NeilBrown13765122013-07-04 16:41:53 +10002930 mddev->reshape_position == MaxSector &&
2931 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002932 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown13765122013-07-04 16:41:53 +10002933 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002934 conf->fullsync == 0) {
2935 *skipped = 1;
NeilBrown13765122013-07-04 16:41:53 +10002936 return mddev->dev_sectors - sector_nr;
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002937 }
2938
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 skipped:
Andre Noll58c0fed2009-03-31 14:33:13 +11002940 max_sector = mddev->dev_sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10002941 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
2942 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 max_sector = mddev->resync_max_sectors;
2944 if (sector_nr >= max_sector) {
NeilBrown6cce3b232006-01-06 00:20:16 -08002945 /* If we aborted, we need to abort the
2946 * sync on the 'current' bitmap chucks (there can
2947 * be several when recovering multiple devices).
2948 * as we may have started syncing it but not finished.
2949 * We can find the current address in
2950 * mddev->curr_resync, but for recovery,
2951 * we need to convert that to several
2952 * virtual addresses.
2953 */
NeilBrown3ea7daa2012-05-22 13:53:47 +10002954 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
2955 end_reshape(conf);
NeilBrownb3968552014-08-18 13:59:50 +10002956 close_sync(conf);
NeilBrown3ea7daa2012-05-22 13:53:47 +10002957 return 0;
2958 }
2959
NeilBrown6cce3b232006-01-06 00:20:16 -08002960 if (mddev->curr_resync < max_sector) { /* aborted */
2961 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2962 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2963 &sync_blocks, 1);
NeilBrown5cf00fc2012-05-21 09:28:20 +10002964 else for (i = 0; i < conf->geo.raid_disks; i++) {
NeilBrown6cce3b232006-01-06 00:20:16 -08002965 sector_t sect =
2966 raid10_find_virt(conf, mddev->curr_resync, i);
2967 bitmap_end_sync(mddev->bitmap, sect,
2968 &sync_blocks, 1);
2969 }
NeilBrown9ad1aef2011-12-23 10:17:55 +11002970 } else {
2971 /* completed sync */
2972 if ((!mddev->bitmap || conf->fullsync)
2973 && conf->have_replacement
2974 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2975 /* Completed a full sync so the replacements
2976 * are now fully recovered.
2977 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10002978 for (i = 0; i < conf->geo.raid_disks; i++)
NeilBrown9ad1aef2011-12-23 10:17:55 +11002979 if (conf->mirrors[i].replacement)
2980 conf->mirrors[i].replacement
2981 ->recovery_offset
2982 = MaxSector;
2983 }
NeilBrown6cce3b232006-01-06 00:20:16 -08002984 conf->fullsync = 0;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002985 }
NeilBrown6cce3b232006-01-06 00:20:16 -08002986 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 close_sync(conf);
NeilBrown57afd892005-06-21 17:17:13 -07002988 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 return sectors_skipped;
2990 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10002991
2992 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2993 return reshape_request(mddev, sector_nr, skipped);
2994
NeilBrown5cf00fc2012-05-21 09:28:20 +10002995 if (chunks_skipped >= conf->geo.raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 /* if there has been nothing to do on any drive,
2997 * then there is nothing to do at all..
2998 */
NeilBrown57afd892005-06-21 17:17:13 -07002999 *skipped = 1;
3000 return (max_sector - sector_nr) + sectors_skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 }
3002
NeilBrownc6207272008-02-06 01:39:52 -08003003 if (max_sector > mddev->resync_max)
3004 max_sector = mddev->resync_max; /* Don't do IO beyond here */
3005
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 /* make sure whole request will fit in a chunk - if chunks
3007 * are meaningful
3008 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003009 if (conf->geo.near_copies < conf->geo.raid_disks &&
3010 max_sector > (sector_nr | chunk_mask))
3011 max_sector = (sector_nr | chunk_mask) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 /*
3013 * If there is non-resync activity waiting for us then
3014 * put in a delay to throttle resync.
3015 */
NeilBrown0a27ec92006-01-06 00:20:13 -08003016 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 msleep_interruptible(1000);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018
3019 /* Again, very different code for resync and recovery.
3020 * Both must result in an r10bio with a list of bios that
3021 * have bi_end_io, bi_sector, bi_bdev set,
3022 * and bi_private set to the r10bio.
3023 * For recovery, we may actually create several r10bios
3024 * with 2 bios in each, that correspond to the bios in the main one.
3025 * In this case, the subordinate r10bios link back through a
3026 * borrowed master_bio pointer, and the counter in the master
3027 * includes a ref from each subordinate.
3028 */
3029 /* First, we decide what to do and set ->bi_end_io
3030 * To end_sync_read if we want to read, and
3031 * end_sync_write if we will want to write.
3032 */
3033
NeilBrown6cce3b232006-01-06 00:20:16 -08003034 max_sync = RESYNC_PAGES << (PAGE_SHIFT-9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3036 /* recovery... the complicated one */
NeilBrowne875ece2011-07-28 11:39:24 +10003037 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 r10_bio = NULL;
3039
NeilBrown5cf00fc2012-05-21 09:28:20 +10003040 for (i = 0 ; i < conf->geo.raid_disks; i++) {
NeilBrownab9d47e2011-05-11 14:54:41 +10003041 int still_degraded;
NeilBrown9f2c9d12011-10-11 16:48:43 +11003042 struct r10bio *rb2;
NeilBrownab9d47e2011-05-11 14:54:41 +10003043 sector_t sect;
3044 int must_sync;
NeilBrowne875ece2011-07-28 11:39:24 +10003045 int any_working;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003046 struct raid10_info *mirror = &conf->mirrors[i];
NeilBrownab9d47e2011-05-11 14:54:41 +10003047
NeilBrown24afd802011-12-23 10:17:55 +11003048 if ((mirror->rdev == NULL ||
3049 test_bit(In_sync, &mirror->rdev->flags))
3050 &&
3051 (mirror->replacement == NULL ||
3052 test_bit(Faulty,
3053 &mirror->replacement->flags)))
NeilBrownab9d47e2011-05-11 14:54:41 +10003054 continue;
3055
3056 still_degraded = 0;
3057 /* want to reconstruct this device */
3058 rb2 = r10_bio;
3059 sect = raid10_find_virt(conf, sector_nr, i);
NeilBrownfc448a12012-07-03 10:37:30 +10003060 if (sect >= mddev->resync_max_sectors) {
3061 /* last stripe is not complete - don't
3062 * try to recover this sector.
3063 */
3064 continue;
3065 }
NeilBrown24afd802011-12-23 10:17:55 +11003066 /* Unless we are doing a full sync, or a replacement
3067 * we only need to recover the block if it is set in
3068 * the bitmap
NeilBrownab9d47e2011-05-11 14:54:41 +10003069 */
3070 must_sync = bitmap_start_sync(mddev->bitmap, sect,
3071 &sync_blocks, 1);
3072 if (sync_blocks < max_sync)
3073 max_sync = sync_blocks;
3074 if (!must_sync &&
NeilBrown24afd802011-12-23 10:17:55 +11003075 mirror->replacement == NULL &&
NeilBrownab9d47e2011-05-11 14:54:41 +10003076 !conf->fullsync) {
3077 /* yep, skip the sync_blocks here, but don't assume
3078 * that there will never be anything to do here
NeilBrown6cce3b232006-01-06 00:20:16 -08003079 */
NeilBrownab9d47e2011-05-11 14:54:41 +10003080 chunks_skipped = -1;
3081 continue;
3082 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
NeilBrownab9d47e2011-05-11 14:54:41 +10003084 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
NeilBrowncb8b12b2014-08-18 14:38:45 +10003085 r10_bio->state = 0;
NeilBrownab9d47e2011-05-11 14:54:41 +10003086 raise_barrier(conf, rb2 != NULL);
3087 atomic_set(&r10_bio->remaining, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
NeilBrownab9d47e2011-05-11 14:54:41 +10003089 r10_bio->master_bio = (struct bio*)rb2;
3090 if (rb2)
3091 atomic_inc(&rb2->remaining);
3092 r10_bio->mddev = mddev;
3093 set_bit(R10BIO_IsRecover, &r10_bio->state);
3094 r10_bio->sector = sect;
NeilBrown6cce3b232006-01-06 00:20:16 -08003095
NeilBrownab9d47e2011-05-11 14:54:41 +10003096 raid10_find_phys(conf, r10_bio);
NeilBrown18055562009-05-07 12:48:10 +10003097
NeilBrownab9d47e2011-05-11 14:54:41 +10003098 /* Need to check if the array will still be
3099 * degraded
3100 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003101 for (j = 0; j < conf->geo.raid_disks; j++)
NeilBrownab9d47e2011-05-11 14:54:41 +10003102 if (conf->mirrors[j].rdev == NULL ||
3103 test_bit(Faulty, &conf->mirrors[j].rdev->flags)) {
3104 still_degraded = 1;
NeilBrown87fc7672005-09-09 16:24:04 -07003105 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003107
3108 must_sync = bitmap_start_sync(mddev->bitmap, sect,
3109 &sync_blocks, still_degraded);
3110
NeilBrowne875ece2011-07-28 11:39:24 +10003111 any_working = 0;
NeilBrownab9d47e2011-05-11 14:54:41 +10003112 for (j=0; j<conf->copies;j++) {
NeilBrowne875ece2011-07-28 11:39:24 +10003113 int k;
NeilBrownab9d47e2011-05-11 14:54:41 +10003114 int d = r10_bio->devs[j].devnum;
NeilBrown5e570282011-07-28 11:39:25 +10003115 sector_t from_addr, to_addr;
NeilBrown3cb03002011-10-11 16:45:26 +11003116 struct md_rdev *rdev;
NeilBrown40c356c2011-07-28 11:39:24 +10003117 sector_t sector, first_bad;
3118 int bad_sectors;
NeilBrownab9d47e2011-05-11 14:54:41 +10003119 if (!conf->mirrors[d].rdev ||
3120 !test_bit(In_sync, &conf->mirrors[d].rdev->flags))
3121 continue;
3122 /* This is where we read from */
NeilBrowne875ece2011-07-28 11:39:24 +10003123 any_working = 1;
NeilBrown40c356c2011-07-28 11:39:24 +10003124 rdev = conf->mirrors[d].rdev;
3125 sector = r10_bio->devs[j].addr;
3126
3127 if (is_badblock(rdev, sector, max_sync,
3128 &first_bad, &bad_sectors)) {
3129 if (first_bad > sector)
3130 max_sync = first_bad - sector;
3131 else {
3132 bad_sectors -= (sector
3133 - first_bad);
3134 if (max_sync > bad_sectors)
3135 max_sync = bad_sectors;
3136 continue;
3137 }
3138 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003139 bio = r10_bio->devs[0].bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003140 bio_reset(bio);
NeilBrownab9d47e2011-05-11 14:54:41 +10003141 bio->bi_next = biolist;
3142 biolist = bio;
3143 bio->bi_private = r10_bio;
3144 bio->bi_end_io = end_sync_read;
3145 bio->bi_rw = READ;
NeilBrown5e570282011-07-28 11:39:25 +10003146 from_addr = r10_bio->devs[j].addr;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003147 bio->bi_iter.bi_sector = from_addr +
3148 rdev->data_offset;
NeilBrown24afd802011-12-23 10:17:55 +11003149 bio->bi_bdev = rdev->bdev;
3150 atomic_inc(&rdev->nr_pending);
3151 /* and we write to 'i' (if not in_sync) */
NeilBrownab9d47e2011-05-11 14:54:41 +10003152
3153 for (k=0; k<conf->copies; k++)
3154 if (r10_bio->devs[k].devnum == i)
3155 break;
3156 BUG_ON(k == conf->copies);
NeilBrown5e570282011-07-28 11:39:25 +10003157 to_addr = r10_bio->devs[k].addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003158 r10_bio->devs[0].devnum = d;
NeilBrown5e570282011-07-28 11:39:25 +10003159 r10_bio->devs[0].addr = from_addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003160 r10_bio->devs[1].devnum = i;
NeilBrown5e570282011-07-28 11:39:25 +10003161 r10_bio->devs[1].addr = to_addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003162
NeilBrown24afd802011-12-23 10:17:55 +11003163 rdev = mirror->rdev;
3164 if (!test_bit(In_sync, &rdev->flags)) {
3165 bio = r10_bio->devs[1].bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003166 bio_reset(bio);
NeilBrown24afd802011-12-23 10:17:55 +11003167 bio->bi_next = biolist;
3168 biolist = bio;
3169 bio->bi_private = r10_bio;
3170 bio->bi_end_io = end_sync_write;
3171 bio->bi_rw = WRITE;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003172 bio->bi_iter.bi_sector = to_addr
NeilBrown24afd802011-12-23 10:17:55 +11003173 + rdev->data_offset;
3174 bio->bi_bdev = rdev->bdev;
3175 atomic_inc(&r10_bio->remaining);
3176 } else
3177 r10_bio->devs[1].bio->bi_end_io = NULL;
3178
3179 /* and maybe write to replacement */
3180 bio = r10_bio->devs[1].repl_bio;
3181 if (bio)
3182 bio->bi_end_io = NULL;
3183 rdev = mirror->replacement;
3184 /* Note: if rdev != NULL, then bio
3185 * cannot be NULL as r10buf_pool_alloc will
3186 * have allocated it.
3187 * So the second test here is pointless.
3188 * But it keeps semantic-checkers happy, and
3189 * this comment keeps human reviewers
3190 * happy.
3191 */
3192 if (rdev == NULL || bio == NULL ||
3193 test_bit(Faulty, &rdev->flags))
3194 break;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003195 bio_reset(bio);
NeilBrown24afd802011-12-23 10:17:55 +11003196 bio->bi_next = biolist;
3197 biolist = bio;
3198 bio->bi_private = r10_bio;
3199 bio->bi_end_io = end_sync_write;
3200 bio->bi_rw = WRITE;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003201 bio->bi_iter.bi_sector = to_addr +
3202 rdev->data_offset;
NeilBrown24afd802011-12-23 10:17:55 +11003203 bio->bi_bdev = rdev->bdev;
3204 atomic_inc(&r10_bio->remaining);
NeilBrownab9d47e2011-05-11 14:54:41 +10003205 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003207 if (j == conf->copies) {
NeilBrowne875ece2011-07-28 11:39:24 +10003208 /* Cannot recover, so abort the recovery or
3209 * record a bad block */
NeilBrowne875ece2011-07-28 11:39:24 +10003210 if (any_working) {
3211 /* problem is that there are bad blocks
3212 * on other device(s)
3213 */
3214 int k;
3215 for (k = 0; k < conf->copies; k++)
3216 if (r10_bio->devs[k].devnum == i)
3217 break;
NeilBrown24afd802011-12-23 10:17:55 +11003218 if (!test_bit(In_sync,
3219 &mirror->rdev->flags)
3220 && !rdev_set_badblocks(
3221 mirror->rdev,
3222 r10_bio->devs[k].addr,
3223 max_sync, 0))
3224 any_working = 0;
3225 if (mirror->replacement &&
3226 !rdev_set_badblocks(
3227 mirror->replacement,
NeilBrowne875ece2011-07-28 11:39:24 +10003228 r10_bio->devs[k].addr,
3229 max_sync, 0))
3230 any_working = 0;
3231 }
3232 if (!any_working) {
3233 if (!test_and_set_bit(MD_RECOVERY_INTR,
3234 &mddev->recovery))
3235 printk(KERN_INFO "md/raid10:%s: insufficient "
3236 "working devices for recovery.\n",
3237 mdname(mddev));
NeilBrown24afd802011-12-23 10:17:55 +11003238 mirror->recovery_disabled
NeilBrowne875ece2011-07-28 11:39:24 +10003239 = mddev->recovery_disabled;
3240 }
NeilBrowne8b84912014-01-06 10:35:34 +11003241 put_buf(r10_bio);
3242 if (rb2)
3243 atomic_dec(&rb2->remaining);
3244 r10_bio = rb2;
NeilBrownab9d47e2011-05-11 14:54:41 +10003245 break;
3246 }
3247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 if (biolist == NULL) {
3249 while (r10_bio) {
NeilBrown9f2c9d12011-10-11 16:48:43 +11003250 struct r10bio *rb2 = r10_bio;
3251 r10_bio = (struct r10bio*) rb2->master_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252 rb2->master_bio = NULL;
3253 put_buf(rb2);
3254 }
3255 goto giveup;
3256 }
3257 } else {
3258 /* resync. Schedule a read for every block at this virt offset */
3259 int count = 0;
NeilBrown6cce3b232006-01-06 00:20:16 -08003260
NeilBrown78200d42009-02-25 13:18:47 +11003261 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
3262
NeilBrown6cce3b232006-01-06 00:20:16 -08003263 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
3264 &sync_blocks, mddev->degraded) &&
NeilBrownab9d47e2011-05-11 14:54:41 +10003265 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED,
3266 &mddev->recovery)) {
NeilBrown6cce3b232006-01-06 00:20:16 -08003267 /* We can skip this block */
3268 *skipped = 1;
3269 return sync_blocks + sectors_skipped;
3270 }
3271 if (sync_blocks < max_sync)
3272 max_sync = sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
NeilBrowncb8b12b2014-08-18 14:38:45 +10003274 r10_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 r10_bio->mddev = mddev;
3277 atomic_set(&r10_bio->remaining, 0);
NeilBrown6cce3b232006-01-06 00:20:16 -08003278 raise_barrier(conf, 0);
3279 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280
3281 r10_bio->master_bio = NULL;
3282 r10_bio->sector = sector_nr;
3283 set_bit(R10BIO_IsSync, &r10_bio->state);
3284 raid10_find_phys(conf, r10_bio);
NeilBrown5cf00fc2012-05-21 09:28:20 +10003285 r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286
NeilBrown5cf00fc2012-05-21 09:28:20 +10003287 for (i = 0; i < conf->copies; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 int d = r10_bio->devs[i].devnum;
NeilBrown40c356c2011-07-28 11:39:24 +10003289 sector_t first_bad, sector;
3290 int bad_sectors;
3291
NeilBrown9ad1aef2011-12-23 10:17:55 +11003292 if (r10_bio->devs[i].repl_bio)
3293 r10_bio->devs[i].repl_bio->bi_end_io = NULL;
3294
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 bio = r10_bio->devs[i].bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003296 bio_reset(bio);
NeilBrownaf03b8e2007-06-16 10:16:06 -07003297 clear_bit(BIO_UPTODATE, &bio->bi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 if (conf->mirrors[d].rdev == NULL ||
NeilBrownb2d444d2005-11-08 21:39:31 -08003299 test_bit(Faulty, &conf->mirrors[d].rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 continue;
NeilBrown40c356c2011-07-28 11:39:24 +10003301 sector = r10_bio->devs[i].addr;
3302 if (is_badblock(conf->mirrors[d].rdev,
3303 sector, max_sync,
3304 &first_bad, &bad_sectors)) {
3305 if (first_bad > sector)
3306 max_sync = first_bad - sector;
3307 else {
3308 bad_sectors -= (sector - first_bad);
3309 if (max_sync > bad_sectors)
Dan Carpenter91502f02012-10-11 14:20:58 +11003310 max_sync = bad_sectors;
NeilBrown40c356c2011-07-28 11:39:24 +10003311 continue;
3312 }
3313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
3315 atomic_inc(&r10_bio->remaining);
3316 bio->bi_next = biolist;
3317 biolist = bio;
3318 bio->bi_private = r10_bio;
3319 bio->bi_end_io = end_sync_read;
NeilBrown802ba062006-12-13 00:34:13 -08003320 bio->bi_rw = READ;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003321 bio->bi_iter.bi_sector = sector +
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 conf->mirrors[d].rdev->data_offset;
3323 bio->bi_bdev = conf->mirrors[d].rdev->bdev;
3324 count++;
NeilBrown9ad1aef2011-12-23 10:17:55 +11003325
3326 if (conf->mirrors[d].replacement == NULL ||
3327 test_bit(Faulty,
3328 &conf->mirrors[d].replacement->flags))
3329 continue;
3330
3331 /* Need to set up for writing to the replacement */
3332 bio = r10_bio->devs[i].repl_bio;
Kent Overstreet8be185f2012-09-06 14:14:43 -07003333 bio_reset(bio);
NeilBrown9ad1aef2011-12-23 10:17:55 +11003334 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3335
3336 sector = r10_bio->devs[i].addr;
3337 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
3338 bio->bi_next = biolist;
3339 biolist = bio;
3340 bio->bi_private = r10_bio;
3341 bio->bi_end_io = end_sync_write;
3342 bio->bi_rw = WRITE;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003343 bio->bi_iter.bi_sector = sector +
NeilBrown9ad1aef2011-12-23 10:17:55 +11003344 conf->mirrors[d].replacement->data_offset;
3345 bio->bi_bdev = conf->mirrors[d].replacement->bdev;
3346 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 }
3348
3349 if (count < 2) {
3350 for (i=0; i<conf->copies; i++) {
3351 int d = r10_bio->devs[i].devnum;
3352 if (r10_bio->devs[i].bio->bi_end_io)
NeilBrownab9d47e2011-05-11 14:54:41 +10003353 rdev_dec_pending(conf->mirrors[d].rdev,
3354 mddev);
NeilBrown9ad1aef2011-12-23 10:17:55 +11003355 if (r10_bio->devs[i].repl_bio &&
3356 r10_bio->devs[i].repl_bio->bi_end_io)
3357 rdev_dec_pending(
3358 conf->mirrors[d].replacement,
3359 mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 }
3361 put_buf(r10_bio);
3362 biolist = NULL;
3363 goto giveup;
3364 }
3365 }
3366
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 nr_sectors = 0;
NeilBrown6cce3b232006-01-06 00:20:16 -08003368 if (sector_nr + max_sync < max_sector)
3369 max_sector = sector_nr + max_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 do {
3371 struct page *page;
3372 int len = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 if (sector_nr + (len>>9) > max_sector)
3374 len = (max_sector - sector_nr) << 9;
3375 if (len == 0)
3376 break;
3377 for (bio= biolist ; bio ; bio=bio->bi_next) {
NeilBrownab9d47e2011-05-11 14:54:41 +10003378 struct bio *bio2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
NeilBrownab9d47e2011-05-11 14:54:41 +10003380 if (bio_add_page(bio, page, len, 0))
3381 continue;
3382
3383 /* stop here */
3384 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
3385 for (bio2 = biolist;
3386 bio2 && bio2 != bio;
3387 bio2 = bio2->bi_next) {
3388 /* remove last page from this bio */
3389 bio2->bi_vcnt--;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003390 bio2->bi_iter.bi_size -= len;
NeilBrownab9d47e2011-05-11 14:54:41 +10003391 bio2->bi_flags &= ~(1<< BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003393 goto bio_full;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 }
3395 nr_sectors += len>>9;
3396 sector_nr += len>>9;
3397 } while (biolist->bi_vcnt < RESYNC_PAGES);
3398 bio_full:
3399 r10_bio->sectors = nr_sectors;
3400
3401 while (biolist) {
3402 bio = biolist;
3403 biolist = biolist->bi_next;
3404
3405 bio->bi_next = NULL;
3406 r10_bio = bio->bi_private;
3407 r10_bio->sectors = nr_sectors;
3408
3409 if (bio->bi_end_io == end_sync_read) {
3410 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrown7bb23c42013-07-16 16:50:47 +10003411 set_bit(BIO_UPTODATE, &bio->bi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412 generic_make_request(bio);
3413 }
3414 }
3415
NeilBrown57afd892005-06-21 17:17:13 -07003416 if (sectors_skipped)
3417 /* pretend they weren't skipped, it makes
3418 * no important difference in this case
3419 */
3420 md_done_sync(mddev, sectors_skipped, 1);
3421
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 return sectors_skipped + nr_sectors;
3423 giveup:
3424 /* There is nowhere to write, so all non-sync
NeilBrowne875ece2011-07-28 11:39:24 +10003425 * drives must be failed or in resync, all drives
3426 * have a bad block, so try the next chunk...
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 */
NeilBrown09b40682009-02-25 13:18:47 +11003428 if (sector_nr + max_sync < max_sector)
3429 max_sector = sector_nr + max_sync;
3430
3431 sectors_skipped += (max_sector - sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 chunks_skipped ++;
3433 sector_nr = max_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 goto skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435}
3436
Dan Williams80c3a6c2009-03-17 18:10:40 -07003437static sector_t
NeilBrownfd01b882011-10-11 16:47:53 +11003438raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07003439{
3440 sector_t size;
NeilBrowne879a872011-10-11 16:49:02 +11003441 struct r10conf *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003442
3443 if (!raid_disks)
NeilBrown3ea7daa2012-05-22 13:53:47 +10003444 raid_disks = min(conf->geo.raid_disks,
3445 conf->prev.raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003446 if (!sectors)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003447 sectors = conf->dev_sectors;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003448
NeilBrown5cf00fc2012-05-21 09:28:20 +10003449 size = sectors >> conf->geo.chunk_shift;
3450 sector_div(size, conf->geo.far_copies);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003451 size = size * raid_disks;
NeilBrown5cf00fc2012-05-21 09:28:20 +10003452 sector_div(size, conf->geo.near_copies);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003453
NeilBrown5cf00fc2012-05-21 09:28:20 +10003454 return size << conf->geo.chunk_shift;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003455}
3456
NeilBrown6508fdb2012-05-17 10:08:45 +10003457static void calc_sectors(struct r10conf *conf, sector_t size)
3458{
3459 /* Calculate the number of sectors-per-device that will
3460 * actually be used, and set conf->dev_sectors and
3461 * conf->stride
3462 */
3463
NeilBrown5cf00fc2012-05-21 09:28:20 +10003464 size = size >> conf->geo.chunk_shift;
3465 sector_div(size, conf->geo.far_copies);
3466 size = size * conf->geo.raid_disks;
3467 sector_div(size, conf->geo.near_copies);
NeilBrown6508fdb2012-05-17 10:08:45 +10003468 /* 'size' is now the number of chunks in the array */
3469 /* calculate "used chunks per device" */
3470 size = size * conf->copies;
3471
3472 /* We need to round up when dividing by raid_disks to
3473 * get the stride size.
3474 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003475 size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks);
NeilBrown6508fdb2012-05-17 10:08:45 +10003476
NeilBrown5cf00fc2012-05-21 09:28:20 +10003477 conf->dev_sectors = size << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003478
NeilBrown5cf00fc2012-05-21 09:28:20 +10003479 if (conf->geo.far_offset)
3480 conf->geo.stride = 1 << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003481 else {
NeilBrown5cf00fc2012-05-21 09:28:20 +10003482 sector_div(size, conf->geo.far_copies);
3483 conf->geo.stride = size << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003484 }
3485}
Trela, Maciejdab8b292010-03-08 16:02:45 +11003486
NeilBrowndeb200d2012-05-21 09:28:33 +10003487enum geo_type {geo_new, geo_old, geo_start};
3488static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new)
3489{
3490 int nc, fc, fo;
3491 int layout, chunk, disks;
3492 switch (new) {
3493 case geo_old:
3494 layout = mddev->layout;
3495 chunk = mddev->chunk_sectors;
3496 disks = mddev->raid_disks - mddev->delta_disks;
3497 break;
3498 case geo_new:
3499 layout = mddev->new_layout;
3500 chunk = mddev->new_chunk_sectors;
3501 disks = mddev->raid_disks;
3502 break;
3503 default: /* avoid 'may be unused' warnings */
3504 case geo_start: /* new when starting reshape - raid_disks not
3505 * updated yet. */
3506 layout = mddev->new_layout;
3507 chunk = mddev->new_chunk_sectors;
3508 disks = mddev->raid_disks + mddev->delta_disks;
3509 break;
3510 }
Jonathan Brassow475901a2013-02-21 13:28:10 +11003511 if (layout >> 18)
NeilBrowndeb200d2012-05-21 09:28:33 +10003512 return -1;
3513 if (chunk < (PAGE_SIZE >> 9) ||
3514 !is_power_of_2(chunk))
3515 return -2;
3516 nc = layout & 255;
3517 fc = (layout >> 8) & 255;
3518 fo = layout & (1<<16);
3519 geo->raid_disks = disks;
3520 geo->near_copies = nc;
3521 geo->far_copies = fc;
3522 geo->far_offset = fo;
Jonathan Brassow475901a2013-02-21 13:28:10 +11003523 geo->far_set_size = (layout & (1<<17)) ? disks / fc : disks;
NeilBrowndeb200d2012-05-21 09:28:33 +10003524 geo->chunk_mask = chunk - 1;
3525 geo->chunk_shift = ffz(~chunk);
3526 return nc*fc;
3527}
3528
NeilBrowne879a872011-10-11 16:49:02 +11003529static struct r10conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530{
NeilBrowne879a872011-10-11 16:49:02 +11003531 struct r10conf *conf = NULL;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003532 int err = -EINVAL;
NeilBrowndeb200d2012-05-21 09:28:33 +10003533 struct geom geo;
3534 int copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535
NeilBrowndeb200d2012-05-21 09:28:33 +10003536 copies = setup_geo(&geo, mddev, geo_new);
3537
3538 if (copies == -2) {
NeilBrown128595e2010-05-03 14:47:14 +10003539 printk(KERN_ERR "md/raid10:%s: chunk size must be "
3540 "at least PAGE_SIZE(%ld) and be a power of 2.\n",
3541 mdname(mddev), PAGE_SIZE);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003542 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543 }
NeilBrown2604b702006-01-06 00:20:36 -08003544
NeilBrowndeb200d2012-05-21 09:28:33 +10003545 if (copies < 2 || copies > mddev->raid_disks) {
NeilBrown128595e2010-05-03 14:47:14 +10003546 printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n",
Maciej Trelaf73ea872010-06-16 11:46:29 +01003547 mdname(mddev), mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 goto out;
3549 }
Trela, Maciejdab8b292010-03-08 16:02:45 +11003550
3551 err = -ENOMEM;
NeilBrowne879a872011-10-11 16:49:02 +11003552 conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003553 if (!conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554 goto out;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003555
NeilBrown3ea7daa2012-05-22 13:53:47 +10003556 /* FIXME calc properly */
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003557 conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks +
NeilBrown78eaa0d2013-07-02 15:58:05 +10003558 max(0,-mddev->delta_disks)),
Trela, Maciejdab8b292010-03-08 16:02:45 +11003559 GFP_KERNEL);
3560 if (!conf->mirrors)
3561 goto out;
NeilBrown4443ae12006-01-06 00:20:28 -08003562
3563 conf->tmppage = alloc_page(GFP_KERNEL);
3564 if (!conf->tmppage)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003565 goto out;
3566
NeilBrowndeb200d2012-05-21 09:28:33 +10003567 conf->geo = geo;
3568 conf->copies = copies;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003569 conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc,
3570 r10bio_pool_free, conf);
3571 if (!conf->r10bio_pool)
3572 goto out;
3573
NeilBrown6508fdb2012-05-17 10:08:45 +10003574 calc_sectors(conf, mddev->dev_sectors);
NeilBrown3ea7daa2012-05-22 13:53:47 +10003575 if (mddev->reshape_position == MaxSector) {
3576 conf->prev = conf->geo;
3577 conf->reshape_progress = MaxSector;
3578 } else {
3579 if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) {
3580 err = -EINVAL;
3581 goto out;
3582 }
3583 conf->reshape_progress = mddev->reshape_position;
3584 if (conf->prev.far_offset)
3585 conf->prev.stride = 1 << conf->prev.chunk_shift;
3586 else
3587 /* far_copies must be 1 */
3588 conf->prev.stride = conf->dev_sectors;
3589 }
Neil Browne7e72bf2008-05-14 16:05:54 -07003590 spin_lock_init(&conf->device_lock);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003591 INIT_LIST_HEAD(&conf->retry_list);
3592
3593 spin_lock_init(&conf->resync_lock);
3594 init_waitqueue_head(&conf->wait_barrier);
3595
NeilBrown02326052012-07-03 15:56:52 +10003596 conf->thread = md_register_thread(raid10d, mddev, "raid10");
Trela, Maciejdab8b292010-03-08 16:02:45 +11003597 if (!conf->thread)
3598 goto out;
3599
Trela, Maciejdab8b292010-03-08 16:02:45 +11003600 conf->mddev = mddev;
3601 return conf;
3602
3603 out:
NeilBrown3ea7daa2012-05-22 13:53:47 +10003604 if (err == -ENOMEM)
3605 printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n",
3606 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003607 if (conf) {
3608 if (conf->r10bio_pool)
3609 mempool_destroy(conf->r10bio_pool);
3610 kfree(conf->mirrors);
3611 safe_put_page(conf->tmppage);
3612 kfree(conf);
3613 }
3614 return ERR_PTR(err);
3615}
3616
NeilBrownfd01b882011-10-11 16:47:53 +11003617static int run(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003618{
NeilBrowne879a872011-10-11 16:49:02 +11003619 struct r10conf *conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003620 int i, disk_idx, chunk_size;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003621 struct raid10_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11003622 struct md_rdev *rdev;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003623 sector_t size;
NeilBrown3ea7daa2012-05-22 13:53:47 +10003624 sector_t min_offset_diff = 0;
3625 int first = 1;
Shaohua Li532a2a32012-10-11 13:30:52 +11003626 bool discard_supported = false;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003627
3628 if (mddev->private == NULL) {
3629 conf = setup_conf(mddev);
3630 if (IS_ERR(conf))
3631 return PTR_ERR(conf);
3632 mddev->private = conf;
3633 }
3634 conf = mddev->private;
3635 if (!conf)
3636 goto out;
3637
Trela, Maciejdab8b292010-03-08 16:02:45 +11003638 mddev->thread = conf->thread;
3639 conf->thread = NULL;
3640
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10003641 chunk_size = mddev->chunk_sectors << 9;
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003642 if (mddev->queue) {
Shaohua Li532a2a32012-10-11 13:30:52 +11003643 blk_queue_max_discard_sectors(mddev->queue,
3644 mddev->chunk_sectors);
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07003645 blk_queue_max_write_same_sectors(mddev->queue, 0);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003646 blk_queue_io_min(mddev->queue, chunk_size);
3647 if (conf->geo.raid_disks % conf->geo.near_copies)
3648 blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks);
3649 else
3650 blk_queue_io_opt(mddev->queue, chunk_size *
3651 (conf->geo.raid_disks / conf->geo.near_copies));
3652 }
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10003653
NeilBrowndafb20f2012-03-19 12:46:39 +11003654 rdev_for_each(rdev, mddev) {
NeilBrown3ea7daa2012-05-22 13:53:47 +10003655 long long diff;
NeilBrownaba336b2012-05-31 15:39:11 +10003656 struct request_queue *q;
NeilBrown34b343c2011-07-28 11:31:47 +10003657
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 disk_idx = rdev->raid_disk;
NeilBrownf8c9e742012-05-21 09:28:33 +10003659 if (disk_idx < 0)
3660 continue;
3661 if (disk_idx >= conf->geo.raid_disks &&
3662 disk_idx >= conf->prev.raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 continue;
3664 disk = conf->mirrors + disk_idx;
3665
NeilBrown56a25592011-12-23 10:17:55 +11003666 if (test_bit(Replacement, &rdev->flags)) {
3667 if (disk->replacement)
3668 goto out_free_conf;
3669 disk->replacement = rdev;
3670 } else {
3671 if (disk->rdev)
3672 goto out_free_conf;
3673 disk->rdev = rdev;
3674 }
NeilBrownaba336b2012-05-31 15:39:11 +10003675 q = bdev_get_queue(rdev->bdev);
3676 if (q->merge_bvec_fn)
3677 mddev->merge_check_needed = 1;
NeilBrown3ea7daa2012-05-22 13:53:47 +10003678 diff = (rdev->new_data_offset - rdev->data_offset);
3679 if (!mddev->reshape_backwards)
3680 diff = -diff;
3681 if (diff < 0)
3682 diff = 0;
3683 if (first || diff < min_offset_diff)
3684 min_offset_diff = diff;
NeilBrown56a25592011-12-23 10:17:55 +11003685
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003686 if (mddev->gendisk)
3687 disk_stack_limits(mddev->gendisk, rdev->bdev,
3688 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689
3690 disk->head_position = 0;
Shaohua Li532a2a32012-10-11 13:30:52 +11003691
3692 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
3693 discard_supported = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10003695
Jonathan Brassowed30be02012-10-31 11:42:30 +11003696 if (mddev->queue) {
3697 if (discard_supported)
3698 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
3699 mddev->queue);
3700 else
3701 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
3702 mddev->queue);
3703 }
NeilBrown6d508242005-09-09 16:24:03 -07003704 /* need to check that every block has at least one working mirror */
NeilBrown700c7212011-07-27 11:00:36 +10003705 if (!enough(conf, -1)) {
NeilBrown128595e2010-05-03 14:47:14 +10003706 printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n",
NeilBrown6d508242005-09-09 16:24:03 -07003707 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 goto out_free_conf;
3709 }
3710
NeilBrown3ea7daa2012-05-22 13:53:47 +10003711 if (conf->reshape_progress != MaxSector) {
3712 /* must ensure that shape change is supported */
3713 if (conf->geo.far_copies != 1 &&
3714 conf->geo.far_offset == 0)
3715 goto out_free_conf;
3716 if (conf->prev.far_copies != 1 &&
NeilBrown78eaa0d2013-07-02 15:58:05 +10003717 conf->prev.far_offset == 0)
NeilBrown3ea7daa2012-05-22 13:53:47 +10003718 goto out_free_conf;
3719 }
3720
Linus Torvalds1da177e2005-04-16 15:20:36 -07003721 mddev->degraded = 0;
NeilBrownf8c9e742012-05-21 09:28:33 +10003722 for (i = 0;
3723 i < conf->geo.raid_disks
3724 || i < conf->prev.raid_disks;
3725 i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726
3727 disk = conf->mirrors + i;
3728
NeilBrown56a25592011-12-23 10:17:55 +11003729 if (!disk->rdev && disk->replacement) {
3730 /* The replacement is all we have - use it */
3731 disk->rdev = disk->replacement;
3732 disk->replacement = NULL;
3733 clear_bit(Replacement, &disk->rdev->flags);
3734 }
3735
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003736 if (!disk->rdev ||
NeilBrown2e333e82006-10-21 10:24:07 -07003737 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738 disk->head_position = 0;
3739 mddev->degraded++;
NeilBrown0b59bb62014-01-14 16:30:10 +11003740 if (disk->rdev &&
3741 disk->rdev->saved_raid_disk < 0)
Neil Brown8c2e8702008-06-28 08:30:52 +10003742 conf->fullsync = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 }
NeilBrownd890fa22011-10-26 11:54:39 +11003744 disk->recovery_disabled = mddev->recovery_disabled - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 }
3746
Andre Noll8c6ac862009-06-18 08:48:06 +10003747 if (mddev->recovery_cp != MaxSector)
NeilBrown128595e2010-05-03 14:47:14 +10003748 printk(KERN_NOTICE "md/raid10:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10003749 " -- starting background reconstruction\n",
3750 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 printk(KERN_INFO
NeilBrown128595e2010-05-03 14:47:14 +10003752 "md/raid10:%s: active with %d out of %d devices\n",
NeilBrown5cf00fc2012-05-21 09:28:20 +10003753 mdname(mddev), conf->geo.raid_disks - mddev->degraded,
3754 conf->geo.raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 /*
3756 * Ok, everything is just fine now
3757 */
Trela, Maciejdab8b292010-03-08 16:02:45 +11003758 mddev->dev_sectors = conf->dev_sectors;
3759 size = raid10_size(mddev, 0, 0);
3760 md_set_array_sectors(mddev, size);
3761 mddev->resync_max_sectors = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003763 if (mddev->queue) {
NeilBrown5cf00fc2012-05-21 09:28:20 +10003764 int stripe = conf->geo.raid_disks *
Andre Noll9d8f0362009-06-18 08:45:01 +10003765 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003766 mddev->queue->backing_dev_info.congested_fn = raid10_congested;
3767 mddev->queue->backing_dev_info.congested_data = mddev;
3768
3769 /* Calculate max read-ahead size.
3770 * We need to readahead at least twice a whole stripe....
3771 * maybe...
3772 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003773 stripe /= conf->geo.near_copies;
NeilBrown3ea7daa2012-05-22 13:53:47 +10003774 if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
3775 mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003776 blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777 }
3778
Martin K. Petersena91a2782011-03-17 11:11:05 +01003779
3780 if (md_integrity_register(mddev))
3781 goto out_free_conf;
3782
NeilBrown3ea7daa2012-05-22 13:53:47 +10003783 if (conf->reshape_progress != MaxSector) {
3784 unsigned long before_length, after_length;
3785
3786 before_length = ((1 << conf->prev.chunk_shift) *
3787 conf->prev.far_copies);
3788 after_length = ((1 << conf->geo.chunk_shift) *
3789 conf->geo.far_copies);
3790
3791 if (max(before_length, after_length) > min_offset_diff) {
3792 /* This cannot work */
3793 printk("md/raid10: offset difference not enough to continue reshape\n");
3794 goto out_free_conf;
3795 }
3796 conf->offset_diff = min_offset_diff;
3797
3798 conf->reshape_safe = conf->reshape_progress;
3799 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3800 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3801 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
3802 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3803 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
3804 "reshape");
3805 }
3806
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807 return 0;
3808
3809out_free_conf:
NeilBrown01f96c02011-09-21 15:30:20 +10003810 md_unregister_thread(&mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 if (conf->r10bio_pool)
3812 mempool_destroy(conf->r10bio_pool);
NeilBrown1345b1d2006-01-06 00:20:40 -08003813 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003814 kfree(conf->mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 kfree(conf);
3816 mddev->private = NULL;
3817out:
3818 return -EIO;
3819}
3820
NeilBrownfd01b882011-10-11 16:47:53 +11003821static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822{
NeilBrowne879a872011-10-11 16:49:02 +11003823 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824
NeilBrown409c57f2009-03-31 14:39:39 +11003825 raise_barrier(conf, 0);
3826 lower_barrier(conf);
3827
NeilBrown01f96c02011-09-21 15:30:20 +10003828 md_unregister_thread(&mddev->thread);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003829 if (mddev->queue)
3830 /* the unplug fn references 'conf'*/
3831 blk_sync_queue(mddev->queue);
3832
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833 if (conf->r10bio_pool)
3834 mempool_destroy(conf->r10bio_pool);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10003835 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003836 kfree(conf->mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837 kfree(conf);
3838 mddev->private = NULL;
3839 return 0;
3840}
3841
NeilBrownfd01b882011-10-11 16:47:53 +11003842static void raid10_quiesce(struct mddev *mddev, int state)
NeilBrown6cce3b232006-01-06 00:20:16 -08003843{
NeilBrowne879a872011-10-11 16:49:02 +11003844 struct r10conf *conf = mddev->private;
NeilBrown6cce3b232006-01-06 00:20:16 -08003845
3846 switch(state) {
3847 case 1:
3848 raise_barrier(conf, 0);
3849 break;
3850 case 0:
3851 lower_barrier(conf);
3852 break;
3853 }
NeilBrown6cce3b232006-01-06 00:20:16 -08003854}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855
NeilBrown006a09a2012-03-19 12:46:40 +11003856static int raid10_resize(struct mddev *mddev, sector_t sectors)
3857{
3858 /* Resize of 'far' arrays is not supported.
3859 * For 'near' and 'offset' arrays we can set the
3860 * number of sectors used to be an appropriate multiple
3861 * of the chunk size.
3862 * For 'offset', this is far_copies*chunksize.
3863 * For 'near' the multiplier is the LCM of
3864 * near_copies and raid_disks.
3865 * So if far_copies > 1 && !far_offset, fail.
3866 * Else find LCM(raid_disks, near_copy)*far_copies and
3867 * multiply by chunk_size. Then round to this number.
3868 * This is mostly done by raid10_size()
3869 */
3870 struct r10conf *conf = mddev->private;
3871 sector_t oldsize, size;
3872
NeilBrownf8c9e742012-05-21 09:28:33 +10003873 if (mddev->reshape_position != MaxSector)
3874 return -EBUSY;
3875
NeilBrown5cf00fc2012-05-21 09:28:20 +10003876 if (conf->geo.far_copies > 1 && !conf->geo.far_offset)
NeilBrown006a09a2012-03-19 12:46:40 +11003877 return -EINVAL;
3878
3879 oldsize = raid10_size(mddev, 0, 0);
3880 size = raid10_size(mddev, sectors, 0);
NeilBrowna4a61252012-05-22 13:55:27 +10003881 if (mddev->external_size &&
3882 mddev->array_sectors > size)
NeilBrown006a09a2012-03-19 12:46:40 +11003883 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10003884 if (mddev->bitmap) {
3885 int ret = bitmap_resize(mddev->bitmap, size, 0, 0);
3886 if (ret)
3887 return ret;
3888 }
3889 md_set_array_sectors(mddev, size);
NeilBrown006a09a2012-03-19 12:46:40 +11003890 set_capacity(mddev->gendisk, mddev->array_sectors);
3891 revalidate_disk(mddev->gendisk);
3892 if (sectors > mddev->dev_sectors &&
3893 mddev->recovery_cp > oldsize) {
3894 mddev->recovery_cp = oldsize;
3895 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3896 }
NeilBrown6508fdb2012-05-17 10:08:45 +10003897 calc_sectors(conf, sectors);
3898 mddev->dev_sectors = conf->dev_sectors;
NeilBrown006a09a2012-03-19 12:46:40 +11003899 mddev->resync_max_sectors = size;
3900 return 0;
3901}
3902
NeilBrownfd01b882011-10-11 16:47:53 +11003903static void *raid10_takeover_raid0(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003904{
NeilBrown3cb03002011-10-11 16:45:26 +11003905 struct md_rdev *rdev;
NeilBrowne879a872011-10-11 16:49:02 +11003906 struct r10conf *conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003907
3908 if (mddev->degraded > 0) {
NeilBrown128595e2010-05-03 14:47:14 +10003909 printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n",
3910 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003911 return ERR_PTR(-EINVAL);
3912 }
3913
Trela, Maciejdab8b292010-03-08 16:02:45 +11003914 /* Set new parameters */
3915 mddev->new_level = 10;
3916 /* new layout: far_copies = 1, near_copies = 2 */
3917 mddev->new_layout = (1<<8) + 2;
3918 mddev->new_chunk_sectors = mddev->chunk_sectors;
3919 mddev->delta_disks = mddev->raid_disks;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003920 mddev->raid_disks *= 2;
3921 /* make sure it will be not marked as dirty */
3922 mddev->recovery_cp = MaxSector;
3923
3924 conf = setup_conf(mddev);
Krzysztof Wojcik02214dc2011-02-04 14:18:26 +01003925 if (!IS_ERR(conf)) {
NeilBrowndafb20f2012-03-19 12:46:39 +11003926 rdev_for_each(rdev, mddev)
NeilBrowne93f68a2010-06-15 09:36:03 +01003927 if (rdev->raid_disk >= 0)
3928 rdev->new_raid_disk = rdev->raid_disk * 2;
Krzysztof Wojcik02214dc2011-02-04 14:18:26 +01003929 conf->barrier = 1;
3930 }
3931
Trela, Maciejdab8b292010-03-08 16:02:45 +11003932 return conf;
3933}
3934
NeilBrownfd01b882011-10-11 16:47:53 +11003935static void *raid10_takeover(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003936{
NeilBrowne373ab12011-10-11 16:48:59 +11003937 struct r0conf *raid0_conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003938
3939 /* raid10 can take over:
3940 * raid0 - providing it has only two drives
3941 */
3942 if (mddev->level == 0) {
3943 /* for raid0 takeover only one zone is supported */
NeilBrowne373ab12011-10-11 16:48:59 +11003944 raid0_conf = mddev->private;
3945 if (raid0_conf->nr_strip_zones > 1) {
NeilBrown128595e2010-05-03 14:47:14 +10003946 printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0"
3947 " with more than one zone.\n",
3948 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003949 return ERR_PTR(-EINVAL);
3950 }
3951 return raid10_takeover_raid0(mddev);
3952 }
3953 return ERR_PTR(-EINVAL);
3954}
3955
NeilBrown3ea7daa2012-05-22 13:53:47 +10003956static int raid10_check_reshape(struct mddev *mddev)
3957{
3958 /* Called when there is a request to change
3959 * - layout (to ->new_layout)
3960 * - chunk size (to ->new_chunk_sectors)
3961 * - raid_disks (by delta_disks)
3962 * or when trying to restart a reshape that was ongoing.
3963 *
3964 * We need to validate the request and possibly allocate
3965 * space if that might be an issue later.
3966 *
3967 * Currently we reject any reshape of a 'far' mode array,
3968 * allow chunk size to change if new is generally acceptable,
3969 * allow raid_disks to increase, and allow
3970 * a switch between 'near' mode and 'offset' mode.
3971 */
3972 struct r10conf *conf = mddev->private;
3973 struct geom geo;
3974
3975 if (conf->geo.far_copies != 1 && !conf->geo.far_offset)
3976 return -EINVAL;
3977
3978 if (setup_geo(&geo, mddev, geo_start) != conf->copies)
3979 /* mustn't change number of copies */
3980 return -EINVAL;
3981 if (geo.far_copies > 1 && !geo.far_offset)
3982 /* Cannot switch to 'far' mode */
3983 return -EINVAL;
3984
3985 if (mddev->array_sectors & geo.chunk_mask)
3986 /* not factor of array size */
3987 return -EINVAL;
3988
NeilBrown3ea7daa2012-05-22 13:53:47 +10003989 if (!enough(conf, -1))
3990 return -EINVAL;
3991
3992 kfree(conf->mirrors_new);
3993 conf->mirrors_new = NULL;
3994 if (mddev->delta_disks > 0) {
3995 /* allocate new 'mirrors' list */
3996 conf->mirrors_new = kzalloc(
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003997 sizeof(struct raid10_info)
NeilBrown3ea7daa2012-05-22 13:53:47 +10003998 *(mddev->raid_disks +
3999 mddev->delta_disks),
4000 GFP_KERNEL);
4001 if (!conf->mirrors_new)
4002 return -ENOMEM;
4003 }
4004 return 0;
4005}
4006
4007/*
4008 * Need to check if array has failed when deciding whether to:
4009 * - start an array
4010 * - remove non-faulty devices
4011 * - add a spare
4012 * - allow a reshape
4013 * This determination is simple when no reshape is happening.
4014 * However if there is a reshape, we need to carefully check
4015 * both the before and after sections.
4016 * This is because some failed devices may only affect one
4017 * of the two sections, and some non-in_sync devices may
4018 * be insync in the section most affected by failed devices.
4019 */
4020static int calc_degraded(struct r10conf *conf)
4021{
4022 int degraded, degraded2;
4023 int i;
4024
4025 rcu_read_lock();
4026 degraded = 0;
4027 /* 'prev' section first */
4028 for (i = 0; i < conf->prev.raid_disks; i++) {
4029 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
4030 if (!rdev || test_bit(Faulty, &rdev->flags))
4031 degraded++;
4032 else if (!test_bit(In_sync, &rdev->flags))
4033 /* When we can reduce the number of devices in
4034 * an array, this might not contribute to
4035 * 'degraded'. It does now.
4036 */
4037 degraded++;
4038 }
4039 rcu_read_unlock();
4040 if (conf->geo.raid_disks == conf->prev.raid_disks)
4041 return degraded;
4042 rcu_read_lock();
4043 degraded2 = 0;
4044 for (i = 0; i < conf->geo.raid_disks; i++) {
4045 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
4046 if (!rdev || test_bit(Faulty, &rdev->flags))
4047 degraded2++;
4048 else if (!test_bit(In_sync, &rdev->flags)) {
4049 /* If reshape is increasing the number of devices,
4050 * this section has already been recovered, so
4051 * it doesn't contribute to degraded.
4052 * else it does.
4053 */
4054 if (conf->geo.raid_disks <= conf->prev.raid_disks)
4055 degraded2++;
4056 }
4057 }
4058 rcu_read_unlock();
4059 if (degraded2 > degraded)
4060 return degraded2;
4061 return degraded;
4062}
4063
4064static int raid10_start_reshape(struct mddev *mddev)
4065{
4066 /* A 'reshape' has been requested. This commits
4067 * the various 'new' fields and sets MD_RECOVER_RESHAPE
4068 * This also checks if there are enough spares and adds them
4069 * to the array.
4070 * We currently require enough spares to make the final
4071 * array non-degraded. We also require that the difference
4072 * between old and new data_offset - on each device - is
4073 * enough that we never risk over-writing.
4074 */
4075
4076 unsigned long before_length, after_length;
4077 sector_t min_offset_diff = 0;
4078 int first = 1;
4079 struct geom new;
4080 struct r10conf *conf = mddev->private;
4081 struct md_rdev *rdev;
4082 int spares = 0;
NeilBrownbb63a702012-05-22 13:55:28 +10004083 int ret;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004084
4085 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4086 return -EBUSY;
4087
4088 if (setup_geo(&new, mddev, geo_start) != conf->copies)
4089 return -EINVAL;
4090
4091 before_length = ((1 << conf->prev.chunk_shift) *
4092 conf->prev.far_copies);
4093 after_length = ((1 << conf->geo.chunk_shift) *
4094 conf->geo.far_copies);
4095
4096 rdev_for_each(rdev, mddev) {
4097 if (!test_bit(In_sync, &rdev->flags)
4098 && !test_bit(Faulty, &rdev->flags))
4099 spares++;
4100 if (rdev->raid_disk >= 0) {
4101 long long diff = (rdev->new_data_offset
4102 - rdev->data_offset);
4103 if (!mddev->reshape_backwards)
4104 diff = -diff;
4105 if (diff < 0)
4106 diff = 0;
4107 if (first || diff < min_offset_diff)
4108 min_offset_diff = diff;
4109 }
4110 }
4111
4112 if (max(before_length, after_length) > min_offset_diff)
4113 return -EINVAL;
4114
4115 if (spares < mddev->delta_disks)
4116 return -EINVAL;
4117
4118 conf->offset_diff = min_offset_diff;
4119 spin_lock_irq(&conf->device_lock);
4120 if (conf->mirrors_new) {
4121 memcpy(conf->mirrors_new, conf->mirrors,
Jonathan Brassowdc280d982012-07-31 10:03:52 +10004122 sizeof(struct raid10_info)*conf->prev.raid_disks);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004123 smp_mb();
4124 kfree(conf->mirrors_old); /* FIXME and elsewhere */
4125 conf->mirrors_old = conf->mirrors;
4126 conf->mirrors = conf->mirrors_new;
4127 conf->mirrors_new = NULL;
4128 }
4129 setup_geo(&conf->geo, mddev, geo_start);
4130 smp_mb();
4131 if (mddev->reshape_backwards) {
4132 sector_t size = raid10_size(mddev, 0, 0);
4133 if (size < mddev->array_sectors) {
4134 spin_unlock_irq(&conf->device_lock);
4135 printk(KERN_ERR "md/raid10:%s: array size must be reduce before number of disks\n",
4136 mdname(mddev));
4137 return -EINVAL;
4138 }
4139 mddev->resync_max_sectors = size;
4140 conf->reshape_progress = size;
4141 } else
4142 conf->reshape_progress = 0;
4143 spin_unlock_irq(&conf->device_lock);
4144
NeilBrownbb63a702012-05-22 13:55:28 +10004145 if (mddev->delta_disks && mddev->bitmap) {
4146 ret = bitmap_resize(mddev->bitmap,
4147 raid10_size(mddev, 0,
4148 conf->geo.raid_disks),
4149 0, 0);
4150 if (ret)
4151 goto abort;
4152 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004153 if (mddev->delta_disks > 0) {
4154 rdev_for_each(rdev, mddev)
4155 if (rdev->raid_disk < 0 &&
4156 !test_bit(Faulty, &rdev->flags)) {
4157 if (raid10_add_disk(mddev, rdev) == 0) {
4158 if (rdev->raid_disk >=
4159 conf->prev.raid_disks)
4160 set_bit(In_sync, &rdev->flags);
4161 else
4162 rdev->recovery_offset = 0;
4163
4164 if (sysfs_link_rdev(mddev, rdev))
4165 /* Failure here is OK */;
4166 }
4167 } else if (rdev->raid_disk >= conf->prev.raid_disks
4168 && !test_bit(Faulty, &rdev->flags)) {
4169 /* This is a spare that was manually added */
4170 set_bit(In_sync, &rdev->flags);
4171 }
4172 }
4173 /* When a reshape changes the number of devices,
4174 * ->degraded is measured against the larger of the
4175 * pre and post numbers.
4176 */
4177 spin_lock_irq(&conf->device_lock);
4178 mddev->degraded = calc_degraded(conf);
4179 spin_unlock_irq(&conf->device_lock);
4180 mddev->raid_disks = conf->geo.raid_disks;
4181 mddev->reshape_position = conf->reshape_progress;
4182 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4183
4184 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4185 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4186 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4187 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4188
4189 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
4190 "reshape");
4191 if (!mddev->sync_thread) {
NeilBrownbb63a702012-05-22 13:55:28 +10004192 ret = -EAGAIN;
4193 goto abort;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004194 }
4195 conf->reshape_checkpoint = jiffies;
4196 md_wakeup_thread(mddev->sync_thread);
4197 md_new_event(mddev);
4198 return 0;
NeilBrownbb63a702012-05-22 13:55:28 +10004199
4200abort:
4201 mddev->recovery = 0;
4202 spin_lock_irq(&conf->device_lock);
4203 conf->geo = conf->prev;
4204 mddev->raid_disks = conf->geo.raid_disks;
4205 rdev_for_each(rdev, mddev)
4206 rdev->new_data_offset = rdev->data_offset;
4207 smp_wmb();
4208 conf->reshape_progress = MaxSector;
4209 mddev->reshape_position = MaxSector;
4210 spin_unlock_irq(&conf->device_lock);
4211 return ret;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004212}
4213
4214/* Calculate the last device-address that could contain
4215 * any block from the chunk that includes the array-address 's'
4216 * and report the next address.
4217 * i.e. the address returned will be chunk-aligned and after
4218 * any data that is in the chunk containing 's'.
4219 */
4220static sector_t last_dev_address(sector_t s, struct geom *geo)
4221{
4222 s = (s | geo->chunk_mask) + 1;
4223 s >>= geo->chunk_shift;
4224 s *= geo->near_copies;
4225 s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks);
4226 s *= geo->far_copies;
4227 s <<= geo->chunk_shift;
4228 return s;
4229}
4230
4231/* Calculate the first device-address that could contain
4232 * any block from the chunk that includes the array-address 's'.
4233 * This too will be the start of a chunk
4234 */
4235static sector_t first_dev_address(sector_t s, struct geom *geo)
4236{
4237 s >>= geo->chunk_shift;
4238 s *= geo->near_copies;
4239 sector_div(s, geo->raid_disks);
4240 s *= geo->far_copies;
4241 s <<= geo->chunk_shift;
4242 return s;
4243}
4244
4245static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
4246 int *skipped)
4247{
4248 /* We simply copy at most one chunk (smallest of old and new)
4249 * at a time, possibly less if that exceeds RESYNC_PAGES,
4250 * or we hit a bad block or something.
4251 * This might mean we pause for normal IO in the middle of
4252 * a chunk, but that is not a problem was mddev->reshape_position
4253 * can record any location.
4254 *
4255 * If we will want to write to a location that isn't
4256 * yet recorded as 'safe' (i.e. in metadata on disk) then
4257 * we need to flush all reshape requests and update the metadata.
4258 *
4259 * When reshaping forwards (e.g. to more devices), we interpret
4260 * 'safe' as the earliest block which might not have been copied
4261 * down yet. We divide this by previous stripe size and multiply
4262 * by previous stripe length to get lowest device offset that we
4263 * cannot write to yet.
4264 * We interpret 'sector_nr' as an address that we want to write to.
4265 * From this we use last_device_address() to find where we might
4266 * write to, and first_device_address on the 'safe' position.
4267 * If this 'next' write position is after the 'safe' position,
4268 * we must update the metadata to increase the 'safe' position.
4269 *
4270 * When reshaping backwards, we round in the opposite direction
4271 * and perform the reverse test: next write position must not be
4272 * less than current safe position.
4273 *
4274 * In all this the minimum difference in data offsets
4275 * (conf->offset_diff - always positive) allows a bit of slack,
4276 * so next can be after 'safe', but not by more than offset_disk
4277 *
4278 * We need to prepare all the bios here before we start any IO
4279 * to ensure the size we choose is acceptable to all devices.
4280 * The means one for each copy for write-out and an extra one for
4281 * read-in.
4282 * We store the read-in bio in ->master_bio and the others in
4283 * ->devs[x].bio and ->devs[x].repl_bio.
4284 */
4285 struct r10conf *conf = mddev->private;
4286 struct r10bio *r10_bio;
4287 sector_t next, safe, last;
4288 int max_sectors;
4289 int nr_sectors;
4290 int s;
4291 struct md_rdev *rdev;
4292 int need_flush = 0;
4293 struct bio *blist;
4294 struct bio *bio, *read_bio;
4295 int sectors_done = 0;
4296
4297 if (sector_nr == 0) {
4298 /* If restarting in the middle, skip the initial sectors */
4299 if (mddev->reshape_backwards &&
4300 conf->reshape_progress < raid10_size(mddev, 0, 0)) {
4301 sector_nr = (raid10_size(mddev, 0, 0)
4302 - conf->reshape_progress);
4303 } else if (!mddev->reshape_backwards &&
4304 conf->reshape_progress > 0)
4305 sector_nr = conf->reshape_progress;
4306 if (sector_nr) {
4307 mddev->curr_resync_completed = sector_nr;
4308 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
4309 *skipped = 1;
4310 return sector_nr;
4311 }
4312 }
4313
4314 /* We don't use sector_nr to track where we are up to
4315 * as that doesn't work well for ->reshape_backwards.
4316 * So just use ->reshape_progress.
4317 */
4318 if (mddev->reshape_backwards) {
4319 /* 'next' is the earliest device address that we might
4320 * write to for this chunk in the new layout
4321 */
4322 next = first_dev_address(conf->reshape_progress - 1,
4323 &conf->geo);
4324
4325 /* 'safe' is the last device address that we might read from
4326 * in the old layout after a restart
4327 */
4328 safe = last_dev_address(conf->reshape_safe - 1,
4329 &conf->prev);
4330
4331 if (next + conf->offset_diff < safe)
4332 need_flush = 1;
4333
4334 last = conf->reshape_progress - 1;
4335 sector_nr = last & ~(sector_t)(conf->geo.chunk_mask
4336 & conf->prev.chunk_mask);
4337 if (sector_nr + RESYNC_BLOCK_SIZE/512 < last)
4338 sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512;
4339 } else {
4340 /* 'next' is after the last device address that we
4341 * might write to for this chunk in the new layout
4342 */
4343 next = last_dev_address(conf->reshape_progress, &conf->geo);
4344
4345 /* 'safe' is the earliest device address that we might
4346 * read from in the old layout after a restart
4347 */
4348 safe = first_dev_address(conf->reshape_safe, &conf->prev);
4349
4350 /* Need to update metadata if 'next' might be beyond 'safe'
4351 * as that would possibly corrupt data
4352 */
4353 if (next > safe + conf->offset_diff)
4354 need_flush = 1;
4355
4356 sector_nr = conf->reshape_progress;
4357 last = sector_nr | (conf->geo.chunk_mask
4358 & conf->prev.chunk_mask);
4359
4360 if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last)
4361 last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1;
4362 }
4363
4364 if (need_flush ||
4365 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
4366 /* Need to update reshape_position in metadata */
4367 wait_barrier(conf);
4368 mddev->reshape_position = conf->reshape_progress;
4369 if (mddev->reshape_backwards)
4370 mddev->curr_resync_completed = raid10_size(mddev, 0, 0)
4371 - conf->reshape_progress;
4372 else
4373 mddev->curr_resync_completed = conf->reshape_progress;
4374 conf->reshape_checkpoint = jiffies;
4375 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4376 md_wakeup_thread(mddev->thread);
4377 wait_event(mddev->sb_wait, mddev->flags == 0 ||
NeilBrownc91abf52013-11-19 12:02:01 +11004378 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4379 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4380 allow_barrier(conf);
4381 return sectors_done;
4382 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004383 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);
NeilBrowncb8b12b2014-08-18 14:38:45 +10004390 r10_bio->state = 0;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004391 raise_barrier(conf, sectors_done != 0);
4392 atomic_set(&r10_bio->remaining, 0);
4393 r10_bio->mddev = mddev;
4394 r10_bio->sector = sector_nr;
4395 set_bit(R10BIO_IsReshape, &r10_bio->state);
4396 r10_bio->sectors = last - sector_nr + 1;
4397 rdev = read_balance(conf, r10_bio, &max_sectors);
4398 BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state));
4399
4400 if (!rdev) {
4401 /* Cannot read from here, so need to record bad blocks
4402 * on all the target devices.
4403 */
4404 // FIXME
NeilBrowne337aea2014-08-18 14:48:54 +10004405 mempool_free(r10_bio, conf->r10buf_pool);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004406 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4407 return sectors_done;
4408 }
4409
4410 read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev);
4411
4412 read_bio->bi_bdev = rdev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004413 read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr
NeilBrown3ea7daa2012-05-22 13:53:47 +10004414 + rdev->data_offset);
4415 read_bio->bi_private = r10_bio;
4416 read_bio->bi_end_io = end_sync_read;
4417 read_bio->bi_rw = READ;
NeilBrownce0b0a42014-08-18 13:56:38 +10004418 read_bio->bi_flags &= (~0UL << BIO_RESET_BITS);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004419 read_bio->bi_flags |= 1 << BIO_UPTODATE;
4420 read_bio->bi_vcnt = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004421 read_bio->bi_iter.bi_size = 0;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004422 r10_bio->master_bio = read_bio;
4423 r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum;
4424
4425 /* Now find the locations in the new layout */
4426 __raid10_find_phys(&conf->geo, r10_bio);
4427
4428 blist = read_bio;
4429 read_bio->bi_next = NULL;
4430
4431 for (s = 0; s < conf->copies*2; s++) {
4432 struct bio *b;
4433 int d = r10_bio->devs[s/2].devnum;
4434 struct md_rdev *rdev2;
4435 if (s&1) {
4436 rdev2 = conf->mirrors[d].replacement;
4437 b = r10_bio->devs[s/2].repl_bio;
4438 } else {
4439 rdev2 = conf->mirrors[d].rdev;
4440 b = r10_bio->devs[s/2].bio;
4441 }
4442 if (!rdev2 || test_bit(Faulty, &rdev2->flags))
4443 continue;
Kent Overstreet8be185f2012-09-06 14:14:43 -07004444
4445 bio_reset(b);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004446 b->bi_bdev = rdev2->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004447 b->bi_iter.bi_sector = r10_bio->devs[s/2].addr +
4448 rdev2->new_data_offset;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004449 b->bi_private = r10_bio;
4450 b->bi_end_io = end_reshape_write;
4451 b->bi_rw = WRITE;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004452 b->bi_next = blist;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004453 blist = b;
4454 }
4455
4456 /* Now add as many pages as possible to all of these bios. */
4457
4458 nr_sectors = 0;
4459 for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) {
4460 struct page *page = r10_bio->devs[0].bio->bi_io_vec[s/(PAGE_SIZE>>9)].bv_page;
4461 int len = (max_sectors - s) << 9;
4462 if (len > PAGE_SIZE)
4463 len = PAGE_SIZE;
4464 for (bio = blist; bio ; bio = bio->bi_next) {
4465 struct bio *bio2;
4466 if (bio_add_page(bio, page, len, 0))
4467 continue;
4468
4469 /* Didn't fit, must stop */
4470 for (bio2 = blist;
4471 bio2 && bio2 != bio;
4472 bio2 = bio2->bi_next) {
4473 /* Remove last page from this bio */
4474 bio2->bi_vcnt--;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004475 bio2->bi_iter.bi_size -= len;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004476 bio2->bi_flags &= ~(1<<BIO_SEG_VALID);
4477 }
4478 goto bio_full;
4479 }
4480 sector_nr += len >> 9;
4481 nr_sectors += len >> 9;
4482 }
4483bio_full:
4484 r10_bio->sectors = nr_sectors;
4485
4486 /* Now submit the read */
4487 md_sync_acct(read_bio->bi_bdev, r10_bio->sectors);
4488 atomic_inc(&r10_bio->remaining);
4489 read_bio->bi_next = NULL;
4490 generic_make_request(read_bio);
4491 sector_nr += nr_sectors;
4492 sectors_done += nr_sectors;
4493 if (sector_nr <= last)
4494 goto read_more;
4495
4496 /* Now that we have done the whole section we can
4497 * update reshape_progress
4498 */
4499 if (mddev->reshape_backwards)
4500 conf->reshape_progress -= sectors_done;
4501 else
4502 conf->reshape_progress += sectors_done;
4503
4504 return sectors_done;
4505}
4506
4507static void end_reshape_request(struct r10bio *r10_bio);
4508static int handle_reshape_read_error(struct mddev *mddev,
4509 struct r10bio *r10_bio);
4510static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio)
4511{
4512 /* Reshape read completed. Hopefully we have a block
4513 * to write out.
4514 * If we got a read error then we do sync 1-page reads from
4515 * elsewhere until we find the data - or give up.
4516 */
4517 struct r10conf *conf = mddev->private;
4518 int s;
4519
4520 if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
4521 if (handle_reshape_read_error(mddev, r10_bio) < 0) {
4522 /* Reshape has been aborted */
4523 md_done_sync(mddev, r10_bio->sectors, 0);
4524 return;
4525 }
4526
4527 /* We definitely have the data in the pages, schedule the
4528 * writes.
4529 */
4530 atomic_set(&r10_bio->remaining, 1);
4531 for (s = 0; s < conf->copies*2; s++) {
4532 struct bio *b;
4533 int d = r10_bio->devs[s/2].devnum;
4534 struct md_rdev *rdev;
4535 if (s&1) {
4536 rdev = conf->mirrors[d].replacement;
4537 b = r10_bio->devs[s/2].repl_bio;
4538 } else {
4539 rdev = conf->mirrors[d].rdev;
4540 b = r10_bio->devs[s/2].bio;
4541 }
4542 if (!rdev || test_bit(Faulty, &rdev->flags))
4543 continue;
4544 atomic_inc(&rdev->nr_pending);
4545 md_sync_acct(b->bi_bdev, r10_bio->sectors);
4546 atomic_inc(&r10_bio->remaining);
4547 b->bi_next = NULL;
4548 generic_make_request(b);
4549 }
4550 end_reshape_request(r10_bio);
4551}
4552
4553static void end_reshape(struct r10conf *conf)
4554{
4555 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery))
4556 return;
4557
4558 spin_lock_irq(&conf->device_lock);
4559 conf->prev = conf->geo;
4560 md_finish_reshape(conf->mddev);
4561 smp_wmb();
4562 conf->reshape_progress = MaxSector;
4563 spin_unlock_irq(&conf->device_lock);
4564
4565 /* read-ahead size must cover two whole stripes, which is
4566 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4567 */
4568 if (conf->mddev->queue) {
4569 int stripe = conf->geo.raid_disks *
4570 ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE);
4571 stripe /= conf->geo.near_copies;
4572 if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
4573 conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
4574 }
4575 conf->fullsync = 0;
4576}
4577
4578
4579static int handle_reshape_read_error(struct mddev *mddev,
4580 struct r10bio *r10_bio)
4581{
4582 /* Use sync reads to get the blocks from somewhere else */
4583 int sectors = r10_bio->sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004584 struct r10conf *conf = mddev->private;
NeilBrowne0ee7782012-08-18 09:51:42 +10004585 struct {
4586 struct r10bio r10_bio;
4587 struct r10dev devs[conf->copies];
4588 } on_stack;
4589 struct r10bio *r10b = &on_stack.r10_bio;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004590 int slot = 0;
4591 int idx = 0;
4592 struct bio_vec *bvec = r10_bio->master_bio->bi_io_vec;
4593
NeilBrowne0ee7782012-08-18 09:51:42 +10004594 r10b->sector = r10_bio->sector;
4595 __raid10_find_phys(&conf->prev, r10b);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004596
4597 while (sectors) {
4598 int s = sectors;
4599 int success = 0;
4600 int first_slot = slot;
4601
4602 if (s > (PAGE_SIZE >> 9))
4603 s = PAGE_SIZE >> 9;
4604
4605 while (!success) {
NeilBrowne0ee7782012-08-18 09:51:42 +10004606 int d = r10b->devs[slot].devnum;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004607 struct md_rdev *rdev = conf->mirrors[d].rdev;
4608 sector_t addr;
4609 if (rdev == NULL ||
4610 test_bit(Faulty, &rdev->flags) ||
4611 !test_bit(In_sync, &rdev->flags))
4612 goto failed;
4613
NeilBrowne0ee7782012-08-18 09:51:42 +10004614 addr = r10b->devs[slot].addr + idx * PAGE_SIZE;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004615 success = sync_page_io(rdev,
4616 addr,
4617 s << 9,
4618 bvec[idx].bv_page,
4619 READ, false);
4620 if (success)
4621 break;
4622 failed:
4623 slot++;
4624 if (slot >= conf->copies)
4625 slot = 0;
4626 if (slot == first_slot)
4627 break;
4628 }
4629 if (!success) {
4630 /* couldn't read this block, must give up */
4631 set_bit(MD_RECOVERY_INTR,
4632 &mddev->recovery);
4633 return -EIO;
4634 }
4635 sectors -= s;
4636 idx++;
4637 }
4638 return 0;
4639}
4640
4641static void end_reshape_write(struct bio *bio, int error)
4642{
4643 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
4644 struct r10bio *r10_bio = bio->bi_private;
4645 struct mddev *mddev = r10_bio->mddev;
4646 struct r10conf *conf = mddev->private;
4647 int d;
4648 int slot;
4649 int repl;
4650 struct md_rdev *rdev = NULL;
4651
4652 d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
4653 if (repl)
4654 rdev = conf->mirrors[d].replacement;
4655 if (!rdev) {
4656 smp_mb();
4657 rdev = conf->mirrors[d].rdev;
4658 }
4659
4660 if (!uptodate) {
4661 /* FIXME should record badblock */
4662 md_error(mddev, rdev);
4663 }
4664
4665 rdev_dec_pending(rdev, mddev);
4666 end_reshape_request(r10_bio);
4667}
4668
4669static void end_reshape_request(struct r10bio *r10_bio)
4670{
4671 if (!atomic_dec_and_test(&r10_bio->remaining))
4672 return;
4673 md_done_sync(r10_bio->mddev, r10_bio->sectors, 1);
4674 bio_put(r10_bio->master_bio);
4675 put_buf(r10_bio);
4676}
4677
4678static void raid10_finish_reshape(struct mddev *mddev)
4679{
4680 struct r10conf *conf = mddev->private;
4681
4682 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4683 return;
4684
4685 if (mddev->delta_disks > 0) {
4686 sector_t size = raid10_size(mddev, 0, 0);
4687 md_set_array_sectors(mddev, size);
4688 if (mddev->recovery_cp > mddev->resync_max_sectors) {
4689 mddev->recovery_cp = mddev->resync_max_sectors;
4690 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4691 }
4692 mddev->resync_max_sectors = size;
4693 set_capacity(mddev->gendisk, mddev->array_sectors);
4694 revalidate_disk(mddev->gendisk);
NeilBrown63aced62012-05-22 13:55:33 +10004695 } else {
4696 int d;
4697 for (d = conf->geo.raid_disks ;
4698 d < conf->geo.raid_disks - mddev->delta_disks;
4699 d++) {
4700 struct md_rdev *rdev = conf->mirrors[d].rdev;
4701 if (rdev)
4702 clear_bit(In_sync, &rdev->flags);
4703 rdev = conf->mirrors[d].replacement;
4704 if (rdev)
4705 clear_bit(In_sync, &rdev->flags);
4706 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004707 }
4708 mddev->layout = mddev->new_layout;
4709 mddev->chunk_sectors = 1 << conf->geo.chunk_shift;
4710 mddev->reshape_position = MaxSector;
4711 mddev->delta_disks = 0;
4712 mddev->reshape_backwards = 0;
4713}
4714
NeilBrown84fc4b52011-10-11 16:49:58 +11004715static struct md_personality raid10_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716{
4717 .name = "raid10",
NeilBrown2604b702006-01-06 00:20:36 -08004718 .level = 10,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004719 .owner = THIS_MODULE,
4720 .make_request = make_request,
4721 .run = run,
4722 .stop = stop,
4723 .status = status,
4724 .error_handler = error,
4725 .hot_add_disk = raid10_add_disk,
4726 .hot_remove_disk= raid10_remove_disk,
4727 .spare_active = raid10_spare_active,
4728 .sync_request = sync_request,
NeilBrown6cce3b232006-01-06 00:20:16 -08004729 .quiesce = raid10_quiesce,
Dan Williams80c3a6c2009-03-17 18:10:40 -07004730 .size = raid10_size,
NeilBrown006a09a2012-03-19 12:46:40 +11004731 .resize = raid10_resize,
Trela, Maciejdab8b292010-03-08 16:02:45 +11004732 .takeover = raid10_takeover,
NeilBrown3ea7daa2012-05-22 13:53:47 +10004733 .check_reshape = raid10_check_reshape,
4734 .start_reshape = raid10_start_reshape,
4735 .finish_reshape = raid10_finish_reshape,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004736};
4737
4738static int __init raid_init(void)
4739{
NeilBrown2604b702006-01-06 00:20:36 -08004740 return register_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004741}
4742
4743static void raid_exit(void)
4744{
NeilBrown2604b702006-01-06 00:20:36 -08004745 unregister_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004746}
4747
4748module_init(raid_init);
4749module_exit(raid_exit);
4750MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11004751MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752MODULE_ALIAS("md-personality-9"); /* RAID10 */
NeilBrownd9d166c2006-01-06 00:20:51 -08004753MODULE_ALIAS("md-raid10");
NeilBrown2604b702006-01-06 00:20:36 -08004754MODULE_ALIAS("md-level-10");
NeilBrown34db0cd2011-10-11 16:50:01 +11004755
4756module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);