blob: 41845bae67be76cdaa1c490df22293cfed23d6cf [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>
NeilBrown109e3762016-11-18 13:22:04 +110028#include <trace/events/block.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110029#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110030#include "raid10.h"
Trela, Maciejdab8b292010-03-08 16:02:45 +110031#include "raid0.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110032#include "bitmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34/*
35 * RAID10 provides a combination of RAID0 and RAID1 functionality.
36 * The layout of data is defined by
37 * chunk_size
38 * raid_disks
39 * near_copies (stored in low byte of layout)
40 * far_copies (stored in second byte of layout)
NeilBrownc93983b2006-06-26 00:27:41 -070041 * far_offset (stored in bit 16 of layout )
Jonathan Brassow475901a2013-02-21 13:28:10 +110042 * use_far_sets (stored in bit 17 of layout )
NeilBrown8bce6d32015-10-22 13:20:15 +110043 * use_far_sets_bugfixed (stored in bit 18 of layout )
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 *
Jonathan Brassow475901a2013-02-21 13:28:10 +110045 * The data to be stored is divided into chunks using chunksize. Each device
46 * is divided into far_copies sections. In each section, chunks are laid out
47 * in a style similar to raid0, but near_copies copies of each chunk is stored
48 * (each on a different drive). The starting device for each section is offset
49 * near_copies from the starting device of the previous section. Thus there
50 * are (near_copies * far_copies) of each chunk, and each is on a different
51 * drive. near_copies and far_copies must be at least one, and their product
52 * is at most raid_disks.
NeilBrownc93983b2006-06-26 00:27:41 -070053 *
54 * If far_offset is true, then the far_copies are handled a bit differently.
Jonathan Brassow475901a2013-02-21 13:28:10 +110055 * The copies are still in different stripes, but instead of being very far
56 * apart on disk, there are adjacent stripes.
57 *
58 * The far and offset algorithms are handled slightly differently if
59 * 'use_far_sets' is true. In this case, the array's devices are grouped into
60 * sets that are (near_copies * far_copies) in size. The far copied stripes
61 * are still shifted by 'near_copies' devices, but this shifting stays confined
62 * to the set rather than the entire array. This is done to improve the number
63 * of device combinations that can fail without causing the array to fail.
64 * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk
65 * on a device):
66 * A B C D A B C D E
67 * ... ...
68 * D A B C E A B C D
69 * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s):
70 * [A B] [C D] [A B] [C D E]
71 * |...| |...| |...| | ... |
72 * [B A] [D C] [B A] [E C D]
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 */
74
75/*
76 * Number of guaranteed r10bios in case of extreme VM load:
77 */
78#define NR_RAID10_BIOS 256
79
Jonathan Brassow473e87c2012-07-31 10:03:52 +100080/* when we get a read error on a read-only array, we redirect to another
81 * device without failing the first device, or trying to over-write to
82 * correct the read error. To keep track of bad blocks on a per-bio
83 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
84 */
85#define IO_BLOCKED ((struct bio *)1)
86/* When we successfully write to a known bad-block, we need to remove the
87 * bad-block marking which must be done from process context. So we record
88 * the success by setting devs[n].bio to IO_MADE_GOOD
89 */
90#define IO_MADE_GOOD ((struct bio *)2)
91
92#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
93
94/* When there are this many requests queued to be written by
NeilBrown34db0cd2011-10-11 16:50:01 +110095 * the raid10 thread, we become 'congested' to provide back-pressure
96 * for writeback.
97 */
98static int max_queued_requests = 1024;
99
NeilBrowne879a872011-10-11 16:49:02 +1100100static void allow_barrier(struct r10conf *conf);
101static void lower_barrier(struct r10conf *conf);
NeilBrown635f6412013-06-11 14:57:09 +1000102static int _enough(struct r10conf *conf, int previous, int ignore);
NeilBrown1919cbb2016-11-18 16:16:12 +1100103static int enough(struct r10conf *conf, int ignore);
NeilBrown3ea7daa2012-05-22 13:53:47 +1000104static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
105 int *skipped);
106static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200107static void end_reshape_write(struct bio *bio);
NeilBrown3ea7daa2012-05-22 13:53:47 +1000108static void end_reshape(struct r10conf *conf);
NeilBrown0a27ec92006-01-06 00:20:13 -0800109
NeilBrown578b54a2016-11-14 16:30:21 +1100110#define raid10_log(md, fmt, args...) \
111 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid10 " fmt, ##args); } while (0)
112
Ming Leif0250612017-03-17 00:12:33 +0800113/*
114 * 'strct resync_pages' stores actual pages used for doing the resync
115 * IO, and it is per-bio, so make .bi_private points to it.
116 */
117static inline struct resync_pages *get_resync_pages(struct bio *bio)
118{
119 return bio->bi_private;
120}
121
122/*
123 * for resync bio, r10bio pointer can be retrieved from the per-bio
124 * 'struct resync_pages'.
125 */
126static inline struct r10bio *get_resync_r10bio(struct bio *bio)
127{
128 return get_resync_pages(bio)->raid_bio;
129}
130
Al Virodd0fc662005-10-07 07:46:04 +0100131static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
NeilBrowne879a872011-10-11 16:49:02 +1100133 struct r10conf *conf = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100134 int size = offsetof(struct r10bio, devs[conf->copies]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
NeilBrown69335ef2011-12-23 10:17:54 +1100136 /* allocate a r10bio with room for raid_disks entries in the
137 * bios array */
Jens Axboe7eaceac2011-03-10 08:52:07 +0100138 return kzalloc(size, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139}
140
141static void r10bio_pool_free(void *r10_bio, void *data)
142{
143 kfree(r10_bio);
144}
145
NeilBrown0310fa22008-08-05 15:54:14 +1000146/* amount of memory to reserve for resync requests */
147#define RESYNC_WINDOW (1024*1024)
148/* maximum number of concurrent requests, memory permitting */
149#define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
151/*
152 * When performing a resync, we need to read and compare, so
153 * we need as many pages are there are copies.
154 * When performing a recovery, we need 2 bios, one for read,
155 * one for write (we recover only one drive per r10buf)
156 *
157 */
Al Virodd0fc662005-10-07 07:46:04 +0100158static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
NeilBrowne879a872011-10-11 16:49:02 +1100160 struct r10conf *conf = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100161 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 struct bio *bio;
Ming Leif0250612017-03-17 00:12:33 +0800163 int j;
164 int nalloc, nalloc_rp;
165 struct resync_pages *rps;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 r10_bio = r10bio_pool_alloc(gfp_flags, conf);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100168 if (!r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
NeilBrown3ea7daa2012-05-22 13:53:47 +1000171 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) ||
172 test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 nalloc = conf->copies; /* resync */
174 else
175 nalloc = 2; /* recovery */
176
Ming Leif0250612017-03-17 00:12:33 +0800177 /* allocate once for all bios */
178 if (!conf->have_replacement)
179 nalloc_rp = nalloc;
180 else
181 nalloc_rp = nalloc * 2;
182 rps = kmalloc(sizeof(struct resync_pages) * nalloc_rp, gfp_flags);
183 if (!rps)
184 goto out_free_r10bio;
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 /*
187 * Allocate bios.
188 */
189 for (j = nalloc ; j-- ; ) {
NeilBrown67465572010-10-26 17:33:54 +1100190 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 if (!bio)
192 goto out_free_bio;
193 r10_bio->devs[j].bio = bio;
NeilBrown69335ef2011-12-23 10:17:54 +1100194 if (!conf->have_replacement)
195 continue;
196 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
197 if (!bio)
198 goto out_free_bio;
199 r10_bio->devs[j].repl_bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
201 /*
202 * Allocate RESYNC_PAGES data pages and attach them
203 * where needed.
204 */
Ming Leif0250612017-03-17 00:12:33 +0800205 for (j = 0; j < nalloc; j++) {
NeilBrown69335ef2011-12-23 10:17:54 +1100206 struct bio *rbio = r10_bio->devs[j].repl_bio;
Ming Leif0250612017-03-17 00:12:33 +0800207 struct resync_pages *rp, *rp_repl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Ming Leif0250612017-03-17 00:12:33 +0800209 rp = &rps[j];
210 if (rbio)
211 rp_repl = &rps[nalloc + j];
212
213 bio = r10_bio->devs[j].bio;
214
215 if (!j || test_bit(MD_RECOVERY_SYNC,
216 &conf->mddev->recovery)) {
217 if (resync_alloc_pages(rp, gfp_flags))
218 goto out_free_pages;
219 } else {
220 memcpy(rp, &rps[0], sizeof(*rp));
221 resync_get_all_pages(rp);
222 }
223
224 rp->idx = 0;
225 rp->raid_bio = r10_bio;
226 bio->bi_private = rp;
227 if (rbio) {
228 memcpy(rp_repl, rp, sizeof(*rp));
229 rbio->bi_private = rp_repl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 }
231 }
232
233 return r10_bio;
234
235out_free_pages:
Ming Leif0250612017-03-17 00:12:33 +0800236 while (--j >= 0)
237 resync_free_pages(&rps[j * 2]);
238
majianpeng5fdd2cf2012-05-22 13:55:03 +1000239 j = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240out_free_bio:
majianpeng5fdd2cf2012-05-22 13:55:03 +1000241 for ( ; j < nalloc; j++) {
242 if (r10_bio->devs[j].bio)
243 bio_put(r10_bio->devs[j].bio);
NeilBrown69335ef2011-12-23 10:17:54 +1100244 if (r10_bio->devs[j].repl_bio)
245 bio_put(r10_bio->devs[j].repl_bio);
246 }
Ming Leif0250612017-03-17 00:12:33 +0800247 kfree(rps);
248out_free_r10bio:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 r10bio_pool_free(r10_bio, conf);
250 return NULL;
251}
252
253static void r10buf_pool_free(void *__r10_bio, void *data)
254{
NeilBrowne879a872011-10-11 16:49:02 +1100255 struct r10conf *conf = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100256 struct r10bio *r10bio = __r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 int j;
Ming Leif0250612017-03-17 00:12:33 +0800258 struct resync_pages *rp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Ming Leif0250612017-03-17 00:12:33 +0800260 for (j = conf->copies; j--; ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 struct bio *bio = r10bio->devs[j].bio;
Ming Leif0250612017-03-17 00:12:33 +0800262
263 rp = get_resync_pages(bio);
264 resync_free_pages(rp);
265 bio_put(bio);
266
NeilBrown69335ef2011-12-23 10:17:54 +1100267 bio = r10bio->devs[j].repl_bio;
268 if (bio)
269 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 }
Ming Leif0250612017-03-17 00:12:33 +0800271
272 /* resync pages array stored in the 1st bio's .bi_private */
273 kfree(rp);
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 r10bio_pool_free(r10bio, conf);
276}
277
NeilBrowne879a872011-10-11 16:49:02 +1100278static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
280 int i;
281
282 for (i = 0; i < conf->copies; i++) {
283 struct bio **bio = & r10_bio->devs[i].bio;
NeilBrown749c55e2011-07-28 11:39:24 +1000284 if (!BIO_SPECIAL(*bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 bio_put(*bio);
286 *bio = NULL;
NeilBrown69335ef2011-12-23 10:17:54 +1100287 bio = &r10_bio->devs[i].repl_bio;
288 if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio))
289 bio_put(*bio);
290 *bio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 }
292}
293
NeilBrown9f2c9d12011-10-11 16:48:43 +1100294static void free_r10bio(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
NeilBrowne879a872011-10-11 16:49:02 +1100296 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 put_all_bios(conf, r10_bio);
299 mempool_free(r10_bio, conf->r10bio_pool);
300}
301
NeilBrown9f2c9d12011-10-11 16:48:43 +1100302static void put_buf(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
NeilBrowne879a872011-10-11 16:49:02 +1100304 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
306 mempool_free(r10_bio, conf->r10buf_pool);
307
NeilBrown0a27ec92006-01-06 00:20:13 -0800308 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
NeilBrown9f2c9d12011-10-11 16:48:43 +1100311static void reschedule_retry(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
313 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +1100314 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +1100315 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 spin_lock_irqsave(&conf->device_lock, flags);
318 list_add(&r10_bio->retry_list, &conf->retry_list);
NeilBrown4443ae12006-01-06 00:20:28 -0800319 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 spin_unlock_irqrestore(&conf->device_lock, flags);
321
Arthur Jones388667b2008-07-25 12:03:38 -0700322 /* wake up frozen array... */
323 wake_up(&conf->wait_barrier);
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 md_wakeup_thread(mddev->thread);
326}
327
328/*
329 * raid_end_bio_io() is called when we have finished servicing a mirrored
330 * operation and are ready to return a success/failure code to the buffer
331 * cache layer.
332 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100333static void raid_end_bio_io(struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 struct bio *bio = r10_bio->master_bio;
NeilBrowne879a872011-10-11 16:49:02 +1100336 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
NeilBrown856e08e2011-07-28 11:39:23 +1000338 if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200339 bio->bi_error = -EIO;
NeilBrownfd16f2e2017-03-15 14:05:13 +1100340
341 bio_endio(bio);
342 /*
343 * Wake up any possible resync thread that waits for the device
344 * to go idle.
345 */
346 allow_barrier(conf);
347
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 free_r10bio(r10_bio);
349}
350
351/*
352 * Update disk head position estimator based on IRQ completion info.
353 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100354static inline void update_head_pos(int slot, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
NeilBrowne879a872011-10-11 16:49:02 +1100356 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 conf->mirrors[r10_bio->devs[slot].devnum].head_position =
359 r10_bio->devs[slot].addr + (r10_bio->sectors);
360}
361
Namhyung Kim778ca012011-07-18 17:38:47 +1000362/*
363 * Find the disk number which triggered given bio
364 */
NeilBrowne879a872011-10-11 16:49:02 +1100365static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio,
NeilBrown69335ef2011-12-23 10:17:54 +1100366 struct bio *bio, int *slotp, int *replp)
Namhyung Kim778ca012011-07-18 17:38:47 +1000367{
368 int slot;
NeilBrown69335ef2011-12-23 10:17:54 +1100369 int repl = 0;
Namhyung Kim778ca012011-07-18 17:38:47 +1000370
NeilBrown69335ef2011-12-23 10:17:54 +1100371 for (slot = 0; slot < conf->copies; slot++) {
Namhyung Kim778ca012011-07-18 17:38:47 +1000372 if (r10_bio->devs[slot].bio == bio)
373 break;
NeilBrown69335ef2011-12-23 10:17:54 +1100374 if (r10_bio->devs[slot].repl_bio == bio) {
375 repl = 1;
376 break;
377 }
378 }
Namhyung Kim778ca012011-07-18 17:38:47 +1000379
380 BUG_ON(slot == conf->copies);
381 update_head_pos(slot, r10_bio);
382
NeilBrown749c55e2011-07-28 11:39:24 +1000383 if (slotp)
384 *slotp = slot;
NeilBrown69335ef2011-12-23 10:17:54 +1100385 if (replp)
386 *replp = repl;
Namhyung Kim778ca012011-07-18 17:38:47 +1000387 return r10_bio->devs[slot].devnum;
388}
389
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200390static void raid10_end_read_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200392 int uptodate = !bio->bi_error;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100393 struct r10bio *r10_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 int slot, dev;
NeilBrownabbf0982011-12-23 10:17:54 +1100395 struct md_rdev *rdev;
NeilBrowne879a872011-10-11 16:49:02 +1100396 struct r10conf *conf = r10_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 slot = r10_bio->read_slot;
399 dev = r10_bio->devs[slot].devnum;
NeilBrownabbf0982011-12-23 10:17:54 +1100400 rdev = r10_bio->devs[slot].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 /*
402 * this branch is our 'one mirror IO has finished' event handler:
403 */
NeilBrown4443ae12006-01-06 00:20:28 -0800404 update_head_pos(slot, r10_bio);
405
406 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 /*
408 * Set R10BIO_Uptodate in our master bio, so that
409 * we will return a good error code to the higher
410 * levels even if IO on some other mirrored buffer fails.
411 *
412 * The 'master' represents the composite IO operation to
413 * user-side. So if something waits for IO, then it will
414 * wait for the 'master' bio.
415 */
416 set_bit(R10BIO_Uptodate, &r10_bio->state);
NeilBrownfae8cc52012-02-14 11:10:10 +1100417 } else {
418 /* If all other devices that store this block have
419 * failed, we want to return the error upwards rather
420 * than fail the last device. Here we redefine
421 * "uptodate" to mean "Don't want to retry"
422 */
NeilBrown635f6412013-06-11 14:57:09 +1000423 if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state),
424 rdev->raid_disk))
NeilBrownfae8cc52012-02-14 11:10:10 +1100425 uptodate = 1;
NeilBrownfae8cc52012-02-14 11:10:10 +1100426 }
427 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 raid_end_bio_io(r10_bio);
NeilBrownabbf0982011-12-23 10:17:54 +1100429 rdev_dec_pending(rdev, conf->mddev);
NeilBrown4443ae12006-01-06 00:20:28 -0800430 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 /*
NeilBrown7c4e06f2011-05-11 14:53:17 +1000432 * oops, read error - keep the refcount on the rdev
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 */
434 char b[BDEVNAME_SIZE];
NeilBrown08464e02016-11-02 14:16:50 +1100435 pr_err_ratelimited("md/raid10:%s: %s: rescheduling sector %llu\n",
Christian Dietrich8bda4702011-07-27 11:00:36 +1000436 mdname(conf->mddev),
NeilBrownabbf0982011-12-23 10:17:54 +1100437 bdevname(rdev->bdev, b),
Christian Dietrich8bda4702011-07-27 11:00:36 +1000438 (unsigned long long)r10_bio->sector);
NeilBrown856e08e2011-07-28 11:39:23 +1000439 set_bit(R10BIO_ReadError, &r10_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 reschedule_retry(r10_bio);
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443
NeilBrown9f2c9d12011-10-11 16:48:43 +1100444static void close_write(struct r10bio *r10_bio)
NeilBrownbd870a12011-07-28 11:39:24 +1000445{
446 /* clear the bitmap if all writes complete successfully */
447 bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector,
448 r10_bio->sectors,
449 !test_bit(R10BIO_Degraded, &r10_bio->state),
450 0);
451 md_write_end(r10_bio->mddev);
452}
453
NeilBrown9f2c9d12011-10-11 16:48:43 +1100454static void one_write_done(struct r10bio *r10_bio)
NeilBrown19d5f832011-09-10 17:21:17 +1000455{
456 if (atomic_dec_and_test(&r10_bio->remaining)) {
457 if (test_bit(R10BIO_WriteError, &r10_bio->state))
458 reschedule_retry(r10_bio);
459 else {
460 close_write(r10_bio);
461 if (test_bit(R10BIO_MadeGood, &r10_bio->state))
462 reschedule_retry(r10_bio);
463 else
464 raid_end_bio_io(r10_bio);
465 }
466 }
467}
468
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200469static void raid10_end_write_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
NeilBrown9f2c9d12011-10-11 16:48:43 +1100471 struct r10bio *r10_bio = bio->bi_private;
Namhyung Kim778ca012011-07-18 17:38:47 +1000472 int dev;
NeilBrown749c55e2011-07-28 11:39:24 +1000473 int dec_rdev = 1;
NeilBrowne879a872011-10-11 16:49:02 +1100474 struct r10conf *conf = r10_bio->mddev->private;
NeilBrown475b0322011-12-23 10:17:55 +1100475 int slot, repl;
NeilBrown4ca40c22011-12-23 10:17:55 +1100476 struct md_rdev *rdev = NULL;
NeilBrown1919cbb2016-11-18 16:16:12 +1100477 struct bio *to_put = NULL;
Shaohua Li579ed342016-10-06 14:13:52 -0700478 bool discard_error;
479
480 discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
NeilBrown475b0322011-12-23 10:17:55 +1100482 dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
NeilBrown475b0322011-12-23 10:17:55 +1100484 if (repl)
485 rdev = conf->mirrors[dev].replacement;
NeilBrown4ca40c22011-12-23 10:17:55 +1100486 if (!rdev) {
487 smp_rmb();
488 repl = 0;
NeilBrown475b0322011-12-23 10:17:55 +1100489 rdev = conf->mirrors[dev].rdev;
NeilBrown4ca40c22011-12-23 10:17:55 +1100490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 /*
492 * this branch is our 'one mirror IO has finished' event handler:
493 */
Shaohua Li579ed342016-10-06 14:13:52 -0700494 if (bio->bi_error && !discard_error) {
NeilBrown475b0322011-12-23 10:17:55 +1100495 if (repl)
496 /* Never record new bad blocks to replacement,
497 * just fail it.
498 */
499 md_error(rdev->mddev, rdev);
500 else {
501 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +1100502 if (!test_and_set_bit(WantReplacement, &rdev->flags))
503 set_bit(MD_RECOVERY_NEEDED,
504 &rdev->mddev->recovery);
NeilBrown1919cbb2016-11-18 16:16:12 +1100505
NeilBrown475b0322011-12-23 10:17:55 +1100506 dec_rdev = 0;
NeilBrown1919cbb2016-11-18 16:16:12 +1100507 if (test_bit(FailFast, &rdev->flags) &&
508 (bio->bi_opf & MD_FAILFAST)) {
509 md_error(rdev->mddev, rdev);
510 if (!test_bit(Faulty, &rdev->flags))
511 /* This is the only remaining device,
512 * We need to retry the write without
513 * FailFast
514 */
515 set_bit(R10BIO_WriteError, &r10_bio->state);
516 else {
517 r10_bio->devs[slot].bio = NULL;
518 to_put = bio;
519 dec_rdev = 1;
520 }
521 } else
522 set_bit(R10BIO_WriteError, &r10_bio->state);
NeilBrown475b0322011-12-23 10:17:55 +1100523 }
NeilBrown749c55e2011-07-28 11:39:24 +1000524 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 /*
526 * Set R10BIO_Uptodate in our master bio, so that
527 * we will return a good error code for to the higher
528 * levels even if IO on some other mirrored buffer fails.
529 *
530 * The 'master' represents the composite IO operation to
531 * user-side. So if something waits for IO, then it will
532 * wait for the 'master' bio.
533 */
NeilBrown749c55e2011-07-28 11:39:24 +1000534 sector_t first_bad;
535 int bad_sectors;
536
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300537 /*
538 * Do not set R10BIO_Uptodate if the current device is
539 * rebuilding or Faulty. This is because we cannot use
540 * such device for properly reading the data back (we could
541 * potentially use it, if the current write would have felt
542 * before rdev->recovery_offset, but for simplicity we don't
543 * check this here.
544 */
545 if (test_bit(In_sync, &rdev->flags) &&
546 !test_bit(Faulty, &rdev->flags))
547 set_bit(R10BIO_Uptodate, &r10_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
NeilBrown749c55e2011-07-28 11:39:24 +1000549 /* Maybe we can clear some bad blocks. */
NeilBrown475b0322011-12-23 10:17:55 +1100550 if (is_badblock(rdev,
NeilBrown749c55e2011-07-28 11:39:24 +1000551 r10_bio->devs[slot].addr,
552 r10_bio->sectors,
Shaohua Li579ed342016-10-06 14:13:52 -0700553 &first_bad, &bad_sectors) && !discard_error) {
NeilBrown749c55e2011-07-28 11:39:24 +1000554 bio_put(bio);
NeilBrown475b0322011-12-23 10:17:55 +1100555 if (repl)
556 r10_bio->devs[slot].repl_bio = IO_MADE_GOOD;
557 else
558 r10_bio->devs[slot].bio = IO_MADE_GOOD;
NeilBrown749c55e2011-07-28 11:39:24 +1000559 dec_rdev = 0;
560 set_bit(R10BIO_MadeGood, &r10_bio->state);
561 }
562 }
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 /*
565 *
566 * Let's see if all mirrored write operations have finished
567 * already.
568 */
NeilBrown19d5f832011-09-10 17:21:17 +1000569 one_write_done(r10_bio);
NeilBrown749c55e2011-07-28 11:39:24 +1000570 if (dec_rdev)
NeilBrown884162d2012-11-22 15:12:09 +1100571 rdev_dec_pending(rdev, conf->mddev);
NeilBrown1919cbb2016-11-18 16:16:12 +1100572 if (to_put)
573 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576/*
577 * RAID10 layout manager
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300578 * As well as the chunksize and raid_disks count, there are two
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * parameters: near_copies and far_copies.
580 * near_copies * far_copies must be <= raid_disks.
581 * Normally one of these will be 1.
582 * If both are 1, we get raid0.
583 * If near_copies == raid_disks, we get raid1.
584 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300585 * Chunks are laid out in raid0 style with near_copies copies of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 * first chunk, followed by near_copies copies of the next chunk and
587 * so on.
588 * If far_copies > 1, then after 1/far_copies of the array has been assigned
589 * as described above, we start again with a device offset of near_copies.
590 * So we effectively have another copy of the whole array further down all
591 * the drives, but with blocks on different drives.
592 * With this layout, and block is never stored twice on the one device.
593 *
594 * raid10_find_phys finds the sector offset of a given virtual sector
NeilBrownc93983b2006-06-26 00:27:41 -0700595 * on each device that it is on.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 *
597 * raid10_find_virt does the reverse mapping, from a device and a
598 * sector offset to a virtual address
599 */
600
NeilBrownf8c9e742012-05-21 09:28:33 +1000601static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
603 int n,f;
604 sector_t sector;
605 sector_t chunk;
606 sector_t stripe;
607 int dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 int slot = 0;
Jonathan Brassow9a3152a2013-02-21 13:28:10 +1100609 int last_far_set_start, last_far_set_size;
610
611 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
612 last_far_set_start *= geo->far_set_size;
613
614 last_far_set_size = geo->far_set_size;
615 last_far_set_size += (geo->raid_disks % geo->far_set_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617 /* now calculate first sector/dev */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000618 chunk = r10bio->sector >> geo->chunk_shift;
619 sector = r10bio->sector & geo->chunk_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
NeilBrown5cf00fc2012-05-21 09:28:20 +1000621 chunk *= geo->near_copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 stripe = chunk;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000623 dev = sector_div(stripe, geo->raid_disks);
624 if (geo->far_offset)
625 stripe *= geo->far_copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
NeilBrown5cf00fc2012-05-21 09:28:20 +1000627 sector += stripe << geo->chunk_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 /* and calculate all the others */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000630 for (n = 0; n < geo->near_copies; n++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 int d = dev;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100632 int set;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 sector_t s = sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 r10bio->devs[slot].devnum = d;
Jonathan Brassow4c0ca262013-02-21 13:28:09 +1100635 r10bio->devs[slot].addr = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 slot++;
637
NeilBrown5cf00fc2012-05-21 09:28:20 +1000638 for (f = 1; f < geo->far_copies; f++) {
Jonathan Brassow475901a2013-02-21 13:28:10 +1100639 set = d / geo->far_set_size;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000640 d += geo->near_copies;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100641
Jonathan Brassow9a3152a2013-02-21 13:28:10 +1100642 if ((geo->raid_disks % geo->far_set_size) &&
643 (d > last_far_set_start)) {
644 d -= last_far_set_start;
645 d %= last_far_set_size;
646 d += last_far_set_start;
647 } else {
648 d %= geo->far_set_size;
649 d += geo->far_set_size * set;
650 }
NeilBrown5cf00fc2012-05-21 09:28:20 +1000651 s += geo->stride;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 r10bio->devs[slot].devnum = d;
653 r10bio->devs[slot].addr = s;
654 slot++;
655 }
656 dev++;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000657 if (dev >= geo->raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 dev = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000659 sector += (geo->chunk_mask + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 }
661 }
NeilBrownf8c9e742012-05-21 09:28:33 +1000662}
663
664static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio)
665{
666 struct geom *geo = &conf->geo;
667
668 if (conf->reshape_progress != MaxSector &&
669 ((r10bio->sector >= conf->reshape_progress) !=
670 conf->mddev->reshape_backwards)) {
671 set_bit(R10BIO_Previous, &r10bio->state);
672 geo = &conf->prev;
673 } else
674 clear_bit(R10BIO_Previous, &r10bio->state);
675
676 __raid10_find_phys(geo, r10bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677}
678
NeilBrowne879a872011-10-11 16:49:02 +1100679static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 sector_t offset, chunk, vchunk;
NeilBrownf8c9e742012-05-21 09:28:33 +1000682 /* Never use conf->prev as this is only called during resync
683 * or recovery, so reshape isn't happening
684 */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000685 struct geom *geo = &conf->geo;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100686 int far_set_start = (dev / geo->far_set_size) * geo->far_set_size;
687 int far_set_size = geo->far_set_size;
Jonathan Brassow9a3152a2013-02-21 13:28:10 +1100688 int last_far_set_start;
689
690 if (geo->raid_disks % geo->far_set_size) {
691 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
692 last_far_set_start *= geo->far_set_size;
693
694 if (dev >= last_far_set_start) {
695 far_set_size = geo->far_set_size;
696 far_set_size += (geo->raid_disks % geo->far_set_size);
697 far_set_start = last_far_set_start;
698 }
699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
NeilBrown5cf00fc2012-05-21 09:28:20 +1000701 offset = sector & geo->chunk_mask;
702 if (geo->far_offset) {
NeilBrownc93983b2006-06-26 00:27:41 -0700703 int fc;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000704 chunk = sector >> geo->chunk_shift;
705 fc = sector_div(chunk, geo->far_copies);
706 dev -= fc * geo->near_copies;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100707 if (dev < far_set_start)
708 dev += far_set_size;
NeilBrownc93983b2006-06-26 00:27:41 -0700709 } else {
NeilBrown5cf00fc2012-05-21 09:28:20 +1000710 while (sector >= geo->stride) {
711 sector -= geo->stride;
Jonathan Brassow475901a2013-02-21 13:28:10 +1100712 if (dev < (geo->near_copies + far_set_start))
713 dev += far_set_size - geo->near_copies;
NeilBrownc93983b2006-06-26 00:27:41 -0700714 else
NeilBrown5cf00fc2012-05-21 09:28:20 +1000715 dev -= geo->near_copies;
NeilBrownc93983b2006-06-26 00:27:41 -0700716 }
NeilBrown5cf00fc2012-05-21 09:28:20 +1000717 chunk = sector >> geo->chunk_shift;
NeilBrownc93983b2006-06-26 00:27:41 -0700718 }
NeilBrown5cf00fc2012-05-21 09:28:20 +1000719 vchunk = chunk * geo->raid_disks + dev;
720 sector_div(vchunk, geo->near_copies);
721 return (vchunk << geo->chunk_shift) + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724/*
725 * This routine returns the disk from which the requested read should
726 * be done. There is a per-array 'next expected sequential IO' sector
727 * number - if this matches on the next IO then we use the last disk.
728 * There is also a per-disk 'last know head position' sector that is
729 * maintained from IRQ contexts, both the normal and the resync IO
730 * completion handlers update this position correctly. If there is no
731 * perfect sequential match then we pick the disk whose head is closest.
732 *
733 * If there are 2 mirrors in the same 2 devices, performance degrades
734 * because position is mirror, not device based.
735 *
736 * The rdev for the device selected will have nr_pending incremented.
737 */
738
739/*
740 * FIXME: possibly should rethink readbalancing and do it differently
741 * depending on near_copies / far_copies geometry.
742 */
NeilBrown96c3fd12011-12-23 10:17:54 +1100743static struct md_rdev *read_balance(struct r10conf *conf,
744 struct r10bio *r10_bio,
745 int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000747 const sector_t this_sector = r10_bio->sector;
NeilBrown56d99122011-05-11 14:27:03 +1000748 int disk, slot;
NeilBrown856e08e2011-07-28 11:39:23 +1000749 int sectors = r10_bio->sectors;
750 int best_good_sectors;
NeilBrown56d99122011-05-11 14:27:03 +1000751 sector_t new_distance, best_dist;
Jonathan Brassow3bbae042012-07-31 10:03:52 +1000752 struct md_rdev *best_rdev, *rdev = NULL;
NeilBrown56d99122011-05-11 14:27:03 +1000753 int do_balance;
754 int best_slot;
NeilBrown5cf00fc2012-05-21 09:28:20 +1000755 struct geom *geo = &conf->geo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 raid10_find_phys(conf, r10_bio);
758 rcu_read_lock();
NeilBrown856e08e2011-07-28 11:39:23 +1000759 sectors = r10_bio->sectors;
NeilBrown56d99122011-05-11 14:27:03 +1000760 best_slot = -1;
NeilBrownabbf0982011-12-23 10:17:54 +1100761 best_rdev = NULL;
NeilBrown56d99122011-05-11 14:27:03 +1000762 best_dist = MaxSector;
NeilBrown856e08e2011-07-28 11:39:23 +1000763 best_good_sectors = 0;
NeilBrown56d99122011-05-11 14:27:03 +1000764 do_balance = 1;
NeilBrown8d3ca832016-11-18 16:16:12 +1100765 clear_bit(R10BIO_FailFast, &r10_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 /*
767 * Check if we can balance. We can balance on the whole
NeilBrown6cce3b22006-01-06 00:20:16 -0800768 * device if no resync is going on (recovery is ok), or below
769 * the resync window. We take the first readable disk when
770 * above the resync window.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 */
772 if (conf->mddev->recovery_cp < MaxSector
NeilBrown56d99122011-05-11 14:27:03 +1000773 && (this_sector + sectors >= conf->next_resync))
774 do_balance = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
NeilBrown56d99122011-05-11 14:27:03 +1000776 for (slot = 0; slot < conf->copies ; slot++) {
NeilBrown856e08e2011-07-28 11:39:23 +1000777 sector_t first_bad;
778 int bad_sectors;
779 sector_t dev_sector;
780
NeilBrown56d99122011-05-11 14:27:03 +1000781 if (r10_bio->devs[slot].bio == IO_BLOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 continue;
NeilBrown56d99122011-05-11 14:27:03 +1000783 disk = r10_bio->devs[slot].devnum;
NeilBrownabbf0982011-12-23 10:17:54 +1100784 rdev = rcu_dereference(conf->mirrors[disk].replacement);
785 if (rdev == NULL || test_bit(Faulty, &rdev->flags) ||
786 r10_bio->devs[slot].addr + sectors > rdev->recovery_offset)
787 rdev = rcu_dereference(conf->mirrors[disk].rdev);
NeilBrown050b6612012-03-19 12:46:39 +1100788 if (rdev == NULL ||
Kent Overstreet8ae12662015-04-27 23:48:34 -0700789 test_bit(Faulty, &rdev->flags))
NeilBrownabbf0982011-12-23 10:17:54 +1100790 continue;
791 if (!test_bit(In_sync, &rdev->flags) &&
792 r10_bio->devs[slot].addr + sectors > rdev->recovery_offset)
NeilBrown56d99122011-05-11 14:27:03 +1000793 continue;
794
NeilBrown856e08e2011-07-28 11:39:23 +1000795 dev_sector = r10_bio->devs[slot].addr;
796 if (is_badblock(rdev, dev_sector, sectors,
797 &first_bad, &bad_sectors)) {
798 if (best_dist < MaxSector)
799 /* Already have a better slot */
800 continue;
801 if (first_bad <= dev_sector) {
802 /* Cannot read here. If this is the
803 * 'primary' device, then we must not read
804 * beyond 'bad_sectors' from another device.
805 */
806 bad_sectors -= (dev_sector - first_bad);
807 if (!do_balance && sectors > bad_sectors)
808 sectors = bad_sectors;
809 if (best_good_sectors > sectors)
810 best_good_sectors = sectors;
811 } else {
812 sector_t good_sectors =
813 first_bad - dev_sector;
814 if (good_sectors > best_good_sectors) {
815 best_good_sectors = good_sectors;
816 best_slot = slot;
NeilBrownabbf0982011-12-23 10:17:54 +1100817 best_rdev = rdev;
NeilBrown856e08e2011-07-28 11:39:23 +1000818 }
819 if (!do_balance)
820 /* Must read from here */
821 break;
822 }
823 continue;
824 } else
825 best_good_sectors = sectors;
826
NeilBrown56d99122011-05-11 14:27:03 +1000827 if (!do_balance)
828 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
NeilBrown8d3ca832016-11-18 16:16:12 +1100830 if (best_slot >= 0)
831 /* At least 2 disks to choose from so failfast is OK */
832 set_bit(R10BIO_FailFast, &r10_bio->state);
NeilBrown22dfdf52005-11-28 13:44:09 -0800833 /* This optimisation is debatable, and completely destroys
834 * sequential read speed for 'far copies' arrays. So only
835 * keep it for 'near' arrays, and review those later.
836 */
NeilBrown5cf00fc2012-05-21 09:28:20 +1000837 if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending))
NeilBrown8d3ca832016-11-18 16:16:12 +1100838 new_distance = 0;
Keld Simonsen8ed3a192008-03-04 14:29:34 -0800839
840 /* for far > 1 always use the lowest address */
NeilBrown8d3ca832016-11-18 16:16:12 +1100841 else if (geo->far_copies > 1)
NeilBrown56d99122011-05-11 14:27:03 +1000842 new_distance = r10_bio->devs[slot].addr;
Keld Simonsen8ed3a192008-03-04 14:29:34 -0800843 else
NeilBrown56d99122011-05-11 14:27:03 +1000844 new_distance = abs(r10_bio->devs[slot].addr -
845 conf->mirrors[disk].head_position);
846 if (new_distance < best_dist) {
847 best_dist = new_distance;
848 best_slot = slot;
NeilBrownabbf0982011-12-23 10:17:54 +1100849 best_rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851 }
NeilBrownabbf0982011-12-23 10:17:54 +1100852 if (slot >= conf->copies) {
NeilBrown56d99122011-05-11 14:27:03 +1000853 slot = best_slot;
NeilBrownabbf0982011-12-23 10:17:54 +1100854 rdev = best_rdev;
855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
NeilBrown56d99122011-05-11 14:27:03 +1000857 if (slot >= 0) {
NeilBrown56d99122011-05-11 14:27:03 +1000858 atomic_inc(&rdev->nr_pending);
NeilBrown56d99122011-05-11 14:27:03 +1000859 r10_bio->read_slot = slot;
860 } else
NeilBrown96c3fd12011-12-23 10:17:54 +1100861 rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 rcu_read_unlock();
NeilBrown856e08e2011-07-28 11:39:23 +1000863 *max_sectors = best_good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
NeilBrown96c3fd12011-12-23 10:17:54 +1100865 return rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866}
867
NeilBrown5c675f82014-12-15 12:56:56 +1100868static int raid10_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700869{
NeilBrowne879a872011-10-11 16:49:02 +1100870 struct r10conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700871 int i, ret = 0;
872
Tejun Heo44522262015-05-22 17:13:26 -0400873 if ((bits & (1 << WB_async_congested)) &&
NeilBrown34db0cd2011-10-11 16:50:01 +1100874 conf->pending_count >= max_queued_requests)
875 return 1;
876
NeilBrown0d129222006-10-03 01:15:54 -0700877 rcu_read_lock();
NeilBrownf8c9e742012-05-21 09:28:33 +1000878 for (i = 0;
879 (i < conf->geo.raid_disks || i < conf->prev.raid_disks)
880 && ret == 0;
881 i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100882 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700883 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200884 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700885
Jan Karadc3b17c2017-02-02 15:56:50 +0100886 ret |= bdi_congested(q->backing_dev_info, bits);
NeilBrown0d129222006-10-03 01:15:54 -0700887 }
888 }
889 rcu_read_unlock();
890 return ret;
891}
892
NeilBrowne879a872011-10-11 16:49:02 +1100893static void flush_pending_writes(struct r10conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800894{
895 /* Any writes that have been queued but are awaiting
896 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800897 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800898 spin_lock_irq(&conf->device_lock);
899
900 if (conf->pending_bio_list.head) {
901 struct bio *bio;
902 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100903 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800904 spin_unlock_irq(&conf->device_lock);
905 /* flush any pending bitmap writes to disk
906 * before proceeding w/ I/O */
907 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100908 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800909
910 while (bio) { /* submit pending writes */
911 struct bio *next = bio->bi_next;
NeilBrowna9ae93c2016-11-04 16:46:03 +1100912 struct md_rdev *rdev = (void*)bio->bi_bdev;
NeilBrowna35e63e2008-03-04 14:29:29 -0800913 bio->bi_next = NULL;
NeilBrowna9ae93c2016-11-04 16:46:03 +1100914 bio->bi_bdev = rdev->bdev;
915 if (test_bit(Faulty, &rdev->flags)) {
916 bio->bi_error = -EIO;
917 bio_endio(bio);
918 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
919 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
Shaohua Li532a2a32012-10-11 13:30:52 +1100920 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200921 bio_endio(bio);
Shaohua Li532a2a32012-10-11 13:30:52 +1100922 else
923 generic_make_request(bio);
NeilBrowna35e63e2008-03-04 14:29:29 -0800924 bio = next;
925 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800926 } else
927 spin_unlock_irq(&conf->device_lock);
NeilBrowna35e63e2008-03-04 14:29:29 -0800928}
Jens Axboe7eaceac2011-03-10 08:52:07 +0100929
NeilBrown0a27ec92006-01-06 00:20:13 -0800930/* Barriers....
931 * Sometimes we need to suspend IO while we do something else,
932 * either some resync/recovery, or reconfigure the array.
933 * To do this we raise a 'barrier'.
934 * The 'barrier' is a counter that can be raised multiple times
935 * to count how many activities are happening which preclude
936 * normal IO.
937 * We can only raise the barrier if there is no pending IO.
938 * i.e. if nr_pending == 0.
939 * We choose only to raise the barrier if no-one is waiting for the
940 * barrier to go down. This means that as soon as an IO request
941 * is ready, no other operations which require a barrier will start
942 * until the IO request has had a chance.
943 *
944 * So: regular IO calls 'wait_barrier'. When that returns there
945 * is no backgroup IO happening, It must arrange to call
946 * allow_barrier when it has finished its IO.
947 * backgroup IO calls must call raise_barrier. Once that returns
948 * there is no normal IO happeing. It must arrange to call
949 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
NeilBrowne879a872011-10-11 16:49:02 +1100952static void raise_barrier(struct r10conf *conf, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
NeilBrown6cce3b22006-01-06 00:20:16 -0800954 BUG_ON(force && !conf->barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 spin_lock_irq(&conf->resync_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
NeilBrown6cce3b22006-01-06 00:20:16 -0800957 /* Wait until no block IO is waiting (unless 'force') */
958 wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting,
Lukas Czernereed8c022012-11-30 11:42:40 +0100959 conf->resync_lock);
NeilBrown0a27ec92006-01-06 00:20:13 -0800960
961 /* block any new IO from starting */
962 conf->barrier++;
963
NeilBrownc3b328a2011-04-18 18:25:43 +1000964 /* Now wait for all pending IO to complete */
NeilBrown0a27ec92006-01-06 00:20:13 -0800965 wait_event_lock_irq(conf->wait_barrier,
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +0200966 !atomic_read(&conf->nr_pending) && conf->barrier < RESYNC_DEPTH,
Lukas Czernereed8c022012-11-30 11:42:40 +0100967 conf->resync_lock);
NeilBrown0a27ec92006-01-06 00:20:13 -0800968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 spin_unlock_irq(&conf->resync_lock);
970}
971
NeilBrowne879a872011-10-11 16:49:02 +1100972static void lower_barrier(struct r10conf *conf)
NeilBrown0a27ec92006-01-06 00:20:13 -0800973{
974 unsigned long flags;
975 spin_lock_irqsave(&conf->resync_lock, flags);
976 conf->barrier--;
977 spin_unlock_irqrestore(&conf->resync_lock, flags);
978 wake_up(&conf->wait_barrier);
979}
980
NeilBrowne879a872011-10-11 16:49:02 +1100981static void wait_barrier(struct r10conf *conf)
NeilBrown0a27ec92006-01-06 00:20:13 -0800982{
983 spin_lock_irq(&conf->resync_lock);
984 if (conf->barrier) {
985 conf->nr_waiting++;
NeilBrownd6b42dc2012-03-19 12:46:38 +1100986 /* Wait for the barrier to drop.
987 * However if there are already pending
988 * requests (preventing the barrier from
989 * rising completely), and the
990 * pre-process bio queue isn't empty,
991 * then don't wait, as we need to empty
992 * that queue to get the nr_pending
993 * count down.
994 */
NeilBrown578b54a2016-11-14 16:30:21 +1100995 raid10_log(conf->mddev, "wait barrier");
NeilBrownd6b42dc2012-03-19 12:46:38 +1100996 wait_event_lock_irq(conf->wait_barrier,
997 !conf->barrier ||
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +0200998 (atomic_read(&conf->nr_pending) &&
NeilBrownd6b42dc2012-03-19 12:46:38 +1100999 current->bio_list &&
NeilBrownf5fe1b52017-03-10 17:00:47 +11001000 (!bio_list_empty(&current->bio_list[0]) ||
1001 !bio_list_empty(&current->bio_list[1]))),
Lukas Czernereed8c022012-11-30 11:42:40 +01001002 conf->resync_lock);
NeilBrown0a27ec92006-01-06 00:20:13 -08001003 conf->nr_waiting--;
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +02001004 if (!conf->nr_waiting)
1005 wake_up(&conf->wait_barrier);
NeilBrown0a27ec92006-01-06 00:20:13 -08001006 }
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +02001007 atomic_inc(&conf->nr_pending);
NeilBrown0a27ec92006-01-06 00:20:13 -08001008 spin_unlock_irq(&conf->resync_lock);
1009}
1010
NeilBrownfd16f2e2017-03-15 14:05:13 +11001011static void inc_pending(struct r10conf *conf)
1012{
1013 /* The current request requires multiple r10_bio, so
1014 * we need to increment the pending count.
1015 */
1016 WARN_ON(!atomic_read(&conf->nr_pending));
1017 atomic_inc(&conf->nr_pending);
1018}
1019
NeilBrowne879a872011-10-11 16:49:02 +11001020static void allow_barrier(struct r10conf *conf)
NeilBrown0a27ec92006-01-06 00:20:13 -08001021{
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +02001022 if ((atomic_dec_and_test(&conf->nr_pending)) ||
1023 (conf->array_freeze_pending))
1024 wake_up(&conf->wait_barrier);
NeilBrown0a27ec92006-01-06 00:20:13 -08001025}
1026
NeilBrowne2d59922013-06-12 11:01:22 +10001027static void freeze_array(struct r10conf *conf, int extra)
NeilBrown4443ae12006-01-06 00:20:28 -08001028{
1029 /* stop syncio and normal IO and wait for everything to
NeilBrownf1885932006-01-06 00:20:42 -08001030 * go quiet.
NeilBrown4443ae12006-01-06 00:20:28 -08001031 * We increment barrier and nr_waiting, and then
NeilBrowne2d59922013-06-12 11:01:22 +10001032 * wait until nr_pending match nr_queued+extra
NeilBrown1c830532008-03-04 14:29:35 -08001033 * This is called in the context of one normal IO request
1034 * that has failed. Thus any sync request that might be pending
1035 * will be blocked by nr_pending, and we need to wait for
1036 * pending IO requests to complete or be queued for re-try.
NeilBrowne2d59922013-06-12 11:01:22 +10001037 * Thus the number queued (nr_queued) plus this request (extra)
NeilBrown1c830532008-03-04 14:29:35 -08001038 * must match the number of pending IOs (nr_pending) before
1039 * we continue.
NeilBrown4443ae12006-01-06 00:20:28 -08001040 */
1041 spin_lock_irq(&conf->resync_lock);
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +02001042 conf->array_freeze_pending++;
NeilBrown4443ae12006-01-06 00:20:28 -08001043 conf->barrier++;
1044 conf->nr_waiting++;
Lukas Czernereed8c022012-11-30 11:42:40 +01001045 wait_event_lock_irq_cmd(conf->wait_barrier,
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +02001046 atomic_read(&conf->nr_pending) == conf->nr_queued+extra,
Lukas Czernereed8c022012-11-30 11:42:40 +01001047 conf->resync_lock,
1048 flush_pending_writes(conf));
NeilBrownc3b328a2011-04-18 18:25:43 +10001049
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +02001050 conf->array_freeze_pending--;
NeilBrown4443ae12006-01-06 00:20:28 -08001051 spin_unlock_irq(&conf->resync_lock);
1052}
1053
NeilBrowne879a872011-10-11 16:49:02 +11001054static void unfreeze_array(struct r10conf *conf)
NeilBrown4443ae12006-01-06 00:20:28 -08001055{
1056 /* reverse the effect of the freeze */
1057 spin_lock_irq(&conf->resync_lock);
1058 conf->barrier--;
1059 conf->nr_waiting--;
1060 wake_up(&conf->wait_barrier);
1061 spin_unlock_irq(&conf->resync_lock);
1062}
1063
NeilBrownf8c9e742012-05-21 09:28:33 +10001064static sector_t choose_data_offset(struct r10bio *r10_bio,
1065 struct md_rdev *rdev)
1066{
1067 if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) ||
1068 test_bit(R10BIO_Previous, &r10_bio->state))
1069 return rdev->data_offset;
1070 else
1071 return rdev->new_data_offset;
1072}
1073
NeilBrown57c67df2012-10-11 13:32:13 +11001074struct raid10_plug_cb {
1075 struct blk_plug_cb cb;
1076 struct bio_list pending;
1077 int pending_cnt;
1078};
1079
1080static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule)
1081{
1082 struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb,
1083 cb);
1084 struct mddev *mddev = plug->cb.data;
1085 struct r10conf *conf = mddev->private;
1086 struct bio *bio;
1087
NeilBrown874807a2012-11-27 12:14:40 +11001088 if (from_schedule || current->bio_list) {
NeilBrown57c67df2012-10-11 13:32:13 +11001089 spin_lock_irq(&conf->device_lock);
1090 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1091 conf->pending_count += plug->pending_cnt;
1092 spin_unlock_irq(&conf->device_lock);
NeilBrownee0b0242013-02-25 12:38:29 +11001093 wake_up(&conf->wait_barrier);
NeilBrown57c67df2012-10-11 13:32:13 +11001094 md_wakeup_thread(mddev->thread);
1095 kfree(plug);
1096 return;
1097 }
1098
1099 /* we aren't scheduling, so we can do the write-out directly. */
1100 bio = bio_list_get(&plug->pending);
1101 bitmap_unplug(mddev->bitmap);
1102 wake_up(&conf->wait_barrier);
1103
1104 while (bio) { /* submit pending writes */
1105 struct bio *next = bio->bi_next;
NeilBrowna9ae93c2016-11-04 16:46:03 +11001106 struct md_rdev *rdev = (void*)bio->bi_bdev;
NeilBrown57c67df2012-10-11 13:32:13 +11001107 bio->bi_next = NULL;
NeilBrowna9ae93c2016-11-04 16:46:03 +11001108 bio->bi_bdev = rdev->bdev;
1109 if (test_bit(Faulty, &rdev->flags)) {
1110 bio->bi_error = -EIO;
1111 bio_endio(bio);
1112 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
1113 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
Shaohua Li32f9f572013-04-28 18:26:38 +08001114 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001115 bio_endio(bio);
Shaohua Li32f9f572013-04-28 18:26:38 +08001116 else
1117 generic_make_request(bio);
NeilBrown57c67df2012-10-11 13:32:13 +11001118 bio = next;
1119 }
1120 kfree(plug);
1121}
1122
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001123static void raid10_read_request(struct mddev *mddev, struct bio *bio,
1124 struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
NeilBrowne879a872011-10-11 16:49:02 +11001126 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 struct bio *read_bio;
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001128 const int op = bio_op(bio);
1129 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001130 int max_sectors;
1131 sector_t sectors;
1132 struct md_rdev *rdev;
1133 int slot;
1134
1135 /*
1136 * Register the new request and wait if the reconstruction
1137 * thread has put up a bar for new requests.
1138 * Continue immediately if no resync is active currently.
1139 */
1140 wait_barrier(conf);
1141
NeilBrownfc9977d2017-04-05 14:05:51 +10001142 sectors = r10_bio->sectors;
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001143 while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
1144 bio->bi_iter.bi_sector < conf->reshape_progress &&
1145 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) {
1146 /*
1147 * IO spans the reshape position. Need to wait for reshape to
1148 * pass
1149 */
1150 raid10_log(conf->mddev, "wait reshape");
1151 allow_barrier(conf);
1152 wait_event(conf->wait_barrier,
1153 conf->reshape_progress <= bio->bi_iter.bi_sector ||
1154 conf->reshape_progress >= bio->bi_iter.bi_sector +
1155 sectors);
1156 wait_barrier(conf);
1157 }
1158
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001159 rdev = read_balance(conf, r10_bio, &max_sectors);
1160 if (!rdev) {
1161 raid_end_bio_io(r10_bio);
1162 return;
1163 }
NeilBrownfc9977d2017-04-05 14:05:51 +10001164 if (max_sectors < bio_sectors(bio)) {
1165 struct bio *split = bio_split(bio, max_sectors,
1166 GFP_NOIO, conf->bio_split);
1167 bio_chain(split, bio);
1168 generic_make_request(bio);
1169 bio = split;
1170 r10_bio->master_bio = bio;
1171 r10_bio->sectors = max_sectors;
1172 }
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001173 slot = r10_bio->read_slot;
1174
Ming Leid7a10302017-02-14 23:29:03 +08001175 read_bio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set);
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001176
1177 r10_bio->devs[slot].bio = read_bio;
1178 r10_bio->devs[slot].rdev = rdev;
1179
1180 read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr +
1181 choose_data_offset(r10_bio, rdev);
1182 read_bio->bi_bdev = rdev->bdev;
1183 read_bio->bi_end_io = raid10_end_read_request;
1184 bio_set_op_attrs(read_bio, op, do_sync);
1185 if (test_bit(FailFast, &rdev->flags) &&
1186 test_bit(R10BIO_FailFast, &r10_bio->state))
1187 read_bio->bi_opf |= MD_FAILFAST;
1188 read_bio->bi_private = r10_bio;
1189
1190 if (mddev->gendisk)
1191 trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev),
1192 read_bio, disk_devt(mddev->gendisk),
1193 r10_bio->sector);
NeilBrownfc9977d2017-04-05 14:05:51 +10001194 generic_make_request(read_bio);
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001195 return;
1196}
1197
Guoqing Jiang27f26a02017-03-20 17:46:04 +08001198static void raid10_write_one_disk(struct mddev *mddev, struct r10bio *r10_bio,
1199 struct bio *bio, bool replacement,
NeilBrownfc9977d2017-04-05 14:05:51 +10001200 int n_copy)
Guoqing Jiang27f26a02017-03-20 17:46:04 +08001201{
1202 const int op = bio_op(bio);
1203 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1204 const unsigned long do_fua = (bio->bi_opf & REQ_FUA);
1205 unsigned long flags;
1206 struct blk_plug_cb *cb;
1207 struct raid10_plug_cb *plug = NULL;
1208 struct r10conf *conf = mddev->private;
1209 struct md_rdev *rdev;
1210 int devnum = r10_bio->devs[n_copy].devnum;
1211 struct bio *mbio;
1212
1213 if (replacement) {
1214 rdev = conf->mirrors[devnum].replacement;
1215 if (rdev == NULL) {
1216 /* Replacement just got moved to main 'rdev' */
1217 smp_mb();
1218 rdev = conf->mirrors[devnum].rdev;
1219 }
1220 } else
1221 rdev = conf->mirrors[devnum].rdev;
1222
1223 mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set);
Guoqing Jiang27f26a02017-03-20 17:46:04 +08001224 if (replacement)
1225 r10_bio->devs[n_copy].repl_bio = mbio;
1226 else
1227 r10_bio->devs[n_copy].bio = mbio;
1228
1229 mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr +
1230 choose_data_offset(r10_bio, rdev));
1231 mbio->bi_bdev = rdev->bdev;
1232 mbio->bi_end_io = raid10_end_write_request;
1233 bio_set_op_attrs(mbio, op, do_sync | do_fua);
1234 if (!replacement && test_bit(FailFast,
1235 &conf->mirrors[devnum].rdev->flags)
1236 && enough(conf, devnum))
1237 mbio->bi_opf |= MD_FAILFAST;
1238 mbio->bi_private = r10_bio;
1239
1240 if (conf->mddev->gendisk)
1241 trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev),
1242 mbio, disk_devt(conf->mddev->gendisk),
1243 r10_bio->sector);
1244 /* flush_pending_writes() needs access to the rdev so...*/
1245 mbio->bi_bdev = (void *)rdev;
1246
1247 atomic_inc(&r10_bio->remaining);
1248
1249 cb = blk_check_plugged(raid10_unplug, mddev, sizeof(*plug));
1250 if (cb)
1251 plug = container_of(cb, struct raid10_plug_cb, cb);
1252 else
1253 plug = NULL;
1254 spin_lock_irqsave(&conf->device_lock, flags);
1255 if (plug) {
1256 bio_list_add(&plug->pending, mbio);
1257 plug->pending_cnt++;
1258 } else {
1259 bio_list_add(&conf->pending_bio_list, mbio);
1260 conf->pending_count++;
1261 }
1262 spin_unlock_irqrestore(&conf->device_lock, flags);
1263 if (!plug)
1264 md_wakeup_thread(mddev->thread);
1265}
1266
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001267static void raid10_write_request(struct mddev *mddev, struct bio *bio,
1268 struct r10bio *r10_bio)
1269{
1270 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11001272 struct md_rdev *blocked_rdev;
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001273 sector_t sectors;
NeilBrownd4432c22011-07-28 11:39:24 +10001274 int max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Tomasz Majchrzak9b622e22016-07-28 10:28:25 +02001276 md_write_start(mddev, bio);
1277
NeilBrowncc13b1d2014-05-05 13:34:37 +10001278 /*
1279 * Register the new request and wait if the reconstruction
1280 * thread has put up a bar for new requests.
1281 * Continue immediately if no resync is active currently.
1282 */
1283 wait_barrier(conf);
1284
NeilBrownfc9977d2017-04-05 14:05:51 +10001285 sectors = r10_bio->sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10001286 while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
Kent Overstreet4f024f32013-10-11 15:44:27 -07001287 bio->bi_iter.bi_sector < conf->reshape_progress &&
1288 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) {
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001289 /*
1290 * IO spans the reshape position. Need to wait for reshape to
1291 * pass
NeilBrown3ea7daa2012-05-22 13:53:47 +10001292 */
NeilBrown578b54a2016-11-14 16:30:21 +11001293 raid10_log(conf->mddev, "wait reshape");
NeilBrown3ea7daa2012-05-22 13:53:47 +10001294 allow_barrier(conf);
1295 wait_event(conf->wait_barrier,
Kent Overstreet4f024f32013-10-11 15:44:27 -07001296 conf->reshape_progress <= bio->bi_iter.bi_sector ||
1297 conf->reshape_progress >= bio->bi_iter.bi_sector +
1298 sectors);
NeilBrown3ea7daa2012-05-22 13:53:47 +10001299 wait_barrier(conf);
1300 }
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001301
NeilBrown3ea7daa2012-05-22 13:53:47 +10001302 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
NeilBrown3ea7daa2012-05-22 13:53:47 +10001303 (mddev->reshape_backwards
Kent Overstreet4f024f32013-10-11 15:44:27 -07001304 ? (bio->bi_iter.bi_sector < conf->reshape_safe &&
1305 bio->bi_iter.bi_sector + sectors > conf->reshape_progress)
1306 : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe &&
1307 bio->bi_iter.bi_sector < conf->reshape_progress))) {
NeilBrown3ea7daa2012-05-22 13:53:47 +10001308 /* Need to update reshape_position in metadata */
1309 mddev->reshape_position = conf->reshape_progress;
Shaohua Li29530792016-12-08 15:48:19 -08001310 set_mask_bits(&mddev->sb_flags, 0,
1311 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
NeilBrown3ea7daa2012-05-22 13:53:47 +10001312 md_wakeup_thread(mddev->thread);
NeilBrown578b54a2016-11-14 16:30:21 +11001313 raid10_log(conf->mddev, "wait reshape metadata");
NeilBrown3ea7daa2012-05-22 13:53:47 +10001314 wait_event(mddev->sb_wait,
Shaohua Li29530792016-12-08 15:48:19 -08001315 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
NeilBrown3ea7daa2012-05-22 13:53:47 +10001316
1317 conf->reshape_safe = mddev->reshape_position;
1318 }
1319
NeilBrown34db0cd2011-10-11 16:50:01 +11001320 if (conf->pending_count >= max_queued_requests) {
1321 md_wakeup_thread(mddev->thread);
NeilBrown578b54a2016-11-14 16:30:21 +11001322 raid10_log(mddev, "wait queued");
NeilBrown34db0cd2011-10-11 16:50:01 +11001323 wait_event(conf->wait_barrier,
1324 conf->pending_count < max_queued_requests);
1325 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07001326 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 * inc refcount on their rdev. Record them by setting
1328 * bios[x] to bio
NeilBrownd4432c22011-07-28 11:39:24 +10001329 * If there are known/acknowledged bad blocks on any device
1330 * on which we have seen a write error, we want to avoid
1331 * writing to those blocks. This potentially requires several
1332 * writes to write around the bad blocks. Each set of writes
NeilBrownfd16f2e2017-03-15 14:05:13 +11001333 * gets its own r10_bio with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001335
NeilBrown69335ef2011-12-23 10:17:54 +11001336 r10_bio->read_slot = -1; /* make sure repl_bio gets freed */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 raid10_find_phys(conf, r10_bio);
NeilBrownd4432c22011-07-28 11:39:24 +10001338retry_write:
Harvey Harrisoncb6969e2008-05-06 20:42:32 -07001339 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 rcu_read_lock();
NeilBrownd4432c22011-07-28 11:39:24 +10001341 max_sectors = r10_bio->sectors;
1342
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 for (i = 0; i < conf->copies; i++) {
1344 int d = r10_bio->devs[i].devnum;
NeilBrown3cb03002011-10-11 16:45:26 +11001345 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown475b0322011-12-23 10:17:55 +11001346 struct md_rdev *rrdev = rcu_dereference(
1347 conf->mirrors[d].replacement);
NeilBrown4ca40c22011-12-23 10:17:55 +11001348 if (rdev == rrdev)
1349 rrdev = NULL;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001350 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1351 atomic_inc(&rdev->nr_pending);
1352 blocked_rdev = rdev;
1353 break;
1354 }
NeilBrown475b0322011-12-23 10:17:55 +11001355 if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) {
1356 atomic_inc(&rrdev->nr_pending);
1357 blocked_rdev = rrdev;
1358 break;
1359 }
Kent Overstreet8ae12662015-04-27 23:48:34 -07001360 if (rdev && (test_bit(Faulty, &rdev->flags)))
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001361 rdev = NULL;
Kent Overstreet8ae12662015-04-27 23:48:34 -07001362 if (rrdev && (test_bit(Faulty, &rrdev->flags)))
NeilBrown475b0322011-12-23 10:17:55 +11001363 rrdev = NULL;
1364
NeilBrownd4432c22011-07-28 11:39:24 +10001365 r10_bio->devs[i].bio = NULL;
NeilBrown475b0322011-12-23 10:17:55 +11001366 r10_bio->devs[i].repl_bio = NULL;
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001367
1368 if (!rdev && !rrdev) {
NeilBrown6cce3b22006-01-06 00:20:16 -08001369 set_bit(R10BIO_Degraded, &r10_bio->state);
NeilBrownd4432c22011-07-28 11:39:24 +10001370 continue;
NeilBrown6cce3b22006-01-06 00:20:16 -08001371 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001372 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) {
NeilBrownd4432c22011-07-28 11:39:24 +10001373 sector_t first_bad;
1374 sector_t dev_sector = r10_bio->devs[i].addr;
1375 int bad_sectors;
1376 int is_bad;
1377
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001378 is_bad = is_badblock(rdev, dev_sector, max_sectors,
NeilBrownd4432c22011-07-28 11:39:24 +10001379 &first_bad, &bad_sectors);
1380 if (is_bad < 0) {
1381 /* Mustn't write here until the bad block
1382 * is acknowledged
1383 */
1384 atomic_inc(&rdev->nr_pending);
1385 set_bit(BlockedBadBlocks, &rdev->flags);
1386 blocked_rdev = rdev;
1387 break;
1388 }
1389 if (is_bad && first_bad <= dev_sector) {
1390 /* Cannot write here at all */
1391 bad_sectors -= (dev_sector - first_bad);
1392 if (bad_sectors < max_sectors)
1393 /* Mustn't write more than bad_sectors
1394 * to other devices yet
1395 */
1396 max_sectors = bad_sectors;
1397 /* We don't set R10BIO_Degraded as that
1398 * only applies if the disk is missing,
1399 * so it might be re-added, and we want to
1400 * know to recover this chunk.
1401 * In this case the device is here, and the
1402 * fact that this chunk is not in-sync is
1403 * recorded in the bad block log.
1404 */
1405 continue;
1406 }
1407 if (is_bad) {
1408 int good_sectors = first_bad - dev_sector;
1409 if (good_sectors < max_sectors)
1410 max_sectors = good_sectors;
1411 }
1412 }
NeilBrowne7c0c3f2012-11-22 14:42:49 +11001413 if (rdev) {
1414 r10_bio->devs[i].bio = bio;
1415 atomic_inc(&rdev->nr_pending);
1416 }
NeilBrown475b0322011-12-23 10:17:55 +11001417 if (rrdev) {
1418 r10_bio->devs[i].repl_bio = bio;
1419 atomic_inc(&rrdev->nr_pending);
1420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 }
1422 rcu_read_unlock();
1423
Dan Williams6bfe0b42008-04-30 00:52:32 -07001424 if (unlikely(blocked_rdev)) {
1425 /* Have to wait for this device to get unblocked, then retry */
1426 int j;
1427 int d;
1428
NeilBrown475b0322011-12-23 10:17:55 +11001429 for (j = 0; j < i; j++) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07001430 if (r10_bio->devs[j].bio) {
1431 d = r10_bio->devs[j].devnum;
1432 rdev_dec_pending(conf->mirrors[d].rdev, mddev);
1433 }
NeilBrown475b0322011-12-23 10:17:55 +11001434 if (r10_bio->devs[j].repl_bio) {
NeilBrown4ca40c22011-12-23 10:17:55 +11001435 struct md_rdev *rdev;
NeilBrown475b0322011-12-23 10:17:55 +11001436 d = r10_bio->devs[j].devnum;
NeilBrown4ca40c22011-12-23 10:17:55 +11001437 rdev = conf->mirrors[d].replacement;
1438 if (!rdev) {
1439 /* Race with remove_disk */
1440 smp_mb();
1441 rdev = conf->mirrors[d].rdev;
1442 }
1443 rdev_dec_pending(rdev, mddev);
NeilBrown475b0322011-12-23 10:17:55 +11001444 }
1445 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07001446 allow_barrier(conf);
NeilBrown578b54a2016-11-14 16:30:21 +11001447 raid10_log(conf->mddev, "wait rdev %d blocked", blocked_rdev->raid_disk);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001448 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1449 wait_barrier(conf);
1450 goto retry_write;
1451 }
1452
NeilBrown6b6c8112017-03-15 14:05:13 +11001453 if (max_sectors < r10_bio->sectors)
NeilBrownd4432c22011-07-28 11:39:24 +10001454 r10_bio->sectors = max_sectors;
NeilBrownfc9977d2017-04-05 14:05:51 +10001455
1456 if (r10_bio->sectors < bio_sectors(bio)) {
1457 struct bio *split = bio_split(bio, r10_bio->sectors,
1458 GFP_NOIO, conf->bio_split);
1459 bio_chain(split, bio);
1460 generic_make_request(bio);
1461 bio = split;
1462 r10_bio->master_bio = bio;
1463 }
NeilBrownd4432c22011-07-28 11:39:24 +10001464
NeilBrown4e780642010-10-19 12:54:01 +11001465 atomic_set(&r10_bio->remaining, 1);
NeilBrownd4432c22011-07-28 11:39:24 +10001466 bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0);
NeilBrown06d91a52005-06-21 17:17:12 -07001467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 for (i = 0; i < conf->copies; i++) {
Guoqing Jiang27f26a02017-03-20 17:46:04 +08001469 if (r10_bio->devs[i].bio)
NeilBrownfc9977d2017-04-05 14:05:51 +10001470 raid10_write_one_disk(mddev, r10_bio, bio, false, i);
Guoqing Jiang27f26a02017-03-20 17:46:04 +08001471 if (r10_bio->devs[i].repl_bio)
NeilBrownfc9977d2017-04-05 14:05:51 +10001472 raid10_write_one_disk(mddev, r10_bio, bio, true, i);
NeilBrownd4432c22011-07-28 11:39:24 +10001473 }
NeilBrown079fa162011-09-10 17:21:23 +10001474 one_write_done(r10_bio);
Kent Overstreet20d01892013-11-23 18:21:01 -08001475}
1476
NeilBrownfc9977d2017-04-05 14:05:51 +10001477static void __make_request(struct mddev *mddev, struct bio *bio, int sectors)
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001478{
1479 struct r10conf *conf = mddev->private;
1480 struct r10bio *r10_bio;
1481
1482 r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
1483
1484 r10_bio->master_bio = bio;
NeilBrownfc9977d2017-04-05 14:05:51 +10001485 r10_bio->sectors = sectors;
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001486
1487 r10_bio->mddev = mddev;
1488 r10_bio->sector = bio->bi_iter.bi_sector;
1489 r10_bio->state = 0;
1490
Robert LeBlancbb5f1ed2016-12-05 13:02:58 -07001491 if (bio_data_dir(bio) == READ)
1492 raid10_read_request(mddev, bio, r10_bio);
1493 else
1494 raid10_write_request(mddev, bio, r10_bio);
1495}
1496
Shaohua Li849674e2016-01-20 13:52:20 -08001497static void raid10_make_request(struct mddev *mddev, struct bio *bio)
Kent Overstreet20d01892013-11-23 18:21:01 -08001498{
1499 struct r10conf *conf = mddev->private;
1500 sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask);
1501 int chunk_sects = chunk_mask + 1;
NeilBrownfc9977d2017-04-05 14:05:51 +10001502 int sectors = bio_sectors(bio);
Kent Overstreet20d01892013-11-23 18:21:01 -08001503
Jens Axboe1eff9d32016-08-05 15:35:16 -06001504 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
Kent Overstreet20d01892013-11-23 18:21:01 -08001505 md_flush_request(mddev, bio);
1506 return;
1507 }
1508
NeilBrownfc9977d2017-04-05 14:05:51 +10001509 /*
1510 * If this request crosses a chunk boundary, we need to split
1511 * it.
1512 */
1513 if (unlikely((bio->bi_iter.bi_sector & chunk_mask) +
1514 sectors > chunk_sects
1515 && (conf->geo.near_copies < conf->geo.raid_disks
1516 || conf->prev.near_copies <
1517 conf->prev.raid_disks)))
1518 sectors = chunk_sects -
1519 (bio->bi_iter.bi_sector &
1520 (chunk_sects - 1));
1521 __make_request(mddev, bio, sectors);
NeilBrown079fa162011-09-10 17:21:23 +10001522
1523 /* In case raid10d snuck in to freeze_array */
1524 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525}
1526
Shaohua Li849674e2016-01-20 13:52:20 -08001527static void raid10_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528{
NeilBrowne879a872011-10-11 16:49:02 +11001529 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 int i;
1531
NeilBrown5cf00fc2012-05-21 09:28:20 +10001532 if (conf->geo.near_copies < conf->geo.raid_disks)
Andre Noll9d8f0362009-06-18 08:45:01 +10001533 seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2);
NeilBrown5cf00fc2012-05-21 09:28:20 +10001534 if (conf->geo.near_copies > 1)
1535 seq_printf(seq, " %d near-copies", conf->geo.near_copies);
1536 if (conf->geo.far_copies > 1) {
1537 if (conf->geo.far_offset)
1538 seq_printf(seq, " %d offset-copies", conf->geo.far_copies);
NeilBrownc93983b2006-06-26 00:27:41 -07001539 else
NeilBrown5cf00fc2012-05-21 09:28:20 +10001540 seq_printf(seq, " %d far-copies", conf->geo.far_copies);
NeilBrown8bce6d32015-10-22 13:20:15 +11001541 if (conf->geo.far_set_size != conf->geo.raid_disks)
1542 seq_printf(seq, " %d devices per set", conf->geo.far_set_size);
NeilBrownc93983b2006-06-26 00:27:41 -07001543 }
NeilBrown5cf00fc2012-05-21 09:28:20 +10001544 seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks,
1545 conf->geo.raid_disks - mddev->degraded);
NeilBrownd44b0a92016-06-02 16:19:52 +10001546 rcu_read_lock();
1547 for (i = 0; i < conf->geo.raid_disks; i++) {
1548 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1549 seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1550 }
1551 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 seq_printf(seq, "]");
1553}
1554
NeilBrown700c7212011-07-27 11:00:36 +10001555/* check if there are enough drives for
1556 * every block to appear on atleast one.
1557 * Don't consider the device numbered 'ignore'
1558 * as we might be about to remove it.
1559 */
NeilBrown635f6412013-06-11 14:57:09 +10001560static int _enough(struct r10conf *conf, int previous, int ignore)
NeilBrown700c7212011-07-27 11:00:36 +10001561{
1562 int first = 0;
NeilBrown725d6e52013-06-11 15:08:03 +10001563 int has_enough = 0;
NeilBrown635f6412013-06-11 14:57:09 +10001564 int disks, ncopies;
1565 if (previous) {
1566 disks = conf->prev.raid_disks;
1567 ncopies = conf->prev.near_copies;
1568 } else {
1569 disks = conf->geo.raid_disks;
1570 ncopies = conf->geo.near_copies;
1571 }
NeilBrown700c7212011-07-27 11:00:36 +10001572
NeilBrown725d6e52013-06-11 15:08:03 +10001573 rcu_read_lock();
NeilBrown700c7212011-07-27 11:00:36 +10001574 do {
1575 int n = conf->copies;
1576 int cnt = 0;
NeilBrown80b48122012-09-27 12:35:21 +10001577 int this = first;
NeilBrown700c7212011-07-27 11:00:36 +10001578 while (n--) {
NeilBrown725d6e52013-06-11 15:08:03 +10001579 struct md_rdev *rdev;
1580 if (this != ignore &&
1581 (rdev = rcu_dereference(conf->mirrors[this].rdev)) &&
1582 test_bit(In_sync, &rdev->flags))
NeilBrown700c7212011-07-27 11:00:36 +10001583 cnt++;
NeilBrown635f6412013-06-11 14:57:09 +10001584 this = (this+1) % disks;
NeilBrown700c7212011-07-27 11:00:36 +10001585 }
1586 if (cnt == 0)
NeilBrown725d6e52013-06-11 15:08:03 +10001587 goto out;
NeilBrown635f6412013-06-11 14:57:09 +10001588 first = (first + ncopies) % disks;
NeilBrown700c7212011-07-27 11:00:36 +10001589 } while (first != 0);
NeilBrown725d6e52013-06-11 15:08:03 +10001590 has_enough = 1;
1591out:
1592 rcu_read_unlock();
1593 return has_enough;
NeilBrown700c7212011-07-27 11:00:36 +10001594}
1595
NeilBrownf8c9e742012-05-21 09:28:33 +10001596static int enough(struct r10conf *conf, int ignore)
1597{
NeilBrown635f6412013-06-11 14:57:09 +10001598 /* when calling 'enough', both 'prev' and 'geo' must
1599 * be stable.
1600 * This is ensured if ->reconfig_mutex or ->device_lock
1601 * is held.
1602 */
1603 return _enough(conf, 0, ignore) &&
1604 _enough(conf, 1, ignore);
NeilBrownf8c9e742012-05-21 09:28:33 +10001605}
1606
Shaohua Li849674e2016-01-20 13:52:20 -08001607static void raid10_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608{
1609 char b[BDEVNAME_SIZE];
NeilBrowne879a872011-10-11 16:49:02 +11001610 struct r10conf *conf = mddev->private;
NeilBrown635f6412013-06-11 14:57:09 +10001611 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
1613 /*
1614 * If it is not operational, then we have already marked it as dead
1615 * else if it is the last working disks, ignore the error, let the
1616 * next level up know.
1617 * else mark the drive as failed
1618 */
NeilBrown635f6412013-06-11 14:57:09 +10001619 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownb2d444d2005-11-08 21:39:31 -08001620 if (test_bit(In_sync, &rdev->flags)
NeilBrown635f6412013-06-11 14:57:09 +10001621 && !enough(conf, rdev->raid_disk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 /*
1623 * Don't fail the drive, just return an IO error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 */
NeilBrownc04be0a2006-10-03 01:15:53 -07001625 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown635f6412013-06-11 14:57:09 +10001626 return;
1627 }
NeilBrown2446dba2014-07-31 10:16:29 +10001628 if (test_and_clear_bit(In_sync, &rdev->flags))
NeilBrown635f6412013-06-11 14:57:09 +10001629 mddev->degraded++;
NeilBrown2446dba2014-07-31 10:16:29 +10001630 /*
1631 * If recovery is running, make sure it aborts.
1632 */
1633 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrownde393cd2011-07-28 11:31:48 +10001634 set_bit(Blocked, &rdev->flags);
NeilBrownb2d444d2005-11-08 21:39:31 -08001635 set_bit(Faulty, &rdev->flags);
Shaohua Li29530792016-12-08 15:48:19 -08001636 set_mask_bits(&mddev->sb_flags, 0,
1637 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
NeilBrown635f6412013-06-11 14:57:09 +10001638 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrown08464e02016-11-02 14:16:50 +11001639 pr_crit("md/raid10:%s: Disk failure on %s, disabling device.\n"
1640 "md/raid10:%s: Operation continuing on %d devices.\n",
1641 mdname(mddev), bdevname(rdev->bdev, b),
1642 mdname(mddev), conf->geo.raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643}
1644
NeilBrowne879a872011-10-11 16:49:02 +11001645static void print_conf(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646{
1647 int i;
NeilBrown4056ca52016-06-02 16:19:52 +10001648 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
NeilBrown08464e02016-11-02 14:16:50 +11001650 pr_debug("RAID10 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (!conf) {
NeilBrown08464e02016-11-02 14:16:50 +11001652 pr_debug("(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 return;
1654 }
NeilBrown08464e02016-11-02 14:16:50 +11001655 pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded,
1656 conf->geo.raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
NeilBrown4056ca52016-06-02 16:19:52 +10001658 /* This is only called with ->reconfix_mutex held, so
1659 * rcu protection of rdev is not needed */
NeilBrown5cf00fc2012-05-21 09:28:20 +10001660 for (i = 0; i < conf->geo.raid_disks; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 char b[BDEVNAME_SIZE];
NeilBrown4056ca52016-06-02 16:19:52 +10001662 rdev = conf->mirrors[i].rdev;
1663 if (rdev)
NeilBrown08464e02016-11-02 14:16:50 +11001664 pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n",
1665 i, !test_bit(In_sync, &rdev->flags),
1666 !test_bit(Faulty, &rdev->flags),
1667 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 }
1669}
1670
NeilBrowne879a872011-10-11 16:49:02 +11001671static void close_sync(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672{
NeilBrown0a27ec92006-01-06 00:20:13 -08001673 wait_barrier(conf);
1674 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 mempool_destroy(conf->r10buf_pool);
1677 conf->r10buf_pool = NULL;
1678}
1679
NeilBrownfd01b882011-10-11 16:47:53 +11001680static int raid10_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
1682 int i;
NeilBrowne879a872011-10-11 16:49:02 +11001683 struct r10conf *conf = mddev->private;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10001684 struct raid10_info *tmp;
NeilBrown6b965622010-08-18 11:56:59 +10001685 int count = 0;
1686 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
1688 /*
1689 * Find all non-in_sync disks within the RAID10 configuration
1690 * and mark them in_sync
1691 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10001692 for (i = 0; i < conf->geo.raid_disks; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 tmp = conf->mirrors + i;
NeilBrown4ca40c22011-12-23 10:17:55 +11001694 if (tmp->replacement
1695 && tmp->replacement->recovery_offset == MaxSector
1696 && !test_bit(Faulty, &tmp->replacement->flags)
1697 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
1698 /* Replacement has just become active */
1699 if (!tmp->rdev
1700 || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
1701 count++;
1702 if (tmp->rdev) {
1703 /* Replaced device not technically faulty,
1704 * but we need to be sure it gets removed
1705 * and never re-added.
1706 */
1707 set_bit(Faulty, &tmp->rdev->flags);
1708 sysfs_notify_dirent_safe(
1709 tmp->rdev->sysfs_state);
1710 }
1711 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
1712 } else if (tmp->rdev
Lukasz Dorau61e49472013-10-24 12:55:17 +11001713 && tmp->rdev->recovery_offset == MaxSector
NeilBrown4ca40c22011-12-23 10:17:55 +11001714 && !test_bit(Faulty, &tmp->rdev->flags)
1715 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001716 count++;
Jonathan Brassow2863b9e2012-10-11 13:38:58 +11001717 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719 }
NeilBrown6b965622010-08-18 11:56:59 +10001720 spin_lock_irqsave(&conf->device_lock, flags);
1721 mddev->degraded -= count;
1722 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
1724 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001725 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726}
1727
NeilBrownfd01b882011-10-11 16:47:53 +11001728static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729{
NeilBrowne879a872011-10-11 16:49:02 +11001730 struct r10conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001731 int err = -EEXIST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 int mirror;
Neil Brown6c2fce22008-06-28 08:31:31 +10001733 int first = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10001734 int last = conf->geo.raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
1736 if (mddev->recovery_cp < MaxSector)
1737 /* only hot-add to in-sync arrays, as recovery is
1738 * very different from resync
1739 */
Neil Brown199050e2008-06-28 08:31:33 +10001740 return -EBUSY;
NeilBrown635f6412013-06-11 14:57:09 +10001741 if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1))
Neil Brown199050e2008-06-28 08:31:33 +10001742 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Dan Williams1501efa2016-01-13 16:00:07 -08001744 if (md_integrity_add_rdev(rdev, mddev))
1745 return -ENXIO;
1746
NeilBrowna53a6c82008-11-06 17:28:20 +11001747 if (rdev->raid_disk >= 0)
Neil Brown6c2fce22008-06-28 08:31:31 +10001748 first = last = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
Namhyung Kim2c4193d2011-07-18 17:38:43 +10001750 if (rdev->saved_raid_disk >= first &&
NeilBrown6cce3b22006-01-06 00:20:16 -08001751 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1752 mirror = rdev->saved_raid_disk;
1753 else
Neil Brown6c2fce22008-06-28 08:31:31 +10001754 mirror = first;
NeilBrown2bb77732011-07-27 11:00:36 +10001755 for ( ; mirror <= last ; mirror++) {
Jonathan Brassowdc280d982012-07-31 10:03:52 +10001756 struct raid10_info *p = &conf->mirrors[mirror];
NeilBrown2bb77732011-07-27 11:00:36 +10001757 if (p->recovery_disabled == mddev->recovery_disabled)
1758 continue;
NeilBrownb7044d42011-12-23 10:17:56 +11001759 if (p->rdev) {
1760 if (!test_bit(WantReplacement, &p->rdev->flags) ||
1761 p->replacement != NULL)
1762 continue;
1763 clear_bit(In_sync, &rdev->flags);
1764 set_bit(Replacement, &rdev->flags);
1765 rdev->raid_disk = mirror;
1766 err = 0;
Jonathan Brassow9092c022013-05-02 14:19:24 -05001767 if (mddev->gendisk)
1768 disk_stack_limits(mddev->gendisk, rdev->bdev,
1769 rdev->data_offset << 9);
NeilBrownb7044d42011-12-23 10:17:56 +11001770 conf->fullsync = 1;
1771 rcu_assign_pointer(p->replacement, rdev);
1772 break;
1773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Jonathan Brassow9092c022013-05-02 14:19:24 -05001775 if (mddev->gendisk)
1776 disk_stack_limits(mddev->gendisk, rdev->bdev,
1777 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
NeilBrown2bb77732011-07-27 11:00:36 +10001779 p->head_position = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11001780 p->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown2bb77732011-07-27 11:00:36 +10001781 rdev->raid_disk = mirror;
1782 err = 0;
1783 if (rdev->saved_raid_disk != mirror)
1784 conf->fullsync = 1;
1785 rcu_assign_pointer(p->rdev, rdev);
1786 break;
1787 }
Jonathan Brassowed30be02012-10-31 11:42:30 +11001788 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Shaohua Li532a2a32012-10-11 13:30:52 +11001789 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001792 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793}
1794
NeilBrownb8321b62011-12-23 10:17:51 +11001795static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796{
NeilBrowne879a872011-10-11 16:49:02 +11001797 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001799 int number = rdev->raid_disk;
NeilBrownc8ab9032011-12-23 10:17:54 +11001800 struct md_rdev **rdevp;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10001801 struct raid10_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803 print_conf(conf);
NeilBrownc8ab9032011-12-23 10:17:54 +11001804 if (rdev == p->rdev)
1805 rdevp = &p->rdev;
1806 else if (rdev == p->replacement)
1807 rdevp = &p->replacement;
1808 else
1809 return 0;
1810
1811 if (test_bit(In_sync, &rdev->flags) ||
1812 atomic_read(&rdev->nr_pending)) {
1813 err = -EBUSY;
1814 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 }
NeilBrownd787be42016-06-02 16:19:53 +10001816 /* Only remove non-faulty devices if recovery
NeilBrownc8ab9032011-12-23 10:17:54 +11001817 * is not possible.
1818 */
1819 if (!test_bit(Faulty, &rdev->flags) &&
1820 mddev->recovery_disabled != p->recovery_disabled &&
NeilBrown4ca40c22011-12-23 10:17:55 +11001821 (!p->replacement || p->replacement == rdev) &&
NeilBrown63aced62012-05-22 13:55:33 +10001822 number < conf->geo.raid_disks &&
NeilBrownc8ab9032011-12-23 10:17:54 +11001823 enough(conf, -1)) {
1824 err = -EBUSY;
1825 goto abort;
1826 }
1827 *rdevp = NULL;
NeilBrownd787be42016-06-02 16:19:53 +10001828 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1829 synchronize_rcu();
1830 if (atomic_read(&rdev->nr_pending)) {
1831 /* lost the race, try later */
1832 err = -EBUSY;
1833 *rdevp = rdev;
1834 goto abort;
1835 }
1836 }
1837 if (p->replacement) {
NeilBrown4ca40c22011-12-23 10:17:55 +11001838 /* We must have just cleared 'rdev' */
1839 p->rdev = p->replacement;
1840 clear_bit(Replacement, &p->replacement->flags);
1841 smp_mb(); /* Make sure other CPUs may see both as identical
1842 * but will never see neither -- if they are careful.
1843 */
1844 p->replacement = NULL;
1845 clear_bit(WantReplacement, &rdev->flags);
1846 } else
1847 /* We might have just remove the Replacement as faulty
1848 * Clear the flag just in case
1849 */
1850 clear_bit(WantReplacement, &rdev->flags);
1851
NeilBrownc8ab9032011-12-23 10:17:54 +11001852 err = md_integrity_register(mddev);
1853
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854abort:
1855
1856 print_conf(conf);
1857 return err;
1858}
1859
Ming Lei81fa1522017-03-17 00:12:32 +08001860static void __end_sync_read(struct r10bio *r10_bio, struct bio *bio, int d)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861{
NeilBrowne879a872011-10-11 16:49:02 +11001862 struct r10conf *conf = r10_bio->mddev->private;
NeilBrown0eb3ff12006-01-06 00:20:29 -08001863
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001864 if (!bio->bi_error)
NeilBrown0eb3ff12006-01-06 00:20:29 -08001865 set_bit(R10BIO_Uptodate, &r10_bio->state);
NeilBrowne684e412011-07-28 11:39:25 +10001866 else
1867 /* The write handler will notice the lack of
1868 * R10BIO_Uptodate and record any errors etc
1869 */
NeilBrown4dbcdc72006-01-06 00:20:52 -08001870 atomic_add(r10_bio->sectors,
1871 &conf->mirrors[d].rdev->corrected_errors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873 /* for reconstruct, we always reschedule after a read.
1874 * for resync, only after all reads
1875 */
NeilBrown73d5c382009-02-25 13:18:47 +11001876 rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
1878 atomic_dec_and_test(&r10_bio->remaining)) {
1879 /* we have read all the blocks,
1880 * do the comparison in process context in raid10d
1881 */
1882 reschedule_retry(r10_bio);
1883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884}
1885
Ming Lei81fa1522017-03-17 00:12:32 +08001886static void end_sync_read(struct bio *bio)
1887{
Ming Leif0250612017-03-17 00:12:33 +08001888 struct r10bio *r10_bio = get_resync_r10bio(bio);
Ming Lei81fa1522017-03-17 00:12:32 +08001889 struct r10conf *conf = r10_bio->mddev->private;
1890 int d = find_bio_disk(conf, r10_bio, bio, NULL, NULL);
1891
1892 __end_sync_read(r10_bio, bio, d);
1893}
1894
1895static void end_reshape_read(struct bio *bio)
1896{
Ming Leif0250612017-03-17 00:12:33 +08001897 /* reshape read bio isn't allocated from r10buf_pool */
Ming Lei81fa1522017-03-17 00:12:32 +08001898 struct r10bio *r10_bio = bio->bi_private;
1899
1900 __end_sync_read(r10_bio, bio, r10_bio->read_slot);
1901}
1902
NeilBrown9f2c9d12011-10-11 16:48:43 +11001903static void end_sync_request(struct r10bio *r10_bio)
NeilBrown5e570282011-07-28 11:39:25 +10001904{
NeilBrownfd01b882011-10-11 16:47:53 +11001905 struct mddev *mddev = r10_bio->mddev;
NeilBrown5e570282011-07-28 11:39:25 +10001906
1907 while (atomic_dec_and_test(&r10_bio->remaining)) {
1908 if (r10_bio->master_bio == NULL) {
1909 /* the primary of several recovery bios */
1910 sector_t s = r10_bio->sectors;
1911 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
1912 test_bit(R10BIO_WriteError, &r10_bio->state))
1913 reschedule_retry(r10_bio);
1914 else
1915 put_buf(r10_bio);
1916 md_done_sync(mddev, s, 1);
1917 break;
1918 } else {
NeilBrown9f2c9d12011-10-11 16:48:43 +11001919 struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio;
NeilBrown5e570282011-07-28 11:39:25 +10001920 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
1921 test_bit(R10BIO_WriteError, &r10_bio->state))
1922 reschedule_retry(r10_bio);
1923 else
1924 put_buf(r10_bio);
1925 r10_bio = r10_bio2;
1926 }
1927 }
1928}
1929
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001930static void end_sync_write(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931{
Ming Leif0250612017-03-17 00:12:33 +08001932 struct r10bio *r10_bio = get_resync_r10bio(bio);
NeilBrownfd01b882011-10-11 16:47:53 +11001933 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11001934 struct r10conf *conf = mddev->private;
Namhyung Kim778ca012011-07-18 17:38:47 +10001935 int d;
NeilBrown749c55e2011-07-28 11:39:24 +10001936 sector_t first_bad;
1937 int bad_sectors;
1938 int slot;
NeilBrown9ad1aef2011-12-23 10:17:55 +11001939 int repl;
NeilBrown4ca40c22011-12-23 10:17:55 +11001940 struct md_rdev *rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
NeilBrown9ad1aef2011-12-23 10:17:55 +11001942 d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
1943 if (repl)
1944 rdev = conf->mirrors[d].replacement;
NeilBrown547414d2012-03-13 11:21:20 +11001945 else
NeilBrown9ad1aef2011-12-23 10:17:55 +11001946 rdev = conf->mirrors[d].rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001948 if (bio->bi_error) {
NeilBrown9ad1aef2011-12-23 10:17:55 +11001949 if (repl)
1950 md_error(mddev, rdev);
1951 else {
1952 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11001953 if (!test_and_set_bit(WantReplacement, &rdev->flags))
1954 set_bit(MD_RECOVERY_NEEDED,
1955 &rdev->mddev->recovery);
NeilBrown9ad1aef2011-12-23 10:17:55 +11001956 set_bit(R10BIO_WriteError, &r10_bio->state);
1957 }
1958 } else if (is_badblock(rdev,
NeilBrown749c55e2011-07-28 11:39:24 +10001959 r10_bio->devs[slot].addr,
1960 r10_bio->sectors,
1961 &first_bad, &bad_sectors))
1962 set_bit(R10BIO_MadeGood, &r10_bio->state);
NeilBrowndfc70642008-05-23 13:04:39 -07001963
NeilBrown9ad1aef2011-12-23 10:17:55 +11001964 rdev_dec_pending(rdev, mddev);
NeilBrown5e570282011-07-28 11:39:25 +10001965
1966 end_sync_request(r10_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967}
1968
1969/*
1970 * Note: sync and recover and handled very differently for raid10
1971 * This code is for resync.
1972 * For resync, we read through virtual addresses and read all blocks.
1973 * If there is any error, we schedule a write. The lowest numbered
1974 * drive is authoritative.
1975 * However requests come for physical address, so we need to map.
1976 * For every physical address there are raid_disks/copies virtual addresses,
1977 * which is always are least one, but is not necessarly an integer.
1978 * This means that a physical address can span multiple chunks, so we may
1979 * have to submit multiple io requests for a single sync request.
1980 */
1981/*
1982 * We check if all blocks are in-sync and only write to blocks that
1983 * aren't in sync
1984 */
NeilBrown9f2c9d12011-10-11 16:48:43 +11001985static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986{
NeilBrowne879a872011-10-11 16:49:02 +11001987 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 int i, first;
1989 struct bio *tbio, *fbio;
majianpengf4380a92012-04-12 16:04:47 +10001990 int vcnt;
Ming Leicdb76be2017-03-17 00:12:34 +08001991 struct page **tpages, **fpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992
1993 atomic_set(&r10_bio->remaining, 1);
1994
1995 /* find the first device with a block */
1996 for (i=0; i<conf->copies; i++)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001997 if (!r10_bio->devs[i].bio->bi_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 break;
1999
2000 if (i == conf->copies)
2001 goto done;
2002
2003 first = i;
2004 fbio = r10_bio->devs[i].bio;
Artur Paszkiewiczcc578582015-12-18 15:19:16 +11002005 fbio->bi_iter.bi_size = r10_bio->sectors << 9;
2006 fbio->bi_iter.bi_idx = 0;
Ming Leicdb76be2017-03-17 00:12:34 +08002007 fpages = get_resync_pages(fbio)->pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
majianpengf4380a92012-04-12 16:04:47 +10002009 vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 /* now find blocks with errors */
NeilBrown0eb3ff12006-01-06 00:20:29 -08002011 for (i=0 ; i < conf->copies ; i++) {
2012 int j, d;
NeilBrown8d3ca832016-11-18 16:16:12 +11002013 struct md_rdev *rdev;
Ming Leif0250612017-03-17 00:12:33 +08002014 struct resync_pages *rp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 tbio = r10_bio->devs[i].bio;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002017
2018 if (tbio->bi_end_io != end_sync_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 continue;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002020 if (i == first)
2021 continue;
Ming Leicdb76be2017-03-17 00:12:34 +08002022
2023 tpages = get_resync_pages(tbio)->pages;
NeilBrown8d3ca832016-11-18 16:16:12 +11002024 d = r10_bio->devs[i].devnum;
2025 rdev = conf->mirrors[d].rdev;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002026 if (!r10_bio->devs[i].bio->bi_error) {
NeilBrown0eb3ff12006-01-06 00:20:29 -08002027 /* We know that the bi_io_vec layout is the same for
2028 * both 'first' and 'i', so we just compare them.
2029 * All vec entries are PAGE_SIZE;
2030 */
NeilBrown7bb23c42013-07-16 16:50:47 +10002031 int sectors = r10_bio->sectors;
2032 for (j = 0; j < vcnt; j++) {
2033 int len = PAGE_SIZE;
2034 if (sectors < (len / 512))
2035 len = sectors * 512;
Ming Leicdb76be2017-03-17 00:12:34 +08002036 if (memcmp(page_address(fpages[j]),
2037 page_address(tpages[j]),
NeilBrown7bb23c42013-07-16 16:50:47 +10002038 len))
NeilBrown0eb3ff12006-01-06 00:20:29 -08002039 break;
NeilBrown7bb23c42013-07-16 16:50:47 +10002040 sectors -= len/512;
2041 }
NeilBrown0eb3ff12006-01-06 00:20:29 -08002042 if (j == vcnt)
2043 continue;
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11002044 atomic64_add(r10_bio->sectors, &mddev->resync_mismatches);
NeilBrownf84ee362011-07-28 11:39:25 +10002045 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2046 /* Don't fix anything. */
2047 continue;
NeilBrown8d3ca832016-11-18 16:16:12 +11002048 } else if (test_bit(FailFast, &rdev->flags)) {
2049 /* Just give up on this device */
2050 md_error(rdev->mddev, rdev);
2051 continue;
NeilBrown0eb3ff12006-01-06 00:20:29 -08002052 }
NeilBrownf84ee362011-07-28 11:39:25 +10002053 /* Ok, we need to write this bio, either to correct an
2054 * inconsistency or to correct an unreadable block.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 * First we need to fixup bv_offset, bv_len and
2056 * bi_vecs, as the read request might have corrupted these
2057 */
Ming Leif0250612017-03-17 00:12:33 +08002058 rp = get_resync_pages(tbio);
Kent Overstreet8be185f2012-09-06 14:14:43 -07002059 bio_reset(tbio);
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 tbio->bi_vcnt = vcnt;
Artur Paszkiewiczcc578582015-12-18 15:19:16 +11002062 tbio->bi_iter.bi_size = fbio->bi_iter.bi_size;
Ming Leif0250612017-03-17 00:12:33 +08002063 rp->raid_bio = r10_bio;
2064 tbio->bi_private = rp;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002065 tbio->bi_iter.bi_sector = r10_bio->devs[i].addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 tbio->bi_end_io = end_sync_write;
Mike Christie796a5cf2016-06-05 14:32:07 -05002067 bio_set_op_attrs(tbio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
Kent Overstreetc31df252015-05-06 23:34:20 -07002069 bio_copy_data(tbio, fbio);
2070
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
2072 atomic_inc(&r10_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002073 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
NeilBrown1919cbb2016-11-18 16:16:12 +11002075 if (test_bit(FailFast, &conf->mirrors[d].rdev->flags))
2076 tbio->bi_opf |= MD_FAILFAST;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002077 tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 tbio->bi_bdev = conf->mirrors[d].rdev->bdev;
2079 generic_make_request(tbio);
2080 }
2081
NeilBrown9ad1aef2011-12-23 10:17:55 +11002082 /* Now write out to any replacement devices
2083 * that are active
2084 */
2085 for (i = 0; i < conf->copies; i++) {
Kent Overstreetc31df252015-05-06 23:34:20 -07002086 int d;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002087
2088 tbio = r10_bio->devs[i].repl_bio;
2089 if (!tbio || !tbio->bi_end_io)
2090 continue;
2091 if (r10_bio->devs[i].bio->bi_end_io != end_sync_write
2092 && r10_bio->devs[i].bio != fbio)
Kent Overstreetc31df252015-05-06 23:34:20 -07002093 bio_copy_data(tbio, fbio);
NeilBrown9ad1aef2011-12-23 10:17:55 +11002094 d = r10_bio->devs[i].devnum;
2095 atomic_inc(&r10_bio->remaining);
2096 md_sync_acct(conf->mirrors[d].replacement->bdev,
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002097 bio_sectors(tbio));
NeilBrown9ad1aef2011-12-23 10:17:55 +11002098 generic_make_request(tbio);
2099 }
2100
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101done:
2102 if (atomic_dec_and_test(&r10_bio->remaining)) {
2103 md_done_sync(mddev, r10_bio->sectors, 1);
2104 put_buf(r10_bio);
2105 }
2106}
2107
2108/*
2109 * Now for the recovery code.
2110 * Recovery happens across physical sectors.
2111 * We recover all non-is_sync drives by finding the virtual address of
2112 * each, and then choose a working drive that also has that virt address.
2113 * There is a separate r10_bio for each non-in_sync drive.
2114 * Only the first two slots are in use. The first for reading,
2115 * The second for writing.
2116 *
2117 */
NeilBrown9f2c9d12011-10-11 16:48:43 +11002118static void fix_recovery_read_error(struct r10bio *r10_bio)
NeilBrown5e570282011-07-28 11:39:25 +10002119{
2120 /* We got a read error during recovery.
2121 * We repeat the read in smaller page-sized sections.
2122 * If a read succeeds, write it to the new device or record
2123 * a bad block if we cannot.
2124 * If a read fails, record a bad block on both old and
2125 * new devices.
2126 */
NeilBrownfd01b882011-10-11 16:47:53 +11002127 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11002128 struct r10conf *conf = mddev->private;
NeilBrown5e570282011-07-28 11:39:25 +10002129 struct bio *bio = r10_bio->devs[0].bio;
2130 sector_t sect = 0;
2131 int sectors = r10_bio->sectors;
2132 int idx = 0;
2133 int dr = r10_bio->devs[0].devnum;
2134 int dw = r10_bio->devs[1].devnum;
Ming Leicdb76be2017-03-17 00:12:34 +08002135 struct page **pages = get_resync_pages(bio)->pages;
NeilBrown5e570282011-07-28 11:39:25 +10002136
2137 while (sectors) {
2138 int s = sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11002139 struct md_rdev *rdev;
NeilBrown5e570282011-07-28 11:39:25 +10002140 sector_t addr;
2141 int ok;
2142
2143 if (s > (PAGE_SIZE>>9))
2144 s = PAGE_SIZE >> 9;
2145
2146 rdev = conf->mirrors[dr].rdev;
2147 addr = r10_bio->devs[0].addr + sect,
2148 ok = sync_page_io(rdev,
2149 addr,
2150 s << 9,
Ming Leicdb76be2017-03-17 00:12:34 +08002151 pages[idx],
Mike Christie796a5cf2016-06-05 14:32:07 -05002152 REQ_OP_READ, 0, false);
NeilBrown5e570282011-07-28 11:39:25 +10002153 if (ok) {
2154 rdev = conf->mirrors[dw].rdev;
2155 addr = r10_bio->devs[1].addr + sect;
2156 ok = sync_page_io(rdev,
2157 addr,
2158 s << 9,
Ming Leicdb76be2017-03-17 00:12:34 +08002159 pages[idx],
Mike Christie796a5cf2016-06-05 14:32:07 -05002160 REQ_OP_WRITE, 0, false);
NeilBrownb7044d42011-12-23 10:17:56 +11002161 if (!ok) {
NeilBrown5e570282011-07-28 11:39:25 +10002162 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002163 if (!test_and_set_bit(WantReplacement,
2164 &rdev->flags))
2165 set_bit(MD_RECOVERY_NEEDED,
2166 &rdev->mddev->recovery);
2167 }
NeilBrown5e570282011-07-28 11:39:25 +10002168 }
2169 if (!ok) {
2170 /* We don't worry if we cannot set a bad block -
2171 * it really is bad so there is no loss in not
2172 * recording it yet
2173 */
2174 rdev_set_badblocks(rdev, addr, s, 0);
2175
2176 if (rdev != conf->mirrors[dw].rdev) {
2177 /* need bad block on destination too */
NeilBrown3cb03002011-10-11 16:45:26 +11002178 struct md_rdev *rdev2 = conf->mirrors[dw].rdev;
NeilBrown5e570282011-07-28 11:39:25 +10002179 addr = r10_bio->devs[1].addr + sect;
2180 ok = rdev_set_badblocks(rdev2, addr, s, 0);
2181 if (!ok) {
2182 /* just abort the recovery */
NeilBrown08464e02016-11-02 14:16:50 +11002183 pr_notice("md/raid10:%s: recovery aborted due to read error\n",
2184 mdname(mddev));
NeilBrown5e570282011-07-28 11:39:25 +10002185
2186 conf->mirrors[dw].recovery_disabled
2187 = mddev->recovery_disabled;
2188 set_bit(MD_RECOVERY_INTR,
2189 &mddev->recovery);
2190 break;
2191 }
2192 }
2193 }
2194
2195 sectors -= s;
2196 sect += s;
2197 idx++;
2198 }
2199}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
NeilBrown9f2c9d12011-10-11 16:48:43 +11002201static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
NeilBrowne879a872011-10-11 16:49:02 +11002203 struct r10conf *conf = mddev->private;
Namhyung Kimc65060a2011-07-18 17:38:49 +10002204 int d;
NeilBrown24afd802011-12-23 10:17:55 +11002205 struct bio *wbio, *wbio2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206
NeilBrown5e570282011-07-28 11:39:25 +10002207 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) {
2208 fix_recovery_read_error(r10_bio);
2209 end_sync_request(r10_bio);
2210 return;
2211 }
2212
Namhyung Kimc65060a2011-07-18 17:38:49 +10002213 /*
2214 * share the pages with the first bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 * and submit the write request
2216 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 d = r10_bio->devs[1].devnum;
NeilBrown24afd802011-12-23 10:17:55 +11002218 wbio = r10_bio->devs[1].bio;
2219 wbio2 = r10_bio->devs[1].repl_bio;
NeilBrown0eb25bb2013-07-24 15:37:42 +10002220 /* Need to test wbio2->bi_end_io before we call
2221 * generic_make_request as if the former is NULL,
2222 * the latter is free to free wbio2.
2223 */
2224 if (wbio2 && !wbio2->bi_end_io)
2225 wbio2 = NULL;
NeilBrown24afd802011-12-23 10:17:55 +11002226 if (wbio->bi_end_io) {
2227 atomic_inc(&conf->mirrors[d].rdev->nr_pending);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002228 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio));
NeilBrown24afd802011-12-23 10:17:55 +11002229 generic_make_request(wbio);
2230 }
NeilBrown0eb25bb2013-07-24 15:37:42 +10002231 if (wbio2) {
NeilBrown24afd802011-12-23 10:17:55 +11002232 atomic_inc(&conf->mirrors[d].replacement->nr_pending);
2233 md_sync_acct(conf->mirrors[d].replacement->bdev,
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002234 bio_sectors(wbio2));
NeilBrown24afd802011-12-23 10:17:55 +11002235 generic_make_request(wbio2);
2236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237}
2238
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239/*
Robert Becker1e509152009-12-14 12:49:58 +11002240 * Used by fix_read_error() to decay the per rdev read_errors.
2241 * We halve the read error count for every hour that has elapsed
2242 * since the last recorded read error.
2243 *
2244 */
NeilBrownfd01b882011-10-11 16:47:53 +11002245static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev)
Robert Becker1e509152009-12-14 12:49:58 +11002246{
Arnd Bergmann0e3ef492016-06-17 17:33:10 +02002247 long cur_time_mon;
Robert Becker1e509152009-12-14 12:49:58 +11002248 unsigned long hours_since_last;
2249 unsigned int read_errors = atomic_read(&rdev->read_errors);
2250
Arnd Bergmann0e3ef492016-06-17 17:33:10 +02002251 cur_time_mon = ktime_get_seconds();
Robert Becker1e509152009-12-14 12:49:58 +11002252
Arnd Bergmann0e3ef492016-06-17 17:33:10 +02002253 if (rdev->last_read_error == 0) {
Robert Becker1e509152009-12-14 12:49:58 +11002254 /* first time we've seen a read error */
2255 rdev->last_read_error = cur_time_mon;
2256 return;
2257 }
2258
Arnd Bergmann0e3ef492016-06-17 17:33:10 +02002259 hours_since_last = (long)(cur_time_mon -
2260 rdev->last_read_error) / 3600;
Robert Becker1e509152009-12-14 12:49:58 +11002261
2262 rdev->last_read_error = cur_time_mon;
2263
2264 /*
2265 * if hours_since_last is > the number of bits in read_errors
2266 * just set read errors to 0. We do this to avoid
2267 * overflowing the shift of read_errors by hours_since_last.
2268 */
2269 if (hours_since_last >= 8 * sizeof(read_errors))
2270 atomic_set(&rdev->read_errors, 0);
2271 else
2272 atomic_set(&rdev->read_errors, read_errors >> hours_since_last);
2273}
2274
NeilBrown3cb03002011-10-11 16:45:26 +11002275static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrown58c54fc2011-07-28 11:39:25 +10002276 int sectors, struct page *page, int rw)
2277{
2278 sector_t first_bad;
2279 int bad_sectors;
2280
2281 if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors)
2282 && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags)))
2283 return -1;
Mike Christie796a5cf2016-06-05 14:32:07 -05002284 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
NeilBrown58c54fc2011-07-28 11:39:25 +10002285 /* success */
2286 return 1;
NeilBrownb7044d42011-12-23 10:17:56 +11002287 if (rw == WRITE) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002288 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrownb7044d42011-12-23 10:17:56 +11002289 if (!test_and_set_bit(WantReplacement, &rdev->flags))
2290 set_bit(MD_RECOVERY_NEEDED,
2291 &rdev->mddev->recovery);
2292 }
NeilBrown58c54fc2011-07-28 11:39:25 +10002293 /* need to record an error - either for the block or the device */
2294 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
2295 md_error(rdev->mddev, rdev);
2296 return 0;
2297}
2298
Robert Becker1e509152009-12-14 12:49:58 +11002299/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 * This is a kernel thread which:
2301 *
2302 * 1. Retries failed read operations on working mirrors.
2303 * 2. Updates the raid superblock when problems encounter.
NeilBrown6814d532006-10-03 01:15:45 -07002304 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 */
2306
NeilBrowne879a872011-10-11 16:49:02 +11002307static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio)
NeilBrown6814d532006-10-03 01:15:45 -07002308{
2309 int sect = 0; /* Offset from r10_bio->sector */
2310 int sectors = r10_bio->sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11002311 struct md_rdev*rdev;
Robert Becker1e509152009-12-14 12:49:58 +11002312 int max_read_errors = atomic_read(&mddev->max_corr_read_errors);
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002313 int d = r10_bio->devs[r10_bio->read_slot].devnum;
Robert Becker1e509152009-12-14 12:49:58 +11002314
NeilBrown7c4e06f2011-05-11 14:53:17 +10002315 /* still own a reference to this rdev, so it cannot
2316 * have been cleared recently.
2317 */
2318 rdev = conf->mirrors[d].rdev;
Robert Becker1e509152009-12-14 12:49:58 +11002319
NeilBrown7c4e06f2011-05-11 14:53:17 +10002320 if (test_bit(Faulty, &rdev->flags))
2321 /* drive has already been failed, just ignore any
2322 more fix_read_error() attempts */
2323 return;
2324
2325 check_decay_read_errors(mddev, rdev);
2326 atomic_inc(&rdev->read_errors);
2327 if (atomic_read(&rdev->read_errors) > max_read_errors) {
2328 char b[BDEVNAME_SIZE];
Robert Becker1e509152009-12-14 12:49:58 +11002329 bdevname(rdev->bdev, b);
2330
NeilBrown08464e02016-11-02 14:16:50 +11002331 pr_notice("md/raid10:%s: %s: Raid device exceeded read_error threshold [cur %d:max %d]\n",
2332 mdname(mddev), b,
2333 atomic_read(&rdev->read_errors), max_read_errors);
2334 pr_notice("md/raid10:%s: %s: Failing raid device\n",
2335 mdname(mddev), b);
NeilBrownd683c8e2016-06-02 16:19:52 +10002336 md_error(mddev, rdev);
NeilBrownfae8cc52012-02-14 11:10:10 +11002337 r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED;
NeilBrown7c4e06f2011-05-11 14:53:17 +10002338 return;
Robert Becker1e509152009-12-14 12:49:58 +11002339 }
Robert Becker1e509152009-12-14 12:49:58 +11002340
NeilBrown6814d532006-10-03 01:15:45 -07002341 while(sectors) {
2342 int s = sectors;
2343 int sl = r10_bio->read_slot;
2344 int success = 0;
2345 int start;
2346
2347 if (s > (PAGE_SIZE>>9))
2348 s = PAGE_SIZE >> 9;
2349
2350 rcu_read_lock();
2351 do {
NeilBrown8dbed5c2011-07-28 11:39:24 +10002352 sector_t first_bad;
2353 int bad_sectors;
2354
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002355 d = r10_bio->devs[sl].devnum;
NeilBrown6814d532006-10-03 01:15:45 -07002356 rdev = rcu_dereference(conf->mirrors[d].rdev);
2357 if (rdev &&
NeilBrown8dbed5c2011-07-28 11:39:24 +10002358 test_bit(In_sync, &rdev->flags) &&
NeilBrownf5b67ae2016-06-02 16:19:53 +10002359 !test_bit(Faulty, &rdev->flags) &&
NeilBrown8dbed5c2011-07-28 11:39:24 +10002360 is_badblock(rdev, r10_bio->devs[sl].addr + sect, s,
2361 &first_bad, &bad_sectors) == 0) {
NeilBrown6814d532006-10-03 01:15:45 -07002362 atomic_inc(&rdev->nr_pending);
2363 rcu_read_unlock();
NeilBrown2b193362010-10-27 15:16:40 +11002364 success = sync_page_io(rdev,
NeilBrown6814d532006-10-03 01:15:45 -07002365 r10_bio->devs[sl].addr +
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11002366 sect,
NeilBrown6814d532006-10-03 01:15:45 -07002367 s<<9,
Mike Christie796a5cf2016-06-05 14:32:07 -05002368 conf->tmppage,
2369 REQ_OP_READ, 0, false);
NeilBrown6814d532006-10-03 01:15:45 -07002370 rdev_dec_pending(rdev, mddev);
2371 rcu_read_lock();
2372 if (success)
2373 break;
2374 }
2375 sl++;
2376 if (sl == conf->copies)
2377 sl = 0;
2378 } while (!success && sl != r10_bio->read_slot);
2379 rcu_read_unlock();
2380
2381 if (!success) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002382 /* Cannot read from anywhere, just mark the block
2383 * as bad on the first device to discourage future
2384 * reads.
2385 */
NeilBrown6814d532006-10-03 01:15:45 -07002386 int dn = r10_bio->devs[r10_bio->read_slot].devnum;
NeilBrown58c54fc2011-07-28 11:39:25 +10002387 rdev = conf->mirrors[dn].rdev;
2388
2389 if (!rdev_set_badblocks(
2390 rdev,
2391 r10_bio->devs[r10_bio->read_slot].addr
2392 + sect,
NeilBrownfae8cc52012-02-14 11:10:10 +11002393 s, 0)) {
NeilBrown58c54fc2011-07-28 11:39:25 +10002394 md_error(mddev, rdev);
NeilBrownfae8cc52012-02-14 11:10:10 +11002395 r10_bio->devs[r10_bio->read_slot].bio
2396 = IO_BLOCKED;
2397 }
NeilBrown6814d532006-10-03 01:15:45 -07002398 break;
2399 }
2400
2401 start = sl;
2402 /* write it back and re-read */
2403 rcu_read_lock();
2404 while (sl != r10_bio->read_slot) {
Robert Becker67b8dc42009-12-14 12:49:57 +11002405 char b[BDEVNAME_SIZE];
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002406
NeilBrown6814d532006-10-03 01:15:45 -07002407 if (sl==0)
2408 sl = conf->copies;
2409 sl--;
2410 d = r10_bio->devs[sl].devnum;
2411 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown1294b9c2011-07-28 11:39:23 +10002412 if (!rdev ||
NeilBrownf5b67ae2016-06-02 16:19:53 +10002413 test_bit(Faulty, &rdev->flags) ||
NeilBrown1294b9c2011-07-28 11:39:23 +10002414 !test_bit(In_sync, &rdev->flags))
2415 continue;
2416
2417 atomic_inc(&rdev->nr_pending);
2418 rcu_read_unlock();
NeilBrown58c54fc2011-07-28 11:39:25 +10002419 if (r10_sync_page_io(rdev,
2420 r10_bio->devs[sl].addr +
2421 sect,
NeilBrown055d3742012-07-03 15:55:33 +10002422 s, conf->tmppage, WRITE)
NeilBrown1294b9c2011-07-28 11:39:23 +10002423 == 0) {
2424 /* Well, this device is dead */
NeilBrown08464e02016-11-02 14:16:50 +11002425 pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %s)\n",
2426 mdname(mddev), s,
2427 (unsigned long long)(
2428 sect +
2429 choose_data_offset(r10_bio,
2430 rdev)),
2431 bdevname(rdev->bdev, b));
2432 pr_notice("md/raid10:%s: %s: failing drive\n",
2433 mdname(mddev),
2434 bdevname(rdev->bdev, b));
NeilBrown6814d532006-10-03 01:15:45 -07002435 }
NeilBrown1294b9c2011-07-28 11:39:23 +10002436 rdev_dec_pending(rdev, mddev);
2437 rcu_read_lock();
NeilBrown6814d532006-10-03 01:15:45 -07002438 }
2439 sl = start;
2440 while (sl != r10_bio->read_slot) {
NeilBrown1294b9c2011-07-28 11:39:23 +10002441 char b[BDEVNAME_SIZE];
Prasanna S. Panchamukhi0544a212010-06-24 13:31:03 +10002442
NeilBrown6814d532006-10-03 01:15:45 -07002443 if (sl==0)
2444 sl = conf->copies;
2445 sl--;
2446 d = r10_bio->devs[sl].devnum;
2447 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown1294b9c2011-07-28 11:39:23 +10002448 if (!rdev ||
NeilBrownf5b67ae2016-06-02 16:19:53 +10002449 test_bit(Faulty, &rdev->flags) ||
NeilBrown1294b9c2011-07-28 11:39:23 +10002450 !test_bit(In_sync, &rdev->flags))
2451 continue;
Robert Becker67b8dc42009-12-14 12:49:57 +11002452
NeilBrown1294b9c2011-07-28 11:39:23 +10002453 atomic_inc(&rdev->nr_pending);
2454 rcu_read_unlock();
NeilBrown58c54fc2011-07-28 11:39:25 +10002455 switch (r10_sync_page_io(rdev,
2456 r10_bio->devs[sl].addr +
2457 sect,
NeilBrown055d3742012-07-03 15:55:33 +10002458 s, conf->tmppage,
NeilBrown58c54fc2011-07-28 11:39:25 +10002459 READ)) {
2460 case 0:
NeilBrown1294b9c2011-07-28 11:39:23 +10002461 /* Well, this device is dead */
NeilBrown08464e02016-11-02 14:16:50 +11002462 pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %s)\n",
NeilBrown1294b9c2011-07-28 11:39:23 +10002463 mdname(mddev), s,
2464 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002465 sect +
2466 choose_data_offset(r10_bio, rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002467 bdevname(rdev->bdev, b));
NeilBrown08464e02016-11-02 14:16:50 +11002468 pr_notice("md/raid10:%s: %s: failing drive\n",
NeilBrown1294b9c2011-07-28 11:39:23 +10002469 mdname(mddev),
2470 bdevname(rdev->bdev, b));
NeilBrown58c54fc2011-07-28 11:39:25 +10002471 break;
2472 case 1:
NeilBrown08464e02016-11-02 14:16:50 +11002473 pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %s)\n",
NeilBrown1294b9c2011-07-28 11:39:23 +10002474 mdname(mddev), s,
2475 (unsigned long long)(
NeilBrownf8c9e742012-05-21 09:28:33 +10002476 sect +
2477 choose_data_offset(r10_bio, rdev)),
NeilBrown1294b9c2011-07-28 11:39:23 +10002478 bdevname(rdev->bdev, b));
2479 atomic_add(s, &rdev->corrected_errors);
NeilBrown6814d532006-10-03 01:15:45 -07002480 }
NeilBrown1294b9c2011-07-28 11:39:23 +10002481
2482 rdev_dec_pending(rdev, mddev);
2483 rcu_read_lock();
NeilBrown6814d532006-10-03 01:15:45 -07002484 }
2485 rcu_read_unlock();
2486
2487 sectors -= s;
2488 sect += s;
2489 }
2490}
2491
NeilBrown9f2c9d12011-10-11 16:48:43 +11002492static int narrow_write_error(struct r10bio *r10_bio, int i)
NeilBrownbd870a12011-07-28 11:39:24 +10002493{
2494 struct bio *bio = r10_bio->master_bio;
NeilBrownfd01b882011-10-11 16:47:53 +11002495 struct mddev *mddev = r10_bio->mddev;
NeilBrowne879a872011-10-11 16:49:02 +11002496 struct r10conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002497 struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev;
NeilBrownbd870a12011-07-28 11:39:24 +10002498 /* bio has the data to be written to slot 'i' where
2499 * we just recently had a write error.
2500 * We repeatedly clone the bio and trim down to one block,
2501 * then try the write. Where the write fails we record
2502 * a bad block.
2503 * It is conceivable that the bio doesn't exactly align with
2504 * blocks. We must handle this.
2505 *
2506 * We currently own a reference to the rdev.
2507 */
2508
2509 int block_sectors;
2510 sector_t sector;
2511 int sectors;
2512 int sect_to_write = r10_bio->sectors;
2513 int ok = 1;
2514
2515 if (rdev->badblocks.shift < 0)
2516 return 0;
2517
NeilBrownf04ebb02015-02-16 14:51:54 +11002518 block_sectors = roundup(1 << rdev->badblocks.shift,
2519 bdev_logical_block_size(rdev->bdev) >> 9);
NeilBrownbd870a12011-07-28 11:39:24 +10002520 sector = r10_bio->sector;
2521 sectors = ((r10_bio->sector + block_sectors)
2522 & ~(sector_t)(block_sectors - 1))
2523 - sector;
2524
2525 while (sect_to_write) {
2526 struct bio *wbio;
Tomasz Majchrzak27028622016-08-23 10:53:57 +02002527 sector_t wsector;
NeilBrownbd870a12011-07-28 11:39:24 +10002528 if (sectors > sect_to_write)
2529 sectors = sect_to_write;
2530 /* Write at 'sector' for 'sectors' */
Ming Leid7a10302017-02-14 23:29:03 +08002531 wbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002532 bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors);
Tomasz Majchrzak27028622016-08-23 10:53:57 +02002533 wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector);
2534 wbio->bi_iter.bi_sector = wsector +
2535 choose_data_offset(r10_bio, rdev);
NeilBrownbd870a12011-07-28 11:39:24 +10002536 wbio->bi_bdev = rdev->bdev;
Mike Christie796a5cf2016-06-05 14:32:07 -05002537 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
Mike Christie4e49ea42016-06-05 14:31:41 -05002538
2539 if (submit_bio_wait(wbio) < 0)
NeilBrownbd870a12011-07-28 11:39:24 +10002540 /* Failure! */
Tomasz Majchrzak27028622016-08-23 10:53:57 +02002541 ok = rdev_set_badblocks(rdev, wsector,
NeilBrownbd870a12011-07-28 11:39:24 +10002542 sectors, 0)
2543 && ok;
2544
2545 bio_put(wbio);
2546 sect_to_write -= sectors;
2547 sector += sectors;
2548 sectors = block_sectors;
2549 }
2550 return ok;
2551}
2552
NeilBrown9f2c9d12011-10-11 16:48:43 +11002553static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio)
NeilBrown560f8e52011-07-28 11:39:23 +10002554{
2555 int slot = r10_bio->read_slot;
NeilBrown560f8e52011-07-28 11:39:23 +10002556 struct bio *bio;
NeilBrowne879a872011-10-11 16:49:02 +11002557 struct r10conf *conf = mddev->private;
NeilBrownabbf0982011-12-23 10:17:54 +11002558 struct md_rdev *rdev = r10_bio->devs[slot].rdev;
NeilBrown560f8e52011-07-28 11:39:23 +10002559 char b[BDEVNAME_SIZE];
2560 unsigned long do_sync;
NeilBrown856e08e2011-07-28 11:39:23 +10002561 int max_sectors;
NeilBrown109e3762016-11-18 13:22:04 +11002562 dev_t bio_dev;
2563 sector_t bio_last_sector;
NeilBrown560f8e52011-07-28 11:39:23 +10002564
2565 /* we got a read error. Maybe the drive is bad. Maybe just
2566 * the block and we can fix it.
2567 * We freeze all other IO, and try reading the block from
2568 * other devices. When we find one, we re-write
2569 * and check it that fixes the read error.
2570 * This is all done synchronously while the array is
2571 * frozen.
2572 */
NeilBrownfae8cc52012-02-14 11:10:10 +11002573 bio = r10_bio->devs[slot].bio;
2574 bdevname(bio->bi_bdev, b);
NeilBrown109e3762016-11-18 13:22:04 +11002575 bio_dev = bio->bi_bdev->bd_dev;
2576 bio_last_sector = r10_bio->devs[slot].addr + rdev->data_offset + r10_bio->sectors;
NeilBrownfae8cc52012-02-14 11:10:10 +11002577 bio_put(bio);
2578 r10_bio->devs[slot].bio = NULL;
2579
NeilBrown8d3ca832016-11-18 16:16:12 +11002580 if (mddev->ro)
2581 r10_bio->devs[slot].bio = IO_BLOCKED;
2582 else if (!test_bit(FailFast, &rdev->flags)) {
NeilBrowne2d59922013-06-12 11:01:22 +10002583 freeze_array(conf, 1);
NeilBrown560f8e52011-07-28 11:39:23 +10002584 fix_read_error(conf, mddev, r10_bio);
2585 unfreeze_array(conf);
NeilBrownfae8cc52012-02-14 11:10:10 +11002586 } else
NeilBrown8d3ca832016-11-18 16:16:12 +11002587 md_error(mddev, rdev);
NeilBrownfae8cc52012-02-14 11:10:10 +11002588
NeilBrownabbf0982011-12-23 10:17:54 +11002589 rdev_dec_pending(rdev, mddev);
NeilBrown560f8e52011-07-28 11:39:23 +10002590
NeilBrown7399c312011-07-28 11:39:23 +10002591read_more:
NeilBrown96c3fd12011-12-23 10:17:54 +11002592 rdev = read_balance(conf, r10_bio, &max_sectors);
2593 if (rdev == NULL) {
NeilBrown08464e02016-11-02 14:16:50 +11002594 pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n",
2595 mdname(mddev), b,
2596 (unsigned long long)r10_bio->sector);
NeilBrown560f8e52011-07-28 11:39:23 +10002597 raid_end_bio_io(r10_bio);
NeilBrown560f8e52011-07-28 11:39:23 +10002598 return;
2599 }
2600
Jens Axboe1eff9d32016-08-05 15:35:16 -06002601 do_sync = (r10_bio->master_bio->bi_opf & REQ_SYNC);
NeilBrown560f8e52011-07-28 11:39:23 +10002602 slot = r10_bio->read_slot;
NeilBrown08464e02016-11-02 14:16:50 +11002603 pr_err_ratelimited("md/raid10:%s: %s: redirecting sector %llu to another mirror\n",
2604 mdname(mddev),
2605 bdevname(rdev->bdev, b),
2606 (unsigned long long)r10_bio->sector);
Ming Leid7a10302017-02-14 23:29:03 +08002607 bio = bio_clone_fast(r10_bio->master_bio, GFP_NOIO, mddev->bio_set);
Kent Overstreet4f024f32013-10-11 15:44:27 -07002608 bio_trim(bio, r10_bio->sector - bio->bi_iter.bi_sector, max_sectors);
NeilBrown560f8e52011-07-28 11:39:23 +10002609 r10_bio->devs[slot].bio = bio;
NeilBrownabbf0982011-12-23 10:17:54 +11002610 r10_bio->devs[slot].rdev = rdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002611 bio->bi_iter.bi_sector = r10_bio->devs[slot].addr
NeilBrownf8c9e742012-05-21 09:28:33 +10002612 + choose_data_offset(r10_bio, rdev);
NeilBrown560f8e52011-07-28 11:39:23 +10002613 bio->bi_bdev = rdev->bdev;
Mike Christie796a5cf2016-06-05 14:32:07 -05002614 bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
NeilBrown8d3ca832016-11-18 16:16:12 +11002615 if (test_bit(FailFast, &rdev->flags) &&
2616 test_bit(R10BIO_FailFast, &r10_bio->state))
2617 bio->bi_opf |= MD_FAILFAST;
NeilBrown560f8e52011-07-28 11:39:23 +10002618 bio->bi_private = r10_bio;
2619 bio->bi_end_io = raid10_end_read_request;
NeilBrown109e3762016-11-18 13:22:04 +11002620 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev),
2621 bio, bio_dev,
2622 bio_last_sector - r10_bio->sectors);
2623
NeilBrown7399c312011-07-28 11:39:23 +10002624 if (max_sectors < r10_bio->sectors) {
2625 /* Drat - have to split this up more */
2626 struct bio *mbio = r10_bio->master_bio;
2627 int sectors_handled =
2628 r10_bio->sector + max_sectors
Kent Overstreet4f024f32013-10-11 15:44:27 -07002629 - mbio->bi_iter.bi_sector;
NeilBrown7399c312011-07-28 11:39:23 +10002630 r10_bio->sectors = max_sectors;
NeilBrownfd16f2e2017-03-15 14:05:13 +11002631 bio_inc_remaining(mbio);
2632 inc_pending(conf);
NeilBrown7399c312011-07-28 11:39:23 +10002633 generic_make_request(bio);
NeilBrown7399c312011-07-28 11:39:23 +10002634
2635 r10_bio = mempool_alloc(conf->r10bio_pool,
2636 GFP_NOIO);
2637 r10_bio->master_bio = mbio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002638 r10_bio->sectors = bio_sectors(mbio) - sectors_handled;
NeilBrown7399c312011-07-28 11:39:23 +10002639 r10_bio->state = 0;
2640 set_bit(R10BIO_ReadError,
2641 &r10_bio->state);
2642 r10_bio->mddev = mddev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07002643 r10_bio->sector = mbio->bi_iter.bi_sector
NeilBrown7399c312011-07-28 11:39:23 +10002644 + sectors_handled;
2645
2646 goto read_more;
2647 } else
2648 generic_make_request(bio);
NeilBrown560f8e52011-07-28 11:39:23 +10002649}
2650
NeilBrowne879a872011-10-11 16:49:02 +11002651static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio)
NeilBrown749c55e2011-07-28 11:39:24 +10002652{
2653 /* Some sort of write request has finished and it
2654 * succeeded in writing where we thought there was a
2655 * bad block. So forget the bad block.
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002656 * Or possibly if failed and we need to record
2657 * a bad block.
NeilBrown749c55e2011-07-28 11:39:24 +10002658 */
2659 int m;
NeilBrown3cb03002011-10-11 16:45:26 +11002660 struct md_rdev *rdev;
NeilBrown749c55e2011-07-28 11:39:24 +10002661
2662 if (test_bit(R10BIO_IsSync, &r10_bio->state) ||
2663 test_bit(R10BIO_IsRecover, &r10_bio->state)) {
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002664 for (m = 0; m < conf->copies; m++) {
2665 int dev = r10_bio->devs[m].devnum;
2666 rdev = conf->mirrors[dev].rdev;
2667 if (r10_bio->devs[m].bio == NULL)
2668 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002669 if (!r10_bio->devs[m].bio->bi_error) {
NeilBrown749c55e2011-07-28 11:39:24 +10002670 rdev_clear_badblocks(
2671 rdev,
2672 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002673 r10_bio->sectors, 0);
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002674 } else {
2675 if (!rdev_set_badblocks(
2676 rdev,
2677 r10_bio->devs[m].addr,
2678 r10_bio->sectors, 0))
2679 md_error(conf->mddev, rdev);
NeilBrown749c55e2011-07-28 11:39:24 +10002680 }
NeilBrown9ad1aef2011-12-23 10:17:55 +11002681 rdev = conf->mirrors[dev].replacement;
2682 if (r10_bio->devs[m].repl_bio == NULL)
2683 continue;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002684
2685 if (!r10_bio->devs[m].repl_bio->bi_error) {
NeilBrown9ad1aef2011-12-23 10:17:55 +11002686 rdev_clear_badblocks(
2687 rdev,
2688 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002689 r10_bio->sectors, 0);
NeilBrown9ad1aef2011-12-23 10:17:55 +11002690 } else {
2691 if (!rdev_set_badblocks(
2692 rdev,
2693 r10_bio->devs[m].addr,
2694 r10_bio->sectors, 0))
2695 md_error(conf->mddev, rdev);
2696 }
NeilBrown1a0b7cd2011-07-28 11:39:25 +10002697 }
NeilBrown749c55e2011-07-28 11:39:24 +10002698 put_buf(r10_bio);
2699 } else {
NeilBrown95af5872015-08-14 11:26:17 +10002700 bool fail = false;
NeilBrownbd870a12011-07-28 11:39:24 +10002701 for (m = 0; m < conf->copies; m++) {
2702 int dev = r10_bio->devs[m].devnum;
2703 struct bio *bio = r10_bio->devs[m].bio;
2704 rdev = conf->mirrors[dev].rdev;
2705 if (bio == IO_MADE_GOOD) {
NeilBrown749c55e2011-07-28 11:39:24 +10002706 rdev_clear_badblocks(
2707 rdev,
2708 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002709 r10_bio->sectors, 0);
NeilBrown749c55e2011-07-28 11:39:24 +10002710 rdev_dec_pending(rdev, conf->mddev);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002711 } else if (bio != NULL && bio->bi_error) {
NeilBrown95af5872015-08-14 11:26:17 +10002712 fail = true;
NeilBrownbd870a12011-07-28 11:39:24 +10002713 if (!narrow_write_error(r10_bio, m)) {
2714 md_error(conf->mddev, rdev);
2715 set_bit(R10BIO_Degraded,
2716 &r10_bio->state);
2717 }
2718 rdev_dec_pending(rdev, conf->mddev);
NeilBrown749c55e2011-07-28 11:39:24 +10002719 }
NeilBrown475b0322011-12-23 10:17:55 +11002720 bio = r10_bio->devs[m].repl_bio;
2721 rdev = conf->mirrors[dev].replacement;
NeilBrown4ca40c22011-12-23 10:17:55 +11002722 if (rdev && bio == IO_MADE_GOOD) {
NeilBrown475b0322011-12-23 10:17:55 +11002723 rdev_clear_badblocks(
2724 rdev,
2725 r10_bio->devs[m].addr,
NeilBrownc6563a82012-05-21 09:27:00 +10002726 r10_bio->sectors, 0);
NeilBrown475b0322011-12-23 10:17:55 +11002727 rdev_dec_pending(rdev, conf->mddev);
2728 }
NeilBrownbd870a12011-07-28 11:39:24 +10002729 }
NeilBrown95af5872015-08-14 11:26:17 +10002730 if (fail) {
2731 spin_lock_irq(&conf->device_lock);
2732 list_add(&r10_bio->retry_list, &conf->bio_end_io_list);
Shaohua Li23ddba82016-03-14 11:49:32 -07002733 conf->nr_queued++;
NeilBrown95af5872015-08-14 11:26:17 +10002734 spin_unlock_irq(&conf->device_lock);
2735 md_wakeup_thread(conf->mddev->thread);
NeilBrownc3407022015-10-24 16:23:48 +11002736 } else {
2737 if (test_bit(R10BIO_WriteError,
2738 &r10_bio->state))
2739 close_write(r10_bio);
NeilBrown95af5872015-08-14 11:26:17 +10002740 raid_end_bio_io(r10_bio);
NeilBrownc3407022015-10-24 16:23:48 +11002741 }
NeilBrown749c55e2011-07-28 11:39:24 +10002742 }
2743}
2744
Shaohua Li4ed87312012-10-11 13:34:00 +11002745static void raid10d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746{
Shaohua Li4ed87312012-10-11 13:34:00 +11002747 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002748 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 unsigned long flags;
NeilBrowne879a872011-10-11 16:49:02 +11002750 struct r10conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002752 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753
2754 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
NeilBrown95af5872015-08-14 11:26:17 +10002756 if (!list_empty_careful(&conf->bio_end_io_list) &&
Shaohua Li29530792016-12-08 15:48:19 -08002757 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
NeilBrown95af5872015-08-14 11:26:17 +10002758 LIST_HEAD(tmp);
2759 spin_lock_irqsave(&conf->device_lock, flags);
Shaohua Li29530792016-12-08 15:48:19 -08002760 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
Shaohua Li23ddba82016-03-14 11:49:32 -07002761 while (!list_empty(&conf->bio_end_io_list)) {
2762 list_move(conf->bio_end_io_list.prev, &tmp);
2763 conf->nr_queued--;
2764 }
NeilBrown95af5872015-08-14 11:26:17 +10002765 }
2766 spin_unlock_irqrestore(&conf->device_lock, flags);
2767 while (!list_empty(&tmp)) {
Mikulas Patockaa4527442015-10-01 15:17:43 -04002768 r10_bio = list_first_entry(&tmp, struct r10bio,
2769 retry_list);
NeilBrown95af5872015-08-14 11:26:17 +10002770 list_del(&r10_bio->retry_list);
NeilBrownc3407022015-10-24 16:23:48 +11002771 if (mddev->degraded)
2772 set_bit(R10BIO_Degraded, &r10_bio->state);
2773
2774 if (test_bit(R10BIO_WriteError,
2775 &r10_bio->state))
2776 close_write(r10_bio);
NeilBrown95af5872015-08-14 11:26:17 +10002777 raid_end_bio_io(r10_bio);
2778 }
2779 }
2780
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002781 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002783
NeilBrown0021b7b2012-07-31 09:08:14 +02002784 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002785
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002787 if (list_empty(head)) {
NeilBrown6cce3b22006-01-06 00:20:16 -08002788 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002790 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002791 r10_bio = list_entry(head->prev, struct r10bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 list_del(head->prev);
NeilBrown4443ae12006-01-06 00:20:28 -08002793 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 spin_unlock_irqrestore(&conf->device_lock, flags);
2795
2796 mddev = r10_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002797 conf = mddev->private;
NeilBrownbd870a12011-07-28 11:39:24 +10002798 if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
2799 test_bit(R10BIO_WriteError, &r10_bio->state))
NeilBrown749c55e2011-07-28 11:39:24 +10002800 handle_write_completed(conf, r10_bio);
NeilBrown3ea7daa2012-05-22 13:53:47 +10002801 else if (test_bit(R10BIO_IsReshape, &r10_bio->state))
2802 reshape_request_write(mddev, r10_bio);
NeilBrown749c55e2011-07-28 11:39:24 +10002803 else if (test_bit(R10BIO_IsSync, &r10_bio->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 sync_request_write(mddev, r10_bio);
Jens Axboe7eaceac2011-03-10 08:52:07 +01002805 else if (test_bit(R10BIO_IsRecover, &r10_bio->state))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 recovery_request_write(mddev, r10_bio);
NeilBrown856e08e2011-07-28 11:39:23 +10002807 else if (test_bit(R10BIO_ReadError, &r10_bio->state))
NeilBrown560f8e52011-07-28 11:39:23 +10002808 handle_read_error(mddev, r10_bio);
NeilBrownfc9977d2017-04-05 14:05:51 +10002809 else
2810 WARN_ON_ONCE(1);
NeilBrown4443ae12006-01-06 00:20:28 -08002811
NeilBrown1d9d5242009-10-16 15:55:32 +11002812 cond_resched();
Shaohua Li29530792016-12-08 15:48:19 -08002813 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING))
NeilBrownde393cd2011-07-28 11:31:48 +10002814 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002816 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817}
2818
NeilBrowne879a872011-10-11 16:49:02 +11002819static int init_resync(struct r10conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820{
2821 int buffs;
NeilBrown69335ef2011-12-23 10:17:54 +11002822 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823
2824 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhennb6385482006-04-02 13:34:29 +02002825 BUG_ON(conf->r10buf_pool);
NeilBrown69335ef2011-12-23 10:17:54 +11002826 conf->have_replacement = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10002827 for (i = 0; i < conf->geo.raid_disks; i++)
NeilBrown69335ef2011-12-23 10:17:54 +11002828 if (conf->mirrors[i].replacement)
2829 conf->have_replacement = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf);
2831 if (!conf->r10buf_pool)
2832 return -ENOMEM;
2833 conf->next_resync = 0;
2834 return 0;
2835}
2836
2837/*
2838 * perform a "sync" on one "block"
2839 *
2840 * We need to make sure that no normal I/O request - particularly write
2841 * requests - conflict with active sync requests.
2842 *
2843 * This is achieved by tracking pending requests and a 'barrier' concept
2844 * that can be installed to exclude normal IO requests.
2845 *
2846 * Resync and recovery are handled very differently.
2847 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
2848 *
2849 * For resync, we iterate over virtual addresses, read all copies,
2850 * and update if there are differences. If only one copy is live,
2851 * skip it.
2852 * For recovery, we iterate over physical addresses, read a good
2853 * value for each non-in_sync drive, and over-write.
2854 *
2855 * So, for recovery we may have several outstanding complex requests for a
2856 * given address, one for each out-of-sync device. We model this by allocating
2857 * a number of r10_bio structures, one for each out-of-sync device.
2858 * As we setup these structures, we collect all bio's together into a list
2859 * which we then process collectively to add pages, and then process again
2860 * to pass to generic_make_request.
2861 *
2862 * The r10_bio structures are linked using a borrowed master_bio pointer.
2863 * This link is counted in ->remaining. When the r10_bio that points to NULL
2864 * has its remaining count decremented to 0, the whole complex operation
2865 * is complete.
2866 *
2867 */
2868
Shaohua Li849674e2016-01-20 13:52:20 -08002869static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr,
NeilBrown09314792015-02-19 16:04:40 +11002870 int *skipped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871{
NeilBrowne879a872011-10-11 16:49:02 +11002872 struct r10conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002873 struct r10bio *r10_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 struct bio *biolist = NULL, *bio;
2875 sector_t max_sector, nr_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 int i;
NeilBrown6cce3b22006-01-06 00:20:16 -08002877 int max_sync;
NeilBrown57dab0b2010-10-19 10:03:39 +11002878 sector_t sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 sector_t sectors_skipped = 0;
2880 int chunks_skipped = 0;
NeilBrown5cf00fc2012-05-21 09:28:20 +10002881 sector_t chunk_mask = conf->geo.chunk_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882
2883 if (!conf->r10buf_pool)
2884 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002885 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002887 /*
2888 * Allow skipping a full rebuild for incremental assembly
2889 * of a clean array, like RAID1 does.
2890 */
2891 if (mddev->bitmap == NULL &&
2892 mddev->recovery_cp == MaxSector &&
NeilBrown13765122013-07-04 16:41:53 +10002893 mddev->reshape_position == MaxSector &&
2894 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002895 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown13765122013-07-04 16:41:53 +10002896 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002897 conf->fullsync == 0) {
2898 *skipped = 1;
NeilBrown13765122013-07-04 16:41:53 +10002899 return mddev->dev_sectors - sector_nr;
Martin Wilck7e83ccb2013-04-24 11:42:42 +10002900 }
2901
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 skipped:
Andre Noll58c0fed2009-03-31 14:33:13 +11002903 max_sector = mddev->dev_sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10002904 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
2905 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 max_sector = mddev->resync_max_sectors;
2907 if (sector_nr >= max_sector) {
NeilBrown6cce3b22006-01-06 00:20:16 -08002908 /* If we aborted, we need to abort the
2909 * sync on the 'current' bitmap chucks (there can
2910 * be several when recovering multiple devices).
2911 * as we may have started syncing it but not finished.
2912 * We can find the current address in
2913 * mddev->curr_resync, but for recovery,
2914 * we need to convert that to several
2915 * virtual addresses.
2916 */
NeilBrown3ea7daa2012-05-22 13:53:47 +10002917 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
2918 end_reshape(conf);
NeilBrownb3968552014-08-18 13:59:50 +10002919 close_sync(conf);
NeilBrown3ea7daa2012-05-22 13:53:47 +10002920 return 0;
2921 }
2922
NeilBrown6cce3b22006-01-06 00:20:16 -08002923 if (mddev->curr_resync < max_sector) { /* aborted */
2924 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2925 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2926 &sync_blocks, 1);
NeilBrown5cf00fc2012-05-21 09:28:20 +10002927 else for (i = 0; i < conf->geo.raid_disks; i++) {
NeilBrown6cce3b22006-01-06 00:20:16 -08002928 sector_t sect =
2929 raid10_find_virt(conf, mddev->curr_resync, i);
2930 bitmap_end_sync(mddev->bitmap, sect,
2931 &sync_blocks, 1);
2932 }
NeilBrown9ad1aef2011-12-23 10:17:55 +11002933 } else {
2934 /* completed sync */
2935 if ((!mddev->bitmap || conf->fullsync)
2936 && conf->have_replacement
2937 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2938 /* Completed a full sync so the replacements
2939 * are now fully recovered.
2940 */
NeilBrownf90145f2016-06-02 16:19:52 +10002941 rcu_read_lock();
2942 for (i = 0; i < conf->geo.raid_disks; i++) {
2943 struct md_rdev *rdev =
2944 rcu_dereference(conf->mirrors[i].replacement);
2945 if (rdev)
2946 rdev->recovery_offset = MaxSector;
2947 }
2948 rcu_read_unlock();
NeilBrown9ad1aef2011-12-23 10:17:55 +11002949 }
NeilBrown6cce3b22006-01-06 00:20:16 -08002950 conf->fullsync = 0;
NeilBrown9ad1aef2011-12-23 10:17:55 +11002951 }
NeilBrown6cce3b22006-01-06 00:20:16 -08002952 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 close_sync(conf);
NeilBrown57afd892005-06-21 17:17:13 -07002954 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 return sectors_skipped;
2956 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10002957
2958 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2959 return reshape_request(mddev, sector_nr, skipped);
2960
NeilBrown5cf00fc2012-05-21 09:28:20 +10002961 if (chunks_skipped >= conf->geo.raid_disks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 /* if there has been nothing to do on any drive,
2963 * then there is nothing to do at all..
2964 */
NeilBrown57afd892005-06-21 17:17:13 -07002965 *skipped = 1;
2966 return (max_sector - sector_nr) + sectors_skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 }
2968
NeilBrownc6207272008-02-06 01:39:52 -08002969 if (max_sector > mddev->resync_max)
2970 max_sector = mddev->resync_max; /* Don't do IO beyond here */
2971
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 /* make sure whole request will fit in a chunk - if chunks
2973 * are meaningful
2974 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10002975 if (conf->geo.near_copies < conf->geo.raid_disks &&
2976 max_sector > (sector_nr | chunk_mask))
2977 max_sector = (sector_nr | chunk_mask) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978
Tomasz Majchrzak7ac50442016-06-13 15:51:19 +02002979 /*
2980 * If there is non-resync activity waiting for a turn, then let it
2981 * though before starting on this new sync request.
2982 */
2983 if (conf->nr_waiting)
2984 schedule_timeout_uninterruptible(1);
2985
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 /* Again, very different code for resync and recovery.
2987 * Both must result in an r10bio with a list of bios that
2988 * have bi_end_io, bi_sector, bi_bdev set,
2989 * and bi_private set to the r10bio.
2990 * For recovery, we may actually create several r10bios
2991 * with 2 bios in each, that correspond to the bios in the main one.
2992 * In this case, the subordinate r10bios link back through a
2993 * borrowed master_bio pointer, and the counter in the master
2994 * includes a ref from each subordinate.
2995 */
2996 /* First, we decide what to do and set ->bi_end_io
2997 * To end_sync_read if we want to read, and
2998 * end_sync_write if we will want to write.
2999 */
3000
NeilBrown6cce3b22006-01-06 00:20:16 -08003001 max_sync = RESYNC_PAGES << (PAGE_SHIFT-9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3003 /* recovery... the complicated one */
NeilBrowne875ece2011-07-28 11:39:24 +10003004 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 r10_bio = NULL;
3006
NeilBrown5cf00fc2012-05-21 09:28:20 +10003007 for (i = 0 ; i < conf->geo.raid_disks; i++) {
NeilBrownab9d47e2011-05-11 14:54:41 +10003008 int still_degraded;
NeilBrown9f2c9d12011-10-11 16:48:43 +11003009 struct r10bio *rb2;
NeilBrownab9d47e2011-05-11 14:54:41 +10003010 sector_t sect;
3011 int must_sync;
NeilBrowne875ece2011-07-28 11:39:24 +10003012 int any_working;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003013 struct raid10_info *mirror = &conf->mirrors[i];
NeilBrownf90145f2016-06-02 16:19:52 +10003014 struct md_rdev *mrdev, *mreplace;
NeilBrownab9d47e2011-05-11 14:54:41 +10003015
NeilBrownf90145f2016-06-02 16:19:52 +10003016 rcu_read_lock();
3017 mrdev = rcu_dereference(mirror->rdev);
3018 mreplace = rcu_dereference(mirror->replacement);
3019
3020 if ((mrdev == NULL ||
NeilBrownf5b67ae2016-06-02 16:19:53 +10003021 test_bit(Faulty, &mrdev->flags) ||
NeilBrownf90145f2016-06-02 16:19:52 +10003022 test_bit(In_sync, &mrdev->flags)) &&
3023 (mreplace == NULL ||
3024 test_bit(Faulty, &mreplace->flags))) {
3025 rcu_read_unlock();
NeilBrownab9d47e2011-05-11 14:54:41 +10003026 continue;
NeilBrownf90145f2016-06-02 16:19:52 +10003027 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003028
3029 still_degraded = 0;
3030 /* want to reconstruct this device */
3031 rb2 = r10_bio;
3032 sect = raid10_find_virt(conf, sector_nr, i);
NeilBrownfc448a12012-07-03 10:37:30 +10003033 if (sect >= mddev->resync_max_sectors) {
3034 /* last stripe is not complete - don't
3035 * try to recover this sector.
3036 */
NeilBrownf90145f2016-06-02 16:19:52 +10003037 rcu_read_unlock();
NeilBrownfc448a12012-07-03 10:37:30 +10003038 continue;
3039 }
NeilBrownf5b67ae2016-06-02 16:19:53 +10003040 if (mreplace && test_bit(Faulty, &mreplace->flags))
3041 mreplace = NULL;
NeilBrown24afd802011-12-23 10:17:55 +11003042 /* Unless we are doing a full sync, or a replacement
3043 * we only need to recover the block if it is set in
3044 * the bitmap
NeilBrownab9d47e2011-05-11 14:54:41 +10003045 */
3046 must_sync = bitmap_start_sync(mddev->bitmap, sect,
3047 &sync_blocks, 1);
3048 if (sync_blocks < max_sync)
3049 max_sync = sync_blocks;
3050 if (!must_sync &&
NeilBrownf90145f2016-06-02 16:19:52 +10003051 mreplace == NULL &&
NeilBrownab9d47e2011-05-11 14:54:41 +10003052 !conf->fullsync) {
3053 /* yep, skip the sync_blocks here, but don't assume
3054 * that there will never be anything to do here
NeilBrown6cce3b22006-01-06 00:20:16 -08003055 */
NeilBrownab9d47e2011-05-11 14:54:41 +10003056 chunks_skipped = -1;
NeilBrownf90145f2016-06-02 16:19:52 +10003057 rcu_read_unlock();
NeilBrownab9d47e2011-05-11 14:54:41 +10003058 continue;
3059 }
NeilBrownf90145f2016-06-02 16:19:52 +10003060 atomic_inc(&mrdev->nr_pending);
3061 if (mreplace)
3062 atomic_inc(&mreplace->nr_pending);
3063 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064
NeilBrownab9d47e2011-05-11 14:54:41 +10003065 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
NeilBrowncb8b12b2014-08-18 14:38:45 +10003066 r10_bio->state = 0;
NeilBrownab9d47e2011-05-11 14:54:41 +10003067 raise_barrier(conf, rb2 != NULL);
3068 atomic_set(&r10_bio->remaining, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069
NeilBrownab9d47e2011-05-11 14:54:41 +10003070 r10_bio->master_bio = (struct bio*)rb2;
3071 if (rb2)
3072 atomic_inc(&rb2->remaining);
3073 r10_bio->mddev = mddev;
3074 set_bit(R10BIO_IsRecover, &r10_bio->state);
3075 r10_bio->sector = sect;
NeilBrown6cce3b22006-01-06 00:20:16 -08003076
NeilBrownab9d47e2011-05-11 14:54:41 +10003077 raid10_find_phys(conf, r10_bio);
NeilBrown18055562009-05-07 12:48:10 +10003078
NeilBrownab9d47e2011-05-11 14:54:41 +10003079 /* Need to check if the array will still be
3080 * degraded
3081 */
NeilBrownf90145f2016-06-02 16:19:52 +10003082 rcu_read_lock();
3083 for (j = 0; j < conf->geo.raid_disks; j++) {
3084 struct md_rdev *rdev = rcu_dereference(
3085 conf->mirrors[j].rdev);
3086 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) {
NeilBrownab9d47e2011-05-11 14:54:41 +10003087 still_degraded = 1;
NeilBrown87fc7672005-09-09 16:24:04 -07003088 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 }
NeilBrownf90145f2016-06-02 16:19:52 +10003090 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003091
3092 must_sync = bitmap_start_sync(mddev->bitmap, sect,
3093 &sync_blocks, still_degraded);
3094
NeilBrowne875ece2011-07-28 11:39:24 +10003095 any_working = 0;
NeilBrownab9d47e2011-05-11 14:54:41 +10003096 for (j=0; j<conf->copies;j++) {
NeilBrowne875ece2011-07-28 11:39:24 +10003097 int k;
NeilBrownab9d47e2011-05-11 14:54:41 +10003098 int d = r10_bio->devs[j].devnum;
NeilBrown5e570282011-07-28 11:39:25 +10003099 sector_t from_addr, to_addr;
NeilBrownf90145f2016-06-02 16:19:52 +10003100 struct md_rdev *rdev =
3101 rcu_dereference(conf->mirrors[d].rdev);
NeilBrown40c356c2011-07-28 11:39:24 +10003102 sector_t sector, first_bad;
3103 int bad_sectors;
NeilBrownf90145f2016-06-02 16:19:52 +10003104 if (!rdev ||
3105 !test_bit(In_sync, &rdev->flags))
NeilBrownab9d47e2011-05-11 14:54:41 +10003106 continue;
3107 /* This is where we read from */
NeilBrowne875ece2011-07-28 11:39:24 +10003108 any_working = 1;
NeilBrown40c356c2011-07-28 11:39:24 +10003109 sector = r10_bio->devs[j].addr;
3110
3111 if (is_badblock(rdev, sector, max_sync,
3112 &first_bad, &bad_sectors)) {
3113 if (first_bad > sector)
3114 max_sync = first_bad - sector;
3115 else {
3116 bad_sectors -= (sector
3117 - first_bad);
3118 if (max_sync > bad_sectors)
3119 max_sync = bad_sectors;
3120 continue;
3121 }
3122 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003123 bio = r10_bio->devs[0].bio;
3124 bio->bi_next = biolist;
3125 biolist = bio;
NeilBrownab9d47e2011-05-11 14:54:41 +10003126 bio->bi_end_io = end_sync_read;
Mike Christie796a5cf2016-06-05 14:32:07 -05003127 bio_set_op_attrs(bio, REQ_OP_READ, 0);
NeilBrown8d3ca832016-11-18 16:16:12 +11003128 if (test_bit(FailFast, &rdev->flags))
3129 bio->bi_opf |= MD_FAILFAST;
NeilBrown5e570282011-07-28 11:39:25 +10003130 from_addr = r10_bio->devs[j].addr;
Kent Overstreet4f024f32013-10-11 15:44:27 -07003131 bio->bi_iter.bi_sector = from_addr +
3132 rdev->data_offset;
NeilBrown24afd802011-12-23 10:17:55 +11003133 bio->bi_bdev = rdev->bdev;
3134 atomic_inc(&rdev->nr_pending);
3135 /* and we write to 'i' (if not in_sync) */
NeilBrownab9d47e2011-05-11 14:54:41 +10003136
3137 for (k=0; k<conf->copies; k++)
3138 if (r10_bio->devs[k].devnum == i)
3139 break;
3140 BUG_ON(k == conf->copies);
NeilBrown5e570282011-07-28 11:39:25 +10003141 to_addr = r10_bio->devs[k].addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003142 r10_bio->devs[0].devnum = d;
NeilBrown5e570282011-07-28 11:39:25 +10003143 r10_bio->devs[0].addr = from_addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003144 r10_bio->devs[1].devnum = i;
NeilBrown5e570282011-07-28 11:39:25 +10003145 r10_bio->devs[1].addr = to_addr;
NeilBrownab9d47e2011-05-11 14:54:41 +10003146
NeilBrownf90145f2016-06-02 16:19:52 +10003147 if (!test_bit(In_sync, &mrdev->flags)) {
NeilBrown24afd802011-12-23 10:17:55 +11003148 bio = r10_bio->devs[1].bio;
3149 bio->bi_next = biolist;
3150 biolist = bio;
NeilBrown24afd802011-12-23 10:17:55 +11003151 bio->bi_end_io = end_sync_write;
Mike Christie796a5cf2016-06-05 14:32:07 -05003152 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Kent Overstreet4f024f32013-10-11 15:44:27 -07003153 bio->bi_iter.bi_sector = to_addr
NeilBrownf90145f2016-06-02 16:19:52 +10003154 + mrdev->data_offset;
3155 bio->bi_bdev = mrdev->bdev;
NeilBrown24afd802011-12-23 10:17:55 +11003156 atomic_inc(&r10_bio->remaining);
3157 } else
3158 r10_bio->devs[1].bio->bi_end_io = NULL;
3159
3160 /* and maybe write to replacement */
3161 bio = r10_bio->devs[1].repl_bio;
3162 if (bio)
3163 bio->bi_end_io = NULL;
NeilBrownf90145f2016-06-02 16:19:52 +10003164 /* Note: if mreplace != NULL, then bio
NeilBrown24afd802011-12-23 10:17:55 +11003165 * cannot be NULL as r10buf_pool_alloc will
3166 * have allocated it.
3167 * So the second test here is pointless.
3168 * But it keeps semantic-checkers happy, and
3169 * this comment keeps human reviewers
3170 * happy.
3171 */
NeilBrownf90145f2016-06-02 16:19:52 +10003172 if (mreplace == NULL || bio == NULL ||
3173 test_bit(Faulty, &mreplace->flags))
NeilBrown24afd802011-12-23 10:17:55 +11003174 break;
3175 bio->bi_next = biolist;
3176 biolist = bio;
NeilBrown24afd802011-12-23 10:17:55 +11003177 bio->bi_end_io = end_sync_write;
Mike Christie796a5cf2016-06-05 14:32:07 -05003178 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Kent Overstreet4f024f32013-10-11 15:44:27 -07003179 bio->bi_iter.bi_sector = to_addr +
NeilBrownf90145f2016-06-02 16:19:52 +10003180 mreplace->data_offset;
3181 bio->bi_bdev = mreplace->bdev;
NeilBrown24afd802011-12-23 10:17:55 +11003182 atomic_inc(&r10_bio->remaining);
NeilBrownab9d47e2011-05-11 14:54:41 +10003183 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 }
NeilBrownf90145f2016-06-02 16:19:52 +10003185 rcu_read_unlock();
NeilBrownab9d47e2011-05-11 14:54:41 +10003186 if (j == conf->copies) {
NeilBrowne875ece2011-07-28 11:39:24 +10003187 /* Cannot recover, so abort the recovery or
3188 * record a bad block */
NeilBrowne875ece2011-07-28 11:39:24 +10003189 if (any_working) {
3190 /* problem is that there are bad blocks
3191 * on other device(s)
3192 */
3193 int k;
3194 for (k = 0; k < conf->copies; k++)
3195 if (r10_bio->devs[k].devnum == i)
3196 break;
NeilBrown24afd802011-12-23 10:17:55 +11003197 if (!test_bit(In_sync,
NeilBrownf90145f2016-06-02 16:19:52 +10003198 &mrdev->flags)
NeilBrown24afd802011-12-23 10:17:55 +11003199 && !rdev_set_badblocks(
NeilBrownf90145f2016-06-02 16:19:52 +10003200 mrdev,
NeilBrown24afd802011-12-23 10:17:55 +11003201 r10_bio->devs[k].addr,
3202 max_sync, 0))
3203 any_working = 0;
NeilBrownf90145f2016-06-02 16:19:52 +10003204 if (mreplace &&
NeilBrown24afd802011-12-23 10:17:55 +11003205 !rdev_set_badblocks(
NeilBrownf90145f2016-06-02 16:19:52 +10003206 mreplace,
NeilBrowne875ece2011-07-28 11:39:24 +10003207 r10_bio->devs[k].addr,
3208 max_sync, 0))
3209 any_working = 0;
3210 }
3211 if (!any_working) {
3212 if (!test_and_set_bit(MD_RECOVERY_INTR,
3213 &mddev->recovery))
NeilBrown08464e02016-11-02 14:16:50 +11003214 pr_warn("md/raid10:%s: insufficient working devices for recovery.\n",
NeilBrowne875ece2011-07-28 11:39:24 +10003215 mdname(mddev));
NeilBrown24afd802011-12-23 10:17:55 +11003216 mirror->recovery_disabled
NeilBrowne875ece2011-07-28 11:39:24 +10003217 = mddev->recovery_disabled;
3218 }
NeilBrowne8b84912014-01-06 10:35:34 +11003219 put_buf(r10_bio);
3220 if (rb2)
3221 atomic_dec(&rb2->remaining);
3222 r10_bio = rb2;
NeilBrownf90145f2016-06-02 16:19:52 +10003223 rdev_dec_pending(mrdev, mddev);
3224 if (mreplace)
3225 rdev_dec_pending(mreplace, mddev);
NeilBrownab9d47e2011-05-11 14:54:41 +10003226 break;
3227 }
NeilBrownf90145f2016-06-02 16:19:52 +10003228 rdev_dec_pending(mrdev, mddev);
3229 if (mreplace)
3230 rdev_dec_pending(mreplace, mddev);
NeilBrown8d3ca832016-11-18 16:16:12 +11003231 if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) {
3232 /* Only want this if there is elsewhere to
3233 * read from. 'j' is currently the first
3234 * readable copy.
3235 */
3236 int targets = 1;
3237 for (; j < conf->copies; j++) {
3238 int d = r10_bio->devs[j].devnum;
3239 if (conf->mirrors[d].rdev &&
3240 test_bit(In_sync,
3241 &conf->mirrors[d].rdev->flags))
3242 targets++;
3243 }
3244 if (targets == 1)
3245 r10_bio->devs[0].bio->bi_opf
3246 &= ~MD_FAILFAST;
3247 }
NeilBrownab9d47e2011-05-11 14:54:41 +10003248 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 if (biolist == NULL) {
3250 while (r10_bio) {
NeilBrown9f2c9d12011-10-11 16:48:43 +11003251 struct r10bio *rb2 = r10_bio;
3252 r10_bio = (struct r10bio*) rb2->master_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 rb2->master_bio = NULL;
3254 put_buf(rb2);
3255 }
3256 goto giveup;
3257 }
3258 } else {
3259 /* resync. Schedule a read for every block at this virt offset */
3260 int count = 0;
NeilBrown6cce3b22006-01-06 00:20:16 -08003261
Goldwyn Rodriguesc40f3412015-08-19 08:14:42 +10003262 bitmap_cond_end_sync(mddev->bitmap, sector_nr, 0);
NeilBrown78200d42009-02-25 13:18:47 +11003263
NeilBrown6cce3b22006-01-06 00:20:16 -08003264 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
3265 &sync_blocks, mddev->degraded) &&
NeilBrownab9d47e2011-05-11 14:54:41 +10003266 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED,
3267 &mddev->recovery)) {
NeilBrown6cce3b22006-01-06 00:20:16 -08003268 /* We can skip this block */
3269 *skipped = 1;
3270 return sync_blocks + sectors_skipped;
3271 }
3272 if (sync_blocks < max_sync)
3273 max_sync = sync_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
NeilBrowncb8b12b2014-08-18 14:38:45 +10003275 r10_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 r10_bio->mddev = mddev;
3278 atomic_set(&r10_bio->remaining, 0);
NeilBrown6cce3b22006-01-06 00:20:16 -08003279 raise_barrier(conf, 0);
3280 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281
3282 r10_bio->master_bio = NULL;
3283 r10_bio->sector = sector_nr;
3284 set_bit(R10BIO_IsSync, &r10_bio->state);
3285 raid10_find_phys(conf, r10_bio);
NeilBrown5cf00fc2012-05-21 09:28:20 +10003286 r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287
NeilBrown5cf00fc2012-05-21 09:28:20 +10003288 for (i = 0; i < conf->copies; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 int d = r10_bio->devs[i].devnum;
NeilBrown40c356c2011-07-28 11:39:24 +10003290 sector_t first_bad, sector;
3291 int bad_sectors;
NeilBrownf90145f2016-06-02 16:19:52 +10003292 struct md_rdev *rdev;
NeilBrown40c356c2011-07-28 11:39:24 +10003293
NeilBrown9ad1aef2011-12-23 10:17:55 +11003294 if (r10_bio->devs[i].repl_bio)
3295 r10_bio->devs[i].repl_bio->bi_end_io = NULL;
3296
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 bio = r10_bio->devs[i].bio;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003298 bio->bi_error = -EIO;
NeilBrownf90145f2016-06-02 16:19:52 +10003299 rcu_read_lock();
3300 rdev = rcu_dereference(conf->mirrors[d].rdev);
3301 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) {
3302 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 continue;
NeilBrownf90145f2016-06-02 16:19:52 +10003304 }
NeilBrown40c356c2011-07-28 11:39:24 +10003305 sector = r10_bio->devs[i].addr;
NeilBrownf90145f2016-06-02 16:19:52 +10003306 if (is_badblock(rdev, sector, max_sync,
NeilBrown40c356c2011-07-28 11:39:24 +10003307 &first_bad, &bad_sectors)) {
3308 if (first_bad > sector)
3309 max_sync = first_bad - sector;
3310 else {
3311 bad_sectors -= (sector - first_bad);
3312 if (max_sync > bad_sectors)
Dan Carpenter91502f02012-10-11 14:20:58 +11003313 max_sync = bad_sectors;
NeilBrownf90145f2016-06-02 16:19:52 +10003314 rcu_read_unlock();
NeilBrown40c356c2011-07-28 11:39:24 +10003315 continue;
3316 }
3317 }
NeilBrownf90145f2016-06-02 16:19:52 +10003318 atomic_inc(&rdev->nr_pending);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 atomic_inc(&r10_bio->remaining);
3320 bio->bi_next = biolist;
3321 biolist = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 bio->bi_end_io = end_sync_read;
Mike Christie796a5cf2016-06-05 14:32:07 -05003323 bio_set_op_attrs(bio, REQ_OP_READ, 0);
NeilBrown8d3ca832016-11-18 16:16:12 +11003324 if (test_bit(FailFast, &conf->mirrors[d].rdev->flags))
3325 bio->bi_opf |= MD_FAILFAST;
NeilBrownf90145f2016-06-02 16:19:52 +10003326 bio->bi_iter.bi_sector = sector + rdev->data_offset;
3327 bio->bi_bdev = rdev->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 count++;
NeilBrown9ad1aef2011-12-23 10:17:55 +11003329
NeilBrownf90145f2016-06-02 16:19:52 +10003330 rdev = rcu_dereference(conf->mirrors[d].replacement);
3331 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) {
3332 rcu_read_unlock();
NeilBrown9ad1aef2011-12-23 10:17:55 +11003333 continue;
NeilBrownf90145f2016-06-02 16:19:52 +10003334 }
3335 atomic_inc(&rdev->nr_pending);
3336 rcu_read_unlock();
NeilBrown9ad1aef2011-12-23 10:17:55 +11003337
3338 /* Need to set up for writing to the replacement */
3339 bio = r10_bio->devs[i].repl_bio;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003340 bio->bi_error = -EIO;
NeilBrown9ad1aef2011-12-23 10:17:55 +11003341
3342 sector = r10_bio->devs[i].addr;
NeilBrown9ad1aef2011-12-23 10:17:55 +11003343 bio->bi_next = biolist;
3344 biolist = bio;
NeilBrown9ad1aef2011-12-23 10:17:55 +11003345 bio->bi_end_io = end_sync_write;
Mike Christie796a5cf2016-06-05 14:32:07 -05003346 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
NeilBrown1919cbb2016-11-18 16:16:12 +11003347 if (test_bit(FailFast, &conf->mirrors[d].rdev->flags))
3348 bio->bi_opf |= MD_FAILFAST;
NeilBrownf90145f2016-06-02 16:19:52 +10003349 bio->bi_iter.bi_sector = sector + rdev->data_offset;
3350 bio->bi_bdev = rdev->bdev;
NeilBrown9ad1aef2011-12-23 10:17:55 +11003351 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 }
3353
3354 if (count < 2) {
3355 for (i=0; i<conf->copies; i++) {
3356 int d = r10_bio->devs[i].devnum;
3357 if (r10_bio->devs[i].bio->bi_end_io)
NeilBrownab9d47e2011-05-11 14:54:41 +10003358 rdev_dec_pending(conf->mirrors[d].rdev,
3359 mddev);
NeilBrown9ad1aef2011-12-23 10:17:55 +11003360 if (r10_bio->devs[i].repl_bio &&
3361 r10_bio->devs[i].repl_bio->bi_end_io)
3362 rdev_dec_pending(
3363 conf->mirrors[d].replacement,
3364 mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 }
3366 put_buf(r10_bio);
3367 biolist = NULL;
3368 goto giveup;
3369 }
3370 }
3371
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 nr_sectors = 0;
NeilBrown6cce3b22006-01-06 00:20:16 -08003373 if (sector_nr + max_sync < max_sector)
3374 max_sector = sector_nr + max_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 do {
3376 struct page *page;
3377 int len = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378 if (sector_nr + (len>>9) > max_sector)
3379 len = (max_sector - sector_nr) << 9;
3380 if (len == 0)
3381 break;
3382 for (bio= biolist ; bio ; bio=bio->bi_next) {
Ming Leif0250612017-03-17 00:12:33 +08003383 struct resync_pages *rp = get_resync_pages(bio);
3384 page = resync_fetch_page(rp, rp->idx++);
Ming Leic85ba142017-03-17 00:12:22 +08003385 /*
3386 * won't fail because the vec table is big enough
3387 * to hold all these pages
3388 */
3389 bio_add_page(bio, page, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 }
3391 nr_sectors += len>>9;
3392 sector_nr += len>>9;
Ming Leif0250612017-03-17 00:12:33 +08003393 } while (get_resync_pages(biolist)->idx < RESYNC_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 r10_bio->sectors = nr_sectors;
3395
3396 while (biolist) {
3397 bio = biolist;
3398 biolist = biolist->bi_next;
3399
3400 bio->bi_next = NULL;
Ming Leif0250612017-03-17 00:12:33 +08003401 r10_bio = get_resync_r10bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 r10_bio->sectors = nr_sectors;
3403
3404 if (bio->bi_end_io == end_sync_read) {
3405 md_sync_acct(bio->bi_bdev, nr_sectors);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003406 bio->bi_error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 generic_make_request(bio);
3408 }
3409 }
3410
NeilBrown57afd892005-06-21 17:17:13 -07003411 if (sectors_skipped)
3412 /* pretend they weren't skipped, it makes
3413 * no important difference in this case
3414 */
3415 md_done_sync(mddev, sectors_skipped, 1);
3416
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417 return sectors_skipped + nr_sectors;
3418 giveup:
3419 /* There is nowhere to write, so all non-sync
NeilBrowne875ece2011-07-28 11:39:24 +10003420 * drives must be failed or in resync, all drives
3421 * have a bad block, so try the next chunk...
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 */
NeilBrown09b40682009-02-25 13:18:47 +11003423 if (sector_nr + max_sync < max_sector)
3424 max_sector = sector_nr + max_sync;
3425
3426 sectors_skipped += (max_sector - sector_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 chunks_skipped ++;
3428 sector_nr = max_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 goto skipped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430}
3431
Dan Williams80c3a6c2009-03-17 18:10:40 -07003432static sector_t
NeilBrownfd01b882011-10-11 16:47:53 +11003433raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07003434{
3435 sector_t size;
NeilBrowne879a872011-10-11 16:49:02 +11003436 struct r10conf *conf = mddev->private;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003437
3438 if (!raid_disks)
NeilBrown3ea7daa2012-05-22 13:53:47 +10003439 raid_disks = min(conf->geo.raid_disks,
3440 conf->prev.raid_disks);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003441 if (!sectors)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003442 sectors = conf->dev_sectors;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003443
NeilBrown5cf00fc2012-05-21 09:28:20 +10003444 size = sectors >> conf->geo.chunk_shift;
3445 sector_div(size, conf->geo.far_copies);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003446 size = size * raid_disks;
NeilBrown5cf00fc2012-05-21 09:28:20 +10003447 sector_div(size, conf->geo.near_copies);
Dan Williams80c3a6c2009-03-17 18:10:40 -07003448
NeilBrown5cf00fc2012-05-21 09:28:20 +10003449 return size << conf->geo.chunk_shift;
Dan Williams80c3a6c2009-03-17 18:10:40 -07003450}
3451
NeilBrown6508fdb2012-05-17 10:08:45 +10003452static void calc_sectors(struct r10conf *conf, sector_t size)
3453{
3454 /* Calculate the number of sectors-per-device that will
3455 * actually be used, and set conf->dev_sectors and
3456 * conf->stride
3457 */
3458
NeilBrown5cf00fc2012-05-21 09:28:20 +10003459 size = size >> conf->geo.chunk_shift;
3460 sector_div(size, conf->geo.far_copies);
3461 size = size * conf->geo.raid_disks;
3462 sector_div(size, conf->geo.near_copies);
NeilBrown6508fdb2012-05-17 10:08:45 +10003463 /* 'size' is now the number of chunks in the array */
3464 /* calculate "used chunks per device" */
3465 size = size * conf->copies;
3466
3467 /* We need to round up when dividing by raid_disks to
3468 * get the stride size.
3469 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003470 size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks);
NeilBrown6508fdb2012-05-17 10:08:45 +10003471
NeilBrown5cf00fc2012-05-21 09:28:20 +10003472 conf->dev_sectors = size << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003473
NeilBrown5cf00fc2012-05-21 09:28:20 +10003474 if (conf->geo.far_offset)
3475 conf->geo.stride = 1 << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003476 else {
NeilBrown5cf00fc2012-05-21 09:28:20 +10003477 sector_div(size, conf->geo.far_copies);
3478 conf->geo.stride = size << conf->geo.chunk_shift;
NeilBrown6508fdb2012-05-17 10:08:45 +10003479 }
3480}
Trela, Maciejdab8b292010-03-08 16:02:45 +11003481
NeilBrowndeb200d2012-05-21 09:28:33 +10003482enum geo_type {geo_new, geo_old, geo_start};
3483static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new)
3484{
3485 int nc, fc, fo;
3486 int layout, chunk, disks;
3487 switch (new) {
3488 case geo_old:
3489 layout = mddev->layout;
3490 chunk = mddev->chunk_sectors;
3491 disks = mddev->raid_disks - mddev->delta_disks;
3492 break;
3493 case geo_new:
3494 layout = mddev->new_layout;
3495 chunk = mddev->new_chunk_sectors;
3496 disks = mddev->raid_disks;
3497 break;
3498 default: /* avoid 'may be unused' warnings */
3499 case geo_start: /* new when starting reshape - raid_disks not
3500 * updated yet. */
3501 layout = mddev->new_layout;
3502 chunk = mddev->new_chunk_sectors;
3503 disks = mddev->raid_disks + mddev->delta_disks;
3504 break;
3505 }
NeilBrown8bce6d32015-10-22 13:20:15 +11003506 if (layout >> 19)
NeilBrowndeb200d2012-05-21 09:28:33 +10003507 return -1;
3508 if (chunk < (PAGE_SIZE >> 9) ||
3509 !is_power_of_2(chunk))
3510 return -2;
3511 nc = layout & 255;
3512 fc = (layout >> 8) & 255;
3513 fo = layout & (1<<16);
3514 geo->raid_disks = disks;
3515 geo->near_copies = nc;
3516 geo->far_copies = fc;
3517 geo->far_offset = fo;
NeilBrown8bce6d32015-10-22 13:20:15 +11003518 switch (layout >> 17) {
3519 case 0: /* original layout. simple but not always optimal */
3520 geo->far_set_size = disks;
3521 break;
3522 case 1: /* "improved" layout which was buggy. Hopefully no-one is
3523 * actually using this, but leave code here just in case.*/
3524 geo->far_set_size = disks/fc;
3525 WARN(geo->far_set_size < fc,
3526 "This RAID10 layout does not provide data safety - please backup and create new array\n");
3527 break;
3528 case 2: /* "improved" layout fixed to match documentation */
3529 geo->far_set_size = fc * nc;
3530 break;
3531 default: /* Not a valid layout */
3532 return -1;
3533 }
NeilBrowndeb200d2012-05-21 09:28:33 +10003534 geo->chunk_mask = chunk - 1;
3535 geo->chunk_shift = ffz(~chunk);
3536 return nc*fc;
3537}
3538
NeilBrowne879a872011-10-11 16:49:02 +11003539static struct r10conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540{
NeilBrowne879a872011-10-11 16:49:02 +11003541 struct r10conf *conf = NULL;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003542 int err = -EINVAL;
NeilBrowndeb200d2012-05-21 09:28:33 +10003543 struct geom geo;
3544 int copies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545
NeilBrowndeb200d2012-05-21 09:28:33 +10003546 copies = setup_geo(&geo, mddev, geo_new);
3547
3548 if (copies == -2) {
NeilBrown08464e02016-11-02 14:16:50 +11003549 pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n",
3550 mdname(mddev), PAGE_SIZE);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003551 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552 }
NeilBrown2604b702006-01-06 00:20:36 -08003553
NeilBrowndeb200d2012-05-21 09:28:33 +10003554 if (copies < 2 || copies > mddev->raid_disks) {
NeilBrown08464e02016-11-02 14:16:50 +11003555 pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n",
3556 mdname(mddev), mddev->new_layout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557 goto out;
3558 }
Trela, Maciejdab8b292010-03-08 16:02:45 +11003559
3560 err = -ENOMEM;
NeilBrowne879a872011-10-11 16:49:02 +11003561 conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003562 if (!conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 goto out;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003564
NeilBrown3ea7daa2012-05-22 13:53:47 +10003565 /* FIXME calc properly */
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003566 conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks +
NeilBrown78eaa0d2013-07-02 15:58:05 +10003567 max(0,-mddev->delta_disks)),
Trela, Maciejdab8b292010-03-08 16:02:45 +11003568 GFP_KERNEL);
3569 if (!conf->mirrors)
3570 goto out;
NeilBrown4443ae12006-01-06 00:20:28 -08003571
3572 conf->tmppage = alloc_page(GFP_KERNEL);
3573 if (!conf->tmppage)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003574 goto out;
3575
NeilBrowndeb200d2012-05-21 09:28:33 +10003576 conf->geo = geo;
3577 conf->copies = copies;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003578 conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc,
3579 r10bio_pool_free, conf);
3580 if (!conf->r10bio_pool)
3581 goto out;
3582
NeilBrownfc9977d2017-04-05 14:05:51 +10003583 conf->bio_split = bioset_create(BIO_POOL_SIZE, 0);
3584 if (!conf->bio_split)
3585 goto out;
3586
NeilBrown6508fdb2012-05-17 10:08:45 +10003587 calc_sectors(conf, mddev->dev_sectors);
NeilBrown3ea7daa2012-05-22 13:53:47 +10003588 if (mddev->reshape_position == MaxSector) {
3589 conf->prev = conf->geo;
3590 conf->reshape_progress = MaxSector;
3591 } else {
3592 if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) {
3593 err = -EINVAL;
3594 goto out;
3595 }
3596 conf->reshape_progress = mddev->reshape_position;
3597 if (conf->prev.far_offset)
3598 conf->prev.stride = 1 << conf->prev.chunk_shift;
3599 else
3600 /* far_copies must be 1 */
3601 conf->prev.stride = conf->dev_sectors;
3602 }
NeilBrown299b0682015-07-06 17:37:49 +10003603 conf->reshape_safe = conf->reshape_progress;
Neil Browne7e72bf2008-05-14 16:05:54 -07003604 spin_lock_init(&conf->device_lock);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003605 INIT_LIST_HEAD(&conf->retry_list);
NeilBrown95af5872015-08-14 11:26:17 +10003606 INIT_LIST_HEAD(&conf->bio_end_io_list);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003607
3608 spin_lock_init(&conf->resync_lock);
3609 init_waitqueue_head(&conf->wait_barrier);
Tomasz Majchrzak0e5313e2016-06-24 14:20:16 +02003610 atomic_set(&conf->nr_pending, 0);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003611
NeilBrown02326052012-07-03 15:56:52 +10003612 conf->thread = md_register_thread(raid10d, mddev, "raid10");
Trela, Maciejdab8b292010-03-08 16:02:45 +11003613 if (!conf->thread)
3614 goto out;
3615
Trela, Maciejdab8b292010-03-08 16:02:45 +11003616 conf->mddev = mddev;
3617 return conf;
3618
3619 out:
Trela, Maciejdab8b292010-03-08 16:02:45 +11003620 if (conf) {
Julia Lawall644df1a2015-09-13 14:15:10 +02003621 mempool_destroy(conf->r10bio_pool);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003622 kfree(conf->mirrors);
3623 safe_put_page(conf->tmppage);
NeilBrownfc9977d2017-04-05 14:05:51 +10003624 if (conf->bio_split)
3625 bioset_free(conf->bio_split);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003626 kfree(conf);
3627 }
3628 return ERR_PTR(err);
3629}
3630
Shaohua Li849674e2016-01-20 13:52:20 -08003631static int raid10_run(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003632{
NeilBrowne879a872011-10-11 16:49:02 +11003633 struct r10conf *conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003634 int i, disk_idx, chunk_size;
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003635 struct raid10_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11003636 struct md_rdev *rdev;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003637 sector_t size;
NeilBrown3ea7daa2012-05-22 13:53:47 +10003638 sector_t min_offset_diff = 0;
3639 int first = 1;
Shaohua Li532a2a32012-10-11 13:30:52 +11003640 bool discard_supported = false;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003641
3642 if (mddev->private == NULL) {
3643 conf = setup_conf(mddev);
3644 if (IS_ERR(conf))
3645 return PTR_ERR(conf);
3646 mddev->private = conf;
3647 }
3648 conf = mddev->private;
3649 if (!conf)
3650 goto out;
3651
Trela, Maciejdab8b292010-03-08 16:02:45 +11003652 mddev->thread = conf->thread;
3653 conf->thread = NULL;
3654
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10003655 chunk_size = mddev->chunk_sectors << 9;
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003656 if (mddev->queue) {
Shaohua Li532a2a32012-10-11 13:30:52 +11003657 blk_queue_max_discard_sectors(mddev->queue,
3658 mddev->chunk_sectors);
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07003659 blk_queue_max_write_same_sectors(mddev->queue, 0);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003660 blk_queue_io_min(mddev->queue, chunk_size);
3661 if (conf->geo.raid_disks % conf->geo.near_copies)
3662 blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks);
3663 else
3664 blk_queue_io_opt(mddev->queue, chunk_size *
3665 (conf->geo.raid_disks / conf->geo.near_copies));
3666 }
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10003667
NeilBrowndafb20f2012-03-19 12:46:39 +11003668 rdev_for_each(rdev, mddev) {
NeilBrown3ea7daa2012-05-22 13:53:47 +10003669 long long diff;
NeilBrownaba336b2012-05-31 15:39:11 +10003670 struct request_queue *q;
NeilBrown34b343c2011-07-28 11:31:47 +10003671
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 disk_idx = rdev->raid_disk;
NeilBrownf8c9e742012-05-21 09:28:33 +10003673 if (disk_idx < 0)
3674 continue;
3675 if (disk_idx >= conf->geo.raid_disks &&
3676 disk_idx >= conf->prev.raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677 continue;
3678 disk = conf->mirrors + disk_idx;
3679
NeilBrown56a25592011-12-23 10:17:55 +11003680 if (test_bit(Replacement, &rdev->flags)) {
3681 if (disk->replacement)
3682 goto out_free_conf;
3683 disk->replacement = rdev;
3684 } else {
3685 if (disk->rdev)
3686 goto out_free_conf;
3687 disk->rdev = rdev;
3688 }
NeilBrownaba336b2012-05-31 15:39:11 +10003689 q = bdev_get_queue(rdev->bdev);
NeilBrown3ea7daa2012-05-22 13:53:47 +10003690 diff = (rdev->new_data_offset - rdev->data_offset);
3691 if (!mddev->reshape_backwards)
3692 diff = -diff;
3693 if (diff < 0)
3694 diff = 0;
3695 if (first || diff < min_offset_diff)
3696 min_offset_diff = diff;
NeilBrown56a25592011-12-23 10:17:55 +11003697
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003698 if (mddev->gendisk)
3699 disk_stack_limits(mddev->gendisk, rdev->bdev,
3700 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701
3702 disk->head_position = 0;
Shaohua Li532a2a32012-10-11 13:30:52 +11003703
3704 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
3705 discard_supported = true;
Guoqing Jiang6f287ca2017-04-06 09:12:18 +08003706 first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10003708
Jonathan Brassowed30be02012-10-31 11:42:30 +11003709 if (mddev->queue) {
3710 if (discard_supported)
3711 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
3712 mddev->queue);
3713 else
3714 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
3715 mddev->queue);
3716 }
NeilBrown6d508242005-09-09 16:24:03 -07003717 /* need to check that every block has at least one working mirror */
NeilBrown700c7212011-07-27 11:00:36 +10003718 if (!enough(conf, -1)) {
NeilBrown08464e02016-11-02 14:16:50 +11003719 pr_err("md/raid10:%s: not enough operational mirrors.\n",
NeilBrown6d508242005-09-09 16:24:03 -07003720 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003721 goto out_free_conf;
3722 }
3723
NeilBrown3ea7daa2012-05-22 13:53:47 +10003724 if (conf->reshape_progress != MaxSector) {
3725 /* must ensure that shape change is supported */
3726 if (conf->geo.far_copies != 1 &&
3727 conf->geo.far_offset == 0)
3728 goto out_free_conf;
3729 if (conf->prev.far_copies != 1 &&
NeilBrown78eaa0d2013-07-02 15:58:05 +10003730 conf->prev.far_offset == 0)
NeilBrown3ea7daa2012-05-22 13:53:47 +10003731 goto out_free_conf;
3732 }
3733
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 mddev->degraded = 0;
NeilBrownf8c9e742012-05-21 09:28:33 +10003735 for (i = 0;
3736 i < conf->geo.raid_disks
3737 || i < conf->prev.raid_disks;
3738 i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739
3740 disk = conf->mirrors + i;
3741
NeilBrown56a25592011-12-23 10:17:55 +11003742 if (!disk->rdev && disk->replacement) {
3743 /* The replacement is all we have - use it */
3744 disk->rdev = disk->replacement;
3745 disk->replacement = NULL;
3746 clear_bit(Replacement, &disk->rdev->flags);
3747 }
3748
NeilBrown5fd6c1d2006-06-26 00:27:40 -07003749 if (!disk->rdev ||
NeilBrown2e333e82006-10-21 10:24:07 -07003750 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 disk->head_position = 0;
3752 mddev->degraded++;
NeilBrown0b59bb62014-01-14 16:30:10 +11003753 if (disk->rdev &&
3754 disk->rdev->saved_raid_disk < 0)
Neil Brown8c2e8702008-06-28 08:30:52 +10003755 conf->fullsync = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 }
NeilBrownd890fa22011-10-26 11:54:39 +11003757 disk->recovery_disabled = mddev->recovery_disabled - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 }
3759
Andre Noll8c6ac862009-06-18 08:48:06 +10003760 if (mddev->recovery_cp != MaxSector)
NeilBrown08464e02016-11-02 14:16:50 +11003761 pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n",
3762 mdname(mddev));
3763 pr_info("md/raid10:%s: active with %d out of %d devices\n",
NeilBrown5cf00fc2012-05-21 09:28:20 +10003764 mdname(mddev), conf->geo.raid_disks - mddev->degraded,
3765 conf->geo.raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 /*
3767 * Ok, everything is just fine now
3768 */
Trela, Maciejdab8b292010-03-08 16:02:45 +11003769 mddev->dev_sectors = conf->dev_sectors;
3770 size = raid10_size(mddev, 0, 0);
3771 md_set_array_sectors(mddev, size);
3772 mddev->resync_max_sectors = size;
NeilBrown46533ff2016-11-18 16:16:11 +11003773 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003775 if (mddev->queue) {
NeilBrown5cf00fc2012-05-21 09:28:20 +10003776 int stripe = conf->geo.raid_disks *
Andre Noll9d8f0362009-06-18 08:45:01 +10003777 ((mddev->chunk_sectors << 9) / PAGE_SIZE);
Jonathan Brassowcc4d1ef2012-07-31 10:03:53 +10003778
3779 /* Calculate max read-ahead size.
3780 * We need to readahead at least twice a whole stripe....
3781 * maybe...
3782 */
NeilBrown5cf00fc2012-05-21 09:28:20 +10003783 stripe /= conf->geo.near_copies;
Jan Karadc3b17c2017-02-02 15:56:50 +01003784 if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
3785 mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786 }
3787
Martin K. Petersena91a2782011-03-17 11:11:05 +01003788 if (md_integrity_register(mddev))
3789 goto out_free_conf;
3790
NeilBrown3ea7daa2012-05-22 13:53:47 +10003791 if (conf->reshape_progress != MaxSector) {
3792 unsigned long before_length, after_length;
3793
3794 before_length = ((1 << conf->prev.chunk_shift) *
3795 conf->prev.far_copies);
3796 after_length = ((1 << conf->geo.chunk_shift) *
3797 conf->geo.far_copies);
3798
3799 if (max(before_length, after_length) > min_offset_diff) {
3800 /* This cannot work */
NeilBrown08464e02016-11-02 14:16:50 +11003801 pr_warn("md/raid10: offset difference not enough to continue reshape\n");
NeilBrown3ea7daa2012-05-22 13:53:47 +10003802 goto out_free_conf;
3803 }
3804 conf->offset_diff = min_offset_diff;
3805
NeilBrown3ea7daa2012-05-22 13:53:47 +10003806 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3807 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3808 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
3809 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3810 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
3811 "reshape");
3812 }
3813
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814 return 0;
3815
3816out_free_conf:
NeilBrown01f96c02011-09-21 15:30:20 +10003817 md_unregister_thread(&mddev->thread);
Julia Lawall644df1a2015-09-13 14:15:10 +02003818 mempool_destroy(conf->r10bio_pool);
NeilBrown1345b1d2006-01-06 00:20:40 -08003819 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003820 kfree(conf->mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 kfree(conf);
3822 mddev->private = NULL;
3823out:
3824 return -EIO;
3825}
3826
NeilBrownafa0f552014-12-15 12:56:58 +11003827static void raid10_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828{
NeilBrownafa0f552014-12-15 12:56:58 +11003829 struct r10conf *conf = priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830
Julia Lawall644df1a2015-09-13 14:15:10 +02003831 mempool_destroy(conf->r10bio_pool);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10003832 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07003833 kfree(conf->mirrors);
NeilBrownc4796e22014-08-23 20:19:26 +10003834 kfree(conf->mirrors_old);
3835 kfree(conf->mirrors_new);
NeilBrownfc9977d2017-04-05 14:05:51 +10003836 if (conf->bio_split)
3837 bioset_free(conf->bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839}
3840
NeilBrownfd01b882011-10-11 16:47:53 +11003841static void raid10_quiesce(struct mddev *mddev, int state)
NeilBrown6cce3b22006-01-06 00:20:16 -08003842{
NeilBrowne879a872011-10-11 16:49:02 +11003843 struct r10conf *conf = mddev->private;
NeilBrown6cce3b22006-01-06 00:20:16 -08003844
3845 switch(state) {
3846 case 1:
3847 raise_barrier(conf, 0);
3848 break;
3849 case 0:
3850 lower_barrier(conf);
3851 break;
3852 }
NeilBrown6cce3b22006-01-06 00:20:16 -08003853}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854
NeilBrown006a09a2012-03-19 12:46:40 +11003855static int raid10_resize(struct mddev *mddev, sector_t sectors)
3856{
3857 /* Resize of 'far' arrays is not supported.
3858 * For 'near' and 'offset' arrays we can set the
3859 * number of sectors used to be an appropriate multiple
3860 * of the chunk size.
3861 * For 'offset', this is far_copies*chunksize.
3862 * For 'near' the multiplier is the LCM of
3863 * near_copies and raid_disks.
3864 * So if far_copies > 1 && !far_offset, fail.
3865 * Else find LCM(raid_disks, near_copy)*far_copies and
3866 * multiply by chunk_size. Then round to this number.
3867 * This is mostly done by raid10_size()
3868 */
3869 struct r10conf *conf = mddev->private;
3870 sector_t oldsize, size;
3871
NeilBrownf8c9e742012-05-21 09:28:33 +10003872 if (mddev->reshape_position != MaxSector)
3873 return -EBUSY;
3874
NeilBrown5cf00fc2012-05-21 09:28:20 +10003875 if (conf->geo.far_copies > 1 && !conf->geo.far_offset)
NeilBrown006a09a2012-03-19 12:46:40 +11003876 return -EINVAL;
3877
3878 oldsize = raid10_size(mddev, 0, 0);
3879 size = raid10_size(mddev, sectors, 0);
NeilBrowna4a61252012-05-22 13:55:27 +10003880 if (mddev->external_size &&
3881 mddev->array_sectors > size)
NeilBrown006a09a2012-03-19 12:46:40 +11003882 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10003883 if (mddev->bitmap) {
3884 int ret = bitmap_resize(mddev->bitmap, size, 0, 0);
3885 if (ret)
3886 return ret;
3887 }
3888 md_set_array_sectors(mddev, size);
NeilBrown006a09a2012-03-19 12:46:40 +11003889 if (sectors > mddev->dev_sectors &&
3890 mddev->recovery_cp > oldsize) {
3891 mddev->recovery_cp = oldsize;
3892 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3893 }
NeilBrown6508fdb2012-05-17 10:08:45 +10003894 calc_sectors(conf, sectors);
3895 mddev->dev_sectors = conf->dev_sectors;
NeilBrown006a09a2012-03-19 12:46:40 +11003896 mddev->resync_max_sectors = size;
3897 return 0;
3898}
3899
NeilBrown53a6ab42015-02-12 14:09:57 +11003900static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003901{
NeilBrown3cb03002011-10-11 16:45:26 +11003902 struct md_rdev *rdev;
NeilBrowne879a872011-10-11 16:49:02 +11003903 struct r10conf *conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003904
3905 if (mddev->degraded > 0) {
NeilBrown08464e02016-11-02 14:16:50 +11003906 pr_warn("md/raid10:%s: Error: degraded raid0!\n",
3907 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003908 return ERR_PTR(-EINVAL);
3909 }
NeilBrown53a6ab42015-02-12 14:09:57 +11003910 sector_div(size, devs);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003911
Trela, Maciejdab8b292010-03-08 16:02:45 +11003912 /* Set new parameters */
3913 mddev->new_level = 10;
3914 /* new layout: far_copies = 1, near_copies = 2 */
3915 mddev->new_layout = (1<<8) + 2;
3916 mddev->new_chunk_sectors = mddev->chunk_sectors;
3917 mddev->delta_disks = mddev->raid_disks;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003918 mddev->raid_disks *= 2;
3919 /* make sure it will be not marked as dirty */
3920 mddev->recovery_cp = MaxSector;
NeilBrown53a6ab42015-02-12 14:09:57 +11003921 mddev->dev_sectors = size;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003922
3923 conf = setup_conf(mddev);
Krzysztof Wojcik02214dc2011-02-04 14:18:26 +01003924 if (!IS_ERR(conf)) {
NeilBrowndafb20f2012-03-19 12:46:39 +11003925 rdev_for_each(rdev, mddev)
NeilBrown53a6ab42015-02-12 14:09:57 +11003926 if (rdev->raid_disk >= 0) {
NeilBrowne93f68a2010-06-15 09:36:03 +01003927 rdev->new_raid_disk = rdev->raid_disk * 2;
NeilBrown53a6ab42015-02-12 14:09:57 +11003928 rdev->sectors = size;
3929 }
Krzysztof Wojcik02214dc2011-02-04 14:18:26 +01003930 conf->barrier = 1;
3931 }
3932
Trela, Maciejdab8b292010-03-08 16:02:45 +11003933 return conf;
3934}
3935
NeilBrownfd01b882011-10-11 16:47:53 +11003936static void *raid10_takeover(struct mddev *mddev)
Trela, Maciejdab8b292010-03-08 16:02:45 +11003937{
NeilBrowne373ab12011-10-11 16:48:59 +11003938 struct r0conf *raid0_conf;
Trela, Maciejdab8b292010-03-08 16:02:45 +11003939
3940 /* raid10 can take over:
3941 * raid0 - providing it has only two drives
3942 */
3943 if (mddev->level == 0) {
3944 /* for raid0 takeover only one zone is supported */
NeilBrowne373ab12011-10-11 16:48:59 +11003945 raid0_conf = mddev->private;
3946 if (raid0_conf->nr_strip_zones > 1) {
NeilBrown08464e02016-11-02 14:16:50 +11003947 pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n",
3948 mdname(mddev));
Trela, Maciejdab8b292010-03-08 16:02:45 +11003949 return ERR_PTR(-EINVAL);
3950 }
NeilBrown53a6ab42015-02-12 14:09:57 +11003951 return raid10_takeover_raid0(mddev,
3952 raid0_conf->strip_zone->zone_end,
3953 raid0_conf->strip_zone->nb_dev);
Trela, Maciejdab8b292010-03-08 16:02:45 +11003954 }
3955 return ERR_PTR(-EINVAL);
3956}
3957
NeilBrown3ea7daa2012-05-22 13:53:47 +10003958static int raid10_check_reshape(struct mddev *mddev)
3959{
3960 /* Called when there is a request to change
3961 * - layout (to ->new_layout)
3962 * - chunk size (to ->new_chunk_sectors)
3963 * - raid_disks (by delta_disks)
3964 * or when trying to restart a reshape that was ongoing.
3965 *
3966 * We need to validate the request and possibly allocate
3967 * space if that might be an issue later.
3968 *
3969 * Currently we reject any reshape of a 'far' mode array,
3970 * allow chunk size to change if new is generally acceptable,
3971 * allow raid_disks to increase, and allow
3972 * a switch between 'near' mode and 'offset' mode.
3973 */
3974 struct r10conf *conf = mddev->private;
3975 struct geom geo;
3976
3977 if (conf->geo.far_copies != 1 && !conf->geo.far_offset)
3978 return -EINVAL;
3979
3980 if (setup_geo(&geo, mddev, geo_start) != conf->copies)
3981 /* mustn't change number of copies */
3982 return -EINVAL;
3983 if (geo.far_copies > 1 && !geo.far_offset)
3984 /* Cannot switch to 'far' mode */
3985 return -EINVAL;
3986
3987 if (mddev->array_sectors & geo.chunk_mask)
3988 /* not factor of array size */
3989 return -EINVAL;
3990
NeilBrown3ea7daa2012-05-22 13:53:47 +10003991 if (!enough(conf, -1))
3992 return -EINVAL;
3993
3994 kfree(conf->mirrors_new);
3995 conf->mirrors_new = NULL;
3996 if (mddev->delta_disks > 0) {
3997 /* allocate new 'mirrors' list */
3998 conf->mirrors_new = kzalloc(
Jonathan Brassowdc280d982012-07-31 10:03:52 +10003999 sizeof(struct raid10_info)
NeilBrown3ea7daa2012-05-22 13:53:47 +10004000 *(mddev->raid_disks +
4001 mddev->delta_disks),
4002 GFP_KERNEL);
4003 if (!conf->mirrors_new)
4004 return -ENOMEM;
4005 }
4006 return 0;
4007}
4008
4009/*
4010 * Need to check if array has failed when deciding whether to:
4011 * - start an array
4012 * - remove non-faulty devices
4013 * - add a spare
4014 * - allow a reshape
4015 * This determination is simple when no reshape is happening.
4016 * However if there is a reshape, we need to carefully check
4017 * both the before and after sections.
4018 * This is because some failed devices may only affect one
4019 * of the two sections, and some non-in_sync devices may
4020 * be insync in the section most affected by failed devices.
4021 */
4022static int calc_degraded(struct r10conf *conf)
4023{
4024 int degraded, degraded2;
4025 int i;
4026
4027 rcu_read_lock();
4028 degraded = 0;
4029 /* 'prev' section first */
4030 for (i = 0; i < conf->prev.raid_disks; i++) {
4031 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
4032 if (!rdev || test_bit(Faulty, &rdev->flags))
4033 degraded++;
4034 else if (!test_bit(In_sync, &rdev->flags))
4035 /* When we can reduce the number of devices in
4036 * an array, this might not contribute to
4037 * 'degraded'. It does now.
4038 */
4039 degraded++;
4040 }
4041 rcu_read_unlock();
4042 if (conf->geo.raid_disks == conf->prev.raid_disks)
4043 return degraded;
4044 rcu_read_lock();
4045 degraded2 = 0;
4046 for (i = 0; i < conf->geo.raid_disks; i++) {
4047 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
4048 if (!rdev || test_bit(Faulty, &rdev->flags))
4049 degraded2++;
4050 else if (!test_bit(In_sync, &rdev->flags)) {
4051 /* If reshape is increasing the number of devices,
4052 * this section has already been recovered, so
4053 * it doesn't contribute to degraded.
4054 * else it does.
4055 */
4056 if (conf->geo.raid_disks <= conf->prev.raid_disks)
4057 degraded2++;
4058 }
4059 }
4060 rcu_read_unlock();
4061 if (degraded2 > degraded)
4062 return degraded2;
4063 return degraded;
4064}
4065
4066static int raid10_start_reshape(struct mddev *mddev)
4067{
4068 /* A 'reshape' has been requested. This commits
4069 * the various 'new' fields and sets MD_RECOVER_RESHAPE
4070 * This also checks if there are enough spares and adds them
4071 * to the array.
4072 * We currently require enough spares to make the final
4073 * array non-degraded. We also require that the difference
4074 * between old and new data_offset - on each device - is
4075 * enough that we never risk over-writing.
4076 */
4077
4078 unsigned long before_length, after_length;
4079 sector_t min_offset_diff = 0;
4080 int first = 1;
4081 struct geom new;
4082 struct r10conf *conf = mddev->private;
4083 struct md_rdev *rdev;
4084 int spares = 0;
NeilBrownbb63a702012-05-22 13:55:28 +10004085 int ret;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004086
4087 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4088 return -EBUSY;
4089
4090 if (setup_geo(&new, mddev, geo_start) != conf->copies)
4091 return -EINVAL;
4092
4093 before_length = ((1 << conf->prev.chunk_shift) *
4094 conf->prev.far_copies);
4095 after_length = ((1 << conf->geo.chunk_shift) *
4096 conf->geo.far_copies);
4097
4098 rdev_for_each(rdev, mddev) {
4099 if (!test_bit(In_sync, &rdev->flags)
4100 && !test_bit(Faulty, &rdev->flags))
4101 spares++;
4102 if (rdev->raid_disk >= 0) {
4103 long long diff = (rdev->new_data_offset
4104 - rdev->data_offset);
4105 if (!mddev->reshape_backwards)
4106 diff = -diff;
4107 if (diff < 0)
4108 diff = 0;
4109 if (first || diff < min_offset_diff)
4110 min_offset_diff = diff;
4111 }
Guoqing Jiang6f287ca2017-04-06 09:12:18 +08004112 first = 0;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004113 }
4114
4115 if (max(before_length, after_length) > min_offset_diff)
4116 return -EINVAL;
4117
4118 if (spares < mddev->delta_disks)
4119 return -EINVAL;
4120
4121 conf->offset_diff = min_offset_diff;
4122 spin_lock_irq(&conf->device_lock);
4123 if (conf->mirrors_new) {
4124 memcpy(conf->mirrors_new, conf->mirrors,
Jonathan Brassowdc280d982012-07-31 10:03:52 +10004125 sizeof(struct raid10_info)*conf->prev.raid_disks);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004126 smp_mb();
NeilBrownc4796e22014-08-23 20:19:26 +10004127 kfree(conf->mirrors_old);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004128 conf->mirrors_old = conf->mirrors;
4129 conf->mirrors = conf->mirrors_new;
4130 conf->mirrors_new = NULL;
4131 }
4132 setup_geo(&conf->geo, mddev, geo_start);
4133 smp_mb();
4134 if (mddev->reshape_backwards) {
4135 sector_t size = raid10_size(mddev, 0, 0);
4136 if (size < mddev->array_sectors) {
4137 spin_unlock_irq(&conf->device_lock);
NeilBrown08464e02016-11-02 14:16:50 +11004138 pr_warn("md/raid10:%s: array size must be reduce before number of disks\n",
4139 mdname(mddev));
NeilBrown3ea7daa2012-05-22 13:53:47 +10004140 return -EINVAL;
4141 }
4142 mddev->resync_max_sectors = size;
4143 conf->reshape_progress = size;
4144 } else
4145 conf->reshape_progress = 0;
NeilBrown299b0682015-07-06 17:37:49 +10004146 conf->reshape_safe = conf->reshape_progress;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004147 spin_unlock_irq(&conf->device_lock);
4148
NeilBrownbb63a702012-05-22 13:55:28 +10004149 if (mddev->delta_disks && mddev->bitmap) {
4150 ret = bitmap_resize(mddev->bitmap,
4151 raid10_size(mddev, 0,
4152 conf->geo.raid_disks),
4153 0, 0);
4154 if (ret)
4155 goto abort;
4156 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004157 if (mddev->delta_disks > 0) {
4158 rdev_for_each(rdev, mddev)
4159 if (rdev->raid_disk < 0 &&
4160 !test_bit(Faulty, &rdev->flags)) {
4161 if (raid10_add_disk(mddev, rdev) == 0) {
4162 if (rdev->raid_disk >=
4163 conf->prev.raid_disks)
4164 set_bit(In_sync, &rdev->flags);
4165 else
4166 rdev->recovery_offset = 0;
4167
4168 if (sysfs_link_rdev(mddev, rdev))
4169 /* Failure here is OK */;
4170 }
4171 } else if (rdev->raid_disk >= conf->prev.raid_disks
4172 && !test_bit(Faulty, &rdev->flags)) {
4173 /* This is a spare that was manually added */
4174 set_bit(In_sync, &rdev->flags);
4175 }
4176 }
4177 /* When a reshape changes the number of devices,
4178 * ->degraded is measured against the larger of the
4179 * pre and post numbers.
4180 */
4181 spin_lock_irq(&conf->device_lock);
4182 mddev->degraded = calc_degraded(conf);
4183 spin_unlock_irq(&conf->device_lock);
4184 mddev->raid_disks = conf->geo.raid_disks;
4185 mddev->reshape_position = conf->reshape_progress;
Shaohua Li29530792016-12-08 15:48:19 -08004186 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004187
4188 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4189 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
NeilBrownea358cd2015-06-12 20:05:04 +10004190 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004191 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
4192 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4193
4194 mddev->sync_thread = md_register_thread(md_do_sync, mddev,
4195 "reshape");
4196 if (!mddev->sync_thread) {
NeilBrownbb63a702012-05-22 13:55:28 +10004197 ret = -EAGAIN;
4198 goto abort;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004199 }
4200 conf->reshape_checkpoint = jiffies;
4201 md_wakeup_thread(mddev->sync_thread);
4202 md_new_event(mddev);
4203 return 0;
NeilBrownbb63a702012-05-22 13:55:28 +10004204
4205abort:
4206 mddev->recovery = 0;
4207 spin_lock_irq(&conf->device_lock);
4208 conf->geo = conf->prev;
4209 mddev->raid_disks = conf->geo.raid_disks;
4210 rdev_for_each(rdev, mddev)
4211 rdev->new_data_offset = rdev->data_offset;
4212 smp_wmb();
4213 conf->reshape_progress = MaxSector;
NeilBrown299b0682015-07-06 17:37:49 +10004214 conf->reshape_safe = MaxSector;
NeilBrownbb63a702012-05-22 13:55:28 +10004215 mddev->reshape_position = MaxSector;
4216 spin_unlock_irq(&conf->device_lock);
4217 return ret;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004218}
4219
4220/* Calculate the last device-address that could contain
4221 * any block from the chunk that includes the array-address 's'
4222 * and report the next address.
4223 * i.e. the address returned will be chunk-aligned and after
4224 * any data that is in the chunk containing 's'.
4225 */
4226static sector_t last_dev_address(sector_t s, struct geom *geo)
4227{
4228 s = (s | geo->chunk_mask) + 1;
4229 s >>= geo->chunk_shift;
4230 s *= geo->near_copies;
4231 s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks);
4232 s *= geo->far_copies;
4233 s <<= geo->chunk_shift;
4234 return s;
4235}
4236
4237/* Calculate the first device-address that could contain
4238 * any block from the chunk that includes the array-address 's'.
4239 * This too will be the start of a chunk
4240 */
4241static sector_t first_dev_address(sector_t s, struct geom *geo)
4242{
4243 s >>= geo->chunk_shift;
4244 s *= geo->near_copies;
4245 sector_div(s, geo->raid_disks);
4246 s *= geo->far_copies;
4247 s <<= geo->chunk_shift;
4248 return s;
4249}
4250
4251static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
4252 int *skipped)
4253{
4254 /* We simply copy at most one chunk (smallest of old and new)
4255 * at a time, possibly less if that exceeds RESYNC_PAGES,
4256 * or we hit a bad block or something.
4257 * This might mean we pause for normal IO in the middle of
NeilBrown02ec5022015-07-06 16:33:47 +10004258 * a chunk, but that is not a problem as mddev->reshape_position
NeilBrown3ea7daa2012-05-22 13:53:47 +10004259 * can record any location.
4260 *
4261 * If we will want to write to a location that isn't
4262 * yet recorded as 'safe' (i.e. in metadata on disk) then
4263 * we need to flush all reshape requests and update the metadata.
4264 *
4265 * When reshaping forwards (e.g. to more devices), we interpret
4266 * 'safe' as the earliest block which might not have been copied
4267 * down yet. We divide this by previous stripe size and multiply
4268 * by previous stripe length to get lowest device offset that we
4269 * cannot write to yet.
4270 * We interpret 'sector_nr' as an address that we want to write to.
4271 * From this we use last_device_address() to find where we might
4272 * write to, and first_device_address on the 'safe' position.
4273 * If this 'next' write position is after the 'safe' position,
4274 * we must update the metadata to increase the 'safe' position.
4275 *
4276 * When reshaping backwards, we round in the opposite direction
4277 * and perform the reverse test: next write position must not be
4278 * less than current safe position.
4279 *
4280 * In all this the minimum difference in data offsets
4281 * (conf->offset_diff - always positive) allows a bit of slack,
NeilBrown02ec5022015-07-06 16:33:47 +10004282 * so next can be after 'safe', but not by more than offset_diff
NeilBrown3ea7daa2012-05-22 13:53:47 +10004283 *
4284 * We need to prepare all the bios here before we start any IO
4285 * to ensure the size we choose is acceptable to all devices.
4286 * The means one for each copy for write-out and an extra one for
4287 * read-in.
4288 * We store the read-in bio in ->master_bio and the others in
4289 * ->devs[x].bio and ->devs[x].repl_bio.
4290 */
4291 struct r10conf *conf = mddev->private;
4292 struct r10bio *r10_bio;
4293 sector_t next, safe, last;
4294 int max_sectors;
4295 int nr_sectors;
4296 int s;
4297 struct md_rdev *rdev;
4298 int need_flush = 0;
4299 struct bio *blist;
4300 struct bio *bio, *read_bio;
4301 int sectors_done = 0;
Ming Leif0250612017-03-17 00:12:33 +08004302 struct page **pages;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004303
4304 if (sector_nr == 0) {
4305 /* If restarting in the middle, skip the initial sectors */
4306 if (mddev->reshape_backwards &&
4307 conf->reshape_progress < raid10_size(mddev, 0, 0)) {
4308 sector_nr = (raid10_size(mddev, 0, 0)
4309 - conf->reshape_progress);
4310 } else if (!mddev->reshape_backwards &&
4311 conf->reshape_progress > 0)
4312 sector_nr = conf->reshape_progress;
4313 if (sector_nr) {
4314 mddev->curr_resync_completed = sector_nr;
4315 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
4316 *skipped = 1;
4317 return sector_nr;
4318 }
4319 }
4320
4321 /* We don't use sector_nr to track where we are up to
4322 * as that doesn't work well for ->reshape_backwards.
4323 * So just use ->reshape_progress.
4324 */
4325 if (mddev->reshape_backwards) {
4326 /* 'next' is the earliest device address that we might
4327 * write to for this chunk in the new layout
4328 */
4329 next = first_dev_address(conf->reshape_progress - 1,
4330 &conf->geo);
4331
4332 /* 'safe' is the last device address that we might read from
4333 * in the old layout after a restart
4334 */
4335 safe = last_dev_address(conf->reshape_safe - 1,
4336 &conf->prev);
4337
4338 if (next + conf->offset_diff < safe)
4339 need_flush = 1;
4340
4341 last = conf->reshape_progress - 1;
4342 sector_nr = last & ~(sector_t)(conf->geo.chunk_mask
4343 & conf->prev.chunk_mask);
4344 if (sector_nr + RESYNC_BLOCK_SIZE/512 < last)
4345 sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512;
4346 } else {
4347 /* 'next' is after the last device address that we
4348 * might write to for this chunk in the new layout
4349 */
4350 next = last_dev_address(conf->reshape_progress, &conf->geo);
4351
4352 /* 'safe' is the earliest device address that we might
4353 * read from in the old layout after a restart
4354 */
4355 safe = first_dev_address(conf->reshape_safe, &conf->prev);
4356
4357 /* Need to update metadata if 'next' might be beyond 'safe'
4358 * as that would possibly corrupt data
4359 */
4360 if (next > safe + conf->offset_diff)
4361 need_flush = 1;
4362
4363 sector_nr = conf->reshape_progress;
4364 last = sector_nr | (conf->geo.chunk_mask
4365 & conf->prev.chunk_mask);
4366
4367 if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last)
4368 last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1;
4369 }
4370
4371 if (need_flush ||
4372 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
4373 /* Need to update reshape_position in metadata */
4374 wait_barrier(conf);
4375 mddev->reshape_position = conf->reshape_progress;
4376 if (mddev->reshape_backwards)
4377 mddev->curr_resync_completed = raid10_size(mddev, 0, 0)
4378 - conf->reshape_progress;
4379 else
4380 mddev->curr_resync_completed = conf->reshape_progress;
4381 conf->reshape_checkpoint = jiffies;
Shaohua Li29530792016-12-08 15:48:19 -08004382 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004383 md_wakeup_thread(mddev->thread);
Shaohua Li29530792016-12-08 15:48:19 -08004384 wait_event(mddev->sb_wait, mddev->sb_flags == 0 ||
NeilBrownc91abf52013-11-19 12:02:01 +11004385 test_bit(MD_RECOVERY_INTR, &mddev->recovery));
4386 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4387 allow_barrier(conf);
4388 return sectors_done;
4389 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004390 conf->reshape_safe = mddev->reshape_position;
4391 allow_barrier(conf);
4392 }
4393
4394read_more:
4395 /* Now schedule reads for blocks from sector_nr to last */
4396 r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO);
NeilBrowncb8b12b2014-08-18 14:38:45 +10004397 r10_bio->state = 0;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004398 raise_barrier(conf, sectors_done != 0);
4399 atomic_set(&r10_bio->remaining, 0);
4400 r10_bio->mddev = mddev;
4401 r10_bio->sector = sector_nr;
4402 set_bit(R10BIO_IsReshape, &r10_bio->state);
4403 r10_bio->sectors = last - sector_nr + 1;
4404 rdev = read_balance(conf, r10_bio, &max_sectors);
4405 BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state));
4406
4407 if (!rdev) {
4408 /* Cannot read from here, so need to record bad blocks
4409 * on all the target devices.
4410 */
4411 // FIXME
NeilBrowne337aea2014-08-18 14:48:54 +10004412 mempool_free(r10_bio, conf->r10buf_pool);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004413 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4414 return sectors_done;
4415 }
4416
4417 read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev);
4418
4419 read_bio->bi_bdev = rdev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004420 read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr
NeilBrown3ea7daa2012-05-22 13:53:47 +10004421 + rdev->data_offset);
4422 read_bio->bi_private = r10_bio;
Ming Lei81fa1522017-03-17 00:12:32 +08004423 read_bio->bi_end_io = end_reshape_read;
Mike Christie796a5cf2016-06-05 14:32:07 -05004424 bio_set_op_attrs(read_bio, REQ_OP_READ, 0);
NeilBrownce0b0a42014-08-18 13:56:38 +10004425 read_bio->bi_flags &= (~0UL << BIO_RESET_BITS);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02004426 read_bio->bi_error = 0;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004427 read_bio->bi_vcnt = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004428 read_bio->bi_iter.bi_size = 0;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004429 r10_bio->master_bio = read_bio;
4430 r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum;
4431
4432 /* Now find the locations in the new layout */
4433 __raid10_find_phys(&conf->geo, r10_bio);
4434
4435 blist = read_bio;
4436 read_bio->bi_next = NULL;
4437
NeilBrownd094d682016-06-02 16:19:52 +10004438 rcu_read_lock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004439 for (s = 0; s < conf->copies*2; s++) {
4440 struct bio *b;
4441 int d = r10_bio->devs[s/2].devnum;
4442 struct md_rdev *rdev2;
4443 if (s&1) {
NeilBrownd094d682016-06-02 16:19:52 +10004444 rdev2 = rcu_dereference(conf->mirrors[d].replacement);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004445 b = r10_bio->devs[s/2].repl_bio;
4446 } else {
NeilBrownd094d682016-06-02 16:19:52 +10004447 rdev2 = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004448 b = r10_bio->devs[s/2].bio;
4449 }
4450 if (!rdev2 || test_bit(Faulty, &rdev2->flags))
4451 continue;
Kent Overstreet8be185f2012-09-06 14:14:43 -07004452
NeilBrown3ea7daa2012-05-22 13:53:47 +10004453 b->bi_bdev = rdev2->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -07004454 b->bi_iter.bi_sector = r10_bio->devs[s/2].addr +
4455 rdev2->new_data_offset;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004456 b->bi_end_io = end_reshape_write;
Mike Christie796a5cf2016-06-05 14:32:07 -05004457 bio_set_op_attrs(b, REQ_OP_WRITE, 0);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004458 b->bi_next = blist;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004459 blist = b;
4460 }
4461
4462 /* Now add as many pages as possible to all of these bios. */
4463
4464 nr_sectors = 0;
Ming Leif0250612017-03-17 00:12:33 +08004465 pages = get_resync_pages(r10_bio->devs[0].bio)->pages;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004466 for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) {
Ming Leif0250612017-03-17 00:12:33 +08004467 struct page *page = pages[s / (PAGE_SIZE >> 9)];
NeilBrown3ea7daa2012-05-22 13:53:47 +10004468 int len = (max_sectors - s) << 9;
4469 if (len > PAGE_SIZE)
4470 len = PAGE_SIZE;
4471 for (bio = blist; bio ; bio = bio->bi_next) {
Ming Leic85ba142017-03-17 00:12:22 +08004472 /*
4473 * won't fail because the vec table is big enough
4474 * to hold all these pages
4475 */
4476 bio_add_page(bio, page, len, 0);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004477 }
4478 sector_nr += len >> 9;
4479 nr_sectors += len >> 9;
4480 }
NeilBrownd094d682016-06-02 16:19:52 +10004481 rcu_read_unlock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004482 r10_bio->sectors = nr_sectors;
4483
4484 /* Now submit the read */
4485 md_sync_acct(read_bio->bi_bdev, r10_bio->sectors);
4486 atomic_inc(&r10_bio->remaining);
4487 read_bio->bi_next = NULL;
4488 generic_make_request(read_bio);
4489 sector_nr += nr_sectors;
4490 sectors_done += nr_sectors;
4491 if (sector_nr <= last)
4492 goto read_more;
4493
4494 /* Now that we have done the whole section we can
4495 * update reshape_progress
4496 */
4497 if (mddev->reshape_backwards)
4498 conf->reshape_progress -= sectors_done;
4499 else
4500 conf->reshape_progress += sectors_done;
4501
4502 return sectors_done;
4503}
4504
4505static void end_reshape_request(struct r10bio *r10_bio);
4506static int handle_reshape_read_error(struct mddev *mddev,
4507 struct r10bio *r10_bio);
4508static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio)
4509{
4510 /* Reshape read completed. Hopefully we have a block
4511 * to write out.
4512 * If we got a read error then we do sync 1-page reads from
4513 * elsewhere until we find the data - or give up.
4514 */
4515 struct r10conf *conf = mddev->private;
4516 int s;
4517
4518 if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
4519 if (handle_reshape_read_error(mddev, r10_bio) < 0) {
4520 /* Reshape has been aborted */
4521 md_done_sync(mddev, r10_bio->sectors, 0);
4522 return;
4523 }
4524
4525 /* We definitely have the data in the pages, schedule the
4526 * writes.
4527 */
4528 atomic_set(&r10_bio->remaining, 1);
4529 for (s = 0; s < conf->copies*2; s++) {
4530 struct bio *b;
4531 int d = r10_bio->devs[s/2].devnum;
4532 struct md_rdev *rdev;
NeilBrownd094d682016-06-02 16:19:52 +10004533 rcu_read_lock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004534 if (s&1) {
NeilBrownd094d682016-06-02 16:19:52 +10004535 rdev = rcu_dereference(conf->mirrors[d].replacement);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004536 b = r10_bio->devs[s/2].repl_bio;
4537 } else {
NeilBrownd094d682016-06-02 16:19:52 +10004538 rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004539 b = r10_bio->devs[s/2].bio;
4540 }
NeilBrownd094d682016-06-02 16:19:52 +10004541 if (!rdev || test_bit(Faulty, &rdev->flags)) {
4542 rcu_read_unlock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004543 continue;
NeilBrownd094d682016-06-02 16:19:52 +10004544 }
NeilBrown3ea7daa2012-05-22 13:53:47 +10004545 atomic_inc(&rdev->nr_pending);
NeilBrownd094d682016-06-02 16:19:52 +10004546 rcu_read_unlock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004547 md_sync_acct(b->bi_bdev, r10_bio->sectors);
4548 atomic_inc(&r10_bio->remaining);
4549 b->bi_next = NULL;
4550 generic_make_request(b);
4551 }
4552 end_reshape_request(r10_bio);
4553}
4554
4555static void end_reshape(struct r10conf *conf)
4556{
4557 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery))
4558 return;
4559
4560 spin_lock_irq(&conf->device_lock);
4561 conf->prev = conf->geo;
4562 md_finish_reshape(conf->mddev);
4563 smp_wmb();
4564 conf->reshape_progress = MaxSector;
NeilBrown299b0682015-07-06 17:37:49 +10004565 conf->reshape_safe = MaxSector;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004566 spin_unlock_irq(&conf->device_lock);
4567
4568 /* read-ahead size must cover two whole stripes, which is
4569 * 2 * (datadisks) * chunksize where 'n' is the number of raid devices
4570 */
4571 if (conf->mddev->queue) {
4572 int stripe = conf->geo.raid_disks *
4573 ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE);
4574 stripe /= conf->geo.near_copies;
Jan Karadc3b17c2017-02-02 15:56:50 +01004575 if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe)
4576 conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004577 }
4578 conf->fullsync = 0;
4579}
4580
NeilBrown3ea7daa2012-05-22 13:53:47 +10004581static int handle_reshape_read_error(struct mddev *mddev,
4582 struct r10bio *r10_bio)
4583{
4584 /* Use sync reads to get the blocks from somewhere else */
4585 int sectors = r10_bio->sectors;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004586 struct r10conf *conf = mddev->private;
NeilBrowne0ee7782012-08-18 09:51:42 +10004587 struct {
4588 struct r10bio r10_bio;
4589 struct r10dev devs[conf->copies];
4590 } on_stack;
4591 struct r10bio *r10b = &on_stack.r10_bio;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004592 int slot = 0;
4593 int idx = 0;
Ming Lei2d06e3b2017-03-17 00:12:35 +08004594 struct page **pages;
4595
4596 /* reshape IOs share pages from .devs[0].bio */
4597 pages = get_resync_pages(r10_bio->devs[0].bio)->pages;
NeilBrown3ea7daa2012-05-22 13:53:47 +10004598
NeilBrowne0ee7782012-08-18 09:51:42 +10004599 r10b->sector = r10_bio->sector;
4600 __raid10_find_phys(&conf->prev, r10b);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004601
4602 while (sectors) {
4603 int s = sectors;
4604 int success = 0;
4605 int first_slot = slot;
4606
4607 if (s > (PAGE_SIZE >> 9))
4608 s = PAGE_SIZE >> 9;
4609
NeilBrownd094d682016-06-02 16:19:52 +10004610 rcu_read_lock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004611 while (!success) {
NeilBrowne0ee7782012-08-18 09:51:42 +10004612 int d = r10b->devs[slot].devnum;
NeilBrownd094d682016-06-02 16:19:52 +10004613 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004614 sector_t addr;
4615 if (rdev == NULL ||
4616 test_bit(Faulty, &rdev->flags) ||
4617 !test_bit(In_sync, &rdev->flags))
4618 goto failed;
4619
NeilBrowne0ee7782012-08-18 09:51:42 +10004620 addr = r10b->devs[slot].addr + idx * PAGE_SIZE;
NeilBrownd094d682016-06-02 16:19:52 +10004621 atomic_inc(&rdev->nr_pending);
4622 rcu_read_unlock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004623 success = sync_page_io(rdev,
4624 addr,
4625 s << 9,
Ming Lei2d06e3b2017-03-17 00:12:35 +08004626 pages[idx],
Mike Christie796a5cf2016-06-05 14:32:07 -05004627 REQ_OP_READ, 0, false);
NeilBrownd094d682016-06-02 16:19:52 +10004628 rdev_dec_pending(rdev, mddev);
4629 rcu_read_lock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004630 if (success)
4631 break;
4632 failed:
4633 slot++;
4634 if (slot >= conf->copies)
4635 slot = 0;
4636 if (slot == first_slot)
4637 break;
4638 }
NeilBrownd094d682016-06-02 16:19:52 +10004639 rcu_read_unlock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004640 if (!success) {
4641 /* couldn't read this block, must give up */
4642 set_bit(MD_RECOVERY_INTR,
4643 &mddev->recovery);
4644 return -EIO;
4645 }
4646 sectors -= s;
4647 idx++;
4648 }
4649 return 0;
4650}
4651
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02004652static void end_reshape_write(struct bio *bio)
NeilBrown3ea7daa2012-05-22 13:53:47 +10004653{
Ming Leif0250612017-03-17 00:12:33 +08004654 struct r10bio *r10_bio = get_resync_r10bio(bio);
NeilBrown3ea7daa2012-05-22 13:53:47 +10004655 struct mddev *mddev = r10_bio->mddev;
4656 struct r10conf *conf = mddev->private;
4657 int d;
4658 int slot;
4659 int repl;
4660 struct md_rdev *rdev = NULL;
4661
4662 d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
4663 if (repl)
4664 rdev = conf->mirrors[d].replacement;
4665 if (!rdev) {
4666 smp_mb();
4667 rdev = conf->mirrors[d].rdev;
4668 }
4669
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02004670 if (bio->bi_error) {
NeilBrown3ea7daa2012-05-22 13:53:47 +10004671 /* FIXME should record badblock */
4672 md_error(mddev, rdev);
4673 }
4674
4675 rdev_dec_pending(rdev, mddev);
4676 end_reshape_request(r10_bio);
4677}
4678
4679static void end_reshape_request(struct r10bio *r10_bio)
4680{
4681 if (!atomic_dec_and_test(&r10_bio->remaining))
4682 return;
4683 md_done_sync(r10_bio->mddev, r10_bio->sectors, 1);
4684 bio_put(r10_bio->master_bio);
4685 put_buf(r10_bio);
4686}
4687
4688static void raid10_finish_reshape(struct mddev *mddev)
4689{
4690 struct r10conf *conf = mddev->private;
4691
4692 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
4693 return;
4694
4695 if (mddev->delta_disks > 0) {
4696 sector_t size = raid10_size(mddev, 0, 0);
4697 md_set_array_sectors(mddev, size);
4698 if (mddev->recovery_cp > mddev->resync_max_sectors) {
4699 mddev->recovery_cp = mddev->resync_max_sectors;
4700 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4701 }
4702 mddev->resync_max_sectors = size;
Heinz Mauelshagen859644f2016-05-03 19:43:24 +02004703 if (mddev->queue) {
4704 set_capacity(mddev->gendisk, mddev->array_sectors);
4705 revalidate_disk(mddev->gendisk);
4706 }
NeilBrown63aced62012-05-22 13:55:33 +10004707 } else {
4708 int d;
NeilBrownd094d682016-06-02 16:19:52 +10004709 rcu_read_lock();
NeilBrown63aced62012-05-22 13:55:33 +10004710 for (d = conf->geo.raid_disks ;
4711 d < conf->geo.raid_disks - mddev->delta_disks;
4712 d++) {
NeilBrownd094d682016-06-02 16:19:52 +10004713 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev);
NeilBrown63aced62012-05-22 13:55:33 +10004714 if (rdev)
4715 clear_bit(In_sync, &rdev->flags);
NeilBrownd094d682016-06-02 16:19:52 +10004716 rdev = rcu_dereference(conf->mirrors[d].replacement);
NeilBrown63aced62012-05-22 13:55:33 +10004717 if (rdev)
4718 clear_bit(In_sync, &rdev->flags);
4719 }
NeilBrownd094d682016-06-02 16:19:52 +10004720 rcu_read_unlock();
NeilBrown3ea7daa2012-05-22 13:53:47 +10004721 }
4722 mddev->layout = mddev->new_layout;
4723 mddev->chunk_sectors = 1 << conf->geo.chunk_shift;
4724 mddev->reshape_position = MaxSector;
4725 mddev->delta_disks = 0;
4726 mddev->reshape_backwards = 0;
4727}
4728
NeilBrown84fc4b52011-10-11 16:49:58 +11004729static struct md_personality raid10_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07004730{
4731 .name = "raid10",
NeilBrown2604b702006-01-06 00:20:36 -08004732 .level = 10,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004733 .owner = THIS_MODULE,
Shaohua Li849674e2016-01-20 13:52:20 -08004734 .make_request = raid10_make_request,
4735 .run = raid10_run,
NeilBrownafa0f552014-12-15 12:56:58 +11004736 .free = raid10_free,
Shaohua Li849674e2016-01-20 13:52:20 -08004737 .status = raid10_status,
4738 .error_handler = raid10_error,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004739 .hot_add_disk = raid10_add_disk,
4740 .hot_remove_disk= raid10_remove_disk,
4741 .spare_active = raid10_spare_active,
Shaohua Li849674e2016-01-20 13:52:20 -08004742 .sync_request = raid10_sync_request,
NeilBrown6cce3b22006-01-06 00:20:16 -08004743 .quiesce = raid10_quiesce,
Dan Williams80c3a6c2009-03-17 18:10:40 -07004744 .size = raid10_size,
NeilBrown006a09a2012-03-19 12:46:40 +11004745 .resize = raid10_resize,
Trela, Maciejdab8b292010-03-08 16:02:45 +11004746 .takeover = raid10_takeover,
NeilBrown3ea7daa2012-05-22 13:53:47 +10004747 .check_reshape = raid10_check_reshape,
4748 .start_reshape = raid10_start_reshape,
4749 .finish_reshape = raid10_finish_reshape,
NeilBrown5c675f82014-12-15 12:56:56 +11004750 .congested = raid10_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004751};
4752
4753static int __init raid_init(void)
4754{
NeilBrown2604b702006-01-06 00:20:36 -08004755 return register_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004756}
4757
4758static void raid_exit(void)
4759{
NeilBrown2604b702006-01-06 00:20:36 -08004760 unregister_md_personality(&raid10_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004761}
4762
4763module_init(raid_init);
4764module_exit(raid_exit);
4765MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11004766MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004767MODULE_ALIAS("md-personality-9"); /* RAID10 */
NeilBrownd9d166c2006-01-06 00:20:51 -08004768MODULE_ALIAS("md-raid10");
NeilBrown2604b702006-01-06 00:20:36 -08004769MODULE_ALIAS("md-level-10");
NeilBrown34db0cd2011-10-11 16:50:01 +11004770
4771module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);