Merge tag 'v3.10.107' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / raid1.c
CommitLineData
1da177e4
LT
1/*
2 * raid1.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
5 *
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
7 *
8 * RAID-1 management functions.
9 *
10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
11 *
96de0e25 12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
1da177e4
LT
13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
14 *
191ea9b2
N
15 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
16 * bitmapped intelligence in resync:
17 *
18 * - bitmap marked during normal i/o
19 * - bitmap used to skip nondirty blocks during sync
20 *
21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
22 * - persistent bitmap code
23 *
1da177e4
LT
24 * This program is free software; you can redistribute it and/or modify
25 * it under the terms of the GNU General Public License as published by
26 * the Free Software Foundation; either version 2, or (at your option)
27 * any later version.
28 *
29 * You should have received a copy of the GNU General Public License
30 * (for example /usr/src/linux/COPYING); if not, write to the Free
31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
32 */
33
5a0e3ad6 34#include <linux/slab.h>
25570727 35#include <linux/delay.h>
bff61975 36#include <linux/blkdev.h>
056075c7 37#include <linux/module.h>
bff61975 38#include <linux/seq_file.h>
8bda470e 39#include <linux/ratelimit.h>
43b2e5d8 40#include "md.h"
ef740c37
CH
41#include "raid1.h"
42#include "bitmap.h"
191ea9b2 43
1da177e4
LT
44/*
45 * Number of guaranteed r1bios in case of extreme VM load:
46 */
47#define NR_RAID1_BIOS 256
48
473e87ce
JB
49/* when we get a read error on a read-only array, we redirect to another
50 * device without failing the first device, or trying to over-write to
51 * correct the read error. To keep track of bad blocks on a per-bio
52 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
53 */
54#define IO_BLOCKED ((struct bio *)1)
55/* When we successfully write to a known bad-block, we need to remove the
56 * bad-block marking which must be done from process context. So we record
57 * the success by setting devs[n].bio to IO_MADE_GOOD
58 */
59#define IO_MADE_GOOD ((struct bio *)2)
60
61#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
62
34db0cd6
N
63/* When there are this many requests queue to be written by
64 * the raid1 thread, we become 'congested' to provide back-pressure
65 * for writeback.
66 */
67static int max_queued_requests = 1024;
1da177e4 68
e8096360
N
69static void allow_barrier(struct r1conf *conf);
70static void lower_barrier(struct r1conf *conf);
1da177e4 71
dd0fc66f 72static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
1da177e4
LT
73{
74 struct pool_info *pi = data;
9f2c9d12 75 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
1da177e4
LT
76
77 /* allocate a r1bio with room for raid_disks entries in the bios array */
7eaceacc 78 return kzalloc(size, gfp_flags);
1da177e4
LT
79}
80
81static void r1bio_pool_free(void *r1_bio, void *data)
82{
83 kfree(r1_bio);
84}
85
86#define RESYNC_BLOCK_SIZE (64*1024)
87//#define RESYNC_BLOCK_SIZE PAGE_SIZE
88#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
89#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
90#define RESYNC_WINDOW (2048*1024)
91
dd0fc66f 92static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
1da177e4
LT
93{
94 struct pool_info *pi = data;
9f2c9d12 95 struct r1bio *r1_bio;
1da177e4 96 struct bio *bio;
fdc9aa48 97 int need_pages;
1da177e4
LT
98 int i, j;
99
100 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
7eaceacc 101 if (!r1_bio)
1da177e4 102 return NULL;
1da177e4
LT
103
104 /*
105 * Allocate bios : 1 for reading, n-1 for writing
106 */
107 for (j = pi->raid_disks ; j-- ; ) {
6746557f 108 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
1da177e4
LT
109 if (!bio)
110 goto out_free_bio;
111 r1_bio->bios[j] = bio;
112 }
113 /*
114 * Allocate RESYNC_PAGES data pages and attach them to
d11c171e
N
115 * the first bio.
116 * If this is a user-requested check/repair, allocate
117 * RESYNC_PAGES for each bio.
1da177e4 118 */
d11c171e 119 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
fdc9aa48 120 need_pages = pi->raid_disks;
d11c171e 121 else
fdc9aa48
N
122 need_pages = 1;
123 for (j = 0; j < need_pages; j++) {
d11c171e 124 bio = r1_bio->bios[j];
a0787606 125 bio->bi_vcnt = RESYNC_PAGES;
d11c171e 126
a0787606 127 if (bio_alloc_pages(bio, gfp_flags))
fdc9aa48 128 goto out_free_pages;
d11c171e
N
129 }
130 /* If not user-requests, copy the page pointers to all bios */
131 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
132 for (i=0; i<RESYNC_PAGES ; i++)
133 for (j=1; j<pi->raid_disks; j++)
134 r1_bio->bios[j]->bi_io_vec[i].bv_page =
135 r1_bio->bios[0]->bi_io_vec[i].bv_page;
1da177e4
LT
136 }
137
138 r1_bio->master_bio = NULL;
139
140 return r1_bio;
141
fdc9aa48
N
142out_free_pages:
143 while (--j >= 0) {
144 struct bio_vec *bv;
145
146 bio_for_each_segment_all(bv, r1_bio->bios[j], i)
147 __free_page(bv->bv_page);
148 }
149
1da177e4 150out_free_bio:
8f19ccb2 151 while (++j < pi->raid_disks)
1da177e4
LT
152 bio_put(r1_bio->bios[j]);
153 r1bio_pool_free(r1_bio, data);
154 return NULL;
155}
156
157static void r1buf_pool_free(void *__r1_bio, void *data)
158{
159 struct pool_info *pi = data;
d11c171e 160 int i,j;
9f2c9d12 161 struct r1bio *r1bio = __r1_bio;
1da177e4 162
d11c171e
N
163 for (i = 0; i < RESYNC_PAGES; i++)
164 for (j = pi->raid_disks; j-- ;) {
165 if (j == 0 ||
166 r1bio->bios[j]->bi_io_vec[i].bv_page !=
167 r1bio->bios[0]->bi_io_vec[i].bv_page)
1345b1d8 168 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
d11c171e 169 }
1da177e4
LT
170 for (i=0 ; i < pi->raid_disks; i++)
171 bio_put(r1bio->bios[i]);
172
173 r1bio_pool_free(r1bio, data);
174}
175
e8096360 176static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
1da177e4
LT
177{
178 int i;
179
8f19ccb2 180 for (i = 0; i < conf->raid_disks * 2; i++) {
1da177e4 181 struct bio **bio = r1_bio->bios + i;
4367af55 182 if (!BIO_SPECIAL(*bio))
1da177e4
LT
183 bio_put(*bio);
184 *bio = NULL;
185 }
186}
187
9f2c9d12 188static void free_r1bio(struct r1bio *r1_bio)
1da177e4 189{
e8096360 190 struct r1conf *conf = r1_bio->mddev->private;
1da177e4 191
1da177e4
LT
192 put_all_bios(conf, r1_bio);
193 mempool_free(r1_bio, conf->r1bio_pool);
194}
195
9f2c9d12 196static void put_buf(struct r1bio *r1_bio)
1da177e4 197{
e8096360 198 struct r1conf *conf = r1_bio->mddev->private;
3e198f78
N
199 int i;
200
8f19ccb2 201 for (i = 0; i < conf->raid_disks * 2; i++) {
3e198f78
N
202 struct bio *bio = r1_bio->bios[i];
203 if (bio->bi_end_io)
204 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
205 }
1da177e4
LT
206
207 mempool_free(r1_bio, conf->r1buf_pool);
208
17999be4 209 lower_barrier(conf);
1da177e4
LT
210}
211
9f2c9d12 212static void reschedule_retry(struct r1bio *r1_bio)
1da177e4
LT
213{
214 unsigned long flags;
fd01b88c 215 struct mddev *mddev = r1_bio->mddev;
e8096360 216 struct r1conf *conf = mddev->private;
1da177e4
LT
217
218 spin_lock_irqsave(&conf->device_lock, flags);
219 list_add(&r1_bio->retry_list, &conf->retry_list);
ddaf22ab 220 conf->nr_queued ++;
1da177e4
LT
221 spin_unlock_irqrestore(&conf->device_lock, flags);
222
17999be4 223 wake_up(&conf->wait_barrier);
1da177e4
LT
224 md_wakeup_thread(mddev->thread);
225}
226
227/*
228 * raid_end_bio_io() is called when we have finished servicing a mirrored
229 * operation and are ready to return a success/failure code to the buffer
230 * cache layer.
231 */
9f2c9d12 232static void call_bio_endio(struct r1bio *r1_bio)
d2eb35ac
N
233{
234 struct bio *bio = r1_bio->master_bio;
235 int done;
e8096360 236 struct r1conf *conf = r1_bio->mddev->private;
d2eb35ac
N
237
238 if (bio->bi_phys_segments) {
239 unsigned long flags;
240 spin_lock_irqsave(&conf->device_lock, flags);
241 bio->bi_phys_segments--;
242 done = (bio->bi_phys_segments == 0);
243 spin_unlock_irqrestore(&conf->device_lock, flags);
244 } else
245 done = 1;
246
247 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
248 clear_bit(BIO_UPTODATE, &bio->bi_flags);
249 if (done) {
250 bio_endio(bio, 0);
251 /*
252 * Wake up any possible resync thread that waits for the device
253 * to go idle.
254 */
255 allow_barrier(conf);
256 }
257}
258
9f2c9d12 259static void raid_end_bio_io(struct r1bio *r1_bio)
1da177e4
LT
260{
261 struct bio *bio = r1_bio->master_bio;
262
4b6d287f
N
263 /* if nobody has done the final endio yet, do it now */
264 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
36a4e1fe
N
265 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
266 (bio_data_dir(bio) == WRITE) ? "write" : "read",
267 (unsigned long long) bio->bi_sector,
268 (unsigned long long) bio->bi_sector +
aa8b57aa 269 bio_sectors(bio) - 1);
4b6d287f 270
d2eb35ac 271 call_bio_endio(r1_bio);
4b6d287f 272 }
1da177e4
LT
273 free_r1bio(r1_bio);
274}
275
276/*
277 * Update disk head position estimator based on IRQ completion info.
278 */
9f2c9d12 279static inline void update_head_pos(int disk, struct r1bio *r1_bio)
1da177e4 280{
e8096360 281 struct r1conf *conf = r1_bio->mddev->private;
1da177e4
LT
282
283 conf->mirrors[disk].head_position =
284 r1_bio->sector + (r1_bio->sectors);
285}
286
ba3ae3be
NK
287/*
288 * Find the disk number which triggered given bio
289 */
9f2c9d12 290static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
ba3ae3be
NK
291{
292 int mirror;
30194636
N
293 struct r1conf *conf = r1_bio->mddev->private;
294 int raid_disks = conf->raid_disks;
ba3ae3be 295
8f19ccb2 296 for (mirror = 0; mirror < raid_disks * 2; mirror++)
ba3ae3be
NK
297 if (r1_bio->bios[mirror] == bio)
298 break;
299
8f19ccb2 300 BUG_ON(mirror == raid_disks * 2);
ba3ae3be
NK
301 update_head_pos(mirror, r1_bio);
302
303 return mirror;
304}
305
6712ecf8 306static void raid1_end_read_request(struct bio *bio, int error)
1da177e4
LT
307{
308 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
9f2c9d12 309 struct r1bio *r1_bio = bio->bi_private;
1da177e4 310 int mirror;
e8096360 311 struct r1conf *conf = r1_bio->mddev->private;
1da177e4 312
1da177e4
LT
313 mirror = r1_bio->read_disk;
314 /*
315 * this branch is our 'one mirror IO has finished' event handler:
316 */
ddaf22ab
N
317 update_head_pos(mirror, r1_bio);
318
dd00a99e
N
319 if (uptodate)
320 set_bit(R1BIO_Uptodate, &r1_bio->state);
321 else {
322 /* If all other devices have failed, we want to return
323 * the error upwards rather than fail the last device.
324 * Here we redefine "uptodate" to mean "Don't want to retry"
1da177e4 325 */
dd00a99e
N
326 unsigned long flags;
327 spin_lock_irqsave(&conf->device_lock, flags);
328 if (r1_bio->mddev->degraded == conf->raid_disks ||
329 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
d8312285 330 test_bit(In_sync, &conf->mirrors[mirror].rdev->flags)))
dd00a99e
N
331 uptodate = 1;
332 spin_unlock_irqrestore(&conf->device_lock, flags);
333 }
1da177e4 334
7ad4d4a6 335 if (uptodate) {
1da177e4 336 raid_end_bio_io(r1_bio);
7ad4d4a6
N
337 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
338 } else {
1da177e4
LT
339 /*
340 * oops, read error:
341 */
342 char b[BDEVNAME_SIZE];
8bda470e
CD
343 printk_ratelimited(
344 KERN_ERR "md/raid1:%s: %s: "
345 "rescheduling sector %llu\n",
346 mdname(conf->mddev),
347 bdevname(conf->mirrors[mirror].rdev->bdev,
348 b),
349 (unsigned long long)r1_bio->sector);
d2eb35ac 350 set_bit(R1BIO_ReadError, &r1_bio->state);
1da177e4 351 reschedule_retry(r1_bio);
7ad4d4a6 352 /* don't drop the reference on read_disk yet */
1da177e4 353 }
1da177e4
LT
354}
355
9f2c9d12 356static void close_write(struct r1bio *r1_bio)
cd5ff9a1
N
357{
358 /* it really is the end of this request */
359 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
360 /* free extra copy of the data pages */
361 int i = r1_bio->behind_page_count;
362 while (i--)
363 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
364 kfree(r1_bio->behind_bvecs);
365 r1_bio->behind_bvecs = NULL;
366 }
367 /* clear the bitmap if all writes complete successfully */
368 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
369 r1_bio->sectors,
370 !test_bit(R1BIO_Degraded, &r1_bio->state),
371 test_bit(R1BIO_BehindIO, &r1_bio->state));
372 md_write_end(r1_bio->mddev);
373}
374
9f2c9d12 375static void r1_bio_write_done(struct r1bio *r1_bio)
4e78064f 376{
cd5ff9a1
N
377 if (!atomic_dec_and_test(&r1_bio->remaining))
378 return;
379
380 if (test_bit(R1BIO_WriteError, &r1_bio->state))
381 reschedule_retry(r1_bio);
382 else {
383 close_write(r1_bio);
4367af55
N
384 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
385 reschedule_retry(r1_bio);
386 else
387 raid_end_bio_io(r1_bio);
4e78064f
N
388 }
389}
390
6712ecf8 391static void raid1_end_write_request(struct bio *bio, int error)
1da177e4
LT
392{
393 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
9f2c9d12 394 struct r1bio *r1_bio = bio->bi_private;
a9701a30 395 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
e8096360 396 struct r1conf *conf = r1_bio->mddev->private;
04b857f7 397 struct bio *to_put = NULL;
1da177e4 398
ba3ae3be 399 mirror = find_bio_disk(r1_bio, bio);
1da177e4 400
e9c7469b
TH
401 /*
402 * 'one mirror IO has finished' event handler:
403 */
e9c7469b 404 if (!uptodate) {
cd5ff9a1
N
405 set_bit(WriteErrorSeen,
406 &conf->mirrors[mirror].rdev->flags);
19d67169
N
407 if (!test_and_set_bit(WantReplacement,
408 &conf->mirrors[mirror].rdev->flags))
409 set_bit(MD_RECOVERY_NEEDED, &
410 conf->mddev->recovery);
411
cd5ff9a1 412 set_bit(R1BIO_WriteError, &r1_bio->state);
4367af55 413 } else {
1da177e4 414 /*
e9c7469b
TH
415 * Set R1BIO_Uptodate in our master bio, so that we
416 * will return a good error code for to the higher
417 * levels even if IO on some other mirrored buffer
418 * fails.
419 *
420 * The 'master' represents the composite IO operation
421 * to user-side. So if something waits for IO, then it
422 * will wait for the 'master' bio.
1da177e4 423 */
4367af55
N
424 sector_t first_bad;
425 int bad_sectors;
426
cd5ff9a1
N
427 r1_bio->bios[mirror] = NULL;
428 to_put = bio;
3056e3ae
AL
429 /*
430 * Do not set R1BIO_Uptodate if the current device is
431 * rebuilding or Faulty. This is because we cannot use
432 * such device for properly reading the data back (we could
433 * potentially use it, if the current write would have felt
434 * before rdev->recovery_offset, but for simplicity we don't
435 * check this here.
436 */
437 if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) &&
438 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))
439 set_bit(R1BIO_Uptodate, &r1_bio->state);
e9c7469b 440
4367af55
N
441 /* Maybe we can clear some bad blocks. */
442 if (is_badblock(conf->mirrors[mirror].rdev,
443 r1_bio->sector, r1_bio->sectors,
444 &first_bad, &bad_sectors)) {
445 r1_bio->bios[mirror] = IO_MADE_GOOD;
446 set_bit(R1BIO_MadeGood, &r1_bio->state);
447 }
448 }
449
e9c7469b
TH
450 if (behind) {
451 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
452 atomic_dec(&r1_bio->behind_remaining);
453
454 /*
455 * In behind mode, we ACK the master bio once the I/O
456 * has safely reached all non-writemostly
457 * disks. Setting the Returned bit ensures that this
458 * gets done only once -- we don't ever want to return
459 * -EIO here, instead we'll wait
460 */
461 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
462 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
463 /* Maybe we can return now */
464 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
465 struct bio *mbio = r1_bio->master_bio;
36a4e1fe
N
466 pr_debug("raid1: behind end write sectors"
467 " %llu-%llu\n",
468 (unsigned long long) mbio->bi_sector,
469 (unsigned long long) mbio->bi_sector +
aa8b57aa 470 bio_sectors(mbio) - 1);
d2eb35ac 471 call_bio_endio(r1_bio);
4b6d287f
N
472 }
473 }
474 }
4367af55
N
475 if (r1_bio->bios[mirror] == NULL)
476 rdev_dec_pending(conf->mirrors[mirror].rdev,
477 conf->mddev);
e9c7469b 478
1da177e4 479 /*
1da177e4
LT
480 * Let's see if all mirrored write operations have finished
481 * already.
482 */
af6d7b76 483 r1_bio_write_done(r1_bio);
c70810b3 484
04b857f7
N
485 if (to_put)
486 bio_put(to_put);
1da177e4
LT
487}
488
489
490/*
491 * This routine returns the disk from which the requested read should
492 * be done. There is a per-array 'next expected sequential IO' sector
493 * number - if this matches on the next IO then we use the last disk.
494 * There is also a per-disk 'last know head position' sector that is
495 * maintained from IRQ contexts, both the normal and the resync IO
496 * completion handlers update this position correctly. If there is no
497 * perfect sequential match then we pick the disk whose head is closest.
498 *
499 * If there are 2 mirrors in the same 2 devices, performance degrades
500 * because position is mirror, not device based.
501 *
502 * The rdev for the device selected will have nr_pending incremented.
503 */
e8096360 504static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
1da177e4 505{
af3a2cd6 506 const sector_t this_sector = r1_bio->sector;
d2eb35ac
N
507 int sectors;
508 int best_good_sectors;
9dedf603
SL
509 int best_disk, best_dist_disk, best_pending_disk;
510 int has_nonrot_disk;
be4d3280 511 int disk;
76073054 512 sector_t best_dist;
9dedf603 513 unsigned int min_pending;
3cb03002 514 struct md_rdev *rdev;
f3ac8bf7 515 int choose_first;
12cee5a8 516 int choose_next_idle;
1da177e4
LT
517
518 rcu_read_lock();
519 /*
8ddf9efe 520 * Check if we can balance. We can balance on the whole
1da177e4
LT
521 * device if no resync is going on, or below the resync window.
522 * We take the first readable disk when above the resync window.
523 */
524 retry:
d2eb35ac 525 sectors = r1_bio->sectors;
76073054 526 best_disk = -1;
9dedf603 527 best_dist_disk = -1;
76073054 528 best_dist = MaxSector;
9dedf603
SL
529 best_pending_disk = -1;
530 min_pending = UINT_MAX;
d2eb35ac 531 best_good_sectors = 0;
9dedf603 532 has_nonrot_disk = 0;
12cee5a8 533 choose_next_idle = 0;
d2eb35ac 534
1da177e4 535 if (conf->mddev->recovery_cp < MaxSector &&
be4d3280 536 (this_sector + sectors >= conf->next_resync))
f3ac8bf7 537 choose_first = 1;
be4d3280 538 else
f3ac8bf7 539 choose_first = 0;
1da177e4 540
be4d3280 541 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
76073054 542 sector_t dist;
d2eb35ac
N
543 sector_t first_bad;
544 int bad_sectors;
9dedf603 545 unsigned int pending;
12cee5a8 546 bool nonrot;
d2eb35ac 547
f3ac8bf7
N
548 rdev = rcu_dereference(conf->mirrors[disk].rdev);
549 if (r1_bio->bios[disk] == IO_BLOCKED
550 || rdev == NULL
6b740b8d 551 || test_bit(Unmerged, &rdev->flags)
76073054 552 || test_bit(Faulty, &rdev->flags))
f3ac8bf7 553 continue;
76073054
N
554 if (!test_bit(In_sync, &rdev->flags) &&
555 rdev->recovery_offset < this_sector + sectors)
1da177e4 556 continue;
76073054
N
557 if (test_bit(WriteMostly, &rdev->flags)) {
558 /* Don't balance among write-mostly, just
559 * use the first as a last resort */
b581e762 560 if (best_dist_disk < 0) {
307729c8
N
561 if (is_badblock(rdev, this_sector, sectors,
562 &first_bad, &bad_sectors)) {
7a56cc09 563 if (first_bad <= this_sector)
307729c8
N
564 /* Cannot use this */
565 continue;
566 best_good_sectors = first_bad - this_sector;
567 } else
568 best_good_sectors = sectors;
b581e762
TH
569 best_dist_disk = disk;
570 best_pending_disk = disk;
307729c8 571 }
76073054
N
572 continue;
573 }
574 /* This is a reasonable device to use. It might
575 * even be best.
576 */
d2eb35ac
N
577 if (is_badblock(rdev, this_sector, sectors,
578 &first_bad, &bad_sectors)) {
579 if (best_dist < MaxSector)
580 /* already have a better device */
581 continue;
582 if (first_bad <= this_sector) {
583 /* cannot read here. If this is the 'primary'
584 * device, then we must not read beyond
585 * bad_sectors from another device..
586 */
587 bad_sectors -= (this_sector - first_bad);
588 if (choose_first && sectors > bad_sectors)
589 sectors = bad_sectors;
590 if (best_good_sectors > sectors)
591 best_good_sectors = sectors;
592
593 } else {
594 sector_t good_sectors = first_bad - this_sector;
595 if (good_sectors > best_good_sectors) {
596 best_good_sectors = good_sectors;
597 best_disk = disk;
598 }
599 if (choose_first)
600 break;
601 }
602 continue;
603 } else
604 best_good_sectors = sectors;
605
12cee5a8
SL
606 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
607 has_nonrot_disk |= nonrot;
9dedf603 608 pending = atomic_read(&rdev->nr_pending);
76073054 609 dist = abs(this_sector - conf->mirrors[disk].head_position);
12cee5a8 610 if (choose_first) {
76073054 611 best_disk = disk;
1da177e4
LT
612 break;
613 }
12cee5a8
SL
614 /* Don't change to another disk for sequential reads */
615 if (conf->mirrors[disk].next_seq_sect == this_sector
616 || dist == 0) {
617 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
618 struct raid1_info *mirror = &conf->mirrors[disk];
619
620 best_disk = disk;
621 /*
622 * If buffered sequential IO size exceeds optimal
623 * iosize, check if there is idle disk. If yes, choose
624 * the idle disk. read_balance could already choose an
625 * idle disk before noticing it's a sequential IO in
626 * this disk. This doesn't matter because this disk
627 * will idle, next time it will be utilized after the
628 * first disk has IO size exceeds optimal iosize. In
629 * this way, iosize of the first disk will be optimal
630 * iosize at least. iosize of the second disk might be
631 * small, but not a big deal since when the second disk
632 * starts IO, the first disk is likely still busy.
633 */
634 if (nonrot && opt_iosize > 0 &&
635 mirror->seq_start != MaxSector &&
636 mirror->next_seq_sect > opt_iosize &&
637 mirror->next_seq_sect - opt_iosize >=
638 mirror->seq_start) {
639 choose_next_idle = 1;
640 continue;
641 }
642 break;
643 }
644 /* If device is idle, use it */
645 if (pending == 0) {
646 best_disk = disk;
647 break;
648 }
649
650 if (choose_next_idle)
651 continue;
9dedf603
SL
652
653 if (min_pending > pending) {
654 min_pending = pending;
655 best_pending_disk = disk;
656 }
657
76073054
N
658 if (dist < best_dist) {
659 best_dist = dist;
9dedf603 660 best_dist_disk = disk;
1da177e4 661 }
f3ac8bf7 662 }
1da177e4 663
9dedf603
SL
664 /*
665 * If all disks are rotational, choose the closest disk. If any disk is
666 * non-rotational, choose the disk with less pending request even the
667 * disk is rotational, which might/might not be optimal for raids with
668 * mixed ratation/non-rotational disks depending on workload.
669 */
670 if (best_disk == -1) {
671 if (has_nonrot_disk)
672 best_disk = best_pending_disk;
673 else
674 best_disk = best_dist_disk;
675 }
676
76073054
N
677 if (best_disk >= 0) {
678 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
8ddf9efe
N
679 if (!rdev)
680 goto retry;
681 atomic_inc(&rdev->nr_pending);
76073054 682 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT
683 /* cannot risk returning a device that failed
684 * before we inc'ed nr_pending
685 */
03c902e1 686 rdev_dec_pending(rdev, conf->mddev);
1da177e4
LT
687 goto retry;
688 }
d2eb35ac 689 sectors = best_good_sectors;
12cee5a8
SL
690
691 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
692 conf->mirrors[best_disk].seq_start = this_sector;
693
be4d3280 694 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
1da177e4
LT
695 }
696 rcu_read_unlock();
d2eb35ac 697 *max_sectors = sectors;
1da177e4 698
76073054 699 return best_disk;
1da177e4
LT
700}
701
6b740b8d
N
702static int raid1_mergeable_bvec(struct request_queue *q,
703 struct bvec_merge_data *bvm,
704 struct bio_vec *biovec)
705{
706 struct mddev *mddev = q->queuedata;
707 struct r1conf *conf = mddev->private;
708 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
709 int max = biovec->bv_len;
710
711 if (mddev->merge_check_needed) {
712 int disk;
713 rcu_read_lock();
714 for (disk = 0; disk < conf->raid_disks * 2; disk++) {
715 struct md_rdev *rdev = rcu_dereference(
716 conf->mirrors[disk].rdev);
717 if (rdev && !test_bit(Faulty, &rdev->flags)) {
718 struct request_queue *q =
719 bdev_get_queue(rdev->bdev);
720 if (q->merge_bvec_fn) {
721 bvm->bi_sector = sector +
722 rdev->data_offset;
723 bvm->bi_bdev = rdev->bdev;
724 max = min(max, q->merge_bvec_fn(
725 q, bvm, biovec));
726 }
727 }
728 }
729 rcu_read_unlock();
730 }
731 return max;
732
733}
734
fd01b88c 735int md_raid1_congested(struct mddev *mddev, int bits)
0d129228 736{
e8096360 737 struct r1conf *conf = mddev->private;
0d129228
N
738 int i, ret = 0;
739
34db0cd6
N
740 if ((bits & (1 << BDI_async_congested)) &&
741 conf->pending_count >= max_queued_requests)
742 return 1;
743
0d129228 744 rcu_read_lock();
f53e29fc 745 for (i = 0; i < conf->raid_disks * 2; i++) {
3cb03002 746 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
0d129228 747 if (rdev && !test_bit(Faulty, &rdev->flags)) {
165125e1 748 struct request_queue *q = bdev_get_queue(rdev->bdev);
0d129228 749
1ed7242e
JB
750 BUG_ON(!q);
751
0d129228
N
752 /* Note the '|| 1' - when read_balance prefers
753 * non-congested targets, it can be removed
754 */
91a9e99d 755 if ((bits & (1<<BDI_async_congested)) || 1)
0d129228
N
756 ret |= bdi_congested(&q->backing_dev_info, bits);
757 else
758 ret &= bdi_congested(&q->backing_dev_info, bits);
759 }
760 }
761 rcu_read_unlock();
762 return ret;
763}
1ed7242e 764EXPORT_SYMBOL_GPL(md_raid1_congested);
0d129228 765
1ed7242e
JB
766static int raid1_congested(void *data, int bits)
767{
fd01b88c 768 struct mddev *mddev = data;
1ed7242e
JB
769
770 return mddev_congested(mddev, bits) ||
771 md_raid1_congested(mddev, bits);
772}
0d129228 773
e8096360 774static void flush_pending_writes(struct r1conf *conf)
a35e63ef
N
775{
776 /* Any writes that have been queued but are awaiting
777 * bitmap updates get flushed here.
a35e63ef 778 */
a35e63ef
N
779 spin_lock_irq(&conf->device_lock);
780
781 if (conf->pending_bio_list.head) {
782 struct bio *bio;
783 bio = bio_list_get(&conf->pending_bio_list);
34db0cd6 784 conf->pending_count = 0;
a35e63ef
N
785 spin_unlock_irq(&conf->device_lock);
786 /* flush any pending bitmap writes to
787 * disk before proceeding w/ I/O */
788 bitmap_unplug(conf->mddev->bitmap);
34db0cd6 789 wake_up(&conf->wait_barrier);
a35e63ef
N
790
791 while (bio) { /* submit pending writes */
792 struct bio *next = bio->bi_next;
793 bio->bi_next = NULL;
2ff8cc2c
SL
794 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
795 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
796 /* Just ignore it */
797 bio_endio(bio, 0);
798 else
799 generic_make_request(bio);
a35e63ef
N
800 bio = next;
801 }
a35e63ef
N
802 } else
803 spin_unlock_irq(&conf->device_lock);
7eaceacc
JA
804}
805
17999be4
N
806/* Barriers....
807 * Sometimes we need to suspend IO while we do something else,
808 * either some resync/recovery, or reconfigure the array.
809 * To do this we raise a 'barrier'.
810 * The 'barrier' is a counter that can be raised multiple times
811 * to count how many activities are happening which preclude
812 * normal IO.
813 * We can only raise the barrier if there is no pending IO.
814 * i.e. if nr_pending == 0.
815 * We choose only to raise the barrier if no-one is waiting for the
816 * barrier to go down. This means that as soon as an IO request
817 * is ready, no other operations which require a barrier will start
818 * until the IO request has had a chance.
819 *
820 * So: regular IO calls 'wait_barrier'. When that returns there
821 * is no backgroup IO happening, It must arrange to call
822 * allow_barrier when it has finished its IO.
823 * backgroup IO calls must call raise_barrier. Once that returns
824 * there is no normal IO happeing. It must arrange to call
825 * lower_barrier when the particular background IO completes.
1da177e4
LT
826 */
827#define RESYNC_DEPTH 32
828
e8096360 829static void raise_barrier(struct r1conf *conf)
1da177e4
LT
830{
831 spin_lock_irq(&conf->resync_lock);
17999be4
N
832
833 /* Wait until no block IO is waiting */
834 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
eed8c02e 835 conf->resync_lock);
17999be4
N
836
837 /* block any new IO from starting */
838 conf->barrier++;
839
046abeed 840 /* Now wait for all pending IO to complete */
17999be4
N
841 wait_event_lock_irq(conf->wait_barrier,
842 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
eed8c02e 843 conf->resync_lock);
17999be4
N
844
845 spin_unlock_irq(&conf->resync_lock);
846}
847
e8096360 848static void lower_barrier(struct r1conf *conf)
17999be4
N
849{
850 unsigned long flags;
709ae487 851 BUG_ON(conf->barrier <= 0);
17999be4
N
852 spin_lock_irqsave(&conf->resync_lock, flags);
853 conf->barrier--;
854 spin_unlock_irqrestore(&conf->resync_lock, flags);
855 wake_up(&conf->wait_barrier);
856}
857
e8096360 858static void wait_barrier(struct r1conf *conf)
17999be4
N
859{
860 spin_lock_irq(&conf->resync_lock);
861 if (conf->barrier) {
862 conf->nr_waiting++;
d6b42dcb
N
863 /* Wait for the barrier to drop.
864 * However if there are already pending
865 * requests (preventing the barrier from
866 * rising completely), and the
867 * pre-process bio queue isn't empty,
868 * then don't wait, as we need to empty
869 * that queue to get the nr_pending
870 * count down.
871 */
872 wait_event_lock_irq(conf->wait_barrier,
873 !conf->barrier ||
874 (conf->nr_pending &&
875 current->bio_list &&
876 !bio_list_empty(current->bio_list)),
eed8c02e 877 conf->resync_lock);
17999be4 878 conf->nr_waiting--;
1da177e4 879 }
17999be4 880 conf->nr_pending++;
1da177e4
LT
881 spin_unlock_irq(&conf->resync_lock);
882}
883
e8096360 884static void allow_barrier(struct r1conf *conf)
17999be4
N
885{
886 unsigned long flags;
887 spin_lock_irqsave(&conf->resync_lock, flags);
888 conf->nr_pending--;
889 spin_unlock_irqrestore(&conf->resync_lock, flags);
890 wake_up(&conf->wait_barrier);
891}
892
e2d59925 893static void freeze_array(struct r1conf *conf, int extra)
ddaf22ab
N
894{
895 /* stop syncio and normal IO and wait for everything to
896 * go quite.
897 * We increment barrier and nr_waiting, and then
e2d59925 898 * wait until nr_pending match nr_queued+extra
1c830532
N
899 * This is called in the context of one normal IO request
900 * that has failed. Thus any sync request that might be pending
901 * will be blocked by nr_pending, and we need to wait for
902 * pending IO requests to complete or be queued for re-try.
e2d59925 903 * Thus the number queued (nr_queued) plus this request (extra)
1c830532
N
904 * must match the number of pending IOs (nr_pending) before
905 * we continue.
ddaf22ab
N
906 */
907 spin_lock_irq(&conf->resync_lock);
908 conf->barrier++;
909 conf->nr_waiting++;
eed8c02e 910 wait_event_lock_irq_cmd(conf->wait_barrier,
e2d59925 911 conf->nr_pending == conf->nr_queued+extra,
eed8c02e
LC
912 conf->resync_lock,
913 flush_pending_writes(conf));
ddaf22ab
N
914 spin_unlock_irq(&conf->resync_lock);
915}
e8096360 916static void unfreeze_array(struct r1conf *conf)
ddaf22ab
N
917{
918 /* reverse the effect of the freeze */
919 spin_lock_irq(&conf->resync_lock);
920 conf->barrier--;
921 conf->nr_waiting--;
922 wake_up(&conf->wait_barrier);
923 spin_unlock_irq(&conf->resync_lock);
924}
925
17999be4 926
4e78064f 927/* duplicate the data pages for behind I/O
4e78064f 928 */
9f2c9d12 929static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
4b6d287f
N
930{
931 int i;
932 struct bio_vec *bvec;
2ca68f5e 933 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
4b6d287f 934 GFP_NOIO);
2ca68f5e 935 if (unlikely(!bvecs))
af6d7b76 936 return;
4b6d287f 937
cb34e057 938 bio_for_each_segment_all(bvec, bio, i) {
2ca68f5e
N
939 bvecs[i] = *bvec;
940 bvecs[i].bv_page = alloc_page(GFP_NOIO);
941 if (unlikely(!bvecs[i].bv_page))
4b6d287f 942 goto do_sync_io;
2ca68f5e
N
943 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
944 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
945 kunmap(bvecs[i].bv_page);
4b6d287f
N
946 kunmap(bvec->bv_page);
947 }
2ca68f5e 948 r1_bio->behind_bvecs = bvecs;
af6d7b76
N
949 r1_bio->behind_page_count = bio->bi_vcnt;
950 set_bit(R1BIO_BehindIO, &r1_bio->state);
951 return;
4b6d287f
N
952
953do_sync_io:
af6d7b76 954 for (i = 0; i < bio->bi_vcnt; i++)
2ca68f5e
N
955 if (bvecs[i].bv_page)
956 put_page(bvecs[i].bv_page);
957 kfree(bvecs);
36a4e1fe 958 pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
4b6d287f
N
959}
960
f54a9d0e
N
961struct raid1_plug_cb {
962 struct blk_plug_cb cb;
963 struct bio_list pending;
964 int pending_cnt;
965};
966
967static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
968{
969 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
970 cb);
971 struct mddev *mddev = plug->cb.data;
972 struct r1conf *conf = mddev->private;
973 struct bio *bio;
974
874807a8 975 if (from_schedule || current->bio_list) {
f54a9d0e
N
976 spin_lock_irq(&conf->device_lock);
977 bio_list_merge(&conf->pending_bio_list, &plug->pending);
978 conf->pending_count += plug->pending_cnt;
979 spin_unlock_irq(&conf->device_lock);
ee0b0244 980 wake_up(&conf->wait_barrier);
f54a9d0e
N
981 md_wakeup_thread(mddev->thread);
982 kfree(plug);
983 return;
984 }
985
986 /* we aren't scheduling, so we can do the write-out directly. */
987 bio = bio_list_get(&plug->pending);
988 bitmap_unplug(mddev->bitmap);
989 wake_up(&conf->wait_barrier);
990
991 while (bio) { /* submit pending writes */
992 struct bio *next = bio->bi_next;
993 bio->bi_next = NULL;
32f9f570
SL
994 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
995 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
996 /* Just ignore it */
997 bio_endio(bio, 0);
998 else
999 generic_make_request(bio);
f54a9d0e
N
1000 bio = next;
1001 }
1002 kfree(plug);
1003}
1004
b4fdcb02 1005static void make_request(struct mddev *mddev, struct bio * bio)
1da177e4 1006{
e8096360 1007 struct r1conf *conf = mddev->private;
0eaf822c 1008 struct raid1_info *mirror;
9f2c9d12 1009 struct r1bio *r1_bio;
1da177e4 1010 struct bio *read_bio;
1f68f0c4 1011 int i, disks;
84255d10 1012 struct bitmap *bitmap;
191ea9b2 1013 unsigned long flags;
a362357b 1014 const int rw = bio_data_dir(bio);
2c7d46ec 1015 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
e9c7469b 1016 const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
2ff8cc2c
SL
1017 const unsigned long do_discard = (bio->bi_rw
1018 & (REQ_DISCARD | REQ_SECURE));
c8dc9c65 1019 const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME);
3cb03002 1020 struct md_rdev *blocked_rdev;
f54a9d0e
N
1021 struct blk_plug_cb *cb;
1022 struct raid1_plug_cb *plug = NULL;
1f68f0c4
N
1023 int first_clone;
1024 int sectors_handled;
1025 int max_sectors;
191ea9b2 1026
1da177e4
LT
1027 /*
1028 * Register the new request and wait if the reconstruction
1029 * thread has put up a bar for new requests.
1030 * Continue immediately if no resync is active currently.
1031 */
62de608d 1032
3d310eb7
N
1033 md_write_start(mddev, bio); /* wait on superblock update early */
1034
6eef4b21 1035 if (bio_data_dir(bio) == WRITE &&
f73a1c7d 1036 bio_end_sector(bio) > mddev->suspend_lo &&
6eef4b21
N
1037 bio->bi_sector < mddev->suspend_hi) {
1038 /* As the suspend_* range is controlled by
1039 * userspace, we want an interruptible
1040 * wait.
1041 */
1042 DEFINE_WAIT(w);
1043 for (;;) {
1044 flush_signals(current);
1045 prepare_to_wait(&conf->wait_barrier,
1046 &w, TASK_INTERRUPTIBLE);
f73a1c7d 1047 if (bio_end_sector(bio) <= mddev->suspend_lo ||
6eef4b21
N
1048 bio->bi_sector >= mddev->suspend_hi)
1049 break;
1050 schedule();
1051 }
1052 finish_wait(&conf->wait_barrier, &w);
1053 }
62de608d 1054
17999be4 1055 wait_barrier(conf);
1da177e4 1056
84255d10
N
1057 bitmap = mddev->bitmap;
1058
1da177e4
LT
1059 /*
1060 * make_request() can abort the operation when READA is being
1061 * used and no empty request is available.
1062 *
1063 */
1064 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1065
1066 r1_bio->master_bio = bio;
aa8b57aa 1067 r1_bio->sectors = bio_sectors(bio);
191ea9b2 1068 r1_bio->state = 0;
1da177e4
LT
1069 r1_bio->mddev = mddev;
1070 r1_bio->sector = bio->bi_sector;
1071
d2eb35ac
N
1072 /* We might need to issue multiple reads to different
1073 * devices if there are bad blocks around, so we keep
1074 * track of the number of reads in bio->bi_phys_segments.
1075 * If this is 0, there is only one r1_bio and no locking
1076 * will be needed when requests complete. If it is
1077 * non-zero, then it is the number of not-completed requests.
1078 */
1079 bio->bi_phys_segments = 0;
1080 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1081
a362357b 1082 if (rw == READ) {
1da177e4
LT
1083 /*
1084 * read balancing logic:
1085 */
d2eb35ac
N
1086 int rdisk;
1087
1088read_again:
1089 rdisk = read_balance(conf, r1_bio, &max_sectors);
1da177e4
LT
1090
1091 if (rdisk < 0) {
1092 /* couldn't find anywhere to read from */
1093 raid_end_bio_io(r1_bio);
5a7bbad2 1094 return;
1da177e4
LT
1095 }
1096 mirror = conf->mirrors + rdisk;
1097
e555190d
N
1098 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1099 bitmap) {
1100 /* Reading from a write-mostly device must
1101 * take care not to over-take any writes
1102 * that are 'behind'
1103 */
1104 wait_event(bitmap->behind_wait,
1105 atomic_read(&bitmap->behind_writes) == 0);
1106 }
1da177e4
LT
1107 r1_bio->read_disk = rdisk;
1108
a167f663 1109 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
d2eb35ac
N
1110 md_trim_bio(read_bio, r1_bio->sector - bio->bi_sector,
1111 max_sectors);
1da177e4
LT
1112
1113 r1_bio->bios[rdisk] = read_bio;
1114
1115 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
1116 read_bio->bi_bdev = mirror->rdev->bdev;
1117 read_bio->bi_end_io = raid1_end_read_request;
7b6d91da 1118 read_bio->bi_rw = READ | do_sync;
1da177e4
LT
1119 read_bio->bi_private = r1_bio;
1120
d2eb35ac
N
1121 if (max_sectors < r1_bio->sectors) {
1122 /* could not read all from this device, so we will
1123 * need another r1_bio.
1124 */
d2eb35ac
N
1125
1126 sectors_handled = (r1_bio->sector + max_sectors
1127 - bio->bi_sector);
1128 r1_bio->sectors = max_sectors;
1129 spin_lock_irq(&conf->device_lock);
1130 if (bio->bi_phys_segments == 0)
1131 bio->bi_phys_segments = 2;
1132 else
1133 bio->bi_phys_segments++;
1134 spin_unlock_irq(&conf->device_lock);
1135 /* Cannot call generic_make_request directly
1136 * as that will be queued in __make_request
1137 * and subsequent mempool_alloc might block waiting
1138 * for it. So hand bio over to raid1d.
1139 */
1140 reschedule_retry(r1_bio);
1141
1142 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1143
1144 r1_bio->master_bio = bio;
aa8b57aa 1145 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
d2eb35ac
N
1146 r1_bio->state = 0;
1147 r1_bio->mddev = mddev;
1148 r1_bio->sector = bio->bi_sector + sectors_handled;
1149 goto read_again;
1150 } else
1151 generic_make_request(read_bio);
5a7bbad2 1152 return;
1da177e4
LT
1153 }
1154
1155 /*
1156 * WRITE:
1157 */
34db0cd6
N
1158 if (conf->pending_count >= max_queued_requests) {
1159 md_wakeup_thread(mddev->thread);
1160 wait_event(conf->wait_barrier,
1161 conf->pending_count < max_queued_requests);
1162 }
1f68f0c4 1163 /* first select target devices under rcu_lock and
1da177e4
LT
1164 * inc refcount on their rdev. Record them by setting
1165 * bios[x] to bio
1f68f0c4
N
1166 * If there are known/acknowledged bad blocks on any device on
1167 * which we have seen a write error, we want to avoid writing those
1168 * blocks.
1169 * This potentially requires several writes to write around
1170 * the bad blocks. Each set of writes gets it's own r1bio
1171 * with a set of bios attached.
1da177e4 1172 */
c3b328ac 1173
8f19ccb2 1174 disks = conf->raid_disks * 2;
6bfe0b49
DW
1175 retry_write:
1176 blocked_rdev = NULL;
1da177e4 1177 rcu_read_lock();
1f68f0c4 1178 max_sectors = r1_bio->sectors;
1da177e4 1179 for (i = 0; i < disks; i++) {
3cb03002 1180 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
6bfe0b49
DW
1181 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1182 atomic_inc(&rdev->nr_pending);
1183 blocked_rdev = rdev;
1184 break;
1185 }
1f68f0c4 1186 r1_bio->bios[i] = NULL;
6b740b8d
N
1187 if (!rdev || test_bit(Faulty, &rdev->flags)
1188 || test_bit(Unmerged, &rdev->flags)) {
8f19ccb2
N
1189 if (i < conf->raid_disks)
1190 set_bit(R1BIO_Degraded, &r1_bio->state);
1f68f0c4
N
1191 continue;
1192 }
1193
1194 atomic_inc(&rdev->nr_pending);
1195 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1196 sector_t first_bad;
1197 int bad_sectors;
1198 int is_bad;
1199
1200 is_bad = is_badblock(rdev, r1_bio->sector,
1201 max_sectors,
1202 &first_bad, &bad_sectors);
1203 if (is_bad < 0) {
1204 /* mustn't write here until the bad block is
1205 * acknowledged*/
1206 set_bit(BlockedBadBlocks, &rdev->flags);
1207 blocked_rdev = rdev;
1208 break;
1209 }
1210 if (is_bad && first_bad <= r1_bio->sector) {
1211 /* Cannot write here at all */
1212 bad_sectors -= (r1_bio->sector - first_bad);
1213 if (bad_sectors < max_sectors)
1214 /* mustn't write more than bad_sectors
1215 * to other devices yet
1216 */
1217 max_sectors = bad_sectors;
03c902e1 1218 rdev_dec_pending(rdev, mddev);
1f68f0c4
N
1219 /* We don't set R1BIO_Degraded as that
1220 * only applies if the disk is
1221 * missing, so it might be re-added,
1222 * and we want to know to recover this
1223 * chunk.
1224 * In this case the device is here,
1225 * and the fact that this chunk is not
1226 * in-sync is recorded in the bad
1227 * block log
1228 */
1229 continue;
964147d5 1230 }
1f68f0c4
N
1231 if (is_bad) {
1232 int good_sectors = first_bad - r1_bio->sector;
1233 if (good_sectors < max_sectors)
1234 max_sectors = good_sectors;
1235 }
1236 }
1237 r1_bio->bios[i] = bio;
1da177e4
LT
1238 }
1239 rcu_read_unlock();
1240
6bfe0b49
DW
1241 if (unlikely(blocked_rdev)) {
1242 /* Wait for this device to become unblocked */
1243 int j;
1244
1245 for (j = 0; j < i; j++)
1246 if (r1_bio->bios[j])
1247 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
1f68f0c4 1248 r1_bio->state = 0;
6bfe0b49
DW
1249 allow_barrier(conf);
1250 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1251 wait_barrier(conf);
1252 goto retry_write;
1253 }
1254
1f68f0c4
N
1255 if (max_sectors < r1_bio->sectors) {
1256 /* We are splitting this write into multiple parts, so
1257 * we need to prepare for allocating another r1_bio.
1258 */
1259 r1_bio->sectors = max_sectors;
1260 spin_lock_irq(&conf->device_lock);
1261 if (bio->bi_phys_segments == 0)
1262 bio->bi_phys_segments = 2;
1263 else
1264 bio->bi_phys_segments++;
1265 spin_unlock_irq(&conf->device_lock);
191ea9b2 1266 }
1f68f0c4 1267 sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector;
4b6d287f 1268
4e78064f 1269 atomic_set(&r1_bio->remaining, 1);
4b6d287f 1270 atomic_set(&r1_bio->behind_remaining, 0);
06d91a5f 1271
1f68f0c4 1272 first_clone = 1;
1da177e4
LT
1273 for (i = 0; i < disks; i++) {
1274 struct bio *mbio;
1275 if (!r1_bio->bios[i])
1276 continue;
1277
a167f663 1278 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1f68f0c4
N
1279 md_trim_bio(mbio, r1_bio->sector - bio->bi_sector, max_sectors);
1280
1281 if (first_clone) {
1282 /* do behind I/O ?
1283 * Not if there are too many, or cannot
1284 * allocate memory, or a reader on WriteMostly
1285 * is waiting for behind writes to flush */
1286 if (bitmap &&
1287 (atomic_read(&bitmap->behind_writes)
1288 < mddev->bitmap_info.max_write_behind) &&
1289 !waitqueue_active(&bitmap->behind_wait))
1290 alloc_behind_pages(mbio, r1_bio);
1291
1292 bitmap_startwrite(bitmap, r1_bio->sector,
1293 r1_bio->sectors,
1294 test_bit(R1BIO_BehindIO,
1295 &r1_bio->state));
1296 first_clone = 0;
1297 }
2ca68f5e 1298 if (r1_bio->behind_bvecs) {
4b6d287f
N
1299 struct bio_vec *bvec;
1300 int j;
1301
cb34e057
KO
1302 /*
1303 * We trimmed the bio, so _all is legit
4b6d287f 1304 */
d74c6d51 1305 bio_for_each_segment_all(bvec, mbio, j)
2ca68f5e 1306 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
4b6d287f
N
1307 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1308 atomic_inc(&r1_bio->behind_remaining);
1309 }
1310
1f68f0c4
N
1311 r1_bio->bios[i] = mbio;
1312
1313 mbio->bi_sector = (r1_bio->sector +
1314 conf->mirrors[i].rdev->data_offset);
1315 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1316 mbio->bi_end_io = raid1_end_write_request;
c8dc9c65
JL
1317 mbio->bi_rw =
1318 WRITE | do_flush_fua | do_sync | do_discard | do_same;
1f68f0c4
N
1319 mbio->bi_private = r1_bio;
1320
1da177e4 1321 atomic_inc(&r1_bio->remaining);
f54a9d0e
N
1322
1323 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1324 if (cb)
1325 plug = container_of(cb, struct raid1_plug_cb, cb);
1326 else
1327 plug = NULL;
4e78064f 1328 spin_lock_irqsave(&conf->device_lock, flags);
f54a9d0e
N
1329 if (plug) {
1330 bio_list_add(&plug->pending, mbio);
1331 plug->pending_cnt++;
1332 } else {
1333 bio_list_add(&conf->pending_bio_list, mbio);
1334 conf->pending_count++;
1335 }
4e78064f 1336 spin_unlock_irqrestore(&conf->device_lock, flags);
f54a9d0e 1337 if (!plug)
b357f04a 1338 md_wakeup_thread(mddev->thread);
1da177e4 1339 }
079fa166
N
1340 /* Mustn't call r1_bio_write_done before this next test,
1341 * as it could result in the bio being freed.
1342 */
aa8b57aa 1343 if (sectors_handled < bio_sectors(bio)) {
079fa166 1344 r1_bio_write_done(r1_bio);
1f68f0c4
N
1345 /* We need another r1_bio. It has already been counted
1346 * in bio->bi_phys_segments
1347 */
1348 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1349 r1_bio->master_bio = bio;
aa8b57aa 1350 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1f68f0c4
N
1351 r1_bio->state = 0;
1352 r1_bio->mddev = mddev;
1353 r1_bio->sector = bio->bi_sector + sectors_handled;
1354 goto retry_write;
1355 }
1356
079fa166
N
1357 r1_bio_write_done(r1_bio);
1358
1359 /* In case raid1d snuck in to freeze_array */
1360 wake_up(&conf->wait_barrier);
1da177e4
LT
1361}
1362
fd01b88c 1363static void status(struct seq_file *seq, struct mddev *mddev)
1da177e4 1364{
e8096360 1365 struct r1conf *conf = mddev->private;
1da177e4
LT
1366 int i;
1367
1368 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
11ce99e6 1369 conf->raid_disks - mddev->degraded);
ddac7c7e
N
1370 rcu_read_lock();
1371 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 1372 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1da177e4 1373 seq_printf(seq, "%s",
ddac7c7e
N
1374 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1375 }
1376 rcu_read_unlock();
1da177e4
LT
1377 seq_printf(seq, "]");
1378}
1379
1380
fd01b88c 1381static void error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
1382{
1383 char b[BDEVNAME_SIZE];
e8096360 1384 struct r1conf *conf = mddev->private;
c4a6d3f3 1385 unsigned long flags;
1da177e4
LT
1386
1387 /*
1388 * If it is not operational, then we have already marked it as dead
1389 * else if it is the last working disks, ignore the error, let the
1390 * next level up know.
1391 * else mark the drive as failed
1392 */
b2d444d7 1393 if (test_bit(In_sync, &rdev->flags)
4044ba58 1394 && (conf->raid_disks - mddev->degraded) == 1) {
1da177e4
LT
1395 /*
1396 * Don't fail the drive, act as though we were just a
4044ba58
N
1397 * normal single drive.
1398 * However don't try a recovery from this drive as
1399 * it is very likely to fail.
1da177e4 1400 */
5389042f 1401 conf->recovery_disabled = mddev->recovery_disabled;
1da177e4 1402 return;
4044ba58 1403 }
de393cde 1404 set_bit(Blocked, &rdev->flags);
c4a6d3f3 1405 spin_lock_irqsave(&conf->device_lock, flags);
c04be0aa 1406 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1da177e4 1407 mddev->degraded++;
dd00a99e 1408 set_bit(Faulty, &rdev->flags);
dd00a99e
N
1409 } else
1410 set_bit(Faulty, &rdev->flags);
c4a6d3f3 1411 spin_unlock_irqrestore(&conf->device_lock, flags);
b08633de
N
1412 /*
1413 * if recovery is running, make sure it aborts.
1414 */
1415 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
850b2b42 1416 set_bit(MD_CHANGE_DEVS, &mddev->flags);
067032bc
JP
1417 printk(KERN_ALERT
1418 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1419 "md/raid1:%s: Operation continuing on %d devices.\n",
9dd1e2fa
N
1420 mdname(mddev), bdevname(rdev->bdev, b),
1421 mdname(mddev), conf->raid_disks - mddev->degraded);
1da177e4
LT
1422}
1423
e8096360 1424static void print_conf(struct r1conf *conf)
1da177e4
LT
1425{
1426 int i;
1da177e4 1427
9dd1e2fa 1428 printk(KERN_DEBUG "RAID1 conf printout:\n");
1da177e4 1429 if (!conf) {
9dd1e2fa 1430 printk(KERN_DEBUG "(!conf)\n");
1da177e4
LT
1431 return;
1432 }
9dd1e2fa 1433 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1da177e4
LT
1434 conf->raid_disks);
1435
ddac7c7e 1436 rcu_read_lock();
1da177e4
LT
1437 for (i = 0; i < conf->raid_disks; i++) {
1438 char b[BDEVNAME_SIZE];
3cb03002 1439 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
ddac7c7e 1440 if (rdev)
9dd1e2fa 1441 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
ddac7c7e
N
1442 i, !test_bit(In_sync, &rdev->flags),
1443 !test_bit(Faulty, &rdev->flags),
1444 bdevname(rdev->bdev,b));
1da177e4 1445 }
ddac7c7e 1446 rcu_read_unlock();
1da177e4
LT
1447}
1448
e8096360 1449static void close_sync(struct r1conf *conf)
1da177e4 1450{
17999be4
N
1451 wait_barrier(conf);
1452 allow_barrier(conf);
1da177e4
LT
1453
1454 mempool_destroy(conf->r1buf_pool);
1455 conf->r1buf_pool = NULL;
1456}
1457
fd01b88c 1458static int raid1_spare_active(struct mddev *mddev)
1da177e4
LT
1459{
1460 int i;
e8096360 1461 struct r1conf *conf = mddev->private;
6b965620
N
1462 int count = 0;
1463 unsigned long flags;
1da177e4
LT
1464
1465 /*
1466 * Find all failed disks within the RAID1 configuration
ddac7c7e
N
1467 * and mark them readable.
1468 * Called under mddev lock, so rcu protection not needed.
c4a6d3f3
N
1469 * device_lock used to avoid races with raid1_end_read_request
1470 * which expects 'In_sync' flags and ->degraded to be consistent.
1da177e4 1471 */
c4a6d3f3 1472 spin_lock_irqsave(&conf->device_lock, flags);
1da177e4 1473 for (i = 0; i < conf->raid_disks; i++) {
3cb03002 1474 struct md_rdev *rdev = conf->mirrors[i].rdev;
8c7a2c2b
N
1475 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1476 if (repl
1477 && repl->recovery_offset == MaxSector
1478 && !test_bit(Faulty, &repl->flags)
1479 && !test_and_set_bit(In_sync, &repl->flags)) {
1480 /* replacement has just become active */
1481 if (!rdev ||
1482 !test_and_clear_bit(In_sync, &rdev->flags))
1483 count++;
1484 if (rdev) {
1485 /* Replaced device not technically
1486 * faulty, but we need to be sure
1487 * it gets removed and never re-added
1488 */
1489 set_bit(Faulty, &rdev->flags);
1490 sysfs_notify_dirent_safe(
1491 rdev->sysfs_state);
1492 }
1493 }
ddac7c7e 1494 if (rdev
ed840bec 1495 && rdev->recovery_offset == MaxSector
ddac7c7e 1496 && !test_bit(Faulty, &rdev->flags)
c04be0aa 1497 && !test_and_set_bit(In_sync, &rdev->flags)) {
6b965620 1498 count++;
654e8b5a 1499 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT
1500 }
1501 }
6b965620
N
1502 mddev->degraded -= count;
1503 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4
LT
1504
1505 print_conf(conf);
6b965620 1506 return count;
1da177e4
LT
1507}
1508
1509
fd01b88c 1510static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1511{
e8096360 1512 struct r1conf *conf = mddev->private;
199050ea 1513 int err = -EEXIST;
41158c7e 1514 int mirror = 0;
0eaf822c 1515 struct raid1_info *p;
6c2fce2e 1516 int first = 0;
30194636 1517 int last = conf->raid_disks - 1;
6b740b8d 1518 struct request_queue *q = bdev_get_queue(rdev->bdev);
1da177e4 1519
5389042f
N
1520 if (mddev->recovery_disabled == conf->recovery_disabled)
1521 return -EBUSY;
1522
6c2fce2e
NB
1523 if (rdev->raid_disk >= 0)
1524 first = last = rdev->raid_disk;
1525
6b740b8d
N
1526 if (q->merge_bvec_fn) {
1527 set_bit(Unmerged, &rdev->flags);
1528 mddev->merge_check_needed = 1;
1529 }
1530
7ef449d1
N
1531 for (mirror = first; mirror <= last; mirror++) {
1532 p = conf->mirrors+mirror;
1533 if (!p->rdev) {
1da177e4 1534
8f6c2e4b
MP
1535 disk_stack_limits(mddev->gendisk, rdev->bdev,
1536 rdev->data_offset << 9);
1da177e4
LT
1537
1538 p->head_position = 0;
1539 rdev->raid_disk = mirror;
199050ea 1540 err = 0;
6aea114a
N
1541 /* As all devices are equivalent, we don't need a full recovery
1542 * if this was recently any drive of the array
1543 */
1544 if (rdev->saved_raid_disk < 0)
41158c7e 1545 conf->fullsync = 1;
d6065f7b 1546 rcu_assign_pointer(p->rdev, rdev);
1da177e4
LT
1547 break;
1548 }
7ef449d1
N
1549 if (test_bit(WantReplacement, &p->rdev->flags) &&
1550 p[conf->raid_disks].rdev == NULL) {
1551 /* Add this device as a replacement */
1552 clear_bit(In_sync, &rdev->flags);
1553 set_bit(Replacement, &rdev->flags);
1554 rdev->raid_disk = mirror;
1555 err = 0;
1556 conf->fullsync = 1;
1557 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1558 break;
1559 }
1560 }
6b740b8d
N
1561 if (err == 0 && test_bit(Unmerged, &rdev->flags)) {
1562 /* Some requests might not have seen this new
1563 * merge_bvec_fn. We must wait for them to complete
1564 * before merging the device fully.
1565 * First we make sure any code which has tested
1566 * our function has submitted the request, then
1567 * we wait for all outstanding requests to complete.
1568 */
1569 synchronize_sched();
e2d59925
N
1570 freeze_array(conf, 0);
1571 unfreeze_array(conf);
6b740b8d
N
1572 clear_bit(Unmerged, &rdev->flags);
1573 }
ac5e7113 1574 md_integrity_add_rdev(rdev, mddev);
2ff8cc2c
SL
1575 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
1576 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1da177e4 1577 print_conf(conf);
199050ea 1578 return err;
1da177e4
LT
1579}
1580
b8321b68 1581static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1582{
e8096360 1583 struct r1conf *conf = mddev->private;
1da177e4 1584 int err = 0;
b8321b68 1585 int number = rdev->raid_disk;
0eaf822c 1586 struct raid1_info *p = conf->mirrors + number;
1da177e4 1587
b014f14c
N
1588 if (rdev != p->rdev)
1589 p = conf->mirrors + conf->raid_disks + number;
1590
1da177e4 1591 print_conf(conf);
b8321b68 1592 if (rdev == p->rdev) {
b2d444d7 1593 if (test_bit(In_sync, &rdev->flags) ||
1da177e4
LT
1594 atomic_read(&rdev->nr_pending)) {
1595 err = -EBUSY;
1596 goto abort;
1597 }
046abeed 1598 /* Only remove non-faulty devices if recovery
dfc70645
N
1599 * is not possible.
1600 */
1601 if (!test_bit(Faulty, &rdev->flags) &&
5389042f 1602 mddev->recovery_disabled != conf->recovery_disabled &&
dfc70645
N
1603 mddev->degraded < conf->raid_disks) {
1604 err = -EBUSY;
1605 goto abort;
1606 }
1da177e4 1607 p->rdev = NULL;
fbd568a3 1608 synchronize_rcu();
1da177e4
LT
1609 if (atomic_read(&rdev->nr_pending)) {
1610 /* lost the race, try later */
1611 err = -EBUSY;
1612 p->rdev = rdev;
ac5e7113 1613 goto abort;
8c7a2c2b
N
1614 } else if (conf->mirrors[conf->raid_disks + number].rdev) {
1615 /* We just removed a device that is being replaced.
1616 * Move down the replacement. We drain all IO before
1617 * doing this to avoid confusion.
1618 */
1619 struct md_rdev *repl =
1620 conf->mirrors[conf->raid_disks + number].rdev;
e2d59925 1621 freeze_array(conf, 0);
8c7a2c2b
N
1622 clear_bit(Replacement, &repl->flags);
1623 p->rdev = repl;
1624 conf->mirrors[conf->raid_disks + number].rdev = NULL;
e2d59925 1625 unfreeze_array(conf);
8c7a2c2b
N
1626 clear_bit(WantReplacement, &rdev->flags);
1627 } else
b014f14c 1628 clear_bit(WantReplacement, &rdev->flags);
a91a2785 1629 err = md_integrity_register(mddev);
1da177e4
LT
1630 }
1631abort:
1632
1633 print_conf(conf);
1634 return err;
1635}
1636
1637
6712ecf8 1638static void end_sync_read(struct bio *bio, int error)
1da177e4 1639{
9f2c9d12 1640 struct r1bio *r1_bio = bio->bi_private;
1da177e4 1641
0fc280f6 1642 update_head_pos(r1_bio->read_disk, r1_bio);
ba3ae3be 1643
1da177e4
LT
1644 /*
1645 * we have read a block, now it needs to be re-written,
1646 * or re-read if the read failed.
1647 * We don't do much here, just schedule handling by raid1d
1648 */
69382e85 1649 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
1da177e4 1650 set_bit(R1BIO_Uptodate, &r1_bio->state);
d11c171e
N
1651
1652 if (atomic_dec_and_test(&r1_bio->remaining))
1653 reschedule_retry(r1_bio);
1da177e4
LT
1654}
1655
6712ecf8 1656static void end_sync_write(struct bio *bio, int error)
1da177e4
LT
1657{
1658 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
9f2c9d12 1659 struct r1bio *r1_bio = bio->bi_private;
fd01b88c 1660 struct mddev *mddev = r1_bio->mddev;
e8096360 1661 struct r1conf *conf = mddev->private;
1da177e4 1662 int mirror=0;
4367af55
N
1663 sector_t first_bad;
1664 int bad_sectors;
1da177e4 1665
ba3ae3be
NK
1666 mirror = find_bio_disk(r1_bio, bio);
1667
6b1117d5 1668 if (!uptodate) {
57dab0bd 1669 sector_t sync_blocks = 0;
6b1117d5
N
1670 sector_t s = r1_bio->sector;
1671 long sectors_to_go = r1_bio->sectors;
1672 /* make sure these bits doesn't get cleared. */
1673 do {
5e3db645 1674 bitmap_end_sync(mddev->bitmap, s,
6b1117d5
N
1675 &sync_blocks, 1);
1676 s += sync_blocks;
1677 sectors_to_go -= sync_blocks;
1678 } while (sectors_to_go > 0);
d8f05d29
N
1679 set_bit(WriteErrorSeen,
1680 &conf->mirrors[mirror].rdev->flags);
19d67169
N
1681 if (!test_and_set_bit(WantReplacement,
1682 &conf->mirrors[mirror].rdev->flags))
1683 set_bit(MD_RECOVERY_NEEDED, &
1684 mddev->recovery);
d8f05d29 1685 set_bit(R1BIO_WriteError, &r1_bio->state);
4367af55
N
1686 } else if (is_badblock(conf->mirrors[mirror].rdev,
1687 r1_bio->sector,
1688 r1_bio->sectors,
3a9f28a5
N
1689 &first_bad, &bad_sectors) &&
1690 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1691 r1_bio->sector,
1692 r1_bio->sectors,
1693 &first_bad, &bad_sectors)
1694 )
4367af55 1695 set_bit(R1BIO_MadeGood, &r1_bio->state);
e3b9703e 1696
1da177e4 1697 if (atomic_dec_and_test(&r1_bio->remaining)) {
4367af55 1698 int s = r1_bio->sectors;
d8f05d29
N
1699 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1700 test_bit(R1BIO_WriteError, &r1_bio->state))
4367af55
N
1701 reschedule_retry(r1_bio);
1702 else {
1703 put_buf(r1_bio);
1704 md_done_sync(mddev, s, uptodate);
1705 }
1da177e4 1706 }
1da177e4
LT
1707}
1708
3cb03002 1709static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
d8f05d29
N
1710 int sectors, struct page *page, int rw)
1711{
1712 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
1713 /* success */
1714 return 1;
19d67169 1715 if (rw == WRITE) {
d8f05d29 1716 set_bit(WriteErrorSeen, &rdev->flags);
19d67169
N
1717 if (!test_and_set_bit(WantReplacement,
1718 &rdev->flags))
1719 set_bit(MD_RECOVERY_NEEDED, &
1720 rdev->mddev->recovery);
1721 }
d8f05d29
N
1722 /* need to record an error - either for the block or the device */
1723 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1724 md_error(rdev->mddev, rdev);
1725 return 0;
1726}
1727
9f2c9d12 1728static int fix_sync_read_error(struct r1bio *r1_bio)
1da177e4 1729{
a68e5870
N
1730 /* Try some synchronous reads of other devices to get
1731 * good data, much like with normal read errors. Only
1732 * read into the pages we already have so we don't
1733 * need to re-issue the read request.
1734 * We don't need to freeze the array, because being in an
1735 * active sync request, there is no normal IO, and
1736 * no overlapping syncs.
06f60385
N
1737 * We don't need to check is_badblock() again as we
1738 * made sure that anything with a bad block in range
1739 * will have bi_end_io clear.
a68e5870 1740 */
fd01b88c 1741 struct mddev *mddev = r1_bio->mddev;
e8096360 1742 struct r1conf *conf = mddev->private;
a68e5870
N
1743 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1744 sector_t sect = r1_bio->sector;
1745 int sectors = r1_bio->sectors;
1746 int idx = 0;
1747
1748 while(sectors) {
1749 int s = sectors;
1750 int d = r1_bio->read_disk;
1751 int success = 0;
3cb03002 1752 struct md_rdev *rdev;
78d7f5f7 1753 int start;
a68e5870
N
1754
1755 if (s > (PAGE_SIZE>>9))
1756 s = PAGE_SIZE >> 9;
1757 do {
1758 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1759 /* No rcu protection needed here devices
1760 * can only be removed when no resync is
1761 * active, and resync is currently active
1762 */
1763 rdev = conf->mirrors[d].rdev;
9d3d8011 1764 if (sync_page_io(rdev, sect, s<<9,
a68e5870
N
1765 bio->bi_io_vec[idx].bv_page,
1766 READ, false)) {
1767 success = 1;
1768 break;
1769 }
1770 }
1771 d++;
8f19ccb2 1772 if (d == conf->raid_disks * 2)
a68e5870
N
1773 d = 0;
1774 } while (!success && d != r1_bio->read_disk);
1775
78d7f5f7 1776 if (!success) {
a68e5870 1777 char b[BDEVNAME_SIZE];
3a9f28a5
N
1778 int abort = 0;
1779 /* Cannot read from anywhere, this block is lost.
1780 * Record a bad block on each device. If that doesn't
1781 * work just disable and interrupt the recovery.
1782 * Don't fail devices as that won't really help.
1783 */
a68e5870
N
1784 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1785 " for block %llu\n",
1786 mdname(mddev),
1787 bdevname(bio->bi_bdev, b),
1788 (unsigned long long)r1_bio->sector);
8f19ccb2 1789 for (d = 0; d < conf->raid_disks * 2; d++) {
3a9f28a5
N
1790 rdev = conf->mirrors[d].rdev;
1791 if (!rdev || test_bit(Faulty, &rdev->flags))
1792 continue;
1793 if (!rdev_set_badblocks(rdev, sect, s, 0))
1794 abort = 1;
1795 }
1796 if (abort) {
d890fa2b
N
1797 conf->recovery_disabled =
1798 mddev->recovery_disabled;
3a9f28a5
N
1799 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1800 md_done_sync(mddev, r1_bio->sectors, 0);
1801 put_buf(r1_bio);
1802 return 0;
1803 }
1804 /* Try next page */
1805 sectors -= s;
1806 sect += s;
1807 idx++;
1808 continue;
d11c171e 1809 }
78d7f5f7
N
1810
1811 start = d;
1812 /* write it back and re-read */
1813 while (d != r1_bio->read_disk) {
1814 if (d == 0)
8f19ccb2 1815 d = conf->raid_disks * 2;
78d7f5f7
N
1816 d--;
1817 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1818 continue;
1819 rdev = conf->mirrors[d].rdev;
d8f05d29
N
1820 if (r1_sync_page_io(rdev, sect, s,
1821 bio->bi_io_vec[idx].bv_page,
1822 WRITE) == 0) {
78d7f5f7
N
1823 r1_bio->bios[d]->bi_end_io = NULL;
1824 rdev_dec_pending(rdev, mddev);
9d3d8011 1825 }
78d7f5f7
N
1826 }
1827 d = start;
1828 while (d != r1_bio->read_disk) {
1829 if (d == 0)
8f19ccb2 1830 d = conf->raid_disks * 2;
78d7f5f7
N
1831 d--;
1832 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1833 continue;
1834 rdev = conf->mirrors[d].rdev;
d8f05d29
N
1835 if (r1_sync_page_io(rdev, sect, s,
1836 bio->bi_io_vec[idx].bv_page,
1837 READ) != 0)
9d3d8011 1838 atomic_add(s, &rdev->corrected_errors);
78d7f5f7 1839 }
a68e5870
N
1840 sectors -= s;
1841 sect += s;
1842 idx ++;
1843 }
78d7f5f7 1844 set_bit(R1BIO_Uptodate, &r1_bio->state);
7ca78d57 1845 set_bit(BIO_UPTODATE, &bio->bi_flags);
a68e5870
N
1846 return 1;
1847}
1848
9f2c9d12 1849static int process_checks(struct r1bio *r1_bio)
a68e5870
N
1850{
1851 /* We have read all readable devices. If we haven't
1852 * got the block, then there is no hope left.
1853 * If we have, then we want to do a comparison
1854 * and skip the write if everything is the same.
1855 * If any blocks failed to read, then we need to
1856 * attempt an over-write
1857 */
fd01b88c 1858 struct mddev *mddev = r1_bio->mddev;
e8096360 1859 struct r1conf *conf = mddev->private;
a68e5870
N
1860 int primary;
1861 int i;
f4380a91 1862 int vcnt;
a68e5870 1863
8afb90da
N
1864 /* Fix variable parts of all bios */
1865 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1866 for (i = 0; i < conf->raid_disks * 2; i++) {
1867 int j;
1868 int size;
9f2d2899 1869 int uptodate;
8afb90da
N
1870 struct bio *b = r1_bio->bios[i];
1871 if (b->bi_end_io != end_sync_read)
1872 continue;
9f2d2899
N
1873 /* fixup the bio for reuse, but preserve BIO_UPTODATE */
1874 uptodate = test_bit(BIO_UPTODATE, &b->bi_flags);
8afb90da 1875 bio_reset(b);
9f2d2899
N
1876 if (!uptodate)
1877 clear_bit(BIO_UPTODATE, &b->bi_flags);
8afb90da
N
1878 b->bi_vcnt = vcnt;
1879 b->bi_size = r1_bio->sectors << 9;
1880 b->bi_sector = r1_bio->sector +
1881 conf->mirrors[i].rdev->data_offset;
1882 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1883 b->bi_end_io = end_sync_read;
1884 b->bi_private = r1_bio;
1885
1886 size = b->bi_size;
1887 for (j = 0; j < vcnt ; j++) {
1888 struct bio_vec *bi;
1889 bi = &b->bi_io_vec[j];
1890 bi->bv_offset = 0;
1891 if (size > PAGE_SIZE)
1892 bi->bv_len = PAGE_SIZE;
1893 else
1894 bi->bv_len = size;
1895 size -= PAGE_SIZE;
1896 }
1897 }
8f19ccb2 1898 for (primary = 0; primary < conf->raid_disks * 2; primary++)
a68e5870
N
1899 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1900 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1901 r1_bio->bios[primary]->bi_end_io = NULL;
1902 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1903 break;
1904 }
1905 r1_bio->read_disk = primary;
8f19ccb2 1906 for (i = 0; i < conf->raid_disks * 2; i++) {
78d7f5f7 1907 int j;
78d7f5f7
N
1908 struct bio *pbio = r1_bio->bios[primary];
1909 struct bio *sbio = r1_bio->bios[i];
9f2d2899 1910 int uptodate = test_bit(BIO_UPTODATE, &sbio->bi_flags);
a68e5870 1911
2aabaa65 1912 if (sbio->bi_end_io != end_sync_read)
78d7f5f7 1913 continue;
9f2d2899
N
1914 /* Now we can 'fixup' the BIO_UPTODATE flag */
1915 set_bit(BIO_UPTODATE, &sbio->bi_flags);
78d7f5f7 1916
9f2d2899 1917 if (uptodate) {
78d7f5f7
N
1918 for (j = vcnt; j-- ; ) {
1919 struct page *p, *s;
1920 p = pbio->bi_io_vec[j].bv_page;
1921 s = sbio->bi_io_vec[j].bv_page;
1922 if (memcmp(page_address(p),
1923 page_address(s),
5020ad7d 1924 sbio->bi_io_vec[j].bv_len))
78d7f5f7 1925 break;
69382e85 1926 }
78d7f5f7
N
1927 } else
1928 j = 0;
1929 if (j >= 0)
7f7583d4 1930 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
78d7f5f7 1931 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
9f2d2899 1932 && uptodate)) {
78d7f5f7
N
1933 /* No need to write to this device. */
1934 sbio->bi_end_io = NULL;
1935 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1936 continue;
1937 }
d3b45c2a
KO
1938
1939 bio_copy_data(sbio, pbio);
78d7f5f7 1940 }
a68e5870
N
1941 return 0;
1942}
1943
9f2c9d12 1944static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
a68e5870 1945{
e8096360 1946 struct r1conf *conf = mddev->private;
a68e5870 1947 int i;
8f19ccb2 1948 int disks = conf->raid_disks * 2;
a68e5870
N
1949 struct bio *bio, *wbio;
1950
1951 bio = r1_bio->bios[r1_bio->read_disk];
1952
a68e5870
N
1953 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
1954 /* ouch - failed to read all of that. */
1955 if (!fix_sync_read_error(r1_bio))
1956 return;
7ca78d57
N
1957
1958 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1959 if (process_checks(r1_bio) < 0)
1960 return;
d11c171e
N
1961 /*
1962 * schedule writes
1963 */
1da177e4
LT
1964 atomic_set(&r1_bio->remaining, 1);
1965 for (i = 0; i < disks ; i++) {
1966 wbio = r1_bio->bios[i];
3e198f78
N
1967 if (wbio->bi_end_io == NULL ||
1968 (wbio->bi_end_io == end_sync_read &&
1969 (i == r1_bio->read_disk ||
1970 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
1da177e4
LT
1971 continue;
1972
3e198f78
N
1973 wbio->bi_rw = WRITE;
1974 wbio->bi_end_io = end_sync_write;
1da177e4 1975 atomic_inc(&r1_bio->remaining);
aa8b57aa 1976 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
191ea9b2 1977
1da177e4
LT
1978 generic_make_request(wbio);
1979 }
1980
1981 if (atomic_dec_and_test(&r1_bio->remaining)) {
191ea9b2 1982 /* if we're here, all write(s) have completed, so clean up */
58e94ae1
N
1983 int s = r1_bio->sectors;
1984 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1985 test_bit(R1BIO_WriteError, &r1_bio->state))
1986 reschedule_retry(r1_bio);
1987 else {
1988 put_buf(r1_bio);
1989 md_done_sync(mddev, s, 1);
1990 }
1da177e4
LT
1991 }
1992}
1993
1994/*
1995 * This is a kernel thread which:
1996 *
1997 * 1. Retries failed read operations on working mirrors.
1998 * 2. Updates the raid superblock when problems encounter.
d2eb35ac 1999 * 3. Performs writes following reads for array synchronising.
1da177e4
LT
2000 */
2001
e8096360 2002static void fix_read_error(struct r1conf *conf, int read_disk,
867868fb
N
2003 sector_t sect, int sectors)
2004{
fd01b88c 2005 struct mddev *mddev = conf->mddev;
867868fb
N
2006 while(sectors) {
2007 int s = sectors;
2008 int d = read_disk;
2009 int success = 0;
2010 int start;
3cb03002 2011 struct md_rdev *rdev;
867868fb
N
2012
2013 if (s > (PAGE_SIZE>>9))
2014 s = PAGE_SIZE >> 9;
2015
2016 do {
2017 /* Note: no rcu protection needed here
2018 * as this is synchronous in the raid1d thread
2019 * which is the thread that might remove
2020 * a device. If raid1d ever becomes multi-threaded....
2021 */
d2eb35ac
N
2022 sector_t first_bad;
2023 int bad_sectors;
2024
867868fb
N
2025 rdev = conf->mirrors[d].rdev;
2026 if (rdev &&
da8840a7 2027 (test_bit(In_sync, &rdev->flags) ||
2028 (!test_bit(Faulty, &rdev->flags) &&
2029 rdev->recovery_offset >= sect + s)) &&
d2eb35ac
N
2030 is_badblock(rdev, sect, s,
2031 &first_bad, &bad_sectors) == 0 &&
ccebd4c4
JB
2032 sync_page_io(rdev, sect, s<<9,
2033 conf->tmppage, READ, false))
867868fb
N
2034 success = 1;
2035 else {
2036 d++;
8f19ccb2 2037 if (d == conf->raid_disks * 2)
867868fb
N
2038 d = 0;
2039 }
2040 } while (!success && d != read_disk);
2041
2042 if (!success) {
d8f05d29 2043 /* Cannot read from anywhere - mark it bad */
3cb03002 2044 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
d8f05d29
N
2045 if (!rdev_set_badblocks(rdev, sect, s, 0))
2046 md_error(mddev, rdev);
867868fb
N
2047 break;
2048 }
2049 /* write it back and re-read */
2050 start = d;
2051 while (d != read_disk) {
2052 if (d==0)
8f19ccb2 2053 d = conf->raid_disks * 2;
867868fb
N
2054 d--;
2055 rdev = conf->mirrors[d].rdev;
2056 if (rdev &&
a8f165f5 2057 !test_bit(Faulty, &rdev->flags))
d8f05d29
N
2058 r1_sync_page_io(rdev, sect, s,
2059 conf->tmppage, WRITE);
867868fb
N
2060 }
2061 d = start;
2062 while (d != read_disk) {
2063 char b[BDEVNAME_SIZE];
2064 if (d==0)
8f19ccb2 2065 d = conf->raid_disks * 2;
867868fb
N
2066 d--;
2067 rdev = conf->mirrors[d].rdev;
2068 if (rdev &&
a8f165f5 2069 !test_bit(Faulty, &rdev->flags)) {
d8f05d29
N
2070 if (r1_sync_page_io(rdev, sect, s,
2071 conf->tmppage, READ)) {
867868fb
N
2072 atomic_add(s, &rdev->corrected_errors);
2073 printk(KERN_INFO
9dd1e2fa 2074 "md/raid1:%s: read error corrected "
867868fb
N
2075 "(%d sectors at %llu on %s)\n",
2076 mdname(mddev), s,
969b755a
RD
2077 (unsigned long long)(sect +
2078 rdev->data_offset),
867868fb
N
2079 bdevname(rdev->bdev, b));
2080 }
2081 }
2082 }
2083 sectors -= s;
2084 sect += s;
2085 }
2086}
2087
9f2c9d12 2088static int narrow_write_error(struct r1bio *r1_bio, int i)
cd5ff9a1 2089{
fd01b88c 2090 struct mddev *mddev = r1_bio->mddev;
e8096360 2091 struct r1conf *conf = mddev->private;
3cb03002 2092 struct md_rdev *rdev = conf->mirrors[i].rdev;
cd5ff9a1
N
2093
2094 /* bio has the data to be written to device 'i' where
2095 * we just recently had a write error.
2096 * We repeatedly clone the bio and trim down to one block,
2097 * then try the write. Where the write fails we record
2098 * a bad block.
2099 * It is conceivable that the bio doesn't exactly align with
2100 * blocks. We must handle this somehow.
2101 *
2102 * We currently own a reference on the rdev.
2103 */
2104
2105 int block_sectors;
2106 sector_t sector;
2107 int sectors;
2108 int sect_to_write = r1_bio->sectors;
2109 int ok = 1;
2110
2111 if (rdev->badblocks.shift < 0)
2112 return 0;
2113
2114 block_sectors = 1 << rdev->badblocks.shift;
2115 sector = r1_bio->sector;
2116 sectors = ((sector + block_sectors)
2117 & ~(sector_t)(block_sectors - 1))
2118 - sector;
2119
cd5ff9a1
N
2120 while (sect_to_write) {
2121 struct bio *wbio;
2122 if (sectors > sect_to_write)
2123 sectors = sect_to_write;
2124 /* Write at 'sector' for 'sectors'*/
2125
b783863f
KO
2126 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2127 unsigned vcnt = r1_bio->behind_page_count;
2128 struct bio_vec *vec = r1_bio->behind_bvecs;
2129
2130 while (!vec->bv_page) {
2131 vec++;
2132 vcnt--;
2133 }
2134
2135 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2136 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2137
2138 wbio->bi_vcnt = vcnt;
2139 } else {
2140 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2141 }
2142
cd5ff9a1 2143 wbio->bi_rw = WRITE;
b783863f 2144 wbio->bi_sector = r1_bio->sector;
cd5ff9a1 2145 wbio->bi_size = r1_bio->sectors << 9;
cd5ff9a1
N
2146
2147 md_trim_bio(wbio, sector - r1_bio->sector, sectors);
2148 wbio->bi_sector += rdev->data_offset;
2149 wbio->bi_bdev = rdev->bdev;
1a4dfd7d 2150 if (submit_bio_wait(WRITE, wbio) < 0)
cd5ff9a1
N
2151 /* failure! */
2152 ok = rdev_set_badblocks(rdev, sector,
2153 sectors, 0)
2154 && ok;
2155
2156 bio_put(wbio);
2157 sect_to_write -= sectors;
2158 sector += sectors;
2159 sectors = block_sectors;
2160 }
2161 return ok;
2162}
2163
e8096360 2164static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
62096bce
N
2165{
2166 int m;
2167 int s = r1_bio->sectors;
8f19ccb2 2168 for (m = 0; m < conf->raid_disks * 2 ; m++) {
3cb03002 2169 struct md_rdev *rdev = conf->mirrors[m].rdev;
62096bce
N
2170 struct bio *bio = r1_bio->bios[m];
2171 if (bio->bi_end_io == NULL)
2172 continue;
2173 if (test_bit(BIO_UPTODATE, &bio->bi_flags) &&
2174 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
c6563a8c 2175 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
62096bce
N
2176 }
2177 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
2178 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2179 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2180 md_error(conf->mddev, rdev);
2181 }
2182 }
2183 put_buf(r1_bio);
2184 md_done_sync(conf->mddev, s, 1);
2185}
2186
e8096360 2187static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
62096bce
N
2188{
2189 int m;
8f19ccb2 2190 for (m = 0; m < conf->raid_disks * 2 ; m++)
62096bce 2191 if (r1_bio->bios[m] == IO_MADE_GOOD) {
3cb03002 2192 struct md_rdev *rdev = conf->mirrors[m].rdev;
62096bce
N
2193 rdev_clear_badblocks(rdev,
2194 r1_bio->sector,
c6563a8c 2195 r1_bio->sectors, 0);
62096bce
N
2196 rdev_dec_pending(rdev, conf->mddev);
2197 } else if (r1_bio->bios[m] != NULL) {
2198 /* This drive got a write error. We need to
2199 * narrow down and record precise write
2200 * errors.
2201 */
2202 if (!narrow_write_error(r1_bio, m)) {
2203 md_error(conf->mddev,
2204 conf->mirrors[m].rdev);
2205 /* an I/O failed, we can't clear the bitmap */
2206 set_bit(R1BIO_Degraded, &r1_bio->state);
2207 }
2208 rdev_dec_pending(conf->mirrors[m].rdev,
2209 conf->mddev);
2210 }
2211 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2212 close_write(r1_bio);
2213 raid_end_bio_io(r1_bio);
2214}
2215
e8096360 2216static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
62096bce
N
2217{
2218 int disk;
2219 int max_sectors;
fd01b88c 2220 struct mddev *mddev = conf->mddev;
62096bce
N
2221 struct bio *bio;
2222 char b[BDEVNAME_SIZE];
3cb03002 2223 struct md_rdev *rdev;
62096bce
N
2224
2225 clear_bit(R1BIO_ReadError, &r1_bio->state);
2226 /* we got a read error. Maybe the drive is bad. Maybe just
2227 * the block and we can fix it.
2228 * We freeze all other IO, and try reading the block from
2229 * other devices. When we find one, we re-write
2230 * and check it that fixes the read error.
2231 * This is all done synchronously while the array is
2232 * frozen
2233 */
2234 if (mddev->ro == 0) {
e2d59925 2235 freeze_array(conf, 1);
62096bce
N
2236 fix_read_error(conf, r1_bio->read_disk,
2237 r1_bio->sector, r1_bio->sectors);
2238 unfreeze_array(conf);
2239 } else
2240 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
7ad4d4a6 2241 rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
62096bce
N
2242
2243 bio = r1_bio->bios[r1_bio->read_disk];
2244 bdevname(bio->bi_bdev, b);
2245read_more:
2246 disk = read_balance(conf, r1_bio, &max_sectors);
2247 if (disk == -1) {
2248 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2249 " read error for block %llu\n",
2250 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2251 raid_end_bio_io(r1_bio);
2252 } else {
2253 const unsigned long do_sync
2254 = r1_bio->master_bio->bi_rw & REQ_SYNC;
2255 if (bio) {
2256 r1_bio->bios[r1_bio->read_disk] =
2257 mddev->ro ? IO_BLOCKED : NULL;
2258 bio_put(bio);
2259 }
2260 r1_bio->read_disk = disk;
2261 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2262 md_trim_bio(bio, r1_bio->sector - bio->bi_sector, max_sectors);
2263 r1_bio->bios[r1_bio->read_disk] = bio;
2264 rdev = conf->mirrors[disk].rdev;
2265 printk_ratelimited(KERN_ERR
2266 "md/raid1:%s: redirecting sector %llu"
2267 " to other mirror: %s\n",
2268 mdname(mddev),
2269 (unsigned long long)r1_bio->sector,
2270 bdevname(rdev->bdev, b));
2271 bio->bi_sector = r1_bio->sector + rdev->data_offset;
2272 bio->bi_bdev = rdev->bdev;
2273 bio->bi_end_io = raid1_end_read_request;
2274 bio->bi_rw = READ | do_sync;
2275 bio->bi_private = r1_bio;
2276 if (max_sectors < r1_bio->sectors) {
2277 /* Drat - have to split this up more */
2278 struct bio *mbio = r1_bio->master_bio;
2279 int sectors_handled = (r1_bio->sector + max_sectors
2280 - mbio->bi_sector);
2281 r1_bio->sectors = max_sectors;
2282 spin_lock_irq(&conf->device_lock);
2283 if (mbio->bi_phys_segments == 0)
2284 mbio->bi_phys_segments = 2;
2285 else
2286 mbio->bi_phys_segments++;
2287 spin_unlock_irq(&conf->device_lock);
2288 generic_make_request(bio);
2289 bio = NULL;
2290
2291 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2292
2293 r1_bio->master_bio = mbio;
aa8b57aa 2294 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
62096bce
N
2295 r1_bio->state = 0;
2296 set_bit(R1BIO_ReadError, &r1_bio->state);
2297 r1_bio->mddev = mddev;
2298 r1_bio->sector = mbio->bi_sector + sectors_handled;
2299
2300 goto read_more;
2301 } else
2302 generic_make_request(bio);
2303 }
2304}
2305
4ed8731d 2306static void raid1d(struct md_thread *thread)
1da177e4 2307{
4ed8731d 2308 struct mddev *mddev = thread->mddev;
9f2c9d12 2309 struct r1bio *r1_bio;
1da177e4 2310 unsigned long flags;
e8096360 2311 struct r1conf *conf = mddev->private;
1da177e4 2312 struct list_head *head = &conf->retry_list;
e1dfa0a2 2313 struct blk_plug plug;
1da177e4
LT
2314
2315 md_check_recovery(mddev);
e1dfa0a2
N
2316
2317 blk_start_plug(&plug);
1da177e4 2318 for (;;) {
191ea9b2 2319
0021b7bc 2320 flush_pending_writes(conf);
191ea9b2 2321
a35e63ef
N
2322 spin_lock_irqsave(&conf->device_lock, flags);
2323 if (list_empty(head)) {
2324 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 2325 break;
a35e63ef 2326 }
9f2c9d12 2327 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
1da177e4 2328 list_del(head->prev);
ddaf22ab 2329 conf->nr_queued--;
1da177e4
LT
2330 spin_unlock_irqrestore(&conf->device_lock, flags);
2331
2332 mddev = r1_bio->mddev;
070ec55d 2333 conf = mddev->private;
4367af55 2334 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
d8f05d29 2335 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
62096bce
N
2336 test_bit(R1BIO_WriteError, &r1_bio->state))
2337 handle_sync_write_finished(conf, r1_bio);
2338 else
4367af55 2339 sync_request_write(mddev, r1_bio);
cd5ff9a1 2340 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
62096bce
N
2341 test_bit(R1BIO_WriteError, &r1_bio->state))
2342 handle_write_finished(conf, r1_bio);
2343 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2344 handle_read_error(conf, r1_bio);
2345 else
d2eb35ac
N
2346 /* just a partial read to be scheduled from separate
2347 * context
2348 */
2349 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
62096bce 2350
1d9d5241 2351 cond_resched();
de393cde
N
2352 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2353 md_check_recovery(mddev);
1da177e4 2354 }
e1dfa0a2 2355 blk_finish_plug(&plug);
1da177e4
LT
2356}
2357
2358
e8096360 2359static int init_resync(struct r1conf *conf)
1da177e4
LT
2360{
2361 int buffs;
2362
2363 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
9e77c485 2364 BUG_ON(conf->r1buf_pool);
1da177e4
LT
2365 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2366 conf->poolinfo);
2367 if (!conf->r1buf_pool)
2368 return -ENOMEM;
2369 conf->next_resync = 0;
2370 return 0;
2371}
2372
2373/*
2374 * perform a "sync" on one "block"
2375 *
2376 * We need to make sure that no normal I/O request - particularly write
2377 * requests - conflict with active sync requests.
2378 *
2379 * This is achieved by tracking pending requests and a 'barrier' concept
2380 * that can be installed to exclude normal IO requests.
2381 */
2382
fd01b88c 2383static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
1da177e4 2384{
e8096360 2385 struct r1conf *conf = mddev->private;
9f2c9d12 2386 struct r1bio *r1_bio;
1da177e4
LT
2387 struct bio *bio;
2388 sector_t max_sector, nr_sectors;
3e198f78 2389 int disk = -1;
1da177e4 2390 int i;
3e198f78
N
2391 int wonly = -1;
2392 int write_targets = 0, read_targets = 0;
57dab0bd 2393 sector_t sync_blocks;
e3b9703e 2394 int still_degraded = 0;
06f60385
N
2395 int good_sectors = RESYNC_SECTORS;
2396 int min_bad = 0; /* number of sectors that are bad in all devices */
1da177e4
LT
2397
2398 if (!conf->r1buf_pool)
2399 if (init_resync(conf))
57afd89f 2400 return 0;
1da177e4 2401
58c0fed4 2402 max_sector = mddev->dev_sectors;
1da177e4 2403 if (sector_nr >= max_sector) {
191ea9b2
N
2404 /* If we aborted, we need to abort the
2405 * sync on the 'current' bitmap chunk (there will
2406 * only be one in raid1 resync.
2407 * We can find the current addess in mddev->curr_resync
2408 */
6a806c51
N
2409 if (mddev->curr_resync < max_sector) /* aborted */
2410 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
191ea9b2 2411 &sync_blocks, 1);
6a806c51 2412 else /* completed sync */
191ea9b2 2413 conf->fullsync = 0;
6a806c51
N
2414
2415 bitmap_close_sync(mddev->bitmap);
1da177e4
LT
2416 close_sync(conf);
2417 return 0;
2418 }
2419
07d84d10
N
2420 if (mddev->bitmap == NULL &&
2421 mddev->recovery_cp == MaxSector &&
6394cca5 2422 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
07d84d10
N
2423 conf->fullsync == 0) {
2424 *skipped = 1;
2425 return max_sector - sector_nr;
2426 }
6394cca5
N
2427 /* before building a request, check if we can skip these blocks..
2428 * This call the bitmap_start_sync doesn't actually record anything
2429 */
e3b9703e 2430 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
e5de485f 2431 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
191ea9b2
N
2432 /* We can skip this block, and probably several more */
2433 *skipped = 1;
2434 return sync_blocks;
2435 }
1da177e4 2436 /*
17999be4
N
2437 * If there is non-resync activity waiting for a turn,
2438 * and resync is going fast enough,
2439 * then let it though before starting on this new sync request.
1da177e4 2440 */
17999be4 2441 if (!go_faster && conf->nr_waiting)
1da177e4 2442 msleep_interruptible(1000);
17999be4 2443
b47490c9 2444 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
1c4588e9 2445 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
17999be4
N
2446 raise_barrier(conf);
2447
2448 conf->next_resync = sector_nr;
1da177e4 2449
3e198f78 2450 rcu_read_lock();
1da177e4 2451 /*
3e198f78
N
2452 * If we get a correctably read error during resync or recovery,
2453 * we might want to read from a different device. So we
2454 * flag all drives that could conceivably be read from for READ,
2455 * and any others (which will be non-In_sync devices) for WRITE.
2456 * If a read fails, we try reading from something else for which READ
2457 * is OK.
1da177e4 2458 */
1da177e4 2459
1da177e4
LT
2460 r1_bio->mddev = mddev;
2461 r1_bio->sector = sector_nr;
191ea9b2 2462 r1_bio->state = 0;
1da177e4 2463 set_bit(R1BIO_IsSync, &r1_bio->state);
1da177e4 2464
8f19ccb2 2465 for (i = 0; i < conf->raid_disks * 2; i++) {
3cb03002 2466 struct md_rdev *rdev;
1da177e4 2467 bio = r1_bio->bios[i];
2aabaa65 2468 bio_reset(bio);
1da177e4 2469
3e198f78
N
2470 rdev = rcu_dereference(conf->mirrors[i].rdev);
2471 if (rdev == NULL ||
06f60385 2472 test_bit(Faulty, &rdev->flags)) {
8f19ccb2
N
2473 if (i < conf->raid_disks)
2474 still_degraded = 1;
3e198f78 2475 } else if (!test_bit(In_sync, &rdev->flags)) {
1da177e4
LT
2476 bio->bi_rw = WRITE;
2477 bio->bi_end_io = end_sync_write;
2478 write_targets ++;
3e198f78
N
2479 } else {
2480 /* may need to read from here */
06f60385
N
2481 sector_t first_bad = MaxSector;
2482 int bad_sectors;
2483
2484 if (is_badblock(rdev, sector_nr, good_sectors,
2485 &first_bad, &bad_sectors)) {
2486 if (first_bad > sector_nr)
2487 good_sectors = first_bad - sector_nr;
2488 else {
2489 bad_sectors -= (sector_nr - first_bad);
2490 if (min_bad == 0 ||
2491 min_bad > bad_sectors)
2492 min_bad = bad_sectors;
2493 }
2494 }
2495 if (sector_nr < first_bad) {
2496 if (test_bit(WriteMostly, &rdev->flags)) {
2497 if (wonly < 0)
2498 wonly = i;
2499 } else {
2500 if (disk < 0)
2501 disk = i;
2502 }
2503 bio->bi_rw = READ;
2504 bio->bi_end_io = end_sync_read;
2505 read_targets++;
d57368af
AL
2506 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2507 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2508 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2509 /*
2510 * The device is suitable for reading (InSync),
2511 * but has bad block(s) here. Let's try to correct them,
2512 * if we are doing resync or repair. Otherwise, leave
2513 * this device alone for this sync request.
2514 */
2515 bio->bi_rw = WRITE;
2516 bio->bi_end_io = end_sync_write;
2517 write_targets++;
3e198f78 2518 }
3e198f78 2519 }
06f60385
N
2520 if (bio->bi_end_io) {
2521 atomic_inc(&rdev->nr_pending);
2522 bio->bi_sector = sector_nr + rdev->data_offset;
2523 bio->bi_bdev = rdev->bdev;
2524 bio->bi_private = r1_bio;
2525 }
1da177e4 2526 }
3e198f78
N
2527 rcu_read_unlock();
2528 if (disk < 0)
2529 disk = wonly;
2530 r1_bio->read_disk = disk;
191ea9b2 2531
06f60385
N
2532 if (read_targets == 0 && min_bad > 0) {
2533 /* These sectors are bad on all InSync devices, so we
2534 * need to mark them bad on all write targets
2535 */
2536 int ok = 1;
8f19ccb2 2537 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
06f60385 2538 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
a42f9d83 2539 struct md_rdev *rdev = conf->mirrors[i].rdev;
06f60385
N
2540 ok = rdev_set_badblocks(rdev, sector_nr,
2541 min_bad, 0
2542 ) && ok;
2543 }
2544 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2545 *skipped = 1;
2546 put_buf(r1_bio);
2547
2548 if (!ok) {
2549 /* Cannot record the badblocks, so need to
2550 * abort the resync.
2551 * If there are multiple read targets, could just
2552 * fail the really bad ones ???
2553 */
2554 conf->recovery_disabled = mddev->recovery_disabled;
2555 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2556 return 0;
2557 } else
2558 return min_bad;
2559
2560 }
2561 if (min_bad > 0 && min_bad < good_sectors) {
2562 /* only resync enough to reach the next bad->good
2563 * transition */
2564 good_sectors = min_bad;
2565 }
2566
3e198f78
N
2567 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2568 /* extra read targets are also write targets */
2569 write_targets += read_targets-1;
2570
2571 if (write_targets == 0 || read_targets == 0) {
1da177e4
LT
2572 /* There is nowhere to write, so all non-sync
2573 * drives must be failed - so we are finished
2574 */
b7219ccb
N
2575 sector_t rv;
2576 if (min_bad > 0)
2577 max_sector = sector_nr + min_bad;
2578 rv = max_sector - sector_nr;
57afd89f 2579 *skipped = 1;
1da177e4 2580 put_buf(r1_bio);
1da177e4
LT
2581 return rv;
2582 }
2583
c6207277
N
2584 if (max_sector > mddev->resync_max)
2585 max_sector = mddev->resync_max; /* Don't do IO beyond here */
06f60385
N
2586 if (max_sector > sector_nr + good_sectors)
2587 max_sector = sector_nr + good_sectors;
1da177e4 2588 nr_sectors = 0;
289e99e8 2589 sync_blocks = 0;
1da177e4
LT
2590 do {
2591 struct page *page;
2592 int len = PAGE_SIZE;
2593 if (sector_nr + (len>>9) > max_sector)
2594 len = (max_sector - sector_nr) << 9;
2595 if (len == 0)
2596 break;
6a806c51
N
2597 if (sync_blocks == 0) {
2598 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
e5de485f
N
2599 &sync_blocks, still_degraded) &&
2600 !conf->fullsync &&
2601 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6a806c51 2602 break;
9e77c485 2603 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
7571ae88 2604 if ((len >> 9) > sync_blocks)
6a806c51 2605 len = sync_blocks<<9;
ab7a30c7 2606 }
191ea9b2 2607
8f19ccb2 2608 for (i = 0 ; i < conf->raid_disks * 2; i++) {
1da177e4
LT
2609 bio = r1_bio->bios[i];
2610 if (bio->bi_end_io) {
d11c171e 2611 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
1da177e4
LT
2612 if (bio_add_page(bio, page, len, 0) == 0) {
2613 /* stop here */
d11c171e 2614 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
1da177e4
LT
2615 while (i > 0) {
2616 i--;
2617 bio = r1_bio->bios[i];
6a806c51
N
2618 if (bio->bi_end_io==NULL)
2619 continue;
1da177e4
LT
2620 /* remove last page from this bio */
2621 bio->bi_vcnt--;
2622 bio->bi_size -= len;
2623 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
2624 }
2625 goto bio_full;
2626 }
2627 }
2628 }
2629 nr_sectors += len>>9;
2630 sector_nr += len>>9;
191ea9b2 2631 sync_blocks -= (len>>9);
1da177e4
LT
2632 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2633 bio_full:
1da177e4
LT
2634 r1_bio->sectors = nr_sectors;
2635
d11c171e
N
2636 /* For a user-requested sync, we read all readable devices and do a
2637 * compare
2638 */
2639 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2640 atomic_set(&r1_bio->remaining, read_targets);
2d4f4f33 2641 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
d11c171e
N
2642 bio = r1_bio->bios[i];
2643 if (bio->bi_end_io == end_sync_read) {
2d4f4f33 2644 read_targets--;
ddac7c7e 2645 md_sync_acct(bio->bi_bdev, nr_sectors);
d11c171e
N
2646 generic_make_request(bio);
2647 }
2648 }
2649 } else {
2650 atomic_set(&r1_bio->remaining, 1);
2651 bio = r1_bio->bios[r1_bio->read_disk];
ddac7c7e 2652 md_sync_acct(bio->bi_bdev, nr_sectors);
d11c171e 2653 generic_make_request(bio);
1da177e4 2654
d11c171e 2655 }
1da177e4
LT
2656 return nr_sectors;
2657}
2658
fd01b88c 2659static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce
DW
2660{
2661 if (sectors)
2662 return sectors;
2663
2664 return mddev->dev_sectors;
2665}
2666
e8096360 2667static struct r1conf *setup_conf(struct mddev *mddev)
1da177e4 2668{
e8096360 2669 struct r1conf *conf;
709ae487 2670 int i;
0eaf822c 2671 struct raid1_info *disk;
3cb03002 2672 struct md_rdev *rdev;
709ae487 2673 int err = -ENOMEM;
1da177e4 2674
e8096360 2675 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
1da177e4 2676 if (!conf)
709ae487 2677 goto abort;
1da177e4 2678
0eaf822c 2679 conf->mirrors = kzalloc(sizeof(struct raid1_info)
8f19ccb2 2680 * mddev->raid_disks * 2,
1da177e4
LT
2681 GFP_KERNEL);
2682 if (!conf->mirrors)
709ae487 2683 goto abort;
1da177e4 2684
ddaf22ab
N
2685 conf->tmppage = alloc_page(GFP_KERNEL);
2686 if (!conf->tmppage)
709ae487 2687 goto abort;
ddaf22ab 2688
709ae487 2689 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1da177e4 2690 if (!conf->poolinfo)
709ae487 2691 goto abort;
8f19ccb2 2692 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
1da177e4
LT
2693 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2694 r1bio_pool_free,
2695 conf->poolinfo);
2696 if (!conf->r1bio_pool)
709ae487
N
2697 goto abort;
2698
ed9bfdf1 2699 conf->poolinfo->mddev = mddev;
1da177e4 2700
c19d5798 2701 err = -EINVAL;
e7e72bf6 2702 spin_lock_init(&conf->device_lock);
dafb20fa 2703 rdev_for_each(rdev, mddev) {
aba336bd 2704 struct request_queue *q;
709ae487 2705 int disk_idx = rdev->raid_disk;
1da177e4
LT
2706 if (disk_idx >= mddev->raid_disks
2707 || disk_idx < 0)
2708 continue;
c19d5798 2709 if (test_bit(Replacement, &rdev->flags))
02b898f2 2710 disk = conf->mirrors + mddev->raid_disks + disk_idx;
c19d5798
N
2711 else
2712 disk = conf->mirrors + disk_idx;
1da177e4 2713
c19d5798
N
2714 if (disk->rdev)
2715 goto abort;
1da177e4 2716 disk->rdev = rdev;
aba336bd
N
2717 q = bdev_get_queue(rdev->bdev);
2718 if (q->merge_bvec_fn)
2719 mddev->merge_check_needed = 1;
1da177e4
LT
2720
2721 disk->head_position = 0;
12cee5a8 2722 disk->seq_start = MaxSector;
1da177e4
LT
2723 }
2724 conf->raid_disks = mddev->raid_disks;
2725 conf->mddev = mddev;
1da177e4 2726 INIT_LIST_HEAD(&conf->retry_list);
1da177e4
LT
2727
2728 spin_lock_init(&conf->resync_lock);
17999be4 2729 init_waitqueue_head(&conf->wait_barrier);
1da177e4 2730
191ea9b2 2731 bio_list_init(&conf->pending_bio_list);
34db0cd6 2732 conf->pending_count = 0;
d890fa2b 2733 conf->recovery_disabled = mddev->recovery_disabled - 1;
191ea9b2 2734
c19d5798 2735 err = -EIO;
8f19ccb2 2736 for (i = 0; i < conf->raid_disks * 2; i++) {
1da177e4
LT
2737
2738 disk = conf->mirrors + i;
2739
c19d5798
N
2740 if (i < conf->raid_disks &&
2741 disk[conf->raid_disks].rdev) {
2742 /* This slot has a replacement. */
2743 if (!disk->rdev) {
2744 /* No original, just make the replacement
2745 * a recovering spare
2746 */
2747 disk->rdev =
2748 disk[conf->raid_disks].rdev;
2749 disk[conf->raid_disks].rdev = NULL;
2750 } else if (!test_bit(In_sync, &disk->rdev->flags))
2751 /* Original is not in_sync - bad */
2752 goto abort;
2753 }
2754
5fd6c1dc
N
2755 if (!disk->rdev ||
2756 !test_bit(In_sync, &disk->rdev->flags)) {
1da177e4 2757 disk->head_position = 0;
4f0a5e01
JB
2758 if (disk->rdev &&
2759 (disk->rdev->saved_raid_disk < 0))
918f0238 2760 conf->fullsync = 1;
be4d3280 2761 }
1da177e4 2762 }
709ae487 2763
709ae487 2764 err = -ENOMEM;
0232605d 2765 conf->thread = md_register_thread(raid1d, mddev, "raid1");
709ae487
N
2766 if (!conf->thread) {
2767 printk(KERN_ERR
9dd1e2fa 2768 "md/raid1:%s: couldn't allocate thread\n",
709ae487
N
2769 mdname(mddev));
2770 goto abort;
11ce99e6 2771 }
1da177e4 2772
709ae487
N
2773 return conf;
2774
2775 abort:
2776 if (conf) {
2777 if (conf->r1bio_pool)
2778 mempool_destroy(conf->r1bio_pool);
2779 kfree(conf->mirrors);
2780 safe_put_page(conf->tmppage);
2781 kfree(conf->poolinfo);
2782 kfree(conf);
2783 }
2784 return ERR_PTR(err);
2785}
2786
5220ea1e 2787static int stop(struct mddev *mddev);
fd01b88c 2788static int run(struct mddev *mddev)
709ae487 2789{
e8096360 2790 struct r1conf *conf;
709ae487 2791 int i;
3cb03002 2792 struct md_rdev *rdev;
5220ea1e 2793 int ret;
2ff8cc2c 2794 bool discard_supported = false;
709ae487
N
2795
2796 if (mddev->level != 1) {
9dd1e2fa 2797 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
709ae487
N
2798 mdname(mddev), mddev->level);
2799 return -EIO;
2800 }
2801 if (mddev->reshape_position != MaxSector) {
9dd1e2fa 2802 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
709ae487
N
2803 mdname(mddev));
2804 return -EIO;
2805 }
1da177e4 2806 /*
709ae487
N
2807 * copy the already verified devices into our private RAID1
2808 * bookkeeping area. [whatever we allocate in run(),
2809 * should be freed in stop()]
1da177e4 2810 */
709ae487
N
2811 if (mddev->private == NULL)
2812 conf = setup_conf(mddev);
2813 else
2814 conf = mddev->private;
1da177e4 2815
709ae487
N
2816 if (IS_ERR(conf))
2817 return PTR_ERR(conf);
1da177e4 2818
c8dc9c65 2819 if (mddev->queue)
5026d7a9
PA
2820 blk_queue_max_write_same_sectors(mddev->queue, 0);
2821
dafb20fa 2822 rdev_for_each(rdev, mddev) {
1ed7242e
JB
2823 if (!mddev->gendisk)
2824 continue;
709ae487
N
2825 disk_stack_limits(mddev->gendisk, rdev->bdev,
2826 rdev->data_offset << 9);
2ff8cc2c
SL
2827 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2828 discard_supported = true;
1da177e4 2829 }
191ea9b2 2830
709ae487
N
2831 mddev->degraded = 0;
2832 for (i=0; i < conf->raid_disks; i++)
2833 if (conf->mirrors[i].rdev == NULL ||
2834 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2835 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2836 mddev->degraded++;
2837
2838 if (conf->raid_disks - mddev->degraded == 1)
2839 mddev->recovery_cp = MaxSector;
2840
8c6ac868 2841 if (mddev->recovery_cp != MaxSector)
9dd1e2fa 2842 printk(KERN_NOTICE "md/raid1:%s: not clean"
8c6ac868
AN
2843 " -- starting background reconstruction\n",
2844 mdname(mddev));
1da177e4 2845 printk(KERN_INFO
9dd1e2fa 2846 "md/raid1:%s: active with %d out of %d mirrors\n",
1da177e4
LT
2847 mdname(mddev), mddev->raid_disks - mddev->degraded,
2848 mddev->raid_disks);
709ae487 2849
1da177e4
LT
2850 /*
2851 * Ok, everything is just fine now
2852 */
709ae487
N
2853 mddev->thread = conf->thread;
2854 conf->thread = NULL;
2855 mddev->private = conf;
2856
1f403624 2857 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
1da177e4 2858
1ed7242e
JB
2859 if (mddev->queue) {
2860 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2861 mddev->queue->backing_dev_info.congested_data = mddev;
6b740b8d 2862 blk_queue_merge_bvec(mddev->queue, raid1_mergeable_bvec);
2ff8cc2c
SL
2863
2864 if (discard_supported)
2865 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
2866 mddev->queue);
2867 else
2868 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
2869 mddev->queue);
1ed7242e 2870 }
5220ea1e 2871
2872 ret = md_integrity_register(mddev);
2873 if (ret)
2874 stop(mddev);
2875 return ret;
1da177e4
LT
2876}
2877
fd01b88c 2878static int stop(struct mddev *mddev)
1da177e4 2879{
e8096360 2880 struct r1conf *conf = mddev->private;
4b6d287f 2881 struct bitmap *bitmap = mddev->bitmap;
4b6d287f
N
2882
2883 /* wait for behind writes to complete */
e555190d 2884 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
9dd1e2fa
N
2885 printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n",
2886 mdname(mddev));
4b6d287f 2887 /* need to kick something here to make sure I/O goes? */
e555190d
N
2888 wait_event(bitmap->behind_wait,
2889 atomic_read(&bitmap->behind_writes) == 0);
4b6d287f 2890 }
1da177e4 2891
409c57f3
N
2892 raise_barrier(conf);
2893 lower_barrier(conf);
2894
01f96c0a 2895 md_unregister_thread(&mddev->thread);
1da177e4
LT
2896 if (conf->r1bio_pool)
2897 mempool_destroy(conf->r1bio_pool);
990a8baf 2898 kfree(conf->mirrors);
0fea7ed8 2899 safe_put_page(conf->tmppage);
990a8baf 2900 kfree(conf->poolinfo);
1da177e4
LT
2901 kfree(conf);
2902 mddev->private = NULL;
2903 return 0;
2904}
2905
fd01b88c 2906static int raid1_resize(struct mddev *mddev, sector_t sectors)
1da177e4
LT
2907{
2908 /* no resync is happening, and there is enough space
2909 * on all devices, so we can resize.
2910 * We need to make sure resync covers any new space.
2911 * If the array is shrinking we should possibly wait until
2912 * any io in the removed space completes, but it hardly seems
2913 * worth it.
2914 */
a4a6125a
N
2915 sector_t newsize = raid1_size(mddev, sectors, 0);
2916 if (mddev->external_size &&
2917 mddev->array_sectors > newsize)
b522adcd 2918 return -EINVAL;
a4a6125a
N
2919 if (mddev->bitmap) {
2920 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
2921 if (ret)
2922 return ret;
2923 }
2924 md_set_array_sectors(mddev, newsize);
f233ea5c 2925 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 2926 revalidate_disk(mddev->gendisk);
b522adcd 2927 if (sectors > mddev->dev_sectors &&
b098636c 2928 mddev->recovery_cp > mddev->dev_sectors) {
58c0fed4 2929 mddev->recovery_cp = mddev->dev_sectors;
1da177e4
LT
2930 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2931 }
b522adcd 2932 mddev->dev_sectors = sectors;
4b5c7ae8 2933 mddev->resync_max_sectors = sectors;
1da177e4
LT
2934 return 0;
2935}
2936
fd01b88c 2937static int raid1_reshape(struct mddev *mddev)
1da177e4
LT
2938{
2939 /* We need to:
2940 * 1/ resize the r1bio_pool
2941 * 2/ resize conf->mirrors
2942 *
2943 * We allocate a new r1bio_pool if we can.
2944 * Then raise a device barrier and wait until all IO stops.
2945 * Then resize conf->mirrors and swap in the new r1bio pool.
6ea9c07c
N
2946 *
2947 * At the same time, we "pack" the devices so that all the missing
2948 * devices have the higher raid_disk numbers.
1da177e4
LT
2949 */
2950 mempool_t *newpool, *oldpool;
2951 struct pool_info *newpoolinfo;
0eaf822c 2952 struct raid1_info *newmirrors;
e8096360 2953 struct r1conf *conf = mddev->private;
63c70c4f 2954 int cnt, raid_disks;
c04be0aa 2955 unsigned long flags;
b5470dc5 2956 int d, d2, err;
1da177e4 2957
63c70c4f 2958 /* Cannot change chunk_size, layout, or level */
664e7c41 2959 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
63c70c4f
N
2960 mddev->layout != mddev->new_layout ||
2961 mddev->level != mddev->new_level) {
664e7c41 2962 mddev->new_chunk_sectors = mddev->chunk_sectors;
63c70c4f
N
2963 mddev->new_layout = mddev->layout;
2964 mddev->new_level = mddev->level;
2965 return -EINVAL;
2966 }
2967
b5470dc5
DW
2968 err = md_allow_write(mddev);
2969 if (err)
2970 return err;
2a2275d6 2971
63c70c4f
N
2972 raid_disks = mddev->raid_disks + mddev->delta_disks;
2973
6ea9c07c
N
2974 if (raid_disks < conf->raid_disks) {
2975 cnt=0;
2976 for (d= 0; d < conf->raid_disks; d++)
2977 if (conf->mirrors[d].rdev)
2978 cnt++;
2979 if (cnt > raid_disks)
1da177e4 2980 return -EBUSY;
6ea9c07c 2981 }
1da177e4
LT
2982
2983 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2984 if (!newpoolinfo)
2985 return -ENOMEM;
2986 newpoolinfo->mddev = mddev;
8f19ccb2 2987 newpoolinfo->raid_disks = raid_disks * 2;
1da177e4
LT
2988
2989 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2990 r1bio_pool_free, newpoolinfo);
2991 if (!newpool) {
2992 kfree(newpoolinfo);
2993 return -ENOMEM;
2994 }
0eaf822c 2995 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
8f19ccb2 2996 GFP_KERNEL);
1da177e4
LT
2997 if (!newmirrors) {
2998 kfree(newpoolinfo);
2999 mempool_destroy(newpool);
3000 return -ENOMEM;
3001 }
1da177e4 3002
e2d59925 3003 freeze_array(conf, 0);
1da177e4
LT
3004
3005 /* ok, everything is stopped */
3006 oldpool = conf->r1bio_pool;
3007 conf->r1bio_pool = newpool;
6ea9c07c 3008
a88aa786 3009 for (d = d2 = 0; d < conf->raid_disks; d++) {
3cb03002 3010 struct md_rdev *rdev = conf->mirrors[d].rdev;
a88aa786 3011 if (rdev && rdev->raid_disk != d2) {
36fad858 3012 sysfs_unlink_rdev(mddev, rdev);
a88aa786 3013 rdev->raid_disk = d2;
36fad858
NK
3014 sysfs_unlink_rdev(mddev, rdev);
3015 if (sysfs_link_rdev(mddev, rdev))
a88aa786 3016 printk(KERN_WARNING
36fad858
NK
3017 "md/raid1:%s: cannot register rd%d\n",
3018 mdname(mddev), rdev->raid_disk);
6ea9c07c 3019 }
a88aa786
N
3020 if (rdev)
3021 newmirrors[d2++].rdev = rdev;
3022 }
1da177e4
LT
3023 kfree(conf->mirrors);
3024 conf->mirrors = newmirrors;
3025 kfree(conf->poolinfo);
3026 conf->poolinfo = newpoolinfo;
3027
c04be0aa 3028 spin_lock_irqsave(&conf->device_lock, flags);
1da177e4 3029 mddev->degraded += (raid_disks - conf->raid_disks);
c04be0aa 3030 spin_unlock_irqrestore(&conf->device_lock, flags);
1da177e4 3031 conf->raid_disks = mddev->raid_disks = raid_disks;
63c70c4f 3032 mddev->delta_disks = 0;
1da177e4 3033
e2d59925 3034 unfreeze_array(conf);
1da177e4
LT
3035
3036 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3037 md_wakeup_thread(mddev->thread);
3038
3039 mempool_destroy(oldpool);
3040 return 0;
3041}
3042
fd01b88c 3043static void raid1_quiesce(struct mddev *mddev, int state)
36fa3063 3044{
e8096360 3045 struct r1conf *conf = mddev->private;
36fa3063
N
3046
3047 switch(state) {
6eef4b21
N
3048 case 2: /* wake for suspend */
3049 wake_up(&conf->wait_barrier);
3050 break;
9e6603da 3051 case 1:
17999be4 3052 raise_barrier(conf);
36fa3063 3053 break;
9e6603da 3054 case 0:
17999be4 3055 lower_barrier(conf);
36fa3063
N
3056 break;
3057 }
36fa3063
N
3058}
3059
fd01b88c 3060static void *raid1_takeover(struct mddev *mddev)
709ae487
N
3061{
3062 /* raid1 can take over:
3063 * raid5 with 2 devices, any layout or chunk size
3064 */
3065 if (mddev->level == 5 && mddev->raid_disks == 2) {
e8096360 3066 struct r1conf *conf;
709ae487
N
3067 mddev->new_level = 1;
3068 mddev->new_layout = 0;
3069 mddev->new_chunk_sectors = 0;
3070 conf = setup_conf(mddev);
3071 if (!IS_ERR(conf))
3072 conf->barrier = 1;
3073 return conf;
3074 }
3075 return ERR_PTR(-EINVAL);
3076}
1da177e4 3077
84fc4b56 3078static struct md_personality raid1_personality =
1da177e4
LT
3079{
3080 .name = "raid1",
2604b703 3081 .level = 1,
1da177e4
LT
3082 .owner = THIS_MODULE,
3083 .make_request = make_request,
3084 .run = run,
3085 .stop = stop,
3086 .status = status,
3087 .error_handler = error,
3088 .hot_add_disk = raid1_add_disk,
3089 .hot_remove_disk= raid1_remove_disk,
3090 .spare_active = raid1_spare_active,
3091 .sync_request = sync_request,
3092 .resize = raid1_resize,
80c3a6ce 3093 .size = raid1_size,
63c70c4f 3094 .check_reshape = raid1_reshape,
36fa3063 3095 .quiesce = raid1_quiesce,
709ae487 3096 .takeover = raid1_takeover,
1da177e4
LT
3097};
3098
3099static int __init raid_init(void)
3100{
2604b703 3101 return register_md_personality(&raid1_personality);
1da177e4
LT
3102}
3103
3104static void raid_exit(void)
3105{
2604b703 3106 unregister_md_personality(&raid1_personality);
1da177e4
LT
3107}
3108
3109module_init(raid_init);
3110module_exit(raid_exit);
3111MODULE_LICENSE("GPL");
0efb9e61 3112MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
1da177e4 3113MODULE_ALIAS("md-personality-3"); /* RAID1 */
d9d166c2 3114MODULE_ALIAS("md-raid1");
2604b703 3115MODULE_ALIAS("md-level-1");
34db0cd6
N
3116
3117module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);