blob: f33a729960ca3f58e4fd618605ca1c197523ef11 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
14 *
NeilBrown191ea9b2005-06-21 17:17:23 -070015 * 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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * 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
NeilBrown191ea9b2005-06-21 17:17:23 -070034#include "dm-bio-list.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/raid/raid1.h>
NeilBrown191ea9b2005-06-21 17:17:23 -070036#include <linux/raid/bitmap.h>
37
38#define DEBUG 0
39#if DEBUG
40#define PRINTK(x...) printk(x)
41#else
42#define PRINTK(x...)
43#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45/*
46 * Number of guaranteed r1bios in case of extreme VM load:
47 */
48#define NR_RAID1_BIOS 256
49
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51static void unplug_slaves(mddev_t *mddev);
52
NeilBrown17999be2006-01-06 00:20:12 -080053static void allow_barrier(conf_t *conf);
54static void lower_barrier(conf_t *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Al Virodd0fc662005-10-07 07:46:04 +010056static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057{
58 struct pool_info *pi = data;
59 r1bio_t *r1_bio;
60 int size = offsetof(r1bio_t, bios[pi->raid_disks]);
61
62 /* allocate a r1bio with room for raid_disks entries in the bios array */
NeilBrown9ffae0c2006-01-06 00:20:32 -080063 r1_bio = kzalloc(size, gfp_flags);
64 if (!r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 unplug_slaves(pi->mddev);
66
67 return r1_bio;
68}
69
70static void r1bio_pool_free(void *r1_bio, void *data)
71{
72 kfree(r1_bio);
73}
74
75#define RESYNC_BLOCK_SIZE (64*1024)
76//#define RESYNC_BLOCK_SIZE PAGE_SIZE
77#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
78#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
79#define RESYNC_WINDOW (2048*1024)
80
Al Virodd0fc662005-10-07 07:46:04 +010081static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
83 struct pool_info *pi = data;
84 struct page *page;
85 r1bio_t *r1_bio;
86 struct bio *bio;
87 int i, j;
88
89 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
90 if (!r1_bio) {
91 unplug_slaves(pi->mddev);
92 return NULL;
93 }
94
95 /*
96 * Allocate bios : 1 for reading, n-1 for writing
97 */
98 for (j = pi->raid_disks ; j-- ; ) {
99 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
100 if (!bio)
101 goto out_free_bio;
102 r1_bio->bios[j] = bio;
103 }
104 /*
105 * Allocate RESYNC_PAGES data pages and attach them to
NeilBrownd11c1712006-01-06 00:20:26 -0800106 * the first bio.
107 * If this is a user-requested check/repair, allocate
108 * RESYNC_PAGES for each bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 */
NeilBrownd11c1712006-01-06 00:20:26 -0800110 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
111 j = pi->raid_disks;
112 else
113 j = 1;
114 while(j--) {
115 bio = r1_bio->bios[j];
116 for (i = 0; i < RESYNC_PAGES; i++) {
117 page = alloc_page(gfp_flags);
118 if (unlikely(!page))
119 goto out_free_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
NeilBrownd11c1712006-01-06 00:20:26 -0800121 bio->bi_io_vec[i].bv_page = page;
122 }
123 }
124 /* If not user-requests, copy the page pointers to all bios */
125 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
126 for (i=0; i<RESYNC_PAGES ; i++)
127 for (j=1; j<pi->raid_disks; j++)
128 r1_bio->bios[j]->bi_io_vec[i].bv_page =
129 r1_bio->bios[0]->bi_io_vec[i].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 }
131
132 r1_bio->master_bio = NULL;
133
134 return r1_bio;
135
136out_free_pages:
NeilBrownd11c1712006-01-06 00:20:26 -0800137 for (i=0; i < RESYNC_PAGES ; i++)
138 for (j=0 ; j < pi->raid_disks; j++)
NeilBrown1345b1d2006-01-06 00:20:40 -0800139 safe_put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800140 j = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141out_free_bio:
142 while ( ++j < pi->raid_disks )
143 bio_put(r1_bio->bios[j]);
144 r1bio_pool_free(r1_bio, data);
145 return NULL;
146}
147
148static void r1buf_pool_free(void *__r1_bio, void *data)
149{
150 struct pool_info *pi = data;
NeilBrownd11c1712006-01-06 00:20:26 -0800151 int i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 r1bio_t *r1bio = __r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
NeilBrownd11c1712006-01-06 00:20:26 -0800154 for (i = 0; i < RESYNC_PAGES; i++)
155 for (j = pi->raid_disks; j-- ;) {
156 if (j == 0 ||
157 r1bio->bios[j]->bi_io_vec[i].bv_page !=
158 r1bio->bios[0]->bi_io_vec[i].bv_page)
NeilBrown1345b1d2006-01-06 00:20:40 -0800159 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 for (i=0 ; i < pi->raid_disks; i++)
162 bio_put(r1bio->bios[i]);
163
164 r1bio_pool_free(r1bio, data);
165}
166
167static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
168{
169 int i;
170
171 for (i = 0; i < conf->raid_disks; i++) {
172 struct bio **bio = r1_bio->bios + i;
NeilBrowncf30a472006-01-06 00:20:23 -0800173 if (*bio && *bio != IO_BLOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 bio_put(*bio);
175 *bio = NULL;
176 }
177}
178
Arjan van de Ven858119e2006-01-14 13:20:43 -0800179static void free_r1bio(r1bio_t *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 conf_t *conf = mddev_to_conf(r1_bio->mddev);
182
183 /*
184 * Wake up any possible resync thread that waits for the device
185 * to go idle.
186 */
NeilBrown17999be2006-01-06 00:20:12 -0800187 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 put_all_bios(conf, r1_bio);
190 mempool_free(r1_bio, conf->r1bio_pool);
191}
192
Arjan van de Ven858119e2006-01-14 13:20:43 -0800193static void put_buf(r1bio_t *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194{
195 conf_t *conf = mddev_to_conf(r1_bio->mddev);
NeilBrown3e198f72006-01-06 00:20:21 -0800196 int i;
197
198 for (i=0; i<conf->raid_disks; i++) {
199 struct bio *bio = r1_bio->bios[i];
200 if (bio->bi_end_io)
201 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204 mempool_free(r1_bio, conf->r1buf_pool);
205
NeilBrown17999be2006-01-06 00:20:12 -0800206 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209static void reschedule_retry(r1bio_t *r1_bio)
210{
211 unsigned long flags;
212 mddev_t *mddev = r1_bio->mddev;
213 conf_t *conf = mddev_to_conf(mddev);
214
215 spin_lock_irqsave(&conf->device_lock, flags);
216 list_add(&r1_bio->retry_list, &conf->retry_list);
NeilBrownddaf22a2006-01-06 00:20:19 -0800217 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 spin_unlock_irqrestore(&conf->device_lock, flags);
219
NeilBrown17999be2006-01-06 00:20:12 -0800220 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 md_wakeup_thread(mddev->thread);
222}
223
224/*
225 * raid_end_bio_io() is called when we have finished servicing a mirrored
226 * operation and are ready to return a success/failure code to the buffer
227 * cache layer.
228 */
229static void raid_end_bio_io(r1bio_t *r1_bio)
230{
231 struct bio *bio = r1_bio->master_bio;
232
NeilBrown4b6d2872005-09-09 16:23:47 -0700233 /* if nobody has done the final endio yet, do it now */
234 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
235 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
236 (bio_data_dir(bio) == WRITE) ? "write" : "read",
237 (unsigned long long) bio->bi_sector,
238 (unsigned long long) bio->bi_sector +
239 (bio->bi_size >> 9) - 1);
240
241 bio_endio(bio, bio->bi_size,
242 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 free_r1bio(r1_bio);
245}
246
247/*
248 * Update disk head position estimator based on IRQ completion info.
249 */
250static inline void update_head_pos(int disk, r1bio_t *r1_bio)
251{
252 conf_t *conf = mddev_to_conf(r1_bio->mddev);
253
254 conf->mirrors[disk].head_position =
255 r1_bio->sector + (r1_bio->sectors);
256}
257
258static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error)
259{
260 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
261 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
262 int mirror;
263 conf_t *conf = mddev_to_conf(r1_bio->mddev);
264
265 if (bio->bi_size)
266 return 1;
267
268 mirror = r1_bio->read_disk;
269 /*
270 * this branch is our 'one mirror IO has finished' event handler:
271 */
NeilBrownddaf22a2006-01-06 00:20:19 -0800272 update_head_pos(mirror, r1_bio);
273
NeilBrowndd00a992007-05-10 03:15:50 -0700274 if (uptodate)
275 set_bit(R1BIO_Uptodate, &r1_bio->state);
276 else {
277 /* If all other devices have failed, we want to return
278 * the error upwards rather than fail the last device.
279 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 */
NeilBrowndd00a992007-05-10 03:15:50 -0700281 unsigned long flags;
282 spin_lock_irqsave(&conf->device_lock, flags);
283 if (r1_bio->mddev->degraded == conf->raid_disks ||
284 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
285 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)))
286 uptodate = 1;
287 spin_unlock_irqrestore(&conf->device_lock, flags);
288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
NeilBrowndd00a992007-05-10 03:15:50 -0700290 if (uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 raid_end_bio_io(r1_bio);
NeilBrowndd00a992007-05-10 03:15:50 -0700292 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 /*
294 * oops, read error:
295 */
296 char b[BDEVNAME_SIZE];
297 if (printk_ratelimit())
298 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
299 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
300 reschedule_retry(r1_bio);
301 }
302
303 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
304 return 0;
305}
306
307static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
308{
309 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
310 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
NeilBrowna9701a32005-11-08 21:39:34 -0800311 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 conf_t *conf = mddev_to_conf(r1_bio->mddev);
NeilBrown04b857f2006-03-09 17:33:46 -0800313 struct bio *to_put = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315 if (bio->bi_size)
316 return 1;
317
318 for (mirror = 0; mirror < conf->raid_disks; mirror++)
319 if (r1_bio->bios[mirror] == bio)
320 break;
321
NeilBrownbea27712006-05-01 12:15:46 -0700322 if (error == -EOPNOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
NeilBrowna9701a32005-11-08 21:39:34 -0800323 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
324 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
325 r1_bio->mddev->barriers_work = 0;
NeilBrown5e7dd2a2006-05-01 12:15:47 -0700326 /* Don't rdev_dec_pending in this branch - keep it for the retry */
NeilBrowna9701a32005-11-08 21:39:34 -0800327 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 /*
NeilBrowna9701a32005-11-08 21:39:34 -0800329 * this branch is our 'one mirror IO has finished' event handler:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 */
NeilBrowna9701a32005-11-08 21:39:34 -0800331 r1_bio->bios[mirror] = NULL;
NeilBrown04b857f2006-03-09 17:33:46 -0800332 to_put = bio;
NeilBrowna9701a32005-11-08 21:39:34 -0800333 if (!uptodate) {
334 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
335 /* an I/O failed, we can't clear the bitmap */
336 set_bit(R1BIO_Degraded, &r1_bio->state);
337 } else
338 /*
339 * Set R1BIO_Uptodate in our master bio, so that
340 * we will return a good error code for to the higher
341 * levels even if IO on some other mirrored buffer fails.
342 *
343 * The 'master' represents the composite IO operation to
344 * user-side. So if something waits for IO, then it will
345 * wait for the 'master' bio.
346 */
347 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
NeilBrowna9701a32005-11-08 21:39:34 -0800349 update_head_pos(mirror, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
NeilBrowna9701a32005-11-08 21:39:34 -0800351 if (behind) {
352 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
353 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700354
NeilBrowna9701a32005-11-08 21:39:34 -0800355 /* In behind mode, we ACK the master bio once the I/O has safely
356 * reached all non-writemostly disks. Setting the Returned bit
357 * ensures that this gets done only once -- we don't ever want to
358 * return -EIO here, instead we'll wait */
NeilBrown4b6d2872005-09-09 16:23:47 -0700359
NeilBrowna9701a32005-11-08 21:39:34 -0800360 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
361 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
362 /* Maybe we can return now */
363 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
364 struct bio *mbio = r1_bio->master_bio;
365 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
366 (unsigned long long) mbio->bi_sector,
367 (unsigned long long) mbio->bi_sector +
368 (mbio->bi_size >> 9) - 1);
369 bio_endio(mbio, mbio->bi_size, 0);
370 }
NeilBrown4b6d2872005-09-09 16:23:47 -0700371 }
372 }
NeilBrown5e7dd2a2006-05-01 12:15:47 -0700373 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
NeilBrown4b6d2872005-09-09 16:23:47 -0700374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 /*
376 *
377 * Let's see if all mirrored write operations have finished
378 * already.
379 */
380 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrownc70810b2006-06-26 00:27:35 -0700381 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state))
NeilBrowna9701a32005-11-08 21:39:34 -0800382 reschedule_retry(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700383 else {
384 /* it really is the end of this request */
385 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
386 /* free extra copy of the data pages */
387 int i = bio->bi_vcnt;
388 while (i--)
389 safe_put_page(bio->bi_io_vec[i].bv_page);
390 }
391 /* clear the bitmap if all writes complete successfully */
392 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
393 r1_bio->sectors,
394 !test_bit(R1BIO_Degraded, &r1_bio->state),
395 behind);
396 md_write_end(r1_bio->mddev);
397 raid_end_bio_io(r1_bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 }
NeilBrownc70810b2006-06-26 00:27:35 -0700400
NeilBrown04b857f2006-03-09 17:33:46 -0800401 if (to_put)
402 bio_put(to_put);
403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 return 0;
405}
406
407
408/*
409 * This routine returns the disk from which the requested read should
410 * be done. There is a per-array 'next expected sequential IO' sector
411 * number - if this matches on the next IO then we use the last disk.
412 * There is also a per-disk 'last know head position' sector that is
413 * maintained from IRQ contexts, both the normal and the resync IO
414 * completion handlers update this position correctly. If there is no
415 * perfect sequential match then we pick the disk whose head is closest.
416 *
417 * If there are 2 mirrors in the same 2 devices, performance degrades
418 * because position is mirror, not device based.
419 *
420 * The rdev for the device selected will have nr_pending incremented.
421 */
422static int read_balance(conf_t *conf, r1bio_t *r1_bio)
423{
424 const unsigned long this_sector = r1_bio->sector;
425 int new_disk = conf->last_used, disk = new_disk;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700426 int wonly_disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 const int sectors = r1_bio->sectors;
428 sector_t new_distance, current_distance;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700429 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431 rcu_read_lock();
432 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700433 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 * device if no resync is going on, or below the resync window.
435 * We take the first readable disk when above the resync window.
436 */
437 retry:
438 if (conf->mddev->recovery_cp < MaxSector &&
439 (this_sector + sectors >= conf->next_resync)) {
440 /* Choose the first operation device, for consistancy */
441 new_disk = 0;
442
Suzanne Woodd6065f72005-11-08 21:39:27 -0800443 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrowncf30a472006-01-06 00:20:23 -0800444 r1_bio->bios[new_disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800445 !rdev || !test_bit(In_sync, &rdev->flags)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700446 || test_bit(WriteMostly, &rdev->flags);
Suzanne Woodd6065f72005-11-08 21:39:27 -0800447 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700448
NeilBrowncf30a472006-01-06 00:20:23 -0800449 if (rdev && test_bit(In_sync, &rdev->flags) &&
450 r1_bio->bios[new_disk] != IO_BLOCKED)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700451 wonly_disk = new_disk;
452
453 if (new_disk == conf->raid_disks - 1) {
454 new_disk = wonly_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 break;
456 }
457 }
458 goto rb_out;
459 }
460
461
462 /* make sure the disk is operational */
Suzanne Woodd6065f72005-11-08 21:39:27 -0800463 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrowncf30a472006-01-06 00:20:23 -0800464 r1_bio->bios[new_disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800465 !rdev || !test_bit(In_sync, &rdev->flags) ||
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700466 test_bit(WriteMostly, &rdev->flags);
Suzanne Woodd6065f72005-11-08 21:39:27 -0800467 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700468
NeilBrowncf30a472006-01-06 00:20:23 -0800469 if (rdev && test_bit(In_sync, &rdev->flags) &&
470 r1_bio->bios[new_disk] != IO_BLOCKED)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700471 wonly_disk = new_disk;
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 if (new_disk <= 0)
474 new_disk = conf->raid_disks;
475 new_disk--;
476 if (new_disk == disk) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700477 new_disk = wonly_disk;
478 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 }
480 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700481
482 if (new_disk < 0)
483 goto rb_out;
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 disk = new_disk;
486 /* now disk == new_disk == starting point for search */
487
488 /*
489 * Don't change to another disk for sequential reads:
490 */
491 if (conf->next_seq_sect == this_sector)
492 goto rb_out;
493 if (this_sector == conf->mirrors[new_disk].head_position)
494 goto rb_out;
495
496 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
497
498 /* Find the disk whose head is closest */
499
500 do {
501 if (disk <= 0)
502 disk = conf->raid_disks;
503 disk--;
504
Suzanne Woodd6065f72005-11-08 21:39:27 -0800505 rdev = rcu_dereference(conf->mirrors[disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700506
NeilBrowncf30a472006-01-06 00:20:23 -0800507 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800508 !test_bit(In_sync, &rdev->flags) ||
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700509 test_bit(WriteMostly, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 continue;
511
512 if (!atomic_read(&rdev->nr_pending)) {
513 new_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 break;
515 }
516 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
517 if (new_distance < current_distance) {
518 current_distance = new_distance;
519 new_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 }
521 } while (disk != conf->last_used);
522
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700523 rb_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525
526 if (new_disk >= 0) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800527 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700528 if (!rdev)
529 goto retry;
530 atomic_inc(&rdev->nr_pending);
NeilBrownb2d444d2005-11-08 21:39:31 -0800531 if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 /* cannot risk returning a device that failed
533 * before we inc'ed nr_pending
534 */
NeilBrown03c902e2006-01-06 00:20:46 -0800535 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 goto retry;
537 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700538 conf->next_seq_sect = this_sector + sectors;
539 conf->last_used = new_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 }
541 rcu_read_unlock();
542
543 return new_disk;
544}
545
546static void unplug_slaves(mddev_t *mddev)
547{
548 conf_t *conf = mddev_to_conf(mddev);
549 int i;
550
551 rcu_read_lock();
552 for (i=0; i<mddev->raid_disks; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800553 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -0800554 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200555 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557 atomic_inc(&rdev->nr_pending);
558 rcu_read_unlock();
559
560 if (r_queue->unplug_fn)
561 r_queue->unplug_fn(r_queue);
562
563 rdev_dec_pending(rdev, mddev);
564 rcu_read_lock();
565 }
566 }
567 rcu_read_unlock();
568}
569
Jens Axboe165125e2007-07-24 09:28:11 +0200570static void raid1_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
NeilBrown191ea9b2005-06-21 17:17:23 -0700572 mddev_t *mddev = q->queuedata;
573
574 unplug_slaves(mddev);
575 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576}
577
Jens Axboe165125e2007-07-24 09:28:11 +0200578static int raid1_issue_flush(struct request_queue *q, struct gendisk *disk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 sector_t *error_sector)
580{
581 mddev_t *mddev = q->queuedata;
582 conf_t *conf = mddev_to_conf(mddev);
583 int i, ret = 0;
584
585 rcu_read_lock();
586 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800587 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -0800588 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 struct block_device *bdev = rdev->bdev;
Jens Axboe165125e2007-07-24 09:28:11 +0200590 struct request_queue *r_queue = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 if (!r_queue->issue_flush_fn)
593 ret = -EOPNOTSUPP;
594 else {
595 atomic_inc(&rdev->nr_pending);
596 rcu_read_unlock();
597 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
598 error_sector);
599 rdev_dec_pending(rdev, mddev);
600 rcu_read_lock();
601 }
602 }
603 }
604 rcu_read_unlock();
605 return ret;
606}
607
NeilBrown0d129222006-10-03 01:15:54 -0700608static int raid1_congested(void *data, int bits)
609{
610 mddev_t *mddev = data;
611 conf_t *conf = mddev_to_conf(mddev);
612 int i, ret = 0;
613
614 rcu_read_lock();
615 for (i = 0; i < mddev->raid_disks; i++) {
616 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
617 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200618 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700619
620 /* Note the '|| 1' - when read_balance prefers
621 * non-congested targets, it can be removed
622 */
623 if ((bits & (1<<BDI_write_congested)) || 1)
624 ret |= bdi_congested(&q->backing_dev_info, bits);
625 else
626 ret &= bdi_congested(&q->backing_dev_info, bits);
627 }
628 }
629 rcu_read_unlock();
630 return ret;
631}
632
633
NeilBrown17999be2006-01-06 00:20:12 -0800634/* Barriers....
635 * Sometimes we need to suspend IO while we do something else,
636 * either some resync/recovery, or reconfigure the array.
637 * To do this we raise a 'barrier'.
638 * The 'barrier' is a counter that can be raised multiple times
639 * to count how many activities are happening which preclude
640 * normal IO.
641 * We can only raise the barrier if there is no pending IO.
642 * i.e. if nr_pending == 0.
643 * We choose only to raise the barrier if no-one is waiting for the
644 * barrier to go down. This means that as soon as an IO request
645 * is ready, no other operations which require a barrier will start
646 * until the IO request has had a chance.
647 *
648 * So: regular IO calls 'wait_barrier'. When that returns there
649 * is no backgroup IO happening, It must arrange to call
650 * allow_barrier when it has finished its IO.
651 * backgroup IO calls must call raise_barrier. Once that returns
652 * there is no normal IO happeing. It must arrange to call
653 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 */
655#define RESYNC_DEPTH 32
656
NeilBrown17999be2006-01-06 00:20:12 -0800657static void raise_barrier(conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
659 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800660
661 /* Wait until no block IO is waiting */
662 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
663 conf->resync_lock,
664 raid1_unplug(conf->mddev->queue));
665
666 /* block any new IO from starting */
667 conf->barrier++;
668
669 /* No wait for all pending IO to complete */
670 wait_event_lock_irq(conf->wait_barrier,
671 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
672 conf->resync_lock,
673 raid1_unplug(conf->mddev->queue));
674
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 spin_unlock_irq(&conf->resync_lock);
676}
677
NeilBrown17999be2006-01-06 00:20:12 -0800678static void lower_barrier(conf_t *conf)
679{
680 unsigned long flags;
681 spin_lock_irqsave(&conf->resync_lock, flags);
682 conf->barrier--;
683 spin_unlock_irqrestore(&conf->resync_lock, flags);
684 wake_up(&conf->wait_barrier);
685}
686
687static void wait_barrier(conf_t *conf)
688{
689 spin_lock_irq(&conf->resync_lock);
690 if (conf->barrier) {
691 conf->nr_waiting++;
692 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
693 conf->resync_lock,
694 raid1_unplug(conf->mddev->queue));
695 conf->nr_waiting--;
696 }
697 conf->nr_pending++;
698 spin_unlock_irq(&conf->resync_lock);
699}
700
701static void allow_barrier(conf_t *conf)
702{
703 unsigned long flags;
704 spin_lock_irqsave(&conf->resync_lock, flags);
705 conf->nr_pending--;
706 spin_unlock_irqrestore(&conf->resync_lock, flags);
707 wake_up(&conf->wait_barrier);
708}
709
NeilBrownddaf22a2006-01-06 00:20:19 -0800710static void freeze_array(conf_t *conf)
711{
712 /* stop syncio and normal IO and wait for everything to
713 * go quite.
714 * We increment barrier and nr_waiting, and then
715 * wait until barrier+nr_pending match nr_queued+2
716 */
717 spin_lock_irq(&conf->resync_lock);
718 conf->barrier++;
719 conf->nr_waiting++;
720 wait_event_lock_irq(conf->wait_barrier,
721 conf->barrier+conf->nr_pending == conf->nr_queued+2,
722 conf->resync_lock,
723 raid1_unplug(conf->mddev->queue));
724 spin_unlock_irq(&conf->resync_lock);
725}
726static void unfreeze_array(conf_t *conf)
727{
728 /* reverse the effect of the freeze */
729 spin_lock_irq(&conf->resync_lock);
730 conf->barrier--;
731 conf->nr_waiting--;
732 wake_up(&conf->wait_barrier);
733 spin_unlock_irq(&conf->resync_lock);
734}
735
NeilBrown17999be2006-01-06 00:20:12 -0800736
NeilBrown4b6d2872005-09-09 16:23:47 -0700737/* duplicate the data pages for behind I/O */
738static struct page **alloc_behind_pages(struct bio *bio)
739{
740 int i;
741 struct bio_vec *bvec;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800742 struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *),
NeilBrown4b6d2872005-09-09 16:23:47 -0700743 GFP_NOIO);
744 if (unlikely(!pages))
745 goto do_sync_io;
746
NeilBrown4b6d2872005-09-09 16:23:47 -0700747 bio_for_each_segment(bvec, bio, i) {
748 pages[i] = alloc_page(GFP_NOIO);
749 if (unlikely(!pages[i]))
750 goto do_sync_io;
751 memcpy(kmap(pages[i]) + bvec->bv_offset,
752 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
753 kunmap(pages[i]);
754 kunmap(bvec->bv_page);
755 }
756
757 return pages;
758
759do_sync_io:
760 if (pages)
761 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
NeilBrown2d1f3b52006-01-06 00:20:31 -0800762 put_page(pages[i]);
NeilBrown4b6d2872005-09-09 16:23:47 -0700763 kfree(pages);
764 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
765 return NULL;
766}
767
Jens Axboe165125e2007-07-24 09:28:11 +0200768static int make_request(struct request_queue *q, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
770 mddev_t *mddev = q->queuedata;
771 conf_t *conf = mddev_to_conf(mddev);
772 mirror_info_t *mirror;
773 r1bio_t *r1_bio;
774 struct bio *read_bio;
NeilBrown191ea9b2005-06-21 17:17:23 -0700775 int i, targets = 0, disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 mdk_rdev_t *rdev;
NeilBrown191ea9b2005-06-21 17:17:23 -0700777 struct bitmap *bitmap = mddev->bitmap;
778 unsigned long flags;
779 struct bio_list bl;
NeilBrown4b6d2872005-09-09 16:23:47 -0700780 struct page **behind_pages = NULL;
Jens Axboea3623572005-11-01 09:26:16 +0100781 const int rw = bio_data_dir(bio);
Lars Ellenberge3881a62007-01-10 23:15:37 -0800782 const int do_sync = bio_sync(bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800783 int do_barriers;
NeilBrown191ea9b2005-06-21 17:17:23 -0700784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 /*
786 * Register the new request and wait if the reconstruction
787 * thread has put up a bar for new requests.
788 * Continue immediately if no resync is active currently.
NeilBrown62de6082006-05-01 12:15:47 -0700789 * We test barriers_work *after* md_write_start as md_write_start
790 * may cause the first superblock write, and that will check out
791 * if barriers work.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 */
NeilBrown62de6082006-05-01 12:15:47 -0700793
NeilBrown3d310eb2005-06-21 17:17:26 -0700794 md_write_start(mddev, bio); /* wait on superblock update early */
795
NeilBrown62de6082006-05-01 12:15:47 -0700796 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
797 if (rw == WRITE)
798 md_write_end(mddev);
799 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
800 return 0;
801 }
802
NeilBrown17999be2006-01-06 00:20:12 -0800803 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Jens Axboea3623572005-11-01 09:26:16 +0100805 disk_stat_inc(mddev->gendisk, ios[rw]);
806 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
808 /*
809 * make_request() can abort the operation when READA is being
810 * used and no empty request is available.
811 *
812 */
813 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
814
815 r1_bio->master_bio = bio;
816 r1_bio->sectors = bio->bi_size >> 9;
NeilBrown191ea9b2005-06-21 17:17:23 -0700817 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 r1_bio->mddev = mddev;
819 r1_bio->sector = bio->bi_sector;
820
Jens Axboea3623572005-11-01 09:26:16 +0100821 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 /*
823 * read balancing logic:
824 */
825 int rdisk = read_balance(conf, r1_bio);
826
827 if (rdisk < 0) {
828 /* couldn't find anywhere to read from */
829 raid_end_bio_io(r1_bio);
830 return 0;
831 }
832 mirror = conf->mirrors + rdisk;
833
834 r1_bio->read_disk = rdisk;
835
836 read_bio = bio_clone(bio, GFP_NOIO);
837
838 r1_bio->bios[rdisk] = read_bio;
839
840 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
841 read_bio->bi_bdev = mirror->rdev->bdev;
842 read_bio->bi_end_io = raid1_end_read_request;
Lars Ellenberge3881a62007-01-10 23:15:37 -0800843 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 read_bio->bi_private = r1_bio;
845
846 generic_make_request(read_bio);
847 return 0;
848 }
849
850 /*
851 * WRITE:
852 */
853 /* first select target devices under spinlock and
854 * inc refcount on their rdev. Record them by setting
855 * bios[x] to bio
856 */
857 disks = conf->raid_disks;
NeilBrown191ea9b2005-06-21 17:17:23 -0700858#if 0
859 { static int first=1;
860 if (first) printk("First Write sector %llu disks %d\n",
861 (unsigned long long)r1_bio->sector, disks);
862 first = 0;
863 }
864#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 rcu_read_lock();
866 for (i = 0; i < disks; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800867 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
NeilBrownb2d444d2005-11-08 21:39:31 -0800868 !test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 atomic_inc(&rdev->nr_pending);
NeilBrownb2d444d2005-11-08 21:39:31 -0800870 if (test_bit(Faulty, &rdev->flags)) {
NeilBrown03c902e2006-01-06 00:20:46 -0800871 rdev_dec_pending(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 r1_bio->bios[i] = NULL;
873 } else
874 r1_bio->bios[i] = bio;
NeilBrown191ea9b2005-06-21 17:17:23 -0700875 targets++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 } else
877 r1_bio->bios[i] = NULL;
878 }
879 rcu_read_unlock();
880
NeilBrown4b6d2872005-09-09 16:23:47 -0700881 BUG_ON(targets == 0); /* we never fail the last device */
882
NeilBrown191ea9b2005-06-21 17:17:23 -0700883 if (targets < conf->raid_disks) {
884 /* array is degraded, we will not clear the bitmap
885 * on I/O completion (see raid1_end_write_request) */
886 set_bit(R1BIO_Degraded, &r1_bio->state);
887 }
NeilBrown06d91a52005-06-21 17:17:12 -0700888
NeilBrown4b6d2872005-09-09 16:23:47 -0700889 /* do behind I/O ? */
890 if (bitmap &&
891 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
892 (behind_pages = alloc_behind_pages(bio)) != NULL)
893 set_bit(R1BIO_BehindIO, &r1_bio->state);
894
NeilBrown191ea9b2005-06-21 17:17:23 -0700895 atomic_set(&r1_bio->remaining, 0);
NeilBrown4b6d2872005-09-09 16:23:47 -0700896 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -0700897
NeilBrown04b857f2006-03-09 17:33:46 -0800898 do_barriers = bio_barrier(bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800899 if (do_barriers)
900 set_bit(R1BIO_Barrier, &r1_bio->state);
901
NeilBrown191ea9b2005-06-21 17:17:23 -0700902 bio_list_init(&bl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 for (i = 0; i < disks; i++) {
904 struct bio *mbio;
905 if (!r1_bio->bios[i])
906 continue;
907
908 mbio = bio_clone(bio, GFP_NOIO);
909 r1_bio->bios[i] = mbio;
910
911 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
912 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
913 mbio->bi_end_io = raid1_end_write_request;
Lars Ellenberge3881a62007-01-10 23:15:37 -0800914 mbio->bi_rw = WRITE | do_barriers | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 mbio->bi_private = r1_bio;
916
NeilBrown4b6d2872005-09-09 16:23:47 -0700917 if (behind_pages) {
918 struct bio_vec *bvec;
919 int j;
920
921 /* Yes, I really want the '__' version so that
922 * we clear any unused pointer in the io_vec, rather
923 * than leave them unchanged. This is important
924 * because when we come to free the pages, we won't
925 * know the originial bi_idx, so we just free
926 * them all
927 */
928 __bio_for_each_segment(bvec, mbio, j, 0)
929 bvec->bv_page = behind_pages[j];
930 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
931 atomic_inc(&r1_bio->behind_remaining);
932 }
933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 atomic_inc(&r1_bio->remaining);
NeilBrown191ea9b2005-06-21 17:17:23 -0700935
936 bio_list_add(&bl, mbio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 }
NeilBrown4b6d2872005-09-09 16:23:47 -0700938 kfree(behind_pages); /* the behind pages are attached to the bios now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
NeilBrown4b6d2872005-09-09 16:23:47 -0700940 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
941 test_bit(R1BIO_BehindIO, &r1_bio->state));
NeilBrown191ea9b2005-06-21 17:17:23 -0700942 spin_lock_irqsave(&conf->device_lock, flags);
943 bio_list_merge(&conf->pending_bio_list, &bl);
944 bio_list_init(&bl);
945
946 blk_plug_device(mddev->queue);
947 spin_unlock_irqrestore(&conf->device_lock, flags);
948
Lars Ellenberge3881a62007-01-10 23:15:37 -0800949 if (do_sync)
950 md_wakeup_thread(mddev->thread);
NeilBrown191ea9b2005-06-21 17:17:23 -0700951#if 0
952 while ((bio = bio_list_pop(&bl)) != NULL)
953 generic_make_request(bio);
954#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
956 return 0;
957}
958
959static void status(struct seq_file *seq, mddev_t *mddev)
960{
961 conf_t *conf = mddev_to_conf(mddev);
962 int i;
963
964 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -0700965 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -0700966 rcu_read_lock();
967 for (i = 0; i < conf->raid_disks; i++) {
968 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -0700970 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
971 }
972 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 seq_printf(seq, "]");
974}
975
976
977static void error(mddev_t *mddev, mdk_rdev_t *rdev)
978{
979 char b[BDEVNAME_SIZE];
980 conf_t *conf = mddev_to_conf(mddev);
981
982 /*
983 * If it is not operational, then we have already marked it as dead
984 * else if it is the last working disks, ignore the error, let the
985 * next level up know.
986 * else mark the drive as failed
987 */
NeilBrownb2d444d2005-11-08 21:39:31 -0800988 if (test_bit(In_sync, &rdev->flags)
NeilBrown11ce99e2006-10-03 01:15:52 -0700989 && (conf->raid_disks - mddev->degraded) == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 /*
991 * Don't fail the drive, act as though we were just a
992 * normal single drive
993 */
994 return;
NeilBrownc04be0a2006-10-03 01:15:53 -0700995 if (test_and_clear_bit(In_sync, &rdev->flags)) {
996 unsigned long flags;
997 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -0700999 set_bit(Faulty, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001000 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 /*
1002 * if recovery is running, make sure it aborts.
1003 */
1004 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
NeilBrowndd00a992007-05-10 03:15:50 -07001005 } else
1006 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001007 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
1009 " Operation continuing on %d devices\n",
NeilBrown11ce99e2006-10-03 01:15:52 -07001010 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011}
1012
1013static void print_conf(conf_t *conf)
1014{
1015 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
1017 printk("RAID1 conf printout:\n");
1018 if (!conf) {
1019 printk("(!conf)\n");
1020 return;
1021 }
NeilBrown11ce99e2006-10-03 01:15:52 -07001022 printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 conf->raid_disks);
1024
NeilBrownddac7c72006-08-31 21:27:36 -07001025 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 for (i = 0; i < conf->raid_disks; i++) {
1027 char b[BDEVNAME_SIZE];
NeilBrownddac7c72006-08-31 21:27:36 -07001028 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
1029 if (rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001031 i, !test_bit(In_sync, &rdev->flags),
1032 !test_bit(Faulty, &rdev->flags),
1033 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 }
NeilBrownddac7c72006-08-31 21:27:36 -07001035 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
1037
1038static void close_sync(conf_t *conf)
1039{
NeilBrown17999be2006-01-06 00:20:12 -08001040 wait_barrier(conf);
1041 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043 mempool_destroy(conf->r1buf_pool);
1044 conf->r1buf_pool = NULL;
1045}
1046
1047static int raid1_spare_active(mddev_t *mddev)
1048{
1049 int i;
1050 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 /*
1053 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001054 * and mark them readable.
1055 * Called under mddev lock, so rcu protection not needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 */
1057 for (i = 0; i < conf->raid_disks; i++) {
NeilBrownddac7c72006-08-31 21:27:36 -07001058 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
1059 if (rdev
1060 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001061 && !test_and_set_bit(In_sync, &rdev->flags)) {
1062 unsigned long flags;
1063 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 mddev->degraded--;
NeilBrownc04be0a2006-10-03 01:15:53 -07001065 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 }
1067 }
1068
1069 print_conf(conf);
1070 return 0;
1071}
1072
1073
1074static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1075{
1076 conf_t *conf = mddev->private;
1077 int found = 0;
NeilBrown41158c72005-06-21 17:17:25 -07001078 int mirror = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 mirror_info_t *p;
1080
1081 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1082 if ( !(p=conf->mirrors+mirror)->rdev) {
1083
1084 blk_queue_stack_limits(mddev->queue,
1085 rdev->bdev->bd_disk->queue);
1086 /* as we don't honour merge_bvec_fn, we must never risk
1087 * violating it, so limit ->max_sector to one PAGE, as
1088 * a one page request is never in violation.
1089 */
1090 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1091 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1092 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1093
1094 p->head_position = 0;
1095 rdev->raid_disk = mirror;
1096 found = 1;
NeilBrown6aea114a2005-11-28 13:44:13 -08001097 /* As all devices are equivalent, we don't need a full recovery
1098 * if this was recently any drive of the array
1099 */
1100 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001101 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001102 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 break;
1104 }
1105
1106 print_conf(conf);
1107 return found;
1108}
1109
1110static int raid1_remove_disk(mddev_t *mddev, int number)
1111{
1112 conf_t *conf = mddev->private;
1113 int err = 0;
1114 mdk_rdev_t *rdev;
1115 mirror_info_t *p = conf->mirrors+ number;
1116
1117 print_conf(conf);
1118 rdev = p->rdev;
1119 if (rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001120 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 atomic_read(&rdev->nr_pending)) {
1122 err = -EBUSY;
1123 goto abort;
1124 }
1125 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001126 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (atomic_read(&rdev->nr_pending)) {
1128 /* lost the race, try later */
1129 err = -EBUSY;
1130 p->rdev = rdev;
1131 }
1132 }
1133abort:
1134
1135 print_conf(conf);
1136 return err;
1137}
1138
1139
1140static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
1141{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
NeilBrownd11c1712006-01-06 00:20:26 -08001143 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
1145 if (bio->bi_size)
1146 return 1;
1147
NeilBrownd11c1712006-01-06 00:20:26 -08001148 for (i=r1_bio->mddev->raid_disks; i--; )
1149 if (r1_bio->bios[i] == bio)
1150 break;
1151 BUG_ON(i < 0);
1152 update_head_pos(i, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 /*
1154 * we have read a block, now it needs to be re-written,
1155 * or re-read if the read failed.
1156 * We don't do much here, just schedule handling by raid1d
1157 */
NeilBrown69382e82006-01-06 00:20:22 -08001158 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001160
1161 if (atomic_dec_and_test(&r1_bio->remaining))
1162 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 return 0;
1164}
1165
1166static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
1167{
1168 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1169 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1170 mddev_t *mddev = r1_bio->mddev;
1171 conf_t *conf = mddev_to_conf(mddev);
1172 int i;
1173 int mirror=0;
1174
1175 if (bio->bi_size)
1176 return 1;
1177
1178 for (i = 0; i < conf->raid_disks; i++)
1179 if (r1_bio->bios[i] == bio) {
1180 mirror = i;
1181 break;
1182 }
NeilBrown6b1117d2006-03-31 02:31:57 -08001183 if (!uptodate) {
1184 int sync_blocks = 0;
1185 sector_t s = r1_bio->sector;
1186 long sectors_to_go = r1_bio->sectors;
1187 /* make sure these bits doesn't get cleared. */
1188 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001189 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001190 &sync_blocks, 1);
1191 s += sync_blocks;
1192 sectors_to_go -= sync_blocks;
1193 } while (sectors_to_go > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 md_error(mddev, conf->mirrors[mirror].rdev);
NeilBrown6b1117d2006-03-31 02:31:57 -08001195 }
NeilBrowne3b97032005-08-04 12:53:34 -07001196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 update_head_pos(mirror, r1_bio);
1198
1199 if (atomic_dec_and_test(&r1_bio->remaining)) {
1200 md_done_sync(mddev, r1_bio->sectors, uptodate);
1201 put_buf(r1_bio);
1202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return 0;
1204}
1205
1206static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1207{
1208 conf_t *conf = mddev_to_conf(mddev);
1209 int i;
1210 int disks = conf->raid_disks;
1211 struct bio *bio, *wbio;
1212
1213 bio = r1_bio->bios[r1_bio->read_disk];
1214
NeilBrown69382e82006-01-06 00:20:22 -08001215
NeilBrownd11c1712006-01-06 00:20:26 -08001216 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1217 /* We have read all readable devices. If we haven't
1218 * got the block, then there is no hope left.
1219 * If we have, then we want to do a comparison
1220 * and skip the write if everything is the same.
1221 * If any blocks failed to read, then we need to
1222 * attempt an over-write
1223 */
1224 int primary;
1225 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1226 for (i=0; i<mddev->raid_disks; i++)
1227 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1228 md_error(mddev, conf->mirrors[i].rdev);
1229
1230 md_done_sync(mddev, r1_bio->sectors, 1);
1231 put_buf(r1_bio);
1232 return;
1233 }
1234 for (primary=0; primary<mddev->raid_disks; primary++)
1235 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1236 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1237 r1_bio->bios[primary]->bi_end_io = NULL;
NeilBrown03c902e2006-01-06 00:20:46 -08001238 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
NeilBrownd11c1712006-01-06 00:20:26 -08001239 break;
1240 }
1241 r1_bio->read_disk = primary;
1242 for (i=0; i<mddev->raid_disks; i++)
Mike Accettaed456662007-06-16 10:16:07 -07001243 if (r1_bio->bios[i]->bi_end_io == end_sync_read) {
NeilBrownd11c1712006-01-06 00:20:26 -08001244 int j;
1245 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1246 struct bio *pbio = r1_bio->bios[primary];
1247 struct bio *sbio = r1_bio->bios[i];
Mike Accettaed456662007-06-16 10:16:07 -07001248
1249 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1250 for (j = vcnt; j-- ; ) {
1251 struct page *p, *s;
1252 p = pbio->bi_io_vec[j].bv_page;
1253 s = sbio->bi_io_vec[j].bv_page;
1254 if (memcmp(page_address(p),
1255 page_address(s),
1256 PAGE_SIZE))
1257 break;
1258 }
1259 } else
1260 j = 0;
NeilBrownd11c1712006-01-06 00:20:26 -08001261 if (j >= 0)
1262 mddev->resync_mismatches += r1_bio->sectors;
NeilBrown03c902e2006-01-06 00:20:46 -08001263 if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
NeilBrownd11c1712006-01-06 00:20:26 -08001264 sbio->bi_end_io = NULL;
NeilBrown03c902e2006-01-06 00:20:46 -08001265 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1266 } else {
NeilBrownd11c1712006-01-06 00:20:26 -08001267 /* fixup the bio for reuse */
1268 sbio->bi_vcnt = vcnt;
1269 sbio->bi_size = r1_bio->sectors << 9;
1270 sbio->bi_idx = 0;
1271 sbio->bi_phys_segments = 0;
1272 sbio->bi_hw_segments = 0;
1273 sbio->bi_hw_front_size = 0;
1274 sbio->bi_hw_back_size = 0;
1275 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1276 sbio->bi_flags |= 1 << BIO_UPTODATE;
1277 sbio->bi_next = NULL;
1278 sbio->bi_sector = r1_bio->sector +
1279 conf->mirrors[i].rdev->data_offset;
1280 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
NeilBrown3eda22d2007-01-26 00:57:01 -08001281 for (j = 0; j < vcnt ; j++)
1282 memcpy(page_address(sbio->bi_io_vec[j].bv_page),
1283 page_address(pbio->bi_io_vec[j].bv_page),
1284 PAGE_SIZE);
1285
NeilBrownd11c1712006-01-06 00:20:26 -08001286 }
1287 }
1288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
NeilBrown69382e82006-01-06 00:20:22 -08001290 /* ouch - failed to read all of that.
1291 * Try some synchronous reads of other devices to get
1292 * good data, much like with normal read errors. Only
NeilBrownddac7c72006-08-31 21:27:36 -07001293 * read into the pages we already have so we don't
NeilBrown69382e82006-01-06 00:20:22 -08001294 * need to re-issue the read request.
1295 * We don't need to freeze the array, because being in an
1296 * active sync request, there is no normal IO, and
1297 * no overlapping syncs.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 */
NeilBrown69382e82006-01-06 00:20:22 -08001299 sector_t sect = r1_bio->sector;
1300 int sectors = r1_bio->sectors;
1301 int idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
NeilBrown69382e82006-01-06 00:20:22 -08001303 while(sectors) {
1304 int s = sectors;
1305 int d = r1_bio->read_disk;
1306 int success = 0;
1307 mdk_rdev_t *rdev;
1308
1309 if (s > (PAGE_SIZE>>9))
1310 s = PAGE_SIZE >> 9;
1311 do {
1312 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07001313 /* No rcu protection needed here devices
1314 * can only be removed when no resync is
1315 * active, and resync is currently active
1316 */
NeilBrown69382e82006-01-06 00:20:22 -08001317 rdev = conf->mirrors[d].rdev;
1318 if (sync_page_io(rdev->bdev,
1319 sect + rdev->data_offset,
1320 s<<9,
1321 bio->bi_io_vec[idx].bv_page,
1322 READ)) {
1323 success = 1;
1324 break;
1325 }
1326 }
1327 d++;
1328 if (d == conf->raid_disks)
1329 d = 0;
1330 } while (!success && d != r1_bio->read_disk);
1331
1332 if (success) {
NeilBrown097426f2006-01-06 00:20:37 -08001333 int start = d;
NeilBrown69382e82006-01-06 00:20:22 -08001334 /* write it back and re-read */
1335 set_bit(R1BIO_Uptodate, &r1_bio->state);
1336 while (d != r1_bio->read_disk) {
1337 if (d == 0)
1338 d = conf->raid_disks;
1339 d--;
1340 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1341 continue;
1342 rdev = conf->mirrors[d].rdev;
NeilBrown4dbcdc72006-01-06 00:20:52 -08001343 atomic_add(s, &rdev->corrected_errors);
NeilBrown69382e82006-01-06 00:20:22 -08001344 if (sync_page_io(rdev->bdev,
1345 sect + rdev->data_offset,
1346 s<<9,
1347 bio->bi_io_vec[idx].bv_page,
NeilBrown097426f2006-01-06 00:20:37 -08001348 WRITE) == 0)
1349 md_error(mddev, rdev);
1350 }
1351 d = start;
1352 while (d != r1_bio->read_disk) {
1353 if (d == 0)
1354 d = conf->raid_disks;
1355 d--;
1356 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1357 continue;
1358 rdev = conf->mirrors[d].rdev;
1359 if (sync_page_io(rdev->bdev,
NeilBrown69382e82006-01-06 00:20:22 -08001360 sect + rdev->data_offset,
1361 s<<9,
1362 bio->bi_io_vec[idx].bv_page,
NeilBrown097426f2006-01-06 00:20:37 -08001363 READ) == 0)
NeilBrown69382e82006-01-06 00:20:22 -08001364 md_error(mddev, rdev);
NeilBrown69382e82006-01-06 00:20:22 -08001365 }
1366 } else {
1367 char b[BDEVNAME_SIZE];
1368 /* Cannot read from anywhere, array is toast */
1369 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1370 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1371 " for block %llu\n",
1372 bdevname(bio->bi_bdev,b),
1373 (unsigned long long)r1_bio->sector);
1374 md_done_sync(mddev, r1_bio->sectors, 0);
1375 put_buf(r1_bio);
1376 return;
1377 }
1378 sectors -= s;
1379 sect += s;
1380 idx ++;
1381 }
1382 }
NeilBrownd11c1712006-01-06 00:20:26 -08001383
1384 /*
1385 * schedule writes
1386 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 atomic_set(&r1_bio->remaining, 1);
1388 for (i = 0; i < disks ; i++) {
1389 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08001390 if (wbio->bi_end_io == NULL ||
1391 (wbio->bi_end_io == end_sync_read &&
1392 (i == r1_bio->read_disk ||
1393 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 continue;
1395
NeilBrown3e198f72006-01-06 00:20:21 -08001396 wbio->bi_rw = WRITE;
1397 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 atomic_inc(&r1_bio->remaining);
1399 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
NeilBrown191ea9b2005-06-21 17:17:23 -07001400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 generic_make_request(wbio);
1402 }
1403
1404 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001405 /* if we're here, all write(s) have completed, so clean up */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 md_done_sync(mddev, r1_bio->sectors, 1);
1407 put_buf(r1_bio);
1408 }
1409}
1410
1411/*
1412 * This is a kernel thread which:
1413 *
1414 * 1. Retries failed read operations on working mirrors.
1415 * 2. Updates the raid superblock when problems encounter.
1416 * 3. Performs writes following reads for array syncronising.
1417 */
1418
NeilBrown867868f2006-10-03 01:15:51 -07001419static void fix_read_error(conf_t *conf, int read_disk,
1420 sector_t sect, int sectors)
1421{
1422 mddev_t *mddev = conf->mddev;
1423 while(sectors) {
1424 int s = sectors;
1425 int d = read_disk;
1426 int success = 0;
1427 int start;
1428 mdk_rdev_t *rdev;
1429
1430 if (s > (PAGE_SIZE>>9))
1431 s = PAGE_SIZE >> 9;
1432
1433 do {
1434 /* Note: no rcu protection needed here
1435 * as this is synchronous in the raid1d thread
1436 * which is the thread that might remove
1437 * a device. If raid1d ever becomes multi-threaded....
1438 */
1439 rdev = conf->mirrors[d].rdev;
1440 if (rdev &&
1441 test_bit(In_sync, &rdev->flags) &&
1442 sync_page_io(rdev->bdev,
1443 sect + rdev->data_offset,
1444 s<<9,
1445 conf->tmppage, READ))
1446 success = 1;
1447 else {
1448 d++;
1449 if (d == conf->raid_disks)
1450 d = 0;
1451 }
1452 } while (!success && d != read_disk);
1453
1454 if (!success) {
1455 /* Cannot read from anywhere -- bye bye array */
1456 md_error(mddev, conf->mirrors[read_disk].rdev);
1457 break;
1458 }
1459 /* write it back and re-read */
1460 start = d;
1461 while (d != read_disk) {
1462 if (d==0)
1463 d = conf->raid_disks;
1464 d--;
1465 rdev = conf->mirrors[d].rdev;
1466 if (rdev &&
1467 test_bit(In_sync, &rdev->flags)) {
1468 if (sync_page_io(rdev->bdev,
1469 sect + rdev->data_offset,
1470 s<<9, conf->tmppage, WRITE)
1471 == 0)
1472 /* Well, this device is dead */
1473 md_error(mddev, rdev);
1474 }
1475 }
1476 d = start;
1477 while (d != read_disk) {
1478 char b[BDEVNAME_SIZE];
1479 if (d==0)
1480 d = conf->raid_disks;
1481 d--;
1482 rdev = conf->mirrors[d].rdev;
1483 if (rdev &&
1484 test_bit(In_sync, &rdev->flags)) {
1485 if (sync_page_io(rdev->bdev,
1486 sect + rdev->data_offset,
1487 s<<9, conf->tmppage, READ)
1488 == 0)
1489 /* Well, this device is dead */
1490 md_error(mddev, rdev);
1491 else {
1492 atomic_add(s, &rdev->corrected_errors);
1493 printk(KERN_INFO
1494 "raid1:%s: read error corrected "
1495 "(%d sectors at %llu on %s)\n",
1496 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07001497 (unsigned long long)(sect +
1498 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07001499 bdevname(rdev->bdev, b));
1500 }
1501 }
1502 }
1503 sectors -= s;
1504 sect += s;
1505 }
1506}
1507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508static void raid1d(mddev_t *mddev)
1509{
1510 r1bio_t *r1_bio;
1511 struct bio *bio;
1512 unsigned long flags;
1513 conf_t *conf = mddev_to_conf(mddev);
1514 struct list_head *head = &conf->retry_list;
1515 int unplug=0;
1516 mdk_rdev_t *rdev;
1517
1518 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
1520 for (;;) {
1521 char b[BDEVNAME_SIZE];
1522 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown191ea9b2005-06-21 17:17:23 -07001523
1524 if (conf->pending_bio_list.head) {
1525 bio = bio_list_get(&conf->pending_bio_list);
1526 blk_remove_plug(mddev->queue);
1527 spin_unlock_irqrestore(&conf->device_lock, flags);
1528 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
NeilBrown4ad13662007-07-17 04:06:13 -07001529 bitmap_unplug(mddev->bitmap);
NeilBrown191ea9b2005-06-21 17:17:23 -07001530
1531 while (bio) { /* submit pending writes */
1532 struct bio *next = bio->bi_next;
1533 bio->bi_next = NULL;
1534 generic_make_request(bio);
1535 bio = next;
1536 }
1537 unplug = 1;
1538
1539 continue;
1540 }
1541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 if (list_empty(head))
1543 break;
1544 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1545 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08001546 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 spin_unlock_irqrestore(&conf->device_lock, flags);
1548
1549 mddev = r1_bio->mddev;
1550 conf = mddev_to_conf(mddev);
1551 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1552 sync_request_write(mddev, r1_bio);
1553 unplug = 1;
NeilBrowna9701a32005-11-08 21:39:34 -08001554 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1555 /* some requests in the r1bio were BIO_RW_BARRIER
NeilBrownbea27712006-05-01 12:15:46 -07001556 * requests which failed with -EOPNOTSUPP. Hohumm..
NeilBrowna9701a32005-11-08 21:39:34 -08001557 * Better resubmit without the barrier.
1558 * We know which devices to resubmit for, because
1559 * all others have had their bios[] entry cleared.
NeilBrown5e7dd2a2006-05-01 12:15:47 -07001560 * We already have a nr_pending reference on these rdevs.
NeilBrowna9701a32005-11-08 21:39:34 -08001561 */
1562 int i;
Lars Ellenberge3881a62007-01-10 23:15:37 -08001563 const int do_sync = bio_sync(r1_bio->master_bio);
NeilBrowna9701a32005-11-08 21:39:34 -08001564 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1565 clear_bit(R1BIO_Barrier, &r1_bio->state);
1566 for (i=0; i < conf->raid_disks; i++)
NeilBrown2f889122006-03-27 01:18:19 -08001567 if (r1_bio->bios[i])
1568 atomic_inc(&r1_bio->remaining);
1569 for (i=0; i < conf->raid_disks; i++)
NeilBrowna9701a32005-11-08 21:39:34 -08001570 if (r1_bio->bios[i]) {
1571 struct bio_vec *bvec;
1572 int j;
1573
1574 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1575 /* copy pages from the failed bio, as
1576 * this might be a write-behind device */
1577 __bio_for_each_segment(bvec, bio, j, 0)
1578 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1579 bio_put(r1_bio->bios[i]);
1580 bio->bi_sector = r1_bio->sector +
1581 conf->mirrors[i].rdev->data_offset;
1582 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1583 bio->bi_end_io = raid1_end_write_request;
Lars Ellenberge3881a62007-01-10 23:15:37 -08001584 bio->bi_rw = WRITE | do_sync;
NeilBrowna9701a32005-11-08 21:39:34 -08001585 bio->bi_private = r1_bio;
1586 r1_bio->bios[i] = bio;
1587 generic_make_request(bio);
1588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 } else {
1590 int disk;
NeilBrownddaf22a2006-01-06 00:20:19 -08001591
1592 /* we got a read error. Maybe the drive is bad. Maybe just
1593 * the block and we can fix it.
1594 * We freeze all other IO, and try reading the block from
1595 * other devices. When we find one, we re-write
1596 * and check it that fixes the read error.
1597 * This is all done synchronously while the array is
1598 * frozen
1599 */
NeilBrown867868f2006-10-03 01:15:51 -07001600 if (mddev->ro == 0) {
1601 freeze_array(conf);
1602 fix_read_error(conf, r1_bio->read_disk,
1603 r1_bio->sector,
1604 r1_bio->sectors);
1605 unfreeze_array(conf);
NeilBrownddaf22a2006-01-06 00:20:19 -08001606 }
1607
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 bio = r1_bio->bios[r1_bio->read_disk];
1609 if ((disk=read_balance(conf, r1_bio)) == -1) {
1610 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1611 " read error for block %llu\n",
1612 bdevname(bio->bi_bdev,b),
1613 (unsigned long long)r1_bio->sector);
1614 raid_end_bio_io(r1_bio);
1615 } else {
Lars Ellenberge3881a62007-01-10 23:15:37 -08001616 const int do_sync = bio_sync(r1_bio->master_bio);
NeilBrowncf30a472006-01-06 00:20:23 -08001617 r1_bio->bios[r1_bio->read_disk] =
1618 mddev->ro ? IO_BLOCKED : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 r1_bio->read_disk = disk;
1620 bio_put(bio);
1621 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1622 r1_bio->bios[r1_bio->read_disk] = bio;
1623 rdev = conf->mirrors[disk].rdev;
1624 if (printk_ratelimit())
1625 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1626 " another mirror\n",
1627 bdevname(rdev->bdev,b),
1628 (unsigned long long)r1_bio->sector);
1629 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1630 bio->bi_bdev = rdev->bdev;
1631 bio->bi_end_io = raid1_end_read_request;
Lars Ellenberge3881a62007-01-10 23:15:37 -08001632 bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 bio->bi_private = r1_bio;
1634 unplug = 1;
1635 generic_make_request(bio);
1636 }
1637 }
1638 }
1639 spin_unlock_irqrestore(&conf->device_lock, flags);
1640 if (unplug)
1641 unplug_slaves(mddev);
1642}
1643
1644
1645static int init_resync(conf_t *conf)
1646{
1647 int buffs;
1648
1649 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02001650 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1652 conf->poolinfo);
1653 if (!conf->r1buf_pool)
1654 return -ENOMEM;
1655 conf->next_resync = 0;
1656 return 0;
1657}
1658
1659/*
1660 * perform a "sync" on one "block"
1661 *
1662 * We need to make sure that no normal I/O request - particularly write
1663 * requests - conflict with active sync requests.
1664 *
1665 * This is achieved by tracking pending requests and a 'barrier' concept
1666 * that can be installed to exclude normal IO requests.
1667 */
1668
NeilBrown57afd892005-06-21 17:17:13 -07001669static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
1671 conf_t *conf = mddev_to_conf(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 r1bio_t *r1_bio;
1673 struct bio *bio;
1674 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08001675 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08001677 int wonly = -1;
1678 int write_targets = 0, read_targets = 0;
NeilBrown191ea9b2005-06-21 17:17:23 -07001679 int sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07001680 int still_degraded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
1682 if (!conf->r1buf_pool)
NeilBrown191ea9b2005-06-21 17:17:23 -07001683 {
1684/*
1685 printk("sync start - bitmap %p\n", mddev->bitmap);
1686*/
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07001688 return 0;
NeilBrown191ea9b2005-06-21 17:17:23 -07001689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691 max_sector = mddev->size << 1;
1692 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001693 /* If we aborted, we need to abort the
1694 * sync on the 'current' bitmap chunk (there will
1695 * only be one in raid1 resync.
1696 * We can find the current addess in mddev->curr_resync
1697 */
NeilBrown6a806c52005-07-15 03:56:35 -07001698 if (mddev->curr_resync < max_sector) /* aborted */
1699 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07001700 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07001701 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07001702 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07001703
1704 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 close_sync(conf);
1706 return 0;
1707 }
1708
NeilBrown07d84d102006-06-26 00:27:56 -07001709 if (mddev->bitmap == NULL &&
1710 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07001711 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07001712 conf->fullsync == 0) {
1713 *skipped = 1;
1714 return max_sector - sector_nr;
1715 }
NeilBrown6394cca2006-08-27 01:23:50 -07001716 /* before building a request, check if we can skip these blocks..
1717 * This call the bitmap_start_sync doesn't actually record anything
1718 */
NeilBrowne3b97032005-08-04 12:53:34 -07001719 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08001720 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001721 /* We can skip this block, and probably several more */
1722 *skipped = 1;
1723 return sync_blocks;
1724 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 /*
NeilBrown17999be2006-01-06 00:20:12 -08001726 * If there is non-resync activity waiting for a turn,
1727 * and resync is going fast enough,
1728 * then let it though before starting on this new sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 */
NeilBrown17999be2006-01-06 00:20:12 -08001730 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 msleep_interruptible(1000);
NeilBrown17999be2006-01-06 00:20:12 -08001732
1733 raise_barrier(conf);
1734
1735 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown3e198f72006-01-06 00:20:21 -08001738 rcu_read_lock();
1739 /*
1740 * If we get a correctably read error during resync or recovery,
1741 * we might want to read from a different device. So we
1742 * flag all drives that could conceivably be read from for READ,
1743 * and any others (which will be non-In_sync devices) for WRITE.
1744 * If a read fails, we try reading from something else for which READ
1745 * is OK.
1746 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 r1_bio->mddev = mddev;
1749 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07001750 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
1753 for (i=0; i < conf->raid_disks; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -08001754 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 bio = r1_bio->bios[i];
1756
1757 /* take from bio_init */
1758 bio->bi_next = NULL;
1759 bio->bi_flags |= 1 << BIO_UPTODATE;
NeilBrown802ba062006-12-13 00:34:13 -08001760 bio->bi_rw = READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 bio->bi_vcnt = 0;
1762 bio->bi_idx = 0;
1763 bio->bi_phys_segments = 0;
1764 bio->bi_hw_segments = 0;
1765 bio->bi_size = 0;
1766 bio->bi_end_io = NULL;
1767 bio->bi_private = NULL;
1768
NeilBrown3e198f72006-01-06 00:20:21 -08001769 rdev = rcu_dereference(conf->mirrors[i].rdev);
1770 if (rdev == NULL ||
1771 test_bit(Faulty, &rdev->flags)) {
NeilBrowne3b97032005-08-04 12:53:34 -07001772 still_degraded = 1;
1773 continue;
NeilBrown3e198f72006-01-06 00:20:21 -08001774 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 bio->bi_rw = WRITE;
1776 bio->bi_end_io = end_sync_write;
1777 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08001778 } else {
1779 /* may need to read from here */
1780 bio->bi_rw = READ;
1781 bio->bi_end_io = end_sync_read;
1782 if (test_bit(WriteMostly, &rdev->flags)) {
1783 if (wonly < 0)
1784 wonly = i;
1785 } else {
1786 if (disk < 0)
1787 disk = i;
1788 }
1789 read_targets++;
1790 }
1791 atomic_inc(&rdev->nr_pending);
1792 bio->bi_sector = sector_nr + rdev->data_offset;
1793 bio->bi_bdev = rdev->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 bio->bi_private = r1_bio;
1795 }
NeilBrown3e198f72006-01-06 00:20:21 -08001796 rcu_read_unlock();
1797 if (disk < 0)
1798 disk = wonly;
1799 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07001800
NeilBrown3e198f72006-01-06 00:20:21 -08001801 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1802 /* extra read targets are also write targets */
1803 write_targets += read_targets-1;
1804
1805 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 /* There is nowhere to write, so all non-sync
1807 * drives must be failed - so we are finished
1808 */
NeilBrown57afd892005-06-21 17:17:13 -07001809 sector_t rv = max_sector - sector_nr;
1810 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 return rv;
1813 }
1814
1815 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07001816 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 do {
1818 struct page *page;
1819 int len = PAGE_SIZE;
1820 if (sector_nr + (len>>9) > max_sector)
1821 len = (max_sector - sector_nr) << 9;
1822 if (len == 0)
1823 break;
NeilBrown6a806c52005-07-15 03:56:35 -07001824 if (sync_blocks == 0) {
1825 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de485f2005-11-08 21:39:38 -08001826 &sync_blocks, still_degraded) &&
1827 !conf->fullsync &&
1828 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07001829 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02001830 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown6a806c52005-07-15 03:56:35 -07001831 if (len > (sync_blocks<<9))
1832 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07001833 }
NeilBrown191ea9b2005-06-21 17:17:23 -07001834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 for (i=0 ; i < conf->raid_disks; i++) {
1836 bio = r1_bio->bios[i];
1837 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08001838 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 if (bio_add_page(bio, page, len, 0) == 0) {
1840 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08001841 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 while (i > 0) {
1843 i--;
1844 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07001845 if (bio->bi_end_io==NULL)
1846 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 /* remove last page from this bio */
1848 bio->bi_vcnt--;
1849 bio->bi_size -= len;
1850 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1851 }
1852 goto bio_full;
1853 }
1854 }
1855 }
1856 nr_sectors += len>>9;
1857 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07001858 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1860 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 r1_bio->sectors = nr_sectors;
1862
NeilBrownd11c1712006-01-06 00:20:26 -08001863 /* For a user-requested sync, we read all readable devices and do a
1864 * compare
1865 */
1866 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1867 atomic_set(&r1_bio->remaining, read_targets);
1868 for (i=0; i<conf->raid_disks; i++) {
1869 bio = r1_bio->bios[i];
1870 if (bio->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07001871 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08001872 generic_make_request(bio);
1873 }
1874 }
1875 } else {
1876 atomic_set(&r1_bio->remaining, 1);
1877 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07001878 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08001879 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
NeilBrownd11c1712006-01-06 00:20:26 -08001881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 return nr_sectors;
1883}
1884
1885static int run(mddev_t *mddev)
1886{
1887 conf_t *conf;
1888 int i, j, disk_idx;
1889 mirror_info_t *disk;
1890 mdk_rdev_t *rdev;
1891 struct list_head *tmp;
1892
1893 if (mddev->level != 1) {
1894 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1895 mdname(mddev), mddev->level);
1896 goto out;
1897 }
NeilBrownf6705572006-03-27 01:18:11 -08001898 if (mddev->reshape_position != MaxSector) {
1899 printk("raid1: %s: reshape_position set but not supported\n",
1900 mdname(mddev));
1901 goto out;
1902 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 /*
1904 * copy the already verified devices into our private RAID1
1905 * bookkeeping area. [whatever we allocate in run(),
1906 * should be freed in stop()]
1907 */
NeilBrown9ffae0c2006-01-06 00:20:32 -08001908 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 mddev->private = conf;
1910 if (!conf)
1911 goto out_no_mem;
1912
NeilBrown9ffae0c2006-01-06 00:20:32 -08001913 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 GFP_KERNEL);
1915 if (!conf->mirrors)
1916 goto out_no_mem;
1917
NeilBrownddaf22a2006-01-06 00:20:19 -08001918 conf->tmppage = alloc_page(GFP_KERNEL);
1919 if (!conf->tmppage)
1920 goto out_no_mem;
1921
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1923 if (!conf->poolinfo)
1924 goto out_no_mem;
1925 conf->poolinfo->mddev = mddev;
1926 conf->poolinfo->raid_disks = mddev->raid_disks;
1927 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1928 r1bio_pool_free,
1929 conf->poolinfo);
1930 if (!conf->r1bio_pool)
1931 goto out_no_mem;
1932
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 ITERATE_RDEV(mddev, rdev, tmp) {
1934 disk_idx = rdev->raid_disk;
1935 if (disk_idx >= mddev->raid_disks
1936 || disk_idx < 0)
1937 continue;
1938 disk = conf->mirrors + disk_idx;
1939
1940 disk->rdev = rdev;
1941
1942 blk_queue_stack_limits(mddev->queue,
1943 rdev->bdev->bd_disk->queue);
1944 /* as we don't honour merge_bvec_fn, we must never risk
1945 * violating it, so limit ->max_sector to one PAGE, as
1946 * a one page request is never in violation.
1947 */
1948 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1949 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1950 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1951
1952 disk->head_position = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 }
1954 conf->raid_disks = mddev->raid_disks;
1955 conf->mddev = mddev;
1956 spin_lock_init(&conf->device_lock);
1957 INIT_LIST_HEAD(&conf->retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
1959 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08001960 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
NeilBrown191ea9b2005-06-21 17:17:23 -07001962 bio_list_init(&conf->pending_bio_list);
1963 bio_list_init(&conf->flushing_bio_list);
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
1966 mddev->degraded = 0;
1967 for (i = 0; i < conf->raid_disks; i++) {
1968
1969 disk = conf->mirrors + i;
1970
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001971 if (!disk->rdev ||
1972 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 disk->head_position = 0;
1974 mddev->degraded++;
NeilBrown918f0232007-08-22 14:01:52 -07001975 if (disk->rdev)
1976 conf->fullsync = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 }
1978 }
NeilBrown11ce99e2006-10-03 01:15:52 -07001979 if (mddev->degraded == conf->raid_disks) {
1980 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1981 mdname(mddev));
1982 goto out_free_conf;
1983 }
1984 if (conf->raid_disks - mddev->degraded == 1)
1985 mddev->recovery_cp = MaxSector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
1987 /*
1988 * find the first working one and use it as a starting point
1989 * to read balancing.
1990 */
1991 for (j = 0; j < conf->raid_disks &&
1992 (!conf->mirrors[j].rdev ||
NeilBrownb2d444d2005-11-08 21:39:31 -08001993 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 /* nothing */;
1995 conf->last_used = j;
1996
1997
NeilBrown191ea9b2005-06-21 17:17:23 -07001998 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1999 if (!mddev->thread) {
2000 printk(KERN_ERR
2001 "raid1: couldn't allocate thread for %s\n",
2002 mdname(mddev));
2003 goto out_free_conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002005
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 printk(KERN_INFO
2007 "raid1: raid set %s active with %d out of %d mirrors\n",
2008 mdname(mddev), mddev->raid_disks - mddev->degraded,
2009 mddev->raid_disks);
2010 /*
2011 * Ok, everything is just fine now
2012 */
2013 mddev->array_size = mddev->size;
2014
NeilBrown7a5febe2005-05-16 21:53:16 -07002015 mddev->queue->unplug_fn = raid1_unplug;
2016 mddev->queue->issue_flush_fn = raid1_issue_flush;
NeilBrown0d129222006-10-03 01:15:54 -07002017 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2018 mddev->queue->backing_dev_info.congested_data = mddev;
NeilBrown7a5febe2005-05-16 21:53:16 -07002019
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 return 0;
2021
2022out_no_mem:
2023 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
2024 mdname(mddev));
2025
2026out_free_conf:
2027 if (conf) {
2028 if (conf->r1bio_pool)
2029 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002030 kfree(conf->mirrors);
NeilBrown1345b1d2006-01-06 00:20:40 -08002031 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002032 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 kfree(conf);
2034 mddev->private = NULL;
2035 }
2036out:
2037 return -EIO;
2038}
2039
2040static int stop(mddev_t *mddev)
2041{
2042 conf_t *conf = mddev_to_conf(mddev);
NeilBrown4b6d2872005-09-09 16:23:47 -07002043 struct bitmap *bitmap = mddev->bitmap;
2044 int behind_wait = 0;
2045
2046 /* wait for behind writes to complete */
2047 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
2048 behind_wait++;
2049 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
2050 set_current_state(TASK_UNINTERRUPTIBLE);
2051 schedule_timeout(HZ); /* wait a second */
2052 /* need to kick something here to make sure I/O goes? */
2053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
2055 md_unregister_thread(mddev->thread);
2056 mddev->thread = NULL;
2057 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
2058 if (conf->r1bio_pool)
2059 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002060 kfree(conf->mirrors);
2061 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 kfree(conf);
2063 mddev->private = NULL;
2064 return 0;
2065}
2066
2067static int raid1_resize(mddev_t *mddev, sector_t sectors)
2068{
2069 /* no resync is happening, and there is enough space
2070 * on all devices, so we can resize.
2071 * We need to make sure resync covers any new space.
2072 * If the array is shrinking we should possibly wait until
2073 * any io in the removed space completes, but it hardly seems
2074 * worth it.
2075 */
2076 mddev->array_size = sectors>>1;
2077 set_capacity(mddev->gendisk, mddev->array_size << 1);
Linus Torvalds44ce62942007-05-09 18:51:36 -07002078 mddev->changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
2080 mddev->recovery_cp = mddev->size << 1;
2081 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2082 }
2083 mddev->size = mddev->array_size;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002084 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 return 0;
2086}
2087
NeilBrown63c70c42006-03-27 01:18:13 -08002088static int raid1_reshape(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089{
2090 /* We need to:
2091 * 1/ resize the r1bio_pool
2092 * 2/ resize conf->mirrors
2093 *
2094 * We allocate a new r1bio_pool if we can.
2095 * Then raise a device barrier and wait until all IO stops.
2096 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07002097 *
2098 * At the same time, we "pack" the devices so that all the missing
2099 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 */
2101 mempool_t *newpool, *oldpool;
2102 struct pool_info *newpoolinfo;
2103 mirror_info_t *newmirrors;
2104 conf_t *conf = mddev_to_conf(mddev);
NeilBrown63c70c42006-03-27 01:18:13 -08002105 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07002106 unsigned long flags;
NeilBrown6ea9c072005-06-21 17:17:09 -07002107 int d, d2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
NeilBrown63c70c42006-03-27 01:18:13 -08002109 /* Cannot change chunk_size, layout, or level */
2110 if (mddev->chunk_size != mddev->new_chunk ||
2111 mddev->layout != mddev->new_layout ||
2112 mddev->level != mddev->new_level) {
2113 mddev->new_chunk = mddev->chunk_size;
2114 mddev->new_layout = mddev->layout;
2115 mddev->new_level = mddev->level;
2116 return -EINVAL;
2117 }
2118
NeilBrown2a2275d2007-01-26 00:57:11 -08002119 md_allow_write(mddev);
2120
NeilBrown63c70c42006-03-27 01:18:13 -08002121 raid_disks = mddev->raid_disks + mddev->delta_disks;
2122
NeilBrown6ea9c072005-06-21 17:17:09 -07002123 if (raid_disks < conf->raid_disks) {
2124 cnt=0;
2125 for (d= 0; d < conf->raid_disks; d++)
2126 if (conf->mirrors[d].rdev)
2127 cnt++;
2128 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07002130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
2132 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2133 if (!newpoolinfo)
2134 return -ENOMEM;
2135 newpoolinfo->mddev = mddev;
2136 newpoolinfo->raid_disks = raid_disks;
2137
2138 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2139 r1bio_pool_free, newpoolinfo);
2140 if (!newpool) {
2141 kfree(newpoolinfo);
2142 return -ENOMEM;
2143 }
NeilBrown9ffae0c2006-01-06 00:20:32 -08002144 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 if (!newmirrors) {
2146 kfree(newpoolinfo);
2147 mempool_destroy(newpool);
2148 return -ENOMEM;
2149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
NeilBrown17999be2006-01-06 00:20:12 -08002151 raise_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
2153 /* ok, everything is stopped */
2154 oldpool = conf->r1bio_pool;
2155 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07002156
NeilBrowna88aa782007-08-22 14:01:53 -07002157 for (d = d2 = 0; d < conf->raid_disks; d++) {
2158 mdk_rdev_t *rdev = conf->mirrors[d].rdev;
2159 if (rdev && rdev->raid_disk != d2) {
2160 char nm[20];
2161 sprintf(nm, "rd%d", rdev->raid_disk);
2162 sysfs_remove_link(&mddev->kobj, nm);
2163 rdev->raid_disk = d2;
2164 sprintf(nm, "rd%d", rdev->raid_disk);
2165 sysfs_remove_link(&mddev->kobj, nm);
2166 if (sysfs_create_link(&mddev->kobj,
2167 &rdev->kobj, nm))
2168 printk(KERN_WARNING
2169 "md/raid1: cannot register "
2170 "%s for %s\n",
2171 nm, mdname(mddev));
NeilBrown6ea9c072005-06-21 17:17:09 -07002172 }
NeilBrowna88aa782007-08-22 14:01:53 -07002173 if (rdev)
2174 newmirrors[d2++].rdev = rdev;
2175 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 kfree(conf->mirrors);
2177 conf->mirrors = newmirrors;
2178 kfree(conf->poolinfo);
2179 conf->poolinfo = newpoolinfo;
2180
NeilBrownc04be0a2006-10-03 01:15:53 -07002181 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07002183 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08002185 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
NeilBrown6ea9c072005-06-21 17:17:09 -07002187 conf->last_used = 0; /* just make sure it is in-range */
NeilBrown17999be2006-01-06 00:20:12 -08002188 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
2190 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2191 md_wakeup_thread(mddev->thread);
2192
2193 mempool_destroy(oldpool);
2194 return 0;
2195}
2196
NeilBrown500af872005-09-09 16:23:58 -07002197static void raid1_quiesce(mddev_t *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07002198{
2199 conf_t *conf = mddev_to_conf(mddev);
2200
2201 switch(state) {
NeilBrown9e6603d2005-09-09 16:23:48 -07002202 case 1:
NeilBrown17999be2006-01-06 00:20:12 -08002203 raise_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002204 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002205 case 0:
NeilBrown17999be2006-01-06 00:20:12 -08002206 lower_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002207 break;
2208 }
NeilBrown36fa3062005-09-09 16:23:45 -07002209}
2210
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211
NeilBrown2604b702006-01-06 00:20:36 -08002212static struct mdk_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213{
2214 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08002215 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 .owner = THIS_MODULE,
2217 .make_request = make_request,
2218 .run = run,
2219 .stop = stop,
2220 .status = status,
2221 .error_handler = error,
2222 .hot_add_disk = raid1_add_disk,
2223 .hot_remove_disk= raid1_remove_disk,
2224 .spare_active = raid1_spare_active,
2225 .sync_request = sync_request,
2226 .resize = raid1_resize,
NeilBrown63c70c42006-03-27 01:18:13 -08002227 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07002228 .quiesce = raid1_quiesce,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229};
2230
2231static int __init raid_init(void)
2232{
NeilBrown2604b702006-01-06 00:20:36 -08002233 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234}
2235
2236static void raid_exit(void)
2237{
NeilBrown2604b702006-01-06 00:20:36 -08002238 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239}
2240
2241module_init(raid_init);
2242module_exit(raid_exit);
2243MODULE_LICENSE("GPL");
2244MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002245MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08002246MODULE_ALIAS("md-level-1");