blob: 3fc9ec2394786a1664ebd98f1f029e8811465a97 [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
NeilBrown11ce99e2006-10-03 01:15:52 -0700274 if (uptodate || (conf->raid_disks - conf->mddev->degraded) <= 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 /*
276 * Set R1BIO_Uptodate in our master bio, so that
277 * we will return a good error code for to the higher
278 * levels even if IO on some other mirrored buffer fails.
279 *
280 * The 'master' represents the composite IO operation to
281 * user-side. So if something waits for IO, then it will
282 * wait for the 'master' bio.
283 */
NeilBrown220946c2006-01-06 00:20:27 -0800284 if (uptodate)
285 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 raid_end_bio_io(r1_bio);
NeilBrownddaf22a2006-01-06 00:20:19 -0800288 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 /*
290 * oops, read error:
291 */
292 char b[BDEVNAME_SIZE];
293 if (printk_ratelimit())
294 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
295 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
296 reschedule_retry(r1_bio);
297 }
298
299 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
300 return 0;
301}
302
303static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error)
304{
305 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
306 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
NeilBrowna9701a32005-11-08 21:39:34 -0800307 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 conf_t *conf = mddev_to_conf(r1_bio->mddev);
NeilBrown04b857f2006-03-09 17:33:46 -0800309 struct bio *to_put = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
311 if (bio->bi_size)
312 return 1;
313
314 for (mirror = 0; mirror < conf->raid_disks; mirror++)
315 if (r1_bio->bios[mirror] == bio)
316 break;
317
NeilBrownbea27712006-05-01 12:15:46 -0700318 if (error == -EOPNOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
NeilBrowna9701a32005-11-08 21:39:34 -0800319 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
320 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
321 r1_bio->mddev->barriers_work = 0;
NeilBrown5e7dd2a2006-05-01 12:15:47 -0700322 /* Don't rdev_dec_pending in this branch - keep it for the retry */
NeilBrowna9701a32005-11-08 21:39:34 -0800323 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 /*
NeilBrowna9701a32005-11-08 21:39:34 -0800325 * this branch is our 'one mirror IO has finished' event handler:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 */
NeilBrowna9701a32005-11-08 21:39:34 -0800327 r1_bio->bios[mirror] = NULL;
NeilBrown04b857f2006-03-09 17:33:46 -0800328 to_put = bio;
NeilBrowna9701a32005-11-08 21:39:34 -0800329 if (!uptodate) {
330 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
331 /* an I/O failed, we can't clear the bitmap */
332 set_bit(R1BIO_Degraded, &r1_bio->state);
333 } else
334 /*
335 * Set R1BIO_Uptodate in our master bio, so that
336 * we will return a good error code for to the higher
337 * levels even if IO on some other mirrored buffer fails.
338 *
339 * The 'master' represents the composite IO operation to
340 * user-side. So if something waits for IO, then it will
341 * wait for the 'master' bio.
342 */
343 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
NeilBrowna9701a32005-11-08 21:39:34 -0800345 update_head_pos(mirror, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
NeilBrowna9701a32005-11-08 21:39:34 -0800347 if (behind) {
348 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
349 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700350
NeilBrowna9701a32005-11-08 21:39:34 -0800351 /* In behind mode, we ACK the master bio once the I/O has safely
352 * reached all non-writemostly disks. Setting the Returned bit
353 * ensures that this gets done only once -- we don't ever want to
354 * return -EIO here, instead we'll wait */
NeilBrown4b6d2872005-09-09 16:23:47 -0700355
NeilBrowna9701a32005-11-08 21:39:34 -0800356 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
357 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
358 /* Maybe we can return now */
359 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
360 struct bio *mbio = r1_bio->master_bio;
361 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
362 (unsigned long long) mbio->bi_sector,
363 (unsigned long long) mbio->bi_sector +
364 (mbio->bi_size >> 9) - 1);
365 bio_endio(mbio, mbio->bi_size, 0);
366 }
NeilBrown4b6d2872005-09-09 16:23:47 -0700367 }
368 }
NeilBrown5e7dd2a2006-05-01 12:15:47 -0700369 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
NeilBrown4b6d2872005-09-09 16:23:47 -0700370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 /*
372 *
373 * Let's see if all mirrored write operations have finished
374 * already.
375 */
376 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrownc70810b2006-06-26 00:27:35 -0700377 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state))
NeilBrowna9701a32005-11-08 21:39:34 -0800378 reschedule_retry(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700379 else {
380 /* it really is the end of this request */
381 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
382 /* free extra copy of the data pages */
383 int i = bio->bi_vcnt;
384 while (i--)
385 safe_put_page(bio->bi_io_vec[i].bv_page);
386 }
387 /* clear the bitmap if all writes complete successfully */
388 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
389 r1_bio->sectors,
390 !test_bit(R1BIO_Degraded, &r1_bio->state),
391 behind);
392 md_write_end(r1_bio->mddev);
393 raid_end_bio_io(r1_bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
NeilBrownc70810b2006-06-26 00:27:35 -0700396
NeilBrown04b857f2006-03-09 17:33:46 -0800397 if (to_put)
398 bio_put(to_put);
399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 return 0;
401}
402
403
404/*
405 * This routine returns the disk from which the requested read should
406 * be done. There is a per-array 'next expected sequential IO' sector
407 * number - if this matches on the next IO then we use the last disk.
408 * There is also a per-disk 'last know head position' sector that is
409 * maintained from IRQ contexts, both the normal and the resync IO
410 * completion handlers update this position correctly. If there is no
411 * perfect sequential match then we pick the disk whose head is closest.
412 *
413 * If there are 2 mirrors in the same 2 devices, performance degrades
414 * because position is mirror, not device based.
415 *
416 * The rdev for the device selected will have nr_pending incremented.
417 */
418static int read_balance(conf_t *conf, r1bio_t *r1_bio)
419{
420 const unsigned long this_sector = r1_bio->sector;
421 int new_disk = conf->last_used, disk = new_disk;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700422 int wonly_disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 const int sectors = r1_bio->sectors;
424 sector_t new_distance, current_distance;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700425 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
427 rcu_read_lock();
428 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700429 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 * device if no resync is going on, or below the resync window.
431 * We take the first readable disk when above the resync window.
432 */
433 retry:
434 if (conf->mddev->recovery_cp < MaxSector &&
435 (this_sector + sectors >= conf->next_resync)) {
436 /* Choose the first operation device, for consistancy */
437 new_disk = 0;
438
Suzanne Woodd6065f72005-11-08 21:39:27 -0800439 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrowncf30a472006-01-06 00:20:23 -0800440 r1_bio->bios[new_disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800441 !rdev || !test_bit(In_sync, &rdev->flags)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700442 || test_bit(WriteMostly, &rdev->flags);
Suzanne Woodd6065f72005-11-08 21:39:27 -0800443 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700444
NeilBrowncf30a472006-01-06 00:20:23 -0800445 if (rdev && test_bit(In_sync, &rdev->flags) &&
446 r1_bio->bios[new_disk] != IO_BLOCKED)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700447 wonly_disk = new_disk;
448
449 if (new_disk == conf->raid_disks - 1) {
450 new_disk = wonly_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 break;
452 }
453 }
454 goto rb_out;
455 }
456
457
458 /* make sure the disk is operational */
Suzanne Woodd6065f72005-11-08 21:39:27 -0800459 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrowncf30a472006-01-06 00:20:23 -0800460 r1_bio->bios[new_disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800461 !rdev || !test_bit(In_sync, &rdev->flags) ||
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700462 test_bit(WriteMostly, &rdev->flags);
Suzanne Woodd6065f72005-11-08 21:39:27 -0800463 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700464
NeilBrowncf30a472006-01-06 00:20:23 -0800465 if (rdev && test_bit(In_sync, &rdev->flags) &&
466 r1_bio->bios[new_disk] != IO_BLOCKED)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700467 wonly_disk = new_disk;
468
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (new_disk <= 0)
470 new_disk = conf->raid_disks;
471 new_disk--;
472 if (new_disk == disk) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700473 new_disk = wonly_disk;
474 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 }
476 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700477
478 if (new_disk < 0)
479 goto rb_out;
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 disk = new_disk;
482 /* now disk == new_disk == starting point for search */
483
484 /*
485 * Don't change to another disk for sequential reads:
486 */
487 if (conf->next_seq_sect == this_sector)
488 goto rb_out;
489 if (this_sector == conf->mirrors[new_disk].head_position)
490 goto rb_out;
491
492 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
493
494 /* Find the disk whose head is closest */
495
496 do {
497 if (disk <= 0)
498 disk = conf->raid_disks;
499 disk--;
500
Suzanne Woodd6065f72005-11-08 21:39:27 -0800501 rdev = rcu_dereference(conf->mirrors[disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700502
NeilBrowncf30a472006-01-06 00:20:23 -0800503 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800504 !test_bit(In_sync, &rdev->flags) ||
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700505 test_bit(WriteMostly, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 continue;
507
508 if (!atomic_read(&rdev->nr_pending)) {
509 new_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 break;
511 }
512 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
513 if (new_distance < current_distance) {
514 current_distance = new_distance;
515 new_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 }
517 } while (disk != conf->last_used);
518
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700519 rb_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521
522 if (new_disk >= 0) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800523 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700524 if (!rdev)
525 goto retry;
526 atomic_inc(&rdev->nr_pending);
NeilBrownb2d444d2005-11-08 21:39:31 -0800527 if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 /* cannot risk returning a device that failed
529 * before we inc'ed nr_pending
530 */
NeilBrown03c902e2006-01-06 00:20:46 -0800531 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 goto retry;
533 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700534 conf->next_seq_sect = this_sector + sectors;
535 conf->last_used = new_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 }
537 rcu_read_unlock();
538
539 return new_disk;
540}
541
542static void unplug_slaves(mddev_t *mddev)
543{
544 conf_t *conf = mddev_to_conf(mddev);
545 int i;
546
547 rcu_read_lock();
548 for (i=0; i<mddev->raid_disks; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800549 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -0800550 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 request_queue_t *r_queue = bdev_get_queue(rdev->bdev);
552
553 atomic_inc(&rdev->nr_pending);
554 rcu_read_unlock();
555
556 if (r_queue->unplug_fn)
557 r_queue->unplug_fn(r_queue);
558
559 rdev_dec_pending(rdev, mddev);
560 rcu_read_lock();
561 }
562 }
563 rcu_read_unlock();
564}
565
566static void raid1_unplug(request_queue_t *q)
567{
NeilBrown191ea9b2005-06-21 17:17:23 -0700568 mddev_t *mddev = q->queuedata;
569
570 unplug_slaves(mddev);
571 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572}
573
574static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk,
575 sector_t *error_sector)
576{
577 mddev_t *mddev = q->queuedata;
578 conf_t *conf = mddev_to_conf(mddev);
579 int i, ret = 0;
580
581 rcu_read_lock();
582 for (i=0; i<mddev->raid_disks && ret == 0; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800583 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -0800584 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 struct block_device *bdev = rdev->bdev;
586 request_queue_t *r_queue = bdev_get_queue(bdev);
587
588 if (!r_queue->issue_flush_fn)
589 ret = -EOPNOTSUPP;
590 else {
591 atomic_inc(&rdev->nr_pending);
592 rcu_read_unlock();
593 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk,
594 error_sector);
595 rdev_dec_pending(rdev, mddev);
596 rcu_read_lock();
597 }
598 }
599 }
600 rcu_read_unlock();
601 return ret;
602}
603
NeilBrown17999be2006-01-06 00:20:12 -0800604/* Barriers....
605 * Sometimes we need to suspend IO while we do something else,
606 * either some resync/recovery, or reconfigure the array.
607 * To do this we raise a 'barrier'.
608 * The 'barrier' is a counter that can be raised multiple times
609 * to count how many activities are happening which preclude
610 * normal IO.
611 * We can only raise the barrier if there is no pending IO.
612 * i.e. if nr_pending == 0.
613 * We choose only to raise the barrier if no-one is waiting for the
614 * barrier to go down. This means that as soon as an IO request
615 * is ready, no other operations which require a barrier will start
616 * until the IO request has had a chance.
617 *
618 * So: regular IO calls 'wait_barrier'. When that returns there
619 * is no backgroup IO happening, It must arrange to call
620 * allow_barrier when it has finished its IO.
621 * backgroup IO calls must call raise_barrier. Once that returns
622 * there is no normal IO happeing. It must arrange to call
623 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 */
625#define RESYNC_DEPTH 32
626
NeilBrown17999be2006-01-06 00:20:12 -0800627static void raise_barrier(conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800630
631 /* Wait until no block IO is waiting */
632 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
633 conf->resync_lock,
634 raid1_unplug(conf->mddev->queue));
635
636 /* block any new IO from starting */
637 conf->barrier++;
638
639 /* No wait for all pending IO to complete */
640 wait_event_lock_irq(conf->wait_barrier,
641 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
642 conf->resync_lock,
643 raid1_unplug(conf->mddev->queue));
644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 spin_unlock_irq(&conf->resync_lock);
646}
647
NeilBrown17999be2006-01-06 00:20:12 -0800648static void lower_barrier(conf_t *conf)
649{
650 unsigned long flags;
651 spin_lock_irqsave(&conf->resync_lock, flags);
652 conf->barrier--;
653 spin_unlock_irqrestore(&conf->resync_lock, flags);
654 wake_up(&conf->wait_barrier);
655}
656
657static void wait_barrier(conf_t *conf)
658{
659 spin_lock_irq(&conf->resync_lock);
660 if (conf->barrier) {
661 conf->nr_waiting++;
662 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
663 conf->resync_lock,
664 raid1_unplug(conf->mddev->queue));
665 conf->nr_waiting--;
666 }
667 conf->nr_pending++;
668 spin_unlock_irq(&conf->resync_lock);
669}
670
671static void allow_barrier(conf_t *conf)
672{
673 unsigned long flags;
674 spin_lock_irqsave(&conf->resync_lock, flags);
675 conf->nr_pending--;
676 spin_unlock_irqrestore(&conf->resync_lock, flags);
677 wake_up(&conf->wait_barrier);
678}
679
NeilBrownddaf22a2006-01-06 00:20:19 -0800680static void freeze_array(conf_t *conf)
681{
682 /* stop syncio and normal IO and wait for everything to
683 * go quite.
684 * We increment barrier and nr_waiting, and then
685 * wait until barrier+nr_pending match nr_queued+2
686 */
687 spin_lock_irq(&conf->resync_lock);
688 conf->barrier++;
689 conf->nr_waiting++;
690 wait_event_lock_irq(conf->wait_barrier,
691 conf->barrier+conf->nr_pending == conf->nr_queued+2,
692 conf->resync_lock,
693 raid1_unplug(conf->mddev->queue));
694 spin_unlock_irq(&conf->resync_lock);
695}
696static void unfreeze_array(conf_t *conf)
697{
698 /* reverse the effect of the freeze */
699 spin_lock_irq(&conf->resync_lock);
700 conf->barrier--;
701 conf->nr_waiting--;
702 wake_up(&conf->wait_barrier);
703 spin_unlock_irq(&conf->resync_lock);
704}
705
NeilBrown17999be2006-01-06 00:20:12 -0800706
NeilBrown4b6d2872005-09-09 16:23:47 -0700707/* duplicate the data pages for behind I/O */
708static struct page **alloc_behind_pages(struct bio *bio)
709{
710 int i;
711 struct bio_vec *bvec;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800712 struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *),
NeilBrown4b6d2872005-09-09 16:23:47 -0700713 GFP_NOIO);
714 if (unlikely(!pages))
715 goto do_sync_io;
716
NeilBrown4b6d2872005-09-09 16:23:47 -0700717 bio_for_each_segment(bvec, bio, i) {
718 pages[i] = alloc_page(GFP_NOIO);
719 if (unlikely(!pages[i]))
720 goto do_sync_io;
721 memcpy(kmap(pages[i]) + bvec->bv_offset,
722 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
723 kunmap(pages[i]);
724 kunmap(bvec->bv_page);
725 }
726
727 return pages;
728
729do_sync_io:
730 if (pages)
731 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
NeilBrown2d1f3b52006-01-06 00:20:31 -0800732 put_page(pages[i]);
NeilBrown4b6d2872005-09-09 16:23:47 -0700733 kfree(pages);
734 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
735 return NULL;
736}
737
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738static int make_request(request_queue_t *q, struct bio * bio)
739{
740 mddev_t *mddev = q->queuedata;
741 conf_t *conf = mddev_to_conf(mddev);
742 mirror_info_t *mirror;
743 r1bio_t *r1_bio;
744 struct bio *read_bio;
NeilBrown191ea9b2005-06-21 17:17:23 -0700745 int i, targets = 0, disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 mdk_rdev_t *rdev;
NeilBrown191ea9b2005-06-21 17:17:23 -0700747 struct bitmap *bitmap = mddev->bitmap;
748 unsigned long flags;
749 struct bio_list bl;
NeilBrown4b6d2872005-09-09 16:23:47 -0700750 struct page **behind_pages = NULL;
Jens Axboea3623572005-11-01 09:26:16 +0100751 const int rw = bio_data_dir(bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800752 int do_barriers;
NeilBrown191ea9b2005-06-21 17:17:23 -0700753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 /*
755 * Register the new request and wait if the reconstruction
756 * thread has put up a bar for new requests.
757 * Continue immediately if no resync is active currently.
NeilBrown62de6082006-05-01 12:15:47 -0700758 * We test barriers_work *after* md_write_start as md_write_start
759 * may cause the first superblock write, and that will check out
760 * if barriers work.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 */
NeilBrown62de6082006-05-01 12:15:47 -0700762
NeilBrown3d310eb2005-06-21 17:17:26 -0700763 md_write_start(mddev, bio); /* wait on superblock update early */
764
NeilBrown62de6082006-05-01 12:15:47 -0700765 if (unlikely(!mddev->barriers_work && bio_barrier(bio))) {
766 if (rw == WRITE)
767 md_write_end(mddev);
768 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
769 return 0;
770 }
771
NeilBrown17999be2006-01-06 00:20:12 -0800772 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Jens Axboea3623572005-11-01 09:26:16 +0100774 disk_stat_inc(mddev->gendisk, ios[rw]);
775 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 /*
778 * make_request() can abort the operation when READA is being
779 * used and no empty request is available.
780 *
781 */
782 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
783
784 r1_bio->master_bio = bio;
785 r1_bio->sectors = bio->bi_size >> 9;
NeilBrown191ea9b2005-06-21 17:17:23 -0700786 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 r1_bio->mddev = mddev;
788 r1_bio->sector = bio->bi_sector;
789
Jens Axboea3623572005-11-01 09:26:16 +0100790 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 /*
792 * read balancing logic:
793 */
794 int rdisk = read_balance(conf, r1_bio);
795
796 if (rdisk < 0) {
797 /* couldn't find anywhere to read from */
798 raid_end_bio_io(r1_bio);
799 return 0;
800 }
801 mirror = conf->mirrors + rdisk;
802
803 r1_bio->read_disk = rdisk;
804
805 read_bio = bio_clone(bio, GFP_NOIO);
806
807 r1_bio->bios[rdisk] = read_bio;
808
809 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
810 read_bio->bi_bdev = mirror->rdev->bdev;
811 read_bio->bi_end_io = raid1_end_read_request;
812 read_bio->bi_rw = READ;
813 read_bio->bi_private = r1_bio;
814
815 generic_make_request(read_bio);
816 return 0;
817 }
818
819 /*
820 * WRITE:
821 */
822 /* first select target devices under spinlock and
823 * inc refcount on their rdev. Record them by setting
824 * bios[x] to bio
825 */
826 disks = conf->raid_disks;
NeilBrown191ea9b2005-06-21 17:17:23 -0700827#if 0
828 { static int first=1;
829 if (first) printk("First Write sector %llu disks %d\n",
830 (unsigned long long)r1_bio->sector, disks);
831 first = 0;
832 }
833#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 rcu_read_lock();
835 for (i = 0; i < disks; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800836 if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL &&
NeilBrownb2d444d2005-11-08 21:39:31 -0800837 !test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 atomic_inc(&rdev->nr_pending);
NeilBrownb2d444d2005-11-08 21:39:31 -0800839 if (test_bit(Faulty, &rdev->flags)) {
NeilBrown03c902e2006-01-06 00:20:46 -0800840 rdev_dec_pending(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 r1_bio->bios[i] = NULL;
842 } else
843 r1_bio->bios[i] = bio;
NeilBrown191ea9b2005-06-21 17:17:23 -0700844 targets++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 } else
846 r1_bio->bios[i] = NULL;
847 }
848 rcu_read_unlock();
849
NeilBrown4b6d2872005-09-09 16:23:47 -0700850 BUG_ON(targets == 0); /* we never fail the last device */
851
NeilBrown191ea9b2005-06-21 17:17:23 -0700852 if (targets < conf->raid_disks) {
853 /* array is degraded, we will not clear the bitmap
854 * on I/O completion (see raid1_end_write_request) */
855 set_bit(R1BIO_Degraded, &r1_bio->state);
856 }
NeilBrown06d91a52005-06-21 17:17:12 -0700857
NeilBrown4b6d2872005-09-09 16:23:47 -0700858 /* do behind I/O ? */
859 if (bitmap &&
860 atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind &&
861 (behind_pages = alloc_behind_pages(bio)) != NULL)
862 set_bit(R1BIO_BehindIO, &r1_bio->state);
863
NeilBrown191ea9b2005-06-21 17:17:23 -0700864 atomic_set(&r1_bio->remaining, 0);
NeilBrown4b6d2872005-09-09 16:23:47 -0700865 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -0700866
NeilBrown04b857f2006-03-09 17:33:46 -0800867 do_barriers = bio_barrier(bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800868 if (do_barriers)
869 set_bit(R1BIO_Barrier, &r1_bio->state);
870
NeilBrown191ea9b2005-06-21 17:17:23 -0700871 bio_list_init(&bl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 for (i = 0; i < disks; i++) {
873 struct bio *mbio;
874 if (!r1_bio->bios[i])
875 continue;
876
877 mbio = bio_clone(bio, GFP_NOIO);
878 r1_bio->bios[i] = mbio;
879
880 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
881 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
882 mbio->bi_end_io = raid1_end_write_request;
NeilBrowna9701a32005-11-08 21:39:34 -0800883 mbio->bi_rw = WRITE | do_barriers;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 mbio->bi_private = r1_bio;
885
NeilBrown4b6d2872005-09-09 16:23:47 -0700886 if (behind_pages) {
887 struct bio_vec *bvec;
888 int j;
889
890 /* Yes, I really want the '__' version so that
891 * we clear any unused pointer in the io_vec, rather
892 * than leave them unchanged. This is important
893 * because when we come to free the pages, we won't
894 * know the originial bi_idx, so we just free
895 * them all
896 */
897 __bio_for_each_segment(bvec, mbio, j, 0)
898 bvec->bv_page = behind_pages[j];
899 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
900 atomic_inc(&r1_bio->behind_remaining);
901 }
902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 atomic_inc(&r1_bio->remaining);
NeilBrown191ea9b2005-06-21 17:17:23 -0700904
905 bio_list_add(&bl, mbio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 }
NeilBrown4b6d2872005-09-09 16:23:47 -0700907 kfree(behind_pages); /* the behind pages are attached to the bios now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
NeilBrown4b6d2872005-09-09 16:23:47 -0700909 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
910 test_bit(R1BIO_BehindIO, &r1_bio->state));
NeilBrown191ea9b2005-06-21 17:17:23 -0700911 spin_lock_irqsave(&conf->device_lock, flags);
912 bio_list_merge(&conf->pending_bio_list, &bl);
913 bio_list_init(&bl);
914
915 blk_plug_device(mddev->queue);
916 spin_unlock_irqrestore(&conf->device_lock, flags);
917
918#if 0
919 while ((bio = bio_list_pop(&bl)) != NULL)
920 generic_make_request(bio);
921#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 return 0;
924}
925
926static void status(struct seq_file *seq, mddev_t *mddev)
927{
928 conf_t *conf = mddev_to_conf(mddev);
929 int i;
930
931 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -0700932 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -0700933 rcu_read_lock();
934 for (i = 0; i < conf->raid_disks; i++) {
935 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -0700937 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
938 }
939 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 seq_printf(seq, "]");
941}
942
943
944static void error(mddev_t *mddev, mdk_rdev_t *rdev)
945{
946 char b[BDEVNAME_SIZE];
947 conf_t *conf = mddev_to_conf(mddev);
948
949 /*
950 * If it is not operational, then we have already marked it as dead
951 * else if it is the last working disks, ignore the error, let the
952 * next level up know.
953 * else mark the drive as failed
954 */
NeilBrownb2d444d2005-11-08 21:39:31 -0800955 if (test_bit(In_sync, &rdev->flags)
NeilBrown11ce99e2006-10-03 01:15:52 -0700956 && (conf->raid_disks - mddev->degraded) == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 /*
958 * Don't fail the drive, act as though we were just a
959 * normal single drive
960 */
961 return;
NeilBrownb2d444d2005-11-08 21:39:31 -0800962 if (test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 mddev->degraded++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 /*
965 * if recovery is running, make sure it aborts.
966 */
967 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
968 }
NeilBrownb2d444d2005-11-08 21:39:31 -0800969 clear_bit(In_sync, &rdev->flags);
970 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -0700971 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n"
973 " Operation continuing on %d devices\n",
NeilBrown11ce99e2006-10-03 01:15:52 -0700974 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
976
977static void print_conf(conf_t *conf)
978{
979 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 printk("RAID1 conf printout:\n");
982 if (!conf) {
983 printk("(!conf)\n");
984 return;
985 }
NeilBrown11ce99e2006-10-03 01:15:52 -0700986 printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 conf->raid_disks);
988
NeilBrownddac7c72006-08-31 21:27:36 -0700989 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 for (i = 0; i < conf->raid_disks; i++) {
991 char b[BDEVNAME_SIZE];
NeilBrownddac7c72006-08-31 21:27:36 -0700992 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
993 if (rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -0700995 i, !test_bit(In_sync, &rdev->flags),
996 !test_bit(Faulty, &rdev->flags),
997 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 }
NeilBrownddac7c72006-08-31 21:27:36 -0700999 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
1002static void close_sync(conf_t *conf)
1003{
NeilBrown17999be2006-01-06 00:20:12 -08001004 wait_barrier(conf);
1005 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
1007 mempool_destroy(conf->r1buf_pool);
1008 conf->r1buf_pool = NULL;
1009}
1010
1011static int raid1_spare_active(mddev_t *mddev)
1012{
1013 int i;
1014 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 /*
1017 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001018 * and mark them readable.
1019 * Called under mddev lock, so rcu protection not needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 */
1021 for (i = 0; i < conf->raid_disks; i++) {
NeilBrownddac7c72006-08-31 21:27:36 -07001022 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
1023 if (rdev
1024 && !test_bit(Faulty, &rdev->flags)
1025 && !test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 mddev->degraded--;
NeilBrownddac7c72006-08-31 21:27:36 -07001027 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 }
1029 }
1030
1031 print_conf(conf);
1032 return 0;
1033}
1034
1035
1036static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1037{
1038 conf_t *conf = mddev->private;
1039 int found = 0;
NeilBrown41158c72005-06-21 17:17:25 -07001040 int mirror = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 mirror_info_t *p;
1042
1043 for (mirror=0; mirror < mddev->raid_disks; mirror++)
1044 if ( !(p=conf->mirrors+mirror)->rdev) {
1045
1046 blk_queue_stack_limits(mddev->queue,
1047 rdev->bdev->bd_disk->queue);
1048 /* as we don't honour merge_bvec_fn, we must never risk
1049 * violating it, so limit ->max_sector to one PAGE, as
1050 * a one page request is never in violation.
1051 */
1052 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1053 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1054 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1055
1056 p->head_position = 0;
1057 rdev->raid_disk = mirror;
1058 found = 1;
NeilBrown6aea114a2005-11-28 13:44:13 -08001059 /* As all devices are equivalent, we don't need a full recovery
1060 * if this was recently any drive of the array
1061 */
1062 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001063 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001064 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 break;
1066 }
1067
1068 print_conf(conf);
1069 return found;
1070}
1071
1072static int raid1_remove_disk(mddev_t *mddev, int number)
1073{
1074 conf_t *conf = mddev->private;
1075 int err = 0;
1076 mdk_rdev_t *rdev;
1077 mirror_info_t *p = conf->mirrors+ number;
1078
1079 print_conf(conf);
1080 rdev = p->rdev;
1081 if (rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001082 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 atomic_read(&rdev->nr_pending)) {
1084 err = -EBUSY;
1085 goto abort;
1086 }
1087 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001088 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (atomic_read(&rdev->nr_pending)) {
1090 /* lost the race, try later */
1091 err = -EBUSY;
1092 p->rdev = rdev;
1093 }
1094 }
1095abort:
1096
1097 print_conf(conf);
1098 return err;
1099}
1100
1101
1102static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
1103{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
NeilBrownd11c1712006-01-06 00:20:26 -08001105 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
1107 if (bio->bi_size)
1108 return 1;
1109
NeilBrownd11c1712006-01-06 00:20:26 -08001110 for (i=r1_bio->mddev->raid_disks; i--; )
1111 if (r1_bio->bios[i] == bio)
1112 break;
1113 BUG_ON(i < 0);
1114 update_head_pos(i, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 /*
1116 * we have read a block, now it needs to be re-written,
1117 * or re-read if the read failed.
1118 * We don't do much here, just schedule handling by raid1d
1119 */
NeilBrown69382e82006-01-06 00:20:22 -08001120 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001122
1123 if (atomic_dec_and_test(&r1_bio->remaining))
1124 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return 0;
1126}
1127
1128static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
1129{
1130 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1131 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1132 mddev_t *mddev = r1_bio->mddev;
1133 conf_t *conf = mddev_to_conf(mddev);
1134 int i;
1135 int mirror=0;
1136
1137 if (bio->bi_size)
1138 return 1;
1139
1140 for (i = 0; i < conf->raid_disks; i++)
1141 if (r1_bio->bios[i] == bio) {
1142 mirror = i;
1143 break;
1144 }
NeilBrown6b1117d2006-03-31 02:31:57 -08001145 if (!uptodate) {
1146 int sync_blocks = 0;
1147 sector_t s = r1_bio->sector;
1148 long sectors_to_go = r1_bio->sectors;
1149 /* make sure these bits doesn't get cleared. */
1150 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001151 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001152 &sync_blocks, 1);
1153 s += sync_blocks;
1154 sectors_to_go -= sync_blocks;
1155 } while (sectors_to_go > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 md_error(mddev, conf->mirrors[mirror].rdev);
NeilBrown6b1117d2006-03-31 02:31:57 -08001157 }
NeilBrowne3b97032005-08-04 12:53:34 -07001158
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 update_head_pos(mirror, r1_bio);
1160
1161 if (atomic_dec_and_test(&r1_bio->remaining)) {
1162 md_done_sync(mddev, r1_bio->sectors, uptodate);
1163 put_buf(r1_bio);
1164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 return 0;
1166}
1167
1168static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1169{
1170 conf_t *conf = mddev_to_conf(mddev);
1171 int i;
1172 int disks = conf->raid_disks;
1173 struct bio *bio, *wbio;
1174
1175 bio = r1_bio->bios[r1_bio->read_disk];
1176
NeilBrown69382e82006-01-06 00:20:22 -08001177
NeilBrownd11c1712006-01-06 00:20:26 -08001178 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1179 /* We have read all readable devices. If we haven't
1180 * got the block, then there is no hope left.
1181 * If we have, then we want to do a comparison
1182 * and skip the write if everything is the same.
1183 * If any blocks failed to read, then we need to
1184 * attempt an over-write
1185 */
1186 int primary;
1187 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1188 for (i=0; i<mddev->raid_disks; i++)
1189 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1190 md_error(mddev, conf->mirrors[i].rdev);
1191
1192 md_done_sync(mddev, r1_bio->sectors, 1);
1193 put_buf(r1_bio);
1194 return;
1195 }
1196 for (primary=0; primary<mddev->raid_disks; primary++)
1197 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1198 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1199 r1_bio->bios[primary]->bi_end_io = NULL;
NeilBrown03c902e2006-01-06 00:20:46 -08001200 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
NeilBrownd11c1712006-01-06 00:20:26 -08001201 break;
1202 }
1203 r1_bio->read_disk = primary;
1204 for (i=0; i<mddev->raid_disks; i++)
1205 if (r1_bio->bios[i]->bi_end_io == end_sync_read &&
1206 test_bit(BIO_UPTODATE, &r1_bio->bios[i]->bi_flags)) {
1207 int j;
1208 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1209 struct bio *pbio = r1_bio->bios[primary];
1210 struct bio *sbio = r1_bio->bios[i];
1211 for (j = vcnt; j-- ; )
1212 if (memcmp(page_address(pbio->bi_io_vec[j].bv_page),
1213 page_address(sbio->bi_io_vec[j].bv_page),
1214 PAGE_SIZE))
1215 break;
1216 if (j >= 0)
1217 mddev->resync_mismatches += r1_bio->sectors;
NeilBrown03c902e2006-01-06 00:20:46 -08001218 if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
NeilBrownd11c1712006-01-06 00:20:26 -08001219 sbio->bi_end_io = NULL;
NeilBrown03c902e2006-01-06 00:20:46 -08001220 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1221 } else {
NeilBrownd11c1712006-01-06 00:20:26 -08001222 /* fixup the bio for reuse */
1223 sbio->bi_vcnt = vcnt;
1224 sbio->bi_size = r1_bio->sectors << 9;
1225 sbio->bi_idx = 0;
1226 sbio->bi_phys_segments = 0;
1227 sbio->bi_hw_segments = 0;
1228 sbio->bi_hw_front_size = 0;
1229 sbio->bi_hw_back_size = 0;
1230 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1231 sbio->bi_flags |= 1 << BIO_UPTODATE;
1232 sbio->bi_next = NULL;
1233 sbio->bi_sector = r1_bio->sector +
1234 conf->mirrors[i].rdev->data_offset;
1235 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1236 }
1237 }
1238 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
NeilBrown69382e82006-01-06 00:20:22 -08001240 /* ouch - failed to read all of that.
1241 * Try some synchronous reads of other devices to get
1242 * good data, much like with normal read errors. Only
NeilBrownddac7c72006-08-31 21:27:36 -07001243 * read into the pages we already have so we don't
NeilBrown69382e82006-01-06 00:20:22 -08001244 * need to re-issue the read request.
1245 * We don't need to freeze the array, because being in an
1246 * active sync request, there is no normal IO, and
1247 * no overlapping syncs.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 */
NeilBrown69382e82006-01-06 00:20:22 -08001249 sector_t sect = r1_bio->sector;
1250 int sectors = r1_bio->sectors;
1251 int idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
NeilBrown69382e82006-01-06 00:20:22 -08001253 while(sectors) {
1254 int s = sectors;
1255 int d = r1_bio->read_disk;
1256 int success = 0;
1257 mdk_rdev_t *rdev;
1258
1259 if (s > (PAGE_SIZE>>9))
1260 s = PAGE_SIZE >> 9;
1261 do {
1262 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07001263 /* No rcu protection needed here devices
1264 * can only be removed when no resync is
1265 * active, and resync is currently active
1266 */
NeilBrown69382e82006-01-06 00:20:22 -08001267 rdev = conf->mirrors[d].rdev;
1268 if (sync_page_io(rdev->bdev,
1269 sect + rdev->data_offset,
1270 s<<9,
1271 bio->bi_io_vec[idx].bv_page,
1272 READ)) {
1273 success = 1;
1274 break;
1275 }
1276 }
1277 d++;
1278 if (d == conf->raid_disks)
1279 d = 0;
1280 } while (!success && d != r1_bio->read_disk);
1281
1282 if (success) {
NeilBrown097426f2006-01-06 00:20:37 -08001283 int start = d;
NeilBrown69382e82006-01-06 00:20:22 -08001284 /* write it back and re-read */
1285 set_bit(R1BIO_Uptodate, &r1_bio->state);
1286 while (d != r1_bio->read_disk) {
1287 if (d == 0)
1288 d = conf->raid_disks;
1289 d--;
1290 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1291 continue;
1292 rdev = conf->mirrors[d].rdev;
NeilBrown4dbcdc72006-01-06 00:20:52 -08001293 atomic_add(s, &rdev->corrected_errors);
NeilBrown69382e82006-01-06 00:20:22 -08001294 if (sync_page_io(rdev->bdev,
1295 sect + rdev->data_offset,
1296 s<<9,
1297 bio->bi_io_vec[idx].bv_page,
NeilBrown097426f2006-01-06 00:20:37 -08001298 WRITE) == 0)
1299 md_error(mddev, rdev);
1300 }
1301 d = start;
1302 while (d != r1_bio->read_disk) {
1303 if (d == 0)
1304 d = conf->raid_disks;
1305 d--;
1306 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1307 continue;
1308 rdev = conf->mirrors[d].rdev;
1309 if (sync_page_io(rdev->bdev,
NeilBrown69382e82006-01-06 00:20:22 -08001310 sect + rdev->data_offset,
1311 s<<9,
1312 bio->bi_io_vec[idx].bv_page,
NeilBrown097426f2006-01-06 00:20:37 -08001313 READ) == 0)
NeilBrown69382e82006-01-06 00:20:22 -08001314 md_error(mddev, rdev);
NeilBrown69382e82006-01-06 00:20:22 -08001315 }
1316 } else {
1317 char b[BDEVNAME_SIZE];
1318 /* Cannot read from anywhere, array is toast */
1319 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1320 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1321 " for block %llu\n",
1322 bdevname(bio->bi_bdev,b),
1323 (unsigned long long)r1_bio->sector);
1324 md_done_sync(mddev, r1_bio->sectors, 0);
1325 put_buf(r1_bio);
1326 return;
1327 }
1328 sectors -= s;
1329 sect += s;
1330 idx ++;
1331 }
1332 }
NeilBrownd11c1712006-01-06 00:20:26 -08001333
1334 /*
1335 * schedule writes
1336 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 atomic_set(&r1_bio->remaining, 1);
1338 for (i = 0; i < disks ; i++) {
1339 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08001340 if (wbio->bi_end_io == NULL ||
1341 (wbio->bi_end_io == end_sync_read &&
1342 (i == r1_bio->read_disk ||
1343 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 continue;
1345
NeilBrown3e198f72006-01-06 00:20:21 -08001346 wbio->bi_rw = WRITE;
1347 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 atomic_inc(&r1_bio->remaining);
1349 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
NeilBrown191ea9b2005-06-21 17:17:23 -07001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 generic_make_request(wbio);
1352 }
1353
1354 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001355 /* if we're here, all write(s) have completed, so clean up */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 md_done_sync(mddev, r1_bio->sectors, 1);
1357 put_buf(r1_bio);
1358 }
1359}
1360
1361/*
1362 * This is a kernel thread which:
1363 *
1364 * 1. Retries failed read operations on working mirrors.
1365 * 2. Updates the raid superblock when problems encounter.
1366 * 3. Performs writes following reads for array syncronising.
1367 */
1368
NeilBrown867868f2006-10-03 01:15:51 -07001369static void fix_read_error(conf_t *conf, int read_disk,
1370 sector_t sect, int sectors)
1371{
1372 mddev_t *mddev = conf->mddev;
1373 while(sectors) {
1374 int s = sectors;
1375 int d = read_disk;
1376 int success = 0;
1377 int start;
1378 mdk_rdev_t *rdev;
1379
1380 if (s > (PAGE_SIZE>>9))
1381 s = PAGE_SIZE >> 9;
1382
1383 do {
1384 /* Note: no rcu protection needed here
1385 * as this is synchronous in the raid1d thread
1386 * which is the thread that might remove
1387 * a device. If raid1d ever becomes multi-threaded....
1388 */
1389 rdev = conf->mirrors[d].rdev;
1390 if (rdev &&
1391 test_bit(In_sync, &rdev->flags) &&
1392 sync_page_io(rdev->bdev,
1393 sect + rdev->data_offset,
1394 s<<9,
1395 conf->tmppage, READ))
1396 success = 1;
1397 else {
1398 d++;
1399 if (d == conf->raid_disks)
1400 d = 0;
1401 }
1402 } while (!success && d != read_disk);
1403
1404 if (!success) {
1405 /* Cannot read from anywhere -- bye bye array */
1406 md_error(mddev, conf->mirrors[read_disk].rdev);
1407 break;
1408 }
1409 /* write it back and re-read */
1410 start = d;
1411 while (d != read_disk) {
1412 if (d==0)
1413 d = conf->raid_disks;
1414 d--;
1415 rdev = conf->mirrors[d].rdev;
1416 if (rdev &&
1417 test_bit(In_sync, &rdev->flags)) {
1418 if (sync_page_io(rdev->bdev,
1419 sect + rdev->data_offset,
1420 s<<9, conf->tmppage, WRITE)
1421 == 0)
1422 /* Well, this device is dead */
1423 md_error(mddev, rdev);
1424 }
1425 }
1426 d = start;
1427 while (d != read_disk) {
1428 char b[BDEVNAME_SIZE];
1429 if (d==0)
1430 d = conf->raid_disks;
1431 d--;
1432 rdev = conf->mirrors[d].rdev;
1433 if (rdev &&
1434 test_bit(In_sync, &rdev->flags)) {
1435 if (sync_page_io(rdev->bdev,
1436 sect + rdev->data_offset,
1437 s<<9, conf->tmppage, READ)
1438 == 0)
1439 /* Well, this device is dead */
1440 md_error(mddev, rdev);
1441 else {
1442 atomic_add(s, &rdev->corrected_errors);
1443 printk(KERN_INFO
1444 "raid1:%s: read error corrected "
1445 "(%d sectors at %llu on %s)\n",
1446 mdname(mddev), s,
1447 (unsigned long long)sect +
1448 rdev->data_offset,
1449 bdevname(rdev->bdev, b));
1450 }
1451 }
1452 }
1453 sectors -= s;
1454 sect += s;
1455 }
1456}
1457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458static void raid1d(mddev_t *mddev)
1459{
1460 r1bio_t *r1_bio;
1461 struct bio *bio;
1462 unsigned long flags;
1463 conf_t *conf = mddev_to_conf(mddev);
1464 struct list_head *head = &conf->retry_list;
1465 int unplug=0;
1466 mdk_rdev_t *rdev;
1467
1468 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
1470 for (;;) {
1471 char b[BDEVNAME_SIZE];
1472 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrown191ea9b2005-06-21 17:17:23 -07001473
1474 if (conf->pending_bio_list.head) {
1475 bio = bio_list_get(&conf->pending_bio_list);
1476 blk_remove_plug(mddev->queue);
1477 spin_unlock_irqrestore(&conf->device_lock, flags);
1478 /* flush any pending bitmap writes to disk before proceeding w/ I/O */
1479 if (bitmap_unplug(mddev->bitmap) != 0)
1480 printk("%s: bitmap file write failed!\n", mdname(mddev));
1481
1482 while (bio) { /* submit pending writes */
1483 struct bio *next = bio->bi_next;
1484 bio->bi_next = NULL;
1485 generic_make_request(bio);
1486 bio = next;
1487 }
1488 unplug = 1;
1489
1490 continue;
1491 }
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (list_empty(head))
1494 break;
1495 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1496 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08001497 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 spin_unlock_irqrestore(&conf->device_lock, flags);
1499
1500 mddev = r1_bio->mddev;
1501 conf = mddev_to_conf(mddev);
1502 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1503 sync_request_write(mddev, r1_bio);
1504 unplug = 1;
NeilBrowna9701a32005-11-08 21:39:34 -08001505 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1506 /* some requests in the r1bio were BIO_RW_BARRIER
NeilBrownbea27712006-05-01 12:15:46 -07001507 * requests which failed with -EOPNOTSUPP. Hohumm..
NeilBrowna9701a32005-11-08 21:39:34 -08001508 * Better resubmit without the barrier.
1509 * We know which devices to resubmit for, because
1510 * all others have had their bios[] entry cleared.
NeilBrown5e7dd2a2006-05-01 12:15:47 -07001511 * We already have a nr_pending reference on these rdevs.
NeilBrowna9701a32005-11-08 21:39:34 -08001512 */
1513 int i;
1514 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1515 clear_bit(R1BIO_Barrier, &r1_bio->state);
1516 for (i=0; i < conf->raid_disks; i++)
NeilBrown2f889122006-03-27 01:18:19 -08001517 if (r1_bio->bios[i])
1518 atomic_inc(&r1_bio->remaining);
1519 for (i=0; i < conf->raid_disks; i++)
NeilBrowna9701a32005-11-08 21:39:34 -08001520 if (r1_bio->bios[i]) {
1521 struct bio_vec *bvec;
1522 int j;
1523
1524 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1525 /* copy pages from the failed bio, as
1526 * this might be a write-behind device */
1527 __bio_for_each_segment(bvec, bio, j, 0)
1528 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1529 bio_put(r1_bio->bios[i]);
1530 bio->bi_sector = r1_bio->sector +
1531 conf->mirrors[i].rdev->data_offset;
1532 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1533 bio->bi_end_io = raid1_end_write_request;
1534 bio->bi_rw = WRITE;
1535 bio->bi_private = r1_bio;
1536 r1_bio->bios[i] = bio;
1537 generic_make_request(bio);
1538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 } else {
1540 int disk;
NeilBrownddaf22a2006-01-06 00:20:19 -08001541
1542 /* we got a read error. Maybe the drive is bad. Maybe just
1543 * the block and we can fix it.
1544 * We freeze all other IO, and try reading the block from
1545 * other devices. When we find one, we re-write
1546 * and check it that fixes the read error.
1547 * This is all done synchronously while the array is
1548 * frozen
1549 */
NeilBrown867868f2006-10-03 01:15:51 -07001550 if (mddev->ro == 0) {
1551 freeze_array(conf);
1552 fix_read_error(conf, r1_bio->read_disk,
1553 r1_bio->sector,
1554 r1_bio->sectors);
1555 unfreeze_array(conf);
NeilBrownddaf22a2006-01-06 00:20:19 -08001556 }
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 bio = r1_bio->bios[r1_bio->read_disk];
1559 if ((disk=read_balance(conf, r1_bio)) == -1) {
1560 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1561 " read error for block %llu\n",
1562 bdevname(bio->bi_bdev,b),
1563 (unsigned long long)r1_bio->sector);
1564 raid_end_bio_io(r1_bio);
1565 } else {
NeilBrowncf30a472006-01-06 00:20:23 -08001566 r1_bio->bios[r1_bio->read_disk] =
1567 mddev->ro ? IO_BLOCKED : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 r1_bio->read_disk = disk;
1569 bio_put(bio);
1570 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1571 r1_bio->bios[r1_bio->read_disk] = bio;
1572 rdev = conf->mirrors[disk].rdev;
1573 if (printk_ratelimit())
1574 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1575 " another mirror\n",
1576 bdevname(rdev->bdev,b),
1577 (unsigned long long)r1_bio->sector);
1578 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1579 bio->bi_bdev = rdev->bdev;
1580 bio->bi_end_io = raid1_end_read_request;
1581 bio->bi_rw = READ;
1582 bio->bi_private = r1_bio;
1583 unplug = 1;
1584 generic_make_request(bio);
1585 }
1586 }
1587 }
1588 spin_unlock_irqrestore(&conf->device_lock, flags);
1589 if (unplug)
1590 unplug_slaves(mddev);
1591}
1592
1593
1594static int init_resync(conf_t *conf)
1595{
1596 int buffs;
1597
1598 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02001599 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1601 conf->poolinfo);
1602 if (!conf->r1buf_pool)
1603 return -ENOMEM;
1604 conf->next_resync = 0;
1605 return 0;
1606}
1607
1608/*
1609 * perform a "sync" on one "block"
1610 *
1611 * We need to make sure that no normal I/O request - particularly write
1612 * requests - conflict with active sync requests.
1613 *
1614 * This is achieved by tracking pending requests and a 'barrier' concept
1615 * that can be installed to exclude normal IO requests.
1616 */
1617
NeilBrown57afd892005-06-21 17:17:13 -07001618static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
1620 conf_t *conf = mddev_to_conf(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 r1bio_t *r1_bio;
1622 struct bio *bio;
1623 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08001624 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08001626 int wonly = -1;
1627 int write_targets = 0, read_targets = 0;
NeilBrown191ea9b2005-06-21 17:17:23 -07001628 int sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07001629 int still_degraded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 if (!conf->r1buf_pool)
NeilBrown191ea9b2005-06-21 17:17:23 -07001632 {
1633/*
1634 printk("sync start - bitmap %p\n", mddev->bitmap);
1635*/
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07001637 return 0;
NeilBrown191ea9b2005-06-21 17:17:23 -07001638 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
1640 max_sector = mddev->size << 1;
1641 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001642 /* If we aborted, we need to abort the
1643 * sync on the 'current' bitmap chunk (there will
1644 * only be one in raid1 resync.
1645 * We can find the current addess in mddev->curr_resync
1646 */
NeilBrown6a806c52005-07-15 03:56:35 -07001647 if (mddev->curr_resync < max_sector) /* aborted */
1648 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07001649 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07001650 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07001651 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07001652
1653 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 close_sync(conf);
1655 return 0;
1656 }
1657
NeilBrown07d84d102006-06-26 00:27:56 -07001658 if (mddev->bitmap == NULL &&
1659 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07001660 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07001661 conf->fullsync == 0) {
1662 *skipped = 1;
1663 return max_sector - sector_nr;
1664 }
NeilBrown6394cca2006-08-27 01:23:50 -07001665 /* before building a request, check if we can skip these blocks..
1666 * This call the bitmap_start_sync doesn't actually record anything
1667 */
NeilBrowne3b97032005-08-04 12:53:34 -07001668 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08001669 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001670 /* We can skip this block, and probably several more */
1671 *skipped = 1;
1672 return sync_blocks;
1673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 /*
NeilBrown17999be2006-01-06 00:20:12 -08001675 * If there is non-resync activity waiting for a turn,
1676 * and resync is going fast enough,
1677 * then let it though before starting on this new sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 */
NeilBrown17999be2006-01-06 00:20:12 -08001679 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 msleep_interruptible(1000);
NeilBrown17999be2006-01-06 00:20:12 -08001681
1682 raise_barrier(conf);
1683
1684 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown3e198f72006-01-06 00:20:21 -08001687 rcu_read_lock();
1688 /*
1689 * If we get a correctably read error during resync or recovery,
1690 * we might want to read from a different device. So we
1691 * flag all drives that could conceivably be read from for READ,
1692 * and any others (which will be non-In_sync devices) for WRITE.
1693 * If a read fails, we try reading from something else for which READ
1694 * is OK.
1695 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 r1_bio->mddev = mddev;
1698 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07001699 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
1702 for (i=0; i < conf->raid_disks; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -08001703 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 bio = r1_bio->bios[i];
1705
1706 /* take from bio_init */
1707 bio->bi_next = NULL;
1708 bio->bi_flags |= 1 << BIO_UPTODATE;
1709 bio->bi_rw = 0;
1710 bio->bi_vcnt = 0;
1711 bio->bi_idx = 0;
1712 bio->bi_phys_segments = 0;
1713 bio->bi_hw_segments = 0;
1714 bio->bi_size = 0;
1715 bio->bi_end_io = NULL;
1716 bio->bi_private = NULL;
1717
NeilBrown3e198f72006-01-06 00:20:21 -08001718 rdev = rcu_dereference(conf->mirrors[i].rdev);
1719 if (rdev == NULL ||
1720 test_bit(Faulty, &rdev->flags)) {
NeilBrowne3b97032005-08-04 12:53:34 -07001721 still_degraded = 1;
1722 continue;
NeilBrown3e198f72006-01-06 00:20:21 -08001723 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 bio->bi_rw = WRITE;
1725 bio->bi_end_io = end_sync_write;
1726 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08001727 } else {
1728 /* may need to read from here */
1729 bio->bi_rw = READ;
1730 bio->bi_end_io = end_sync_read;
1731 if (test_bit(WriteMostly, &rdev->flags)) {
1732 if (wonly < 0)
1733 wonly = i;
1734 } else {
1735 if (disk < 0)
1736 disk = i;
1737 }
1738 read_targets++;
1739 }
1740 atomic_inc(&rdev->nr_pending);
1741 bio->bi_sector = sector_nr + rdev->data_offset;
1742 bio->bi_bdev = rdev->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 bio->bi_private = r1_bio;
1744 }
NeilBrown3e198f72006-01-06 00:20:21 -08001745 rcu_read_unlock();
1746 if (disk < 0)
1747 disk = wonly;
1748 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07001749
NeilBrown3e198f72006-01-06 00:20:21 -08001750 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1751 /* extra read targets are also write targets */
1752 write_targets += read_targets-1;
1753
1754 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 /* There is nowhere to write, so all non-sync
1756 * drives must be failed - so we are finished
1757 */
NeilBrown57afd892005-06-21 17:17:13 -07001758 sector_t rv = max_sector - sector_nr;
1759 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 return rv;
1762 }
1763
1764 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07001765 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 do {
1767 struct page *page;
1768 int len = PAGE_SIZE;
1769 if (sector_nr + (len>>9) > max_sector)
1770 len = (max_sector - sector_nr) << 9;
1771 if (len == 0)
1772 break;
NeilBrown6a806c52005-07-15 03:56:35 -07001773 if (sync_blocks == 0) {
1774 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de485f2005-11-08 21:39:38 -08001775 &sync_blocks, still_degraded) &&
1776 !conf->fullsync &&
1777 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07001778 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02001779 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown6a806c52005-07-15 03:56:35 -07001780 if (len > (sync_blocks<<9))
1781 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07001782 }
NeilBrown191ea9b2005-06-21 17:17:23 -07001783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 for (i=0 ; i < conf->raid_disks; i++) {
1785 bio = r1_bio->bios[i];
1786 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08001787 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 if (bio_add_page(bio, page, len, 0) == 0) {
1789 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08001790 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 while (i > 0) {
1792 i--;
1793 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07001794 if (bio->bi_end_io==NULL)
1795 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 /* remove last page from this bio */
1797 bio->bi_vcnt--;
1798 bio->bi_size -= len;
1799 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1800 }
1801 goto bio_full;
1802 }
1803 }
1804 }
1805 nr_sectors += len>>9;
1806 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07001807 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1809 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 r1_bio->sectors = nr_sectors;
1811
NeilBrownd11c1712006-01-06 00:20:26 -08001812 /* For a user-requested sync, we read all readable devices and do a
1813 * compare
1814 */
1815 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1816 atomic_set(&r1_bio->remaining, read_targets);
1817 for (i=0; i<conf->raid_disks; i++) {
1818 bio = r1_bio->bios[i];
1819 if (bio->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07001820 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08001821 generic_make_request(bio);
1822 }
1823 }
1824 } else {
1825 atomic_set(&r1_bio->remaining, 1);
1826 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07001827 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08001828 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
NeilBrownd11c1712006-01-06 00:20:26 -08001830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 return nr_sectors;
1832}
1833
1834static int run(mddev_t *mddev)
1835{
1836 conf_t *conf;
1837 int i, j, disk_idx;
1838 mirror_info_t *disk;
1839 mdk_rdev_t *rdev;
1840 struct list_head *tmp;
1841
1842 if (mddev->level != 1) {
1843 printk("raid1: %s: raid level not set to mirroring (%d)\n",
1844 mdname(mddev), mddev->level);
1845 goto out;
1846 }
NeilBrownf6705572006-03-27 01:18:11 -08001847 if (mddev->reshape_position != MaxSector) {
1848 printk("raid1: %s: reshape_position set but not supported\n",
1849 mdname(mddev));
1850 goto out;
1851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 /*
1853 * copy the already verified devices into our private RAID1
1854 * bookkeeping area. [whatever we allocate in run(),
1855 * should be freed in stop()]
1856 */
NeilBrown9ffae0c2006-01-06 00:20:32 -08001857 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 mddev->private = conf;
1859 if (!conf)
1860 goto out_no_mem;
1861
NeilBrown9ffae0c2006-01-06 00:20:32 -08001862 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 GFP_KERNEL);
1864 if (!conf->mirrors)
1865 goto out_no_mem;
1866
NeilBrownddaf22a2006-01-06 00:20:19 -08001867 conf->tmppage = alloc_page(GFP_KERNEL);
1868 if (!conf->tmppage)
1869 goto out_no_mem;
1870
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
1872 if (!conf->poolinfo)
1873 goto out_no_mem;
1874 conf->poolinfo->mddev = mddev;
1875 conf->poolinfo->raid_disks = mddev->raid_disks;
1876 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1877 r1bio_pool_free,
1878 conf->poolinfo);
1879 if (!conf->r1bio_pool)
1880 goto out_no_mem;
1881
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 ITERATE_RDEV(mddev, rdev, tmp) {
1883 disk_idx = rdev->raid_disk;
1884 if (disk_idx >= mddev->raid_disks
1885 || disk_idx < 0)
1886 continue;
1887 disk = conf->mirrors + disk_idx;
1888
1889 disk->rdev = rdev;
1890
1891 blk_queue_stack_limits(mddev->queue,
1892 rdev->bdev->bd_disk->queue);
1893 /* as we don't honour merge_bvec_fn, we must never risk
1894 * violating it, so limit ->max_sector to one PAGE, as
1895 * a one page request is never in violation.
1896 */
1897 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
1898 mddev->queue->max_sectors > (PAGE_SIZE>>9))
1899 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1900
1901 disk->head_position = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 }
1903 conf->raid_disks = mddev->raid_disks;
1904 conf->mddev = mddev;
1905 spin_lock_init(&conf->device_lock);
1906 INIT_LIST_HEAD(&conf->retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
1908 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08001909 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
NeilBrown191ea9b2005-06-21 17:17:23 -07001911 bio_list_init(&conf->pending_bio_list);
1912 bio_list_init(&conf->flushing_bio_list);
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
1915 mddev->degraded = 0;
1916 for (i = 0; i < conf->raid_disks; i++) {
1917
1918 disk = conf->mirrors + i;
1919
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001920 if (!disk->rdev ||
1921 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 disk->head_position = 0;
1923 mddev->degraded++;
1924 }
1925 }
NeilBrown11ce99e2006-10-03 01:15:52 -07001926 if (mddev->degraded == conf->raid_disks) {
1927 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
1928 mdname(mddev));
1929 goto out_free_conf;
1930 }
1931 if (conf->raid_disks - mddev->degraded == 1)
1932 mddev->recovery_cp = MaxSector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
1934 /*
1935 * find the first working one and use it as a starting point
1936 * to read balancing.
1937 */
1938 for (j = 0; j < conf->raid_disks &&
1939 (!conf->mirrors[j].rdev ||
NeilBrownb2d444d2005-11-08 21:39:31 -08001940 !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 /* nothing */;
1942 conf->last_used = j;
1943
1944
NeilBrown191ea9b2005-06-21 17:17:23 -07001945 mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1");
1946 if (!mddev->thread) {
1947 printk(KERN_ERR
1948 "raid1: couldn't allocate thread for %s\n",
1949 mdname(mddev));
1950 goto out_free_conf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 }
NeilBrown191ea9b2005-06-21 17:17:23 -07001952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 printk(KERN_INFO
1954 "raid1: raid set %s active with %d out of %d mirrors\n",
1955 mdname(mddev), mddev->raid_disks - mddev->degraded,
1956 mddev->raid_disks);
1957 /*
1958 * Ok, everything is just fine now
1959 */
1960 mddev->array_size = mddev->size;
1961
NeilBrown7a5febe2005-05-16 21:53:16 -07001962 mddev->queue->unplug_fn = raid1_unplug;
1963 mddev->queue->issue_flush_fn = raid1_issue_flush;
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return 0;
1966
1967out_no_mem:
1968 printk(KERN_ERR "raid1: couldn't allocate memory for %s\n",
1969 mdname(mddev));
1970
1971out_free_conf:
1972 if (conf) {
1973 if (conf->r1bio_pool)
1974 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07001975 kfree(conf->mirrors);
NeilBrown1345b1d2006-01-06 00:20:40 -08001976 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07001977 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 kfree(conf);
1979 mddev->private = NULL;
1980 }
1981out:
1982 return -EIO;
1983}
1984
1985static int stop(mddev_t *mddev)
1986{
1987 conf_t *conf = mddev_to_conf(mddev);
NeilBrown4b6d2872005-09-09 16:23:47 -07001988 struct bitmap *bitmap = mddev->bitmap;
1989 int behind_wait = 0;
1990
1991 /* wait for behind writes to complete */
1992 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
1993 behind_wait++;
1994 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
1995 set_current_state(TASK_UNINTERRUPTIBLE);
1996 schedule_timeout(HZ); /* wait a second */
1997 /* need to kick something here to make sure I/O goes? */
1998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 md_unregister_thread(mddev->thread);
2001 mddev->thread = NULL;
2002 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
2003 if (conf->r1bio_pool)
2004 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002005 kfree(conf->mirrors);
2006 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 kfree(conf);
2008 mddev->private = NULL;
2009 return 0;
2010}
2011
2012static int raid1_resize(mddev_t *mddev, sector_t sectors)
2013{
2014 /* no resync is happening, and there is enough space
2015 * on all devices, so we can resize.
2016 * We need to make sure resync covers any new space.
2017 * If the array is shrinking we should possibly wait until
2018 * any io in the removed space completes, but it hardly seems
2019 * worth it.
2020 */
2021 mddev->array_size = sectors>>1;
2022 set_capacity(mddev->gendisk, mddev->array_size << 1);
2023 mddev->changed = 1;
2024 if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) {
2025 mddev->recovery_cp = mddev->size << 1;
2026 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2027 }
2028 mddev->size = mddev->array_size;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002029 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 return 0;
2031}
2032
NeilBrown63c70c42006-03-27 01:18:13 -08002033static int raid1_reshape(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034{
2035 /* We need to:
2036 * 1/ resize the r1bio_pool
2037 * 2/ resize conf->mirrors
2038 *
2039 * We allocate a new r1bio_pool if we can.
2040 * Then raise a device barrier and wait until all IO stops.
2041 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07002042 *
2043 * At the same time, we "pack" the devices so that all the missing
2044 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 */
2046 mempool_t *newpool, *oldpool;
2047 struct pool_info *newpoolinfo;
2048 mirror_info_t *newmirrors;
2049 conf_t *conf = mddev_to_conf(mddev);
NeilBrown63c70c42006-03-27 01:18:13 -08002050 int cnt, raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
NeilBrown6ea9c072005-06-21 17:17:09 -07002052 int d, d2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
NeilBrown63c70c42006-03-27 01:18:13 -08002054 /* Cannot change chunk_size, layout, or level */
2055 if (mddev->chunk_size != mddev->new_chunk ||
2056 mddev->layout != mddev->new_layout ||
2057 mddev->level != mddev->new_level) {
2058 mddev->new_chunk = mddev->chunk_size;
2059 mddev->new_layout = mddev->layout;
2060 mddev->new_level = mddev->level;
2061 return -EINVAL;
2062 }
2063
2064 raid_disks = mddev->raid_disks + mddev->delta_disks;
2065
NeilBrown6ea9c072005-06-21 17:17:09 -07002066 if (raid_disks < conf->raid_disks) {
2067 cnt=0;
2068 for (d= 0; d < conf->raid_disks; d++)
2069 if (conf->mirrors[d].rdev)
2070 cnt++;
2071 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07002073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
2075 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2076 if (!newpoolinfo)
2077 return -ENOMEM;
2078 newpoolinfo->mddev = mddev;
2079 newpoolinfo->raid_disks = raid_disks;
2080
2081 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2082 r1bio_pool_free, newpoolinfo);
2083 if (!newpool) {
2084 kfree(newpoolinfo);
2085 return -ENOMEM;
2086 }
NeilBrown9ffae0c2006-01-06 00:20:32 -08002087 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 if (!newmirrors) {
2089 kfree(newpoolinfo);
2090 mempool_destroy(newpool);
2091 return -ENOMEM;
2092 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
NeilBrown17999be2006-01-06 00:20:12 -08002094 raise_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
2096 /* ok, everything is stopped */
2097 oldpool = conf->r1bio_pool;
2098 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07002099
2100 for (d=d2=0; d < conf->raid_disks; d++)
2101 if (conf->mirrors[d].rdev) {
2102 conf->mirrors[d].rdev->raid_disk = d2;
2103 newmirrors[d2++].rdev = conf->mirrors[d].rdev;
2104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 kfree(conf->mirrors);
2106 conf->mirrors = newmirrors;
2107 kfree(conf->poolinfo);
2108 conf->poolinfo = newpoolinfo;
2109
2110 mddev->degraded += (raid_disks - conf->raid_disks);
2111 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08002112 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
NeilBrown6ea9c072005-06-21 17:17:09 -07002114 conf->last_used = 0; /* just make sure it is in-range */
NeilBrown17999be2006-01-06 00:20:12 -08002115 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2118 md_wakeup_thread(mddev->thread);
2119
2120 mempool_destroy(oldpool);
2121 return 0;
2122}
2123
NeilBrown500af872005-09-09 16:23:58 -07002124static void raid1_quiesce(mddev_t *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07002125{
2126 conf_t *conf = mddev_to_conf(mddev);
2127
2128 switch(state) {
NeilBrown9e6603d2005-09-09 16:23:48 -07002129 case 1:
NeilBrown17999be2006-01-06 00:20:12 -08002130 raise_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002131 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002132 case 0:
NeilBrown17999be2006-01-06 00:20:12 -08002133 lower_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002134 break;
2135 }
NeilBrown36fa3062005-09-09 16:23:45 -07002136}
2137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
NeilBrown2604b702006-01-06 00:20:36 -08002139static struct mdk_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140{
2141 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08002142 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 .owner = THIS_MODULE,
2144 .make_request = make_request,
2145 .run = run,
2146 .stop = stop,
2147 .status = status,
2148 .error_handler = error,
2149 .hot_add_disk = raid1_add_disk,
2150 .hot_remove_disk= raid1_remove_disk,
2151 .spare_active = raid1_spare_active,
2152 .sync_request = sync_request,
2153 .resize = raid1_resize,
NeilBrown63c70c42006-03-27 01:18:13 -08002154 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07002155 .quiesce = raid1_quiesce,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156};
2157
2158static int __init raid_init(void)
2159{
NeilBrown2604b702006-01-06 00:20:36 -08002160 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161}
2162
2163static void raid_exit(void)
2164{
NeilBrown2604b702006-01-06 00:20:36 -08002165 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166}
2167
2168module_init(raid_init);
2169module_exit(raid_exit);
2170MODULE_LICENSE("GPL");
2171MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002172MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08002173MODULE_ALIAS("md-level-1");