11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * raid1.c : Multiple Devices driver for Linux 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * RAID-1 management functions. 91da177e4SLinus Torvalds * 101da177e4SLinus Torvalds * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 111da177e4SLinus Torvalds * 1296de0e25SJan Engelhardt * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> 131da177e4SLinus Torvalds * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> 141da177e4SLinus Torvalds * 15191ea9b2SNeilBrown * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support 16191ea9b2SNeilBrown * bitmapped intelligence in resync: 17191ea9b2SNeilBrown * 18191ea9b2SNeilBrown * - bitmap marked during normal i/o 19191ea9b2SNeilBrown * - bitmap used to skip nondirty blocks during sync 20191ea9b2SNeilBrown * 21191ea9b2SNeilBrown * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: 22191ea9b2SNeilBrown * - persistent bitmap code 23191ea9b2SNeilBrown * 241da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or modify 251da177e4SLinus Torvalds * it under the terms of the GNU General Public License as published by 261da177e4SLinus Torvalds * the Free Software Foundation; either version 2, or (at your option) 271da177e4SLinus Torvalds * any later version. 281da177e4SLinus Torvalds * 291da177e4SLinus Torvalds * You should have received a copy of the GNU General Public License 301da177e4SLinus Torvalds * (for example /usr/src/linux/COPYING); if not, write to the Free 311da177e4SLinus Torvalds * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 321da177e4SLinus Torvalds */ 331da177e4SLinus Torvalds 34191ea9b2SNeilBrown #include "dm-bio-list.h" 351da177e4SLinus Torvalds #include <linux/raid/raid1.h> 36191ea9b2SNeilBrown #include <linux/raid/bitmap.h> 37191ea9b2SNeilBrown 38191ea9b2SNeilBrown #define DEBUG 0 39191ea9b2SNeilBrown #if DEBUG 40191ea9b2SNeilBrown #define PRINTK(x...) printk(x) 41191ea9b2SNeilBrown #else 42191ea9b2SNeilBrown #define PRINTK(x...) 43191ea9b2SNeilBrown #endif 441da177e4SLinus Torvalds 451da177e4SLinus Torvalds /* 461da177e4SLinus Torvalds * Number of guaranteed r1bios in case of extreme VM load: 471da177e4SLinus Torvalds */ 481da177e4SLinus Torvalds #define NR_RAID1_BIOS 256 491da177e4SLinus Torvalds 501da177e4SLinus Torvalds 511da177e4SLinus Torvalds static void unplug_slaves(mddev_t *mddev); 521da177e4SLinus Torvalds 5317999be4SNeilBrown static void allow_barrier(conf_t *conf); 5417999be4SNeilBrown static void lower_barrier(conf_t *conf); 551da177e4SLinus Torvalds 56dd0fc66fSAl Viro static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) 571da177e4SLinus Torvalds { 581da177e4SLinus Torvalds struct pool_info *pi = data; 591da177e4SLinus Torvalds r1bio_t *r1_bio; 601da177e4SLinus Torvalds int size = offsetof(r1bio_t, bios[pi->raid_disks]); 611da177e4SLinus Torvalds 621da177e4SLinus Torvalds /* allocate a r1bio with room for raid_disks entries in the bios array */ 639ffae0cfSNeilBrown r1_bio = kzalloc(size, gfp_flags); 649ffae0cfSNeilBrown if (!r1_bio) 651da177e4SLinus Torvalds unplug_slaves(pi->mddev); 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds return r1_bio; 681da177e4SLinus Torvalds } 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds static void r1bio_pool_free(void *r1_bio, void *data) 711da177e4SLinus Torvalds { 721da177e4SLinus Torvalds kfree(r1_bio); 731da177e4SLinus Torvalds } 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds #define RESYNC_BLOCK_SIZE (64*1024) 761da177e4SLinus Torvalds //#define RESYNC_BLOCK_SIZE PAGE_SIZE 771da177e4SLinus Torvalds #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) 781da177e4SLinus Torvalds #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) 791da177e4SLinus Torvalds #define RESYNC_WINDOW (2048*1024) 801da177e4SLinus Torvalds 81dd0fc66fSAl Viro static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) 821da177e4SLinus Torvalds { 831da177e4SLinus Torvalds struct pool_info *pi = data; 841da177e4SLinus Torvalds struct page *page; 851da177e4SLinus Torvalds r1bio_t *r1_bio; 861da177e4SLinus Torvalds struct bio *bio; 871da177e4SLinus Torvalds int i, j; 881da177e4SLinus Torvalds 891da177e4SLinus Torvalds r1_bio = r1bio_pool_alloc(gfp_flags, pi); 901da177e4SLinus Torvalds if (!r1_bio) { 911da177e4SLinus Torvalds unplug_slaves(pi->mddev); 921da177e4SLinus Torvalds return NULL; 931da177e4SLinus Torvalds } 941da177e4SLinus Torvalds 951da177e4SLinus Torvalds /* 961da177e4SLinus Torvalds * Allocate bios : 1 for reading, n-1 for writing 971da177e4SLinus Torvalds */ 981da177e4SLinus Torvalds for (j = pi->raid_disks ; j-- ; ) { 991da177e4SLinus Torvalds bio = bio_alloc(gfp_flags, RESYNC_PAGES); 1001da177e4SLinus Torvalds if (!bio) 1011da177e4SLinus Torvalds goto out_free_bio; 1021da177e4SLinus Torvalds r1_bio->bios[j] = bio; 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds /* 1051da177e4SLinus Torvalds * Allocate RESYNC_PAGES data pages and attach them to 106d11c171eSNeilBrown * the first bio. 107d11c171eSNeilBrown * If this is a user-requested check/repair, allocate 108d11c171eSNeilBrown * RESYNC_PAGES for each bio. 1091da177e4SLinus Torvalds */ 110d11c171eSNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) 111d11c171eSNeilBrown j = pi->raid_disks; 112d11c171eSNeilBrown else 113d11c171eSNeilBrown j = 1; 114d11c171eSNeilBrown while(j--) { 115d11c171eSNeilBrown bio = r1_bio->bios[j]; 1161da177e4SLinus Torvalds for (i = 0; i < RESYNC_PAGES; i++) { 1171da177e4SLinus Torvalds page = alloc_page(gfp_flags); 1181da177e4SLinus Torvalds if (unlikely(!page)) 1191da177e4SLinus Torvalds goto out_free_pages; 1201da177e4SLinus Torvalds 1211da177e4SLinus Torvalds bio->bi_io_vec[i].bv_page = page; 1221da177e4SLinus Torvalds } 123d11c171eSNeilBrown } 124d11c171eSNeilBrown /* If not user-requests, copy the page pointers to all bios */ 125d11c171eSNeilBrown if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { 126d11c171eSNeilBrown for (i=0; i<RESYNC_PAGES ; i++) 127d11c171eSNeilBrown for (j=1; j<pi->raid_disks; j++) 128d11c171eSNeilBrown r1_bio->bios[j]->bi_io_vec[i].bv_page = 129d11c171eSNeilBrown r1_bio->bios[0]->bi_io_vec[i].bv_page; 130d11c171eSNeilBrown } 1311da177e4SLinus Torvalds 1321da177e4SLinus Torvalds r1_bio->master_bio = NULL; 1331da177e4SLinus Torvalds 1341da177e4SLinus Torvalds return r1_bio; 1351da177e4SLinus Torvalds 1361da177e4SLinus Torvalds out_free_pages: 137d11c171eSNeilBrown for (i=0; i < RESYNC_PAGES ; i++) 138d11c171eSNeilBrown for (j=0 ; j < pi->raid_disks; j++) 1391345b1d8SNeilBrown safe_put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page); 140d11c171eSNeilBrown j = -1; 1411da177e4SLinus Torvalds out_free_bio: 1421da177e4SLinus Torvalds while ( ++j < pi->raid_disks ) 1431da177e4SLinus Torvalds bio_put(r1_bio->bios[j]); 1441da177e4SLinus Torvalds r1bio_pool_free(r1_bio, data); 1451da177e4SLinus Torvalds return NULL; 1461da177e4SLinus Torvalds } 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds static void r1buf_pool_free(void *__r1_bio, void *data) 1491da177e4SLinus Torvalds { 1501da177e4SLinus Torvalds struct pool_info *pi = data; 151d11c171eSNeilBrown int i,j; 1521da177e4SLinus Torvalds r1bio_t *r1bio = __r1_bio; 1531da177e4SLinus Torvalds 154d11c171eSNeilBrown for (i = 0; i < RESYNC_PAGES; i++) 155d11c171eSNeilBrown for (j = pi->raid_disks; j-- ;) { 156d11c171eSNeilBrown if (j == 0 || 157d11c171eSNeilBrown r1bio->bios[j]->bi_io_vec[i].bv_page != 158d11c171eSNeilBrown r1bio->bios[0]->bi_io_vec[i].bv_page) 1591345b1d8SNeilBrown safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); 1601da177e4SLinus Torvalds } 1611da177e4SLinus Torvalds for (i=0 ; i < pi->raid_disks; i++) 1621da177e4SLinus Torvalds bio_put(r1bio->bios[i]); 1631da177e4SLinus Torvalds 1641da177e4SLinus Torvalds r1bio_pool_free(r1bio, data); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds 1671da177e4SLinus Torvalds static void put_all_bios(conf_t *conf, r1bio_t *r1_bio) 1681da177e4SLinus Torvalds { 1691da177e4SLinus Torvalds int i; 1701da177e4SLinus Torvalds 1711da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 1721da177e4SLinus Torvalds struct bio **bio = r1_bio->bios + i; 173cf30a473SNeilBrown if (*bio && *bio != IO_BLOCKED) 1741da177e4SLinus Torvalds bio_put(*bio); 1751da177e4SLinus Torvalds *bio = NULL; 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds } 1781da177e4SLinus Torvalds 179858119e1SArjan van de Ven static void free_r1bio(r1bio_t *r1_bio) 1801da177e4SLinus Torvalds { 1811da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(r1_bio->mddev); 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* 1841da177e4SLinus Torvalds * Wake up any possible resync thread that waits for the device 1851da177e4SLinus Torvalds * to go idle. 1861da177e4SLinus Torvalds */ 18717999be4SNeilBrown allow_barrier(conf); 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds put_all_bios(conf, r1_bio); 1901da177e4SLinus Torvalds mempool_free(r1_bio, conf->r1bio_pool); 1911da177e4SLinus Torvalds } 1921da177e4SLinus Torvalds 193858119e1SArjan van de Ven static void put_buf(r1bio_t *r1_bio) 1941da177e4SLinus Torvalds { 1951da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(r1_bio->mddev); 1963e198f78SNeilBrown int i; 1973e198f78SNeilBrown 1983e198f78SNeilBrown for (i=0; i<conf->raid_disks; i++) { 1993e198f78SNeilBrown struct bio *bio = r1_bio->bios[i]; 2003e198f78SNeilBrown if (bio->bi_end_io) 2013e198f78SNeilBrown rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); 2023e198f78SNeilBrown } 2031da177e4SLinus Torvalds 2041da177e4SLinus Torvalds mempool_free(r1_bio, conf->r1buf_pool); 2051da177e4SLinus Torvalds 20617999be4SNeilBrown lower_barrier(conf); 2071da177e4SLinus Torvalds } 2081da177e4SLinus Torvalds 2091da177e4SLinus Torvalds static void reschedule_retry(r1bio_t *r1_bio) 2101da177e4SLinus Torvalds { 2111da177e4SLinus Torvalds unsigned long flags; 2121da177e4SLinus Torvalds mddev_t *mddev = r1_bio->mddev; 2131da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 2141da177e4SLinus Torvalds 2151da177e4SLinus Torvalds spin_lock_irqsave(&conf->device_lock, flags); 2161da177e4SLinus Torvalds list_add(&r1_bio->retry_list, &conf->retry_list); 217ddaf22abSNeilBrown conf->nr_queued ++; 2181da177e4SLinus Torvalds spin_unlock_irqrestore(&conf->device_lock, flags); 2191da177e4SLinus Torvalds 22017999be4SNeilBrown wake_up(&conf->wait_barrier); 2211da177e4SLinus Torvalds md_wakeup_thread(mddev->thread); 2221da177e4SLinus Torvalds } 2231da177e4SLinus Torvalds 2241da177e4SLinus Torvalds /* 2251da177e4SLinus Torvalds * raid_end_bio_io() is called when we have finished servicing a mirrored 2261da177e4SLinus Torvalds * operation and are ready to return a success/failure code to the buffer 2271da177e4SLinus Torvalds * cache layer. 2281da177e4SLinus Torvalds */ 2291da177e4SLinus Torvalds static void raid_end_bio_io(r1bio_t *r1_bio) 2301da177e4SLinus Torvalds { 2311da177e4SLinus Torvalds struct bio *bio = r1_bio->master_bio; 2321da177e4SLinus Torvalds 2334b6d287fSNeilBrown /* if nobody has done the final endio yet, do it now */ 2344b6d287fSNeilBrown if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { 2354b6d287fSNeilBrown PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n", 2364b6d287fSNeilBrown (bio_data_dir(bio) == WRITE) ? "write" : "read", 2374b6d287fSNeilBrown (unsigned long long) bio->bi_sector, 2384b6d287fSNeilBrown (unsigned long long) bio->bi_sector + 2394b6d287fSNeilBrown (bio->bi_size >> 9) - 1); 2404b6d287fSNeilBrown 2416712ecf8SNeilBrown bio_endio(bio, 2421da177e4SLinus Torvalds test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO); 2434b6d287fSNeilBrown } 2441da177e4SLinus Torvalds free_r1bio(r1_bio); 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 2471da177e4SLinus Torvalds /* 2481da177e4SLinus Torvalds * Update disk head position estimator based on IRQ completion info. 2491da177e4SLinus Torvalds */ 2501da177e4SLinus Torvalds static inline void update_head_pos(int disk, r1bio_t *r1_bio) 2511da177e4SLinus Torvalds { 2521da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(r1_bio->mddev); 2531da177e4SLinus Torvalds 2541da177e4SLinus Torvalds conf->mirrors[disk].head_position = 2551da177e4SLinus Torvalds r1_bio->sector + (r1_bio->sectors); 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 2586712ecf8SNeilBrown static void raid1_end_read_request(struct bio *bio, int error) 2591da177e4SLinus Torvalds { 2601da177e4SLinus Torvalds int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 2611da177e4SLinus Torvalds r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); 2621da177e4SLinus Torvalds int mirror; 2631da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(r1_bio->mddev); 2641da177e4SLinus Torvalds 2651da177e4SLinus Torvalds mirror = r1_bio->read_disk; 2661da177e4SLinus Torvalds /* 2671da177e4SLinus Torvalds * this branch is our 'one mirror IO has finished' event handler: 2681da177e4SLinus Torvalds */ 269ddaf22abSNeilBrown update_head_pos(mirror, r1_bio); 270ddaf22abSNeilBrown 271220946c9SNeilBrown if (uptodate) 2721da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 273dd00a99eSNeilBrown else { 274dd00a99eSNeilBrown /* If all other devices have failed, we want to return 275dd00a99eSNeilBrown * the error upwards rather than fail the last device. 276dd00a99eSNeilBrown * Here we redefine "uptodate" to mean "Don't want to retry" 277dd00a99eSNeilBrown */ 278dd00a99eSNeilBrown unsigned long flags; 279dd00a99eSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 280dd00a99eSNeilBrown if (r1_bio->mddev->degraded == conf->raid_disks || 281dd00a99eSNeilBrown (r1_bio->mddev->degraded == conf->raid_disks-1 && 282dd00a99eSNeilBrown !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) 283dd00a99eSNeilBrown uptodate = 1; 284dd00a99eSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 285dd00a99eSNeilBrown } 2861da177e4SLinus Torvalds 287dd00a99eSNeilBrown if (uptodate) 2881da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 289dd00a99eSNeilBrown else { 2901da177e4SLinus Torvalds /* 2911da177e4SLinus Torvalds * oops, read error: 2921da177e4SLinus Torvalds */ 2931da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 2941da177e4SLinus Torvalds if (printk_ratelimit()) 2951da177e4SLinus Torvalds printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n", 2961da177e4SLinus Torvalds bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector); 2971da177e4SLinus Torvalds reschedule_retry(r1_bio); 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds 3001da177e4SLinus Torvalds rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); 3011da177e4SLinus Torvalds } 3021da177e4SLinus Torvalds 3036712ecf8SNeilBrown static void raid1_end_write_request(struct bio *bio, int error) 3041da177e4SLinus Torvalds { 3051da177e4SLinus Torvalds int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 3061da177e4SLinus Torvalds r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); 307a9701a30SNeilBrown int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); 3081da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(r1_bio->mddev); 30904b857f7SNeilBrown struct bio *to_put = NULL; 3101da177e4SLinus Torvalds 3111da177e4SLinus Torvalds 3121da177e4SLinus Torvalds for (mirror = 0; mirror < conf->raid_disks; mirror++) 3131da177e4SLinus Torvalds if (r1_bio->bios[mirror] == bio) 3141da177e4SLinus Torvalds break; 3151da177e4SLinus Torvalds 316bea27718SNeilBrown if (error == -EOPNOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) { 317a9701a30SNeilBrown set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags); 318a9701a30SNeilBrown set_bit(R1BIO_BarrierRetry, &r1_bio->state); 319a9701a30SNeilBrown r1_bio->mddev->barriers_work = 0; 3205e7dd2abSNeilBrown /* Don't rdev_dec_pending in this branch - keep it for the retry */ 321a9701a30SNeilBrown } else { 3221da177e4SLinus Torvalds /* 3231da177e4SLinus Torvalds * this branch is our 'one mirror IO has finished' event handler: 3241da177e4SLinus Torvalds */ 325a9701a30SNeilBrown r1_bio->bios[mirror] = NULL; 32604b857f7SNeilBrown to_put = bio; 327191ea9b2SNeilBrown if (!uptodate) { 3281da177e4SLinus Torvalds md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); 329191ea9b2SNeilBrown /* an I/O failed, we can't clear the bitmap */ 330191ea9b2SNeilBrown set_bit(R1BIO_Degraded, &r1_bio->state); 331191ea9b2SNeilBrown } else 3321da177e4SLinus Torvalds /* 3331da177e4SLinus Torvalds * Set R1BIO_Uptodate in our master bio, so that 3341da177e4SLinus Torvalds * we will return a good error code for to the higher 3351da177e4SLinus Torvalds * levels even if IO on some other mirrored buffer fails. 3361da177e4SLinus Torvalds * 3371da177e4SLinus Torvalds * The 'master' represents the composite IO operation to 3381da177e4SLinus Torvalds * user-side. So if something waits for IO, then it will 3391da177e4SLinus Torvalds * wait for the 'master' bio. 3401da177e4SLinus Torvalds */ 3411da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 3421da177e4SLinus Torvalds 3431da177e4SLinus Torvalds update_head_pos(mirror, r1_bio); 3441da177e4SLinus Torvalds 3454b6d287fSNeilBrown if (behind) { 3464b6d287fSNeilBrown if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) 3474b6d287fSNeilBrown atomic_dec(&r1_bio->behind_remaining); 3484b6d287fSNeilBrown 3494b6d287fSNeilBrown /* In behind mode, we ACK the master bio once the I/O has safely 3504b6d287fSNeilBrown * reached all non-writemostly disks. Setting the Returned bit 3514b6d287fSNeilBrown * ensures that this gets done only once -- we don't ever want to 3524b6d287fSNeilBrown * return -EIO here, instead we'll wait */ 3534b6d287fSNeilBrown 3544b6d287fSNeilBrown if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && 3554b6d287fSNeilBrown test_bit(R1BIO_Uptodate, &r1_bio->state)) { 3564b6d287fSNeilBrown /* Maybe we can return now */ 3574b6d287fSNeilBrown if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { 3584b6d287fSNeilBrown struct bio *mbio = r1_bio->master_bio; 3594b6d287fSNeilBrown PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n", 3604b6d287fSNeilBrown (unsigned long long) mbio->bi_sector, 3614b6d287fSNeilBrown (unsigned long long) mbio->bi_sector + 3624b6d287fSNeilBrown (mbio->bi_size >> 9) - 1); 3636712ecf8SNeilBrown bio_endio(mbio, 0); 3644b6d287fSNeilBrown } 3654b6d287fSNeilBrown } 3664b6d287fSNeilBrown } 3675e7dd2abSNeilBrown rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); 368a9701a30SNeilBrown } 3691da177e4SLinus Torvalds /* 3701da177e4SLinus Torvalds * 3711da177e4SLinus Torvalds * Let's see if all mirrored write operations have finished 3721da177e4SLinus Torvalds * already. 3731da177e4SLinus Torvalds */ 3741da177e4SLinus Torvalds if (atomic_dec_and_test(&r1_bio->remaining)) { 375c70810b3SNeilBrown if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) 376a9701a30SNeilBrown reschedule_retry(r1_bio); 377c70810b3SNeilBrown else { 3785e7dd2abSNeilBrown /* it really is the end of this request */ 3794b6d287fSNeilBrown if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { 3804b6d287fSNeilBrown /* free extra copy of the data pages */ 3814b6d287fSNeilBrown int i = bio->bi_vcnt; 3824b6d287fSNeilBrown while (i--) 3831345b1d8SNeilBrown safe_put_page(bio->bi_io_vec[i].bv_page); 3844b6d287fSNeilBrown } 385191ea9b2SNeilBrown /* clear the bitmap if all writes complete successfully */ 386191ea9b2SNeilBrown bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, 387191ea9b2SNeilBrown r1_bio->sectors, 3884b6d287fSNeilBrown !test_bit(R1BIO_Degraded, &r1_bio->state), 3894b6d287fSNeilBrown behind); 3901da177e4SLinus Torvalds md_write_end(r1_bio->mddev); 3911da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 3921da177e4SLinus Torvalds } 393c70810b3SNeilBrown } 394c70810b3SNeilBrown 39504b857f7SNeilBrown if (to_put) 39604b857f7SNeilBrown bio_put(to_put); 3971da177e4SLinus Torvalds } 3981da177e4SLinus Torvalds 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds /* 4011da177e4SLinus Torvalds * This routine returns the disk from which the requested read should 4021da177e4SLinus Torvalds * be done. There is a per-array 'next expected sequential IO' sector 4031da177e4SLinus Torvalds * number - if this matches on the next IO then we use the last disk. 4041da177e4SLinus Torvalds * There is also a per-disk 'last know head position' sector that is 4051da177e4SLinus Torvalds * maintained from IRQ contexts, both the normal and the resync IO 4061da177e4SLinus Torvalds * completion handlers update this position correctly. If there is no 4071da177e4SLinus Torvalds * perfect sequential match then we pick the disk whose head is closest. 4081da177e4SLinus Torvalds * 4091da177e4SLinus Torvalds * If there are 2 mirrors in the same 2 devices, performance degrades 4101da177e4SLinus Torvalds * because position is mirror, not device based. 4111da177e4SLinus Torvalds * 4121da177e4SLinus Torvalds * The rdev for the device selected will have nr_pending incremented. 4131da177e4SLinus Torvalds */ 4141da177e4SLinus Torvalds static int read_balance(conf_t *conf, r1bio_t *r1_bio) 4151da177e4SLinus Torvalds { 4161da177e4SLinus Torvalds const unsigned long this_sector = r1_bio->sector; 4171da177e4SLinus Torvalds int new_disk = conf->last_used, disk = new_disk; 4188ddf9efeSNeilBrown int wonly_disk = -1; 4191da177e4SLinus Torvalds const int sectors = r1_bio->sectors; 4201da177e4SLinus Torvalds sector_t new_distance, current_distance; 4218ddf9efeSNeilBrown mdk_rdev_t *rdev; 4221da177e4SLinus Torvalds 4231da177e4SLinus Torvalds rcu_read_lock(); 4241da177e4SLinus Torvalds /* 4258ddf9efeSNeilBrown * Check if we can balance. We can balance on the whole 4261da177e4SLinus Torvalds * device if no resync is going on, or below the resync window. 4271da177e4SLinus Torvalds * We take the first readable disk when above the resync window. 4281da177e4SLinus Torvalds */ 4291da177e4SLinus Torvalds retry: 4301da177e4SLinus Torvalds if (conf->mddev->recovery_cp < MaxSector && 4311da177e4SLinus Torvalds (this_sector + sectors >= conf->next_resync)) { 4321da177e4SLinus Torvalds /* Choose the first operation device, for consistancy */ 4331da177e4SLinus Torvalds new_disk = 0; 4341da177e4SLinus Torvalds 435d6065f7bSSuzanne Wood for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); 436cf30a473SNeilBrown r1_bio->bios[new_disk] == IO_BLOCKED || 437b2d444d7SNeilBrown !rdev || !test_bit(In_sync, &rdev->flags) 4388ddf9efeSNeilBrown || test_bit(WriteMostly, &rdev->flags); 439d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) { 4408ddf9efeSNeilBrown 441cf30a473SNeilBrown if (rdev && test_bit(In_sync, &rdev->flags) && 442cf30a473SNeilBrown r1_bio->bios[new_disk] != IO_BLOCKED) 4438ddf9efeSNeilBrown wonly_disk = new_disk; 4448ddf9efeSNeilBrown 4458ddf9efeSNeilBrown if (new_disk == conf->raid_disks - 1) { 4468ddf9efeSNeilBrown new_disk = wonly_disk; 4471da177e4SLinus Torvalds break; 4481da177e4SLinus Torvalds } 4491da177e4SLinus Torvalds } 4501da177e4SLinus Torvalds goto rb_out; 4511da177e4SLinus Torvalds } 4521da177e4SLinus Torvalds 4531da177e4SLinus Torvalds 4541da177e4SLinus Torvalds /* make sure the disk is operational */ 455d6065f7bSSuzanne Wood for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); 456cf30a473SNeilBrown r1_bio->bios[new_disk] == IO_BLOCKED || 457b2d444d7SNeilBrown !rdev || !test_bit(In_sync, &rdev->flags) || 4588ddf9efeSNeilBrown test_bit(WriteMostly, &rdev->flags); 459d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) { 4608ddf9efeSNeilBrown 461cf30a473SNeilBrown if (rdev && test_bit(In_sync, &rdev->flags) && 462cf30a473SNeilBrown r1_bio->bios[new_disk] != IO_BLOCKED) 4638ddf9efeSNeilBrown wonly_disk = new_disk; 4648ddf9efeSNeilBrown 4651da177e4SLinus Torvalds if (new_disk <= 0) 4661da177e4SLinus Torvalds new_disk = conf->raid_disks; 4671da177e4SLinus Torvalds new_disk--; 4681da177e4SLinus Torvalds if (new_disk == disk) { 4698ddf9efeSNeilBrown new_disk = wonly_disk; 4708ddf9efeSNeilBrown break; 4718ddf9efeSNeilBrown } 4728ddf9efeSNeilBrown } 4738ddf9efeSNeilBrown 4748ddf9efeSNeilBrown if (new_disk < 0) 4751da177e4SLinus Torvalds goto rb_out; 4768ddf9efeSNeilBrown 4771da177e4SLinus Torvalds disk = new_disk; 4781da177e4SLinus Torvalds /* now disk == new_disk == starting point for search */ 4791da177e4SLinus Torvalds 4801da177e4SLinus Torvalds /* 4811da177e4SLinus Torvalds * Don't change to another disk for sequential reads: 4821da177e4SLinus Torvalds */ 4831da177e4SLinus Torvalds if (conf->next_seq_sect == this_sector) 4841da177e4SLinus Torvalds goto rb_out; 4851da177e4SLinus Torvalds if (this_sector == conf->mirrors[new_disk].head_position) 4861da177e4SLinus Torvalds goto rb_out; 4871da177e4SLinus Torvalds 4881da177e4SLinus Torvalds current_distance = abs(this_sector - conf->mirrors[disk].head_position); 4891da177e4SLinus Torvalds 4901da177e4SLinus Torvalds /* Find the disk whose head is closest */ 4911da177e4SLinus Torvalds 4921da177e4SLinus Torvalds do { 4931da177e4SLinus Torvalds if (disk <= 0) 4941da177e4SLinus Torvalds disk = conf->raid_disks; 4951da177e4SLinus Torvalds disk--; 4961da177e4SLinus Torvalds 497d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[disk].rdev); 4988ddf9efeSNeilBrown 499cf30a473SNeilBrown if (!rdev || r1_bio->bios[disk] == IO_BLOCKED || 500b2d444d7SNeilBrown !test_bit(In_sync, &rdev->flags) || 5018ddf9efeSNeilBrown test_bit(WriteMostly, &rdev->flags)) 5021da177e4SLinus Torvalds continue; 5031da177e4SLinus Torvalds 5041da177e4SLinus Torvalds if (!atomic_read(&rdev->nr_pending)) { 5051da177e4SLinus Torvalds new_disk = disk; 5061da177e4SLinus Torvalds break; 5071da177e4SLinus Torvalds } 5081da177e4SLinus Torvalds new_distance = abs(this_sector - conf->mirrors[disk].head_position); 5091da177e4SLinus Torvalds if (new_distance < current_distance) { 5101da177e4SLinus Torvalds current_distance = new_distance; 5111da177e4SLinus Torvalds new_disk = disk; 5121da177e4SLinus Torvalds } 5131da177e4SLinus Torvalds } while (disk != conf->last_used); 5141da177e4SLinus Torvalds 5151da177e4SLinus Torvalds rb_out: 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds 5181da177e4SLinus Torvalds if (new_disk >= 0) { 519d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[new_disk].rdev); 5208ddf9efeSNeilBrown if (!rdev) 5218ddf9efeSNeilBrown goto retry; 5228ddf9efeSNeilBrown atomic_inc(&rdev->nr_pending); 523b2d444d7SNeilBrown if (!test_bit(In_sync, &rdev->flags)) { 5241da177e4SLinus Torvalds /* cannot risk returning a device that failed 5251da177e4SLinus Torvalds * before we inc'ed nr_pending 5261da177e4SLinus Torvalds */ 52703c902e1SNeilBrown rdev_dec_pending(rdev, conf->mddev); 5281da177e4SLinus Torvalds goto retry; 5291da177e4SLinus Torvalds } 5308ddf9efeSNeilBrown conf->next_seq_sect = this_sector + sectors; 5318ddf9efeSNeilBrown conf->last_used = new_disk; 5321da177e4SLinus Torvalds } 5331da177e4SLinus Torvalds rcu_read_unlock(); 5341da177e4SLinus Torvalds 5351da177e4SLinus Torvalds return new_disk; 5361da177e4SLinus Torvalds } 5371da177e4SLinus Torvalds 5381da177e4SLinus Torvalds static void unplug_slaves(mddev_t *mddev) 5391da177e4SLinus Torvalds { 5401da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 5411da177e4SLinus Torvalds int i; 5421da177e4SLinus Torvalds 5431da177e4SLinus Torvalds rcu_read_lock(); 5441da177e4SLinus Torvalds for (i=0; i<mddev->raid_disks; i++) { 545d6065f7bSSuzanne Wood mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); 546b2d444d7SNeilBrown if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { 547165125e1SJens Axboe struct request_queue *r_queue = bdev_get_queue(rdev->bdev); 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds atomic_inc(&rdev->nr_pending); 5501da177e4SLinus Torvalds rcu_read_unlock(); 5511da177e4SLinus Torvalds 5522ad8b1efSAlan D. Brunelle blk_unplug(r_queue); 5531da177e4SLinus Torvalds 5541da177e4SLinus Torvalds rdev_dec_pending(rdev, mddev); 5551da177e4SLinus Torvalds rcu_read_lock(); 5561da177e4SLinus Torvalds } 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds rcu_read_unlock(); 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds 561165125e1SJens Axboe static void raid1_unplug(struct request_queue *q) 5621da177e4SLinus Torvalds { 563191ea9b2SNeilBrown mddev_t *mddev = q->queuedata; 564191ea9b2SNeilBrown 565191ea9b2SNeilBrown unplug_slaves(mddev); 566191ea9b2SNeilBrown md_wakeup_thread(mddev->thread); 5671da177e4SLinus Torvalds } 5681da177e4SLinus Torvalds 5690d129228SNeilBrown static int raid1_congested(void *data, int bits) 5700d129228SNeilBrown { 5710d129228SNeilBrown mddev_t *mddev = data; 5720d129228SNeilBrown conf_t *conf = mddev_to_conf(mddev); 5730d129228SNeilBrown int i, ret = 0; 5740d129228SNeilBrown 5750d129228SNeilBrown rcu_read_lock(); 5760d129228SNeilBrown for (i = 0; i < mddev->raid_disks; i++) { 5770d129228SNeilBrown mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); 5780d129228SNeilBrown if (rdev && !test_bit(Faulty, &rdev->flags)) { 579165125e1SJens Axboe struct request_queue *q = bdev_get_queue(rdev->bdev); 5800d129228SNeilBrown 5810d129228SNeilBrown /* Note the '|| 1' - when read_balance prefers 5820d129228SNeilBrown * non-congested targets, it can be removed 5830d129228SNeilBrown */ 5840d129228SNeilBrown if ((bits & (1<<BDI_write_congested)) || 1) 5850d129228SNeilBrown ret |= bdi_congested(&q->backing_dev_info, bits); 5860d129228SNeilBrown else 5870d129228SNeilBrown ret &= bdi_congested(&q->backing_dev_info, bits); 5880d129228SNeilBrown } 5890d129228SNeilBrown } 5900d129228SNeilBrown rcu_read_unlock(); 5910d129228SNeilBrown return ret; 5920d129228SNeilBrown } 5930d129228SNeilBrown 5940d129228SNeilBrown 595a35e63efSNeilBrown static int flush_pending_writes(conf_t *conf) 596a35e63efSNeilBrown { 597a35e63efSNeilBrown /* Any writes that have been queued but are awaiting 598a35e63efSNeilBrown * bitmap updates get flushed here. 599a35e63efSNeilBrown * We return 1 if any requests were actually submitted. 600a35e63efSNeilBrown */ 601a35e63efSNeilBrown int rv = 0; 602a35e63efSNeilBrown 603a35e63efSNeilBrown spin_lock_irq(&conf->device_lock); 604a35e63efSNeilBrown 605a35e63efSNeilBrown if (conf->pending_bio_list.head) { 606a35e63efSNeilBrown struct bio *bio; 607a35e63efSNeilBrown bio = bio_list_get(&conf->pending_bio_list); 608a35e63efSNeilBrown blk_remove_plug(conf->mddev->queue); 609a35e63efSNeilBrown spin_unlock_irq(&conf->device_lock); 610a35e63efSNeilBrown /* flush any pending bitmap writes to 611a35e63efSNeilBrown * disk before proceeding w/ I/O */ 612a35e63efSNeilBrown bitmap_unplug(conf->mddev->bitmap); 613a35e63efSNeilBrown 614a35e63efSNeilBrown while (bio) { /* submit pending writes */ 615a35e63efSNeilBrown struct bio *next = bio->bi_next; 616a35e63efSNeilBrown bio->bi_next = NULL; 617a35e63efSNeilBrown generic_make_request(bio); 618a35e63efSNeilBrown bio = next; 619a35e63efSNeilBrown } 620a35e63efSNeilBrown rv = 1; 621a35e63efSNeilBrown } else 622a35e63efSNeilBrown spin_unlock_irq(&conf->device_lock); 623a35e63efSNeilBrown return rv; 624a35e63efSNeilBrown } 625a35e63efSNeilBrown 62617999be4SNeilBrown /* Barriers.... 62717999be4SNeilBrown * Sometimes we need to suspend IO while we do something else, 62817999be4SNeilBrown * either some resync/recovery, or reconfigure the array. 62917999be4SNeilBrown * To do this we raise a 'barrier'. 63017999be4SNeilBrown * The 'barrier' is a counter that can be raised multiple times 63117999be4SNeilBrown * to count how many activities are happening which preclude 63217999be4SNeilBrown * normal IO. 63317999be4SNeilBrown * We can only raise the barrier if there is no pending IO. 63417999be4SNeilBrown * i.e. if nr_pending == 0. 63517999be4SNeilBrown * We choose only to raise the barrier if no-one is waiting for the 63617999be4SNeilBrown * barrier to go down. This means that as soon as an IO request 63717999be4SNeilBrown * is ready, no other operations which require a barrier will start 63817999be4SNeilBrown * until the IO request has had a chance. 63917999be4SNeilBrown * 64017999be4SNeilBrown * So: regular IO calls 'wait_barrier'. When that returns there 64117999be4SNeilBrown * is no backgroup IO happening, It must arrange to call 64217999be4SNeilBrown * allow_barrier when it has finished its IO. 64317999be4SNeilBrown * backgroup IO calls must call raise_barrier. Once that returns 64417999be4SNeilBrown * there is no normal IO happeing. It must arrange to call 64517999be4SNeilBrown * lower_barrier when the particular background IO completes. 6461da177e4SLinus Torvalds */ 6471da177e4SLinus Torvalds #define RESYNC_DEPTH 32 6481da177e4SLinus Torvalds 64917999be4SNeilBrown static void raise_barrier(conf_t *conf) 6501da177e4SLinus Torvalds { 6511da177e4SLinus Torvalds spin_lock_irq(&conf->resync_lock); 6521da177e4SLinus Torvalds 65317999be4SNeilBrown /* Wait until no block IO is waiting */ 65417999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, 65517999be4SNeilBrown conf->resync_lock, 65617999be4SNeilBrown raid1_unplug(conf->mddev->queue)); 65717999be4SNeilBrown 65817999be4SNeilBrown /* block any new IO from starting */ 65917999be4SNeilBrown conf->barrier++; 66017999be4SNeilBrown 66117999be4SNeilBrown /* No wait for all pending IO to complete */ 66217999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, 66317999be4SNeilBrown !conf->nr_pending && conf->barrier < RESYNC_DEPTH, 66417999be4SNeilBrown conf->resync_lock, 66517999be4SNeilBrown raid1_unplug(conf->mddev->queue)); 66617999be4SNeilBrown 6671da177e4SLinus Torvalds spin_unlock_irq(&conf->resync_lock); 6681da177e4SLinus Torvalds } 6691da177e4SLinus Torvalds 67017999be4SNeilBrown static void lower_barrier(conf_t *conf) 67117999be4SNeilBrown { 67217999be4SNeilBrown unsigned long flags; 67317999be4SNeilBrown spin_lock_irqsave(&conf->resync_lock, flags); 67417999be4SNeilBrown conf->barrier--; 67517999be4SNeilBrown spin_unlock_irqrestore(&conf->resync_lock, flags); 67617999be4SNeilBrown wake_up(&conf->wait_barrier); 67717999be4SNeilBrown } 67817999be4SNeilBrown 67917999be4SNeilBrown static void wait_barrier(conf_t *conf) 68017999be4SNeilBrown { 68117999be4SNeilBrown spin_lock_irq(&conf->resync_lock); 68217999be4SNeilBrown if (conf->barrier) { 68317999be4SNeilBrown conf->nr_waiting++; 68417999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, !conf->barrier, 68517999be4SNeilBrown conf->resync_lock, 68617999be4SNeilBrown raid1_unplug(conf->mddev->queue)); 68717999be4SNeilBrown conf->nr_waiting--; 68817999be4SNeilBrown } 68917999be4SNeilBrown conf->nr_pending++; 69017999be4SNeilBrown spin_unlock_irq(&conf->resync_lock); 69117999be4SNeilBrown } 69217999be4SNeilBrown 69317999be4SNeilBrown static void allow_barrier(conf_t *conf) 69417999be4SNeilBrown { 69517999be4SNeilBrown unsigned long flags; 69617999be4SNeilBrown spin_lock_irqsave(&conf->resync_lock, flags); 69717999be4SNeilBrown conf->nr_pending--; 69817999be4SNeilBrown spin_unlock_irqrestore(&conf->resync_lock, flags); 69917999be4SNeilBrown wake_up(&conf->wait_barrier); 70017999be4SNeilBrown } 70117999be4SNeilBrown 702ddaf22abSNeilBrown static void freeze_array(conf_t *conf) 703ddaf22abSNeilBrown { 704ddaf22abSNeilBrown /* stop syncio and normal IO and wait for everything to 705ddaf22abSNeilBrown * go quite. 706ddaf22abSNeilBrown * We increment barrier and nr_waiting, and then 7071c830532SNeilBrown * wait until nr_pending match nr_queued+1 7081c830532SNeilBrown * This is called in the context of one normal IO request 7091c830532SNeilBrown * that has failed. Thus any sync request that might be pending 7101c830532SNeilBrown * will be blocked by nr_pending, and we need to wait for 7111c830532SNeilBrown * pending IO requests to complete or be queued for re-try. 7121c830532SNeilBrown * Thus the number queued (nr_queued) plus this request (1) 7131c830532SNeilBrown * must match the number of pending IOs (nr_pending) before 7141c830532SNeilBrown * we continue. 715ddaf22abSNeilBrown */ 716ddaf22abSNeilBrown spin_lock_irq(&conf->resync_lock); 717ddaf22abSNeilBrown conf->barrier++; 718ddaf22abSNeilBrown conf->nr_waiting++; 719ddaf22abSNeilBrown wait_event_lock_irq(conf->wait_barrier, 7201c830532SNeilBrown conf->nr_pending == conf->nr_queued+1, 721ddaf22abSNeilBrown conf->resync_lock, 722a35e63efSNeilBrown ({ flush_pending_writes(conf); 723a35e63efSNeilBrown raid1_unplug(conf->mddev->queue); })); 724ddaf22abSNeilBrown spin_unlock_irq(&conf->resync_lock); 725ddaf22abSNeilBrown } 726ddaf22abSNeilBrown static void unfreeze_array(conf_t *conf) 727ddaf22abSNeilBrown { 728ddaf22abSNeilBrown /* reverse the effect of the freeze */ 729ddaf22abSNeilBrown spin_lock_irq(&conf->resync_lock); 730ddaf22abSNeilBrown conf->barrier--; 731ddaf22abSNeilBrown conf->nr_waiting--; 732ddaf22abSNeilBrown wake_up(&conf->wait_barrier); 733ddaf22abSNeilBrown spin_unlock_irq(&conf->resync_lock); 734ddaf22abSNeilBrown } 735ddaf22abSNeilBrown 73617999be4SNeilBrown 7374b6d287fSNeilBrown /* duplicate the data pages for behind I/O */ 7384b6d287fSNeilBrown static struct page **alloc_behind_pages(struct bio *bio) 7394b6d287fSNeilBrown { 7404b6d287fSNeilBrown int i; 7414b6d287fSNeilBrown struct bio_vec *bvec; 7429ffae0cfSNeilBrown struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *), 7434b6d287fSNeilBrown GFP_NOIO); 7444b6d287fSNeilBrown if (unlikely(!pages)) 7454b6d287fSNeilBrown goto do_sync_io; 7464b6d287fSNeilBrown 7474b6d287fSNeilBrown bio_for_each_segment(bvec, bio, i) { 7484b6d287fSNeilBrown pages[i] = alloc_page(GFP_NOIO); 7494b6d287fSNeilBrown if (unlikely(!pages[i])) 7504b6d287fSNeilBrown goto do_sync_io; 7514b6d287fSNeilBrown memcpy(kmap(pages[i]) + bvec->bv_offset, 7524b6d287fSNeilBrown kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); 7534b6d287fSNeilBrown kunmap(pages[i]); 7544b6d287fSNeilBrown kunmap(bvec->bv_page); 7554b6d287fSNeilBrown } 7564b6d287fSNeilBrown 7574b6d287fSNeilBrown return pages; 7584b6d287fSNeilBrown 7594b6d287fSNeilBrown do_sync_io: 7604b6d287fSNeilBrown if (pages) 7614b6d287fSNeilBrown for (i = 0; i < bio->bi_vcnt && pages[i]; i++) 7622d1f3b5dSNeilBrown put_page(pages[i]); 7634b6d287fSNeilBrown kfree(pages); 7644b6d287fSNeilBrown PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); 7654b6d287fSNeilBrown return NULL; 7664b6d287fSNeilBrown } 7674b6d287fSNeilBrown 768165125e1SJens Axboe static int make_request(struct request_queue *q, struct bio * bio) 7691da177e4SLinus Torvalds { 7701da177e4SLinus Torvalds mddev_t *mddev = q->queuedata; 7711da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 7721da177e4SLinus Torvalds mirror_info_t *mirror; 7731da177e4SLinus Torvalds r1bio_t *r1_bio; 7741da177e4SLinus Torvalds struct bio *read_bio; 775191ea9b2SNeilBrown int i, targets = 0, disks; 7761da177e4SLinus Torvalds mdk_rdev_t *rdev; 777191ea9b2SNeilBrown struct bitmap *bitmap = mddev->bitmap; 778191ea9b2SNeilBrown unsigned long flags; 779191ea9b2SNeilBrown struct bio_list bl; 7804b6d287fSNeilBrown struct page **behind_pages = NULL; 781a362357bSJens Axboe const int rw = bio_data_dir(bio); 782e3881a68SLars Ellenberg const int do_sync = bio_sync(bio); 783a9701a30SNeilBrown int do_barriers; 784191ea9b2SNeilBrown 7851da177e4SLinus Torvalds /* 7861da177e4SLinus Torvalds * Register the new request and wait if the reconstruction 7871da177e4SLinus Torvalds * thread has put up a bar for new requests. 7881da177e4SLinus Torvalds * Continue immediately if no resync is active currently. 78962de608dSNeilBrown * We test barriers_work *after* md_write_start as md_write_start 79062de608dSNeilBrown * may cause the first superblock write, and that will check out 79162de608dSNeilBrown * if barriers work. 7921da177e4SLinus Torvalds */ 79362de608dSNeilBrown 7943d310eb7SNeilBrown md_write_start(mddev, bio); /* wait on superblock update early */ 7953d310eb7SNeilBrown 79662de608dSNeilBrown if (unlikely(!mddev->barriers_work && bio_barrier(bio))) { 79762de608dSNeilBrown if (rw == WRITE) 79862de608dSNeilBrown md_write_end(mddev); 7996712ecf8SNeilBrown bio_endio(bio, -EOPNOTSUPP); 80062de608dSNeilBrown return 0; 80162de608dSNeilBrown } 80262de608dSNeilBrown 80317999be4SNeilBrown wait_barrier(conf); 8041da177e4SLinus Torvalds 805a362357bSJens Axboe disk_stat_inc(mddev->gendisk, ios[rw]); 806a362357bSJens Axboe disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio)); 8071da177e4SLinus Torvalds 8081da177e4SLinus Torvalds /* 8091da177e4SLinus Torvalds * make_request() can abort the operation when READA is being 8101da177e4SLinus Torvalds * used and no empty request is available. 8111da177e4SLinus Torvalds * 8121da177e4SLinus Torvalds */ 8131da177e4SLinus Torvalds r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); 8141da177e4SLinus Torvalds 8151da177e4SLinus Torvalds r1_bio->master_bio = bio; 8161da177e4SLinus Torvalds r1_bio->sectors = bio->bi_size >> 9; 817191ea9b2SNeilBrown r1_bio->state = 0; 8181da177e4SLinus Torvalds r1_bio->mddev = mddev; 8191da177e4SLinus Torvalds r1_bio->sector = bio->bi_sector; 8201da177e4SLinus Torvalds 821a362357bSJens Axboe if (rw == READ) { 8221da177e4SLinus Torvalds /* 8231da177e4SLinus Torvalds * read balancing logic: 8241da177e4SLinus Torvalds */ 8251da177e4SLinus Torvalds int rdisk = read_balance(conf, r1_bio); 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds if (rdisk < 0) { 8281da177e4SLinus Torvalds /* couldn't find anywhere to read from */ 8291da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 8301da177e4SLinus Torvalds return 0; 8311da177e4SLinus Torvalds } 8321da177e4SLinus Torvalds mirror = conf->mirrors + rdisk; 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds r1_bio->read_disk = rdisk; 8351da177e4SLinus Torvalds 8361da177e4SLinus Torvalds read_bio = bio_clone(bio, GFP_NOIO); 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds r1_bio->bios[rdisk] = read_bio; 8391da177e4SLinus Torvalds 8401da177e4SLinus Torvalds read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; 8411da177e4SLinus Torvalds read_bio->bi_bdev = mirror->rdev->bdev; 8421da177e4SLinus Torvalds read_bio->bi_end_io = raid1_end_read_request; 843e3881a68SLars Ellenberg read_bio->bi_rw = READ | do_sync; 8441da177e4SLinus Torvalds read_bio->bi_private = r1_bio; 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds generic_make_request(read_bio); 8471da177e4SLinus Torvalds return 0; 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds 8501da177e4SLinus Torvalds /* 8511da177e4SLinus Torvalds * WRITE: 8521da177e4SLinus Torvalds */ 8531da177e4SLinus Torvalds /* first select target devices under spinlock and 8541da177e4SLinus Torvalds * inc refcount on their rdev. Record them by setting 8551da177e4SLinus Torvalds * bios[x] to bio 8561da177e4SLinus Torvalds */ 8571da177e4SLinus Torvalds disks = conf->raid_disks; 858191ea9b2SNeilBrown #if 0 859191ea9b2SNeilBrown { static int first=1; 860191ea9b2SNeilBrown if (first) printk("First Write sector %llu disks %d\n", 861191ea9b2SNeilBrown (unsigned long long)r1_bio->sector, disks); 862191ea9b2SNeilBrown first = 0; 863191ea9b2SNeilBrown } 864191ea9b2SNeilBrown #endif 8651da177e4SLinus Torvalds rcu_read_lock(); 8661da177e4SLinus Torvalds for (i = 0; i < disks; i++) { 867d6065f7bSSuzanne Wood if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL && 868b2d444d7SNeilBrown !test_bit(Faulty, &rdev->flags)) { 8691da177e4SLinus Torvalds atomic_inc(&rdev->nr_pending); 870b2d444d7SNeilBrown if (test_bit(Faulty, &rdev->flags)) { 87103c902e1SNeilBrown rdev_dec_pending(rdev, mddev); 8721da177e4SLinus Torvalds r1_bio->bios[i] = NULL; 8731da177e4SLinus Torvalds } else 8741da177e4SLinus Torvalds r1_bio->bios[i] = bio; 875191ea9b2SNeilBrown targets++; 8761da177e4SLinus Torvalds } else 8771da177e4SLinus Torvalds r1_bio->bios[i] = NULL; 8781da177e4SLinus Torvalds } 8791da177e4SLinus Torvalds rcu_read_unlock(); 8801da177e4SLinus Torvalds 8814b6d287fSNeilBrown BUG_ON(targets == 0); /* we never fail the last device */ 8824b6d287fSNeilBrown 883191ea9b2SNeilBrown if (targets < conf->raid_disks) { 884191ea9b2SNeilBrown /* array is degraded, we will not clear the bitmap 885191ea9b2SNeilBrown * on I/O completion (see raid1_end_write_request) */ 886191ea9b2SNeilBrown set_bit(R1BIO_Degraded, &r1_bio->state); 887191ea9b2SNeilBrown } 88806d91a5fSNeilBrown 8894b6d287fSNeilBrown /* do behind I/O ? */ 8904b6d287fSNeilBrown if (bitmap && 8914b6d287fSNeilBrown atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind && 8924b6d287fSNeilBrown (behind_pages = alloc_behind_pages(bio)) != NULL) 8934b6d287fSNeilBrown set_bit(R1BIO_BehindIO, &r1_bio->state); 8944b6d287fSNeilBrown 895191ea9b2SNeilBrown atomic_set(&r1_bio->remaining, 0); 8964b6d287fSNeilBrown atomic_set(&r1_bio->behind_remaining, 0); 897191ea9b2SNeilBrown 89804b857f7SNeilBrown do_barriers = bio_barrier(bio); 899a9701a30SNeilBrown if (do_barriers) 900a9701a30SNeilBrown set_bit(R1BIO_Barrier, &r1_bio->state); 901a9701a30SNeilBrown 902191ea9b2SNeilBrown bio_list_init(&bl); 9031da177e4SLinus Torvalds for (i = 0; i < disks; i++) { 9041da177e4SLinus Torvalds struct bio *mbio; 9051da177e4SLinus Torvalds if (!r1_bio->bios[i]) 9061da177e4SLinus Torvalds continue; 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds mbio = bio_clone(bio, GFP_NOIO); 9091da177e4SLinus Torvalds r1_bio->bios[i] = mbio; 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset; 9121da177e4SLinus Torvalds mbio->bi_bdev = conf->mirrors[i].rdev->bdev; 9131da177e4SLinus Torvalds mbio->bi_end_io = raid1_end_write_request; 914e3881a68SLars Ellenberg mbio->bi_rw = WRITE | do_barriers | do_sync; 9151da177e4SLinus Torvalds mbio->bi_private = r1_bio; 9161da177e4SLinus Torvalds 9174b6d287fSNeilBrown if (behind_pages) { 9184b6d287fSNeilBrown struct bio_vec *bvec; 9194b6d287fSNeilBrown int j; 9204b6d287fSNeilBrown 9214b6d287fSNeilBrown /* Yes, I really want the '__' version so that 9224b6d287fSNeilBrown * we clear any unused pointer in the io_vec, rather 9234b6d287fSNeilBrown * than leave them unchanged. This is important 9244b6d287fSNeilBrown * because when we come to free the pages, we won't 9254b6d287fSNeilBrown * know the originial bi_idx, so we just free 9264b6d287fSNeilBrown * them all 9274b6d287fSNeilBrown */ 9284b6d287fSNeilBrown __bio_for_each_segment(bvec, mbio, j, 0) 9294b6d287fSNeilBrown bvec->bv_page = behind_pages[j]; 9304b6d287fSNeilBrown if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) 9314b6d287fSNeilBrown atomic_inc(&r1_bio->behind_remaining); 9324b6d287fSNeilBrown } 9334b6d287fSNeilBrown 9341da177e4SLinus Torvalds atomic_inc(&r1_bio->remaining); 935191ea9b2SNeilBrown 936191ea9b2SNeilBrown bio_list_add(&bl, mbio); 9371da177e4SLinus Torvalds } 9384b6d287fSNeilBrown kfree(behind_pages); /* the behind pages are attached to the bios now */ 9391da177e4SLinus Torvalds 9404b6d287fSNeilBrown bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors, 9414b6d287fSNeilBrown test_bit(R1BIO_BehindIO, &r1_bio->state)); 942191ea9b2SNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 943191ea9b2SNeilBrown bio_list_merge(&conf->pending_bio_list, &bl); 944191ea9b2SNeilBrown bio_list_init(&bl); 945191ea9b2SNeilBrown 946191ea9b2SNeilBrown blk_plug_device(mddev->queue); 947191ea9b2SNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 948191ea9b2SNeilBrown 949a35e63efSNeilBrown /* In case raid1d snuck into freeze_array */ 950a35e63efSNeilBrown wake_up(&conf->wait_barrier); 951a35e63efSNeilBrown 952e3881a68SLars Ellenberg if (do_sync) 953e3881a68SLars Ellenberg md_wakeup_thread(mddev->thread); 954191ea9b2SNeilBrown #if 0 955191ea9b2SNeilBrown while ((bio = bio_list_pop(&bl)) != NULL) 956191ea9b2SNeilBrown generic_make_request(bio); 957191ea9b2SNeilBrown #endif 9581da177e4SLinus Torvalds 9591da177e4SLinus Torvalds return 0; 9601da177e4SLinus Torvalds } 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds static void status(struct seq_file *seq, mddev_t *mddev) 9631da177e4SLinus Torvalds { 9641da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 9651da177e4SLinus Torvalds int i; 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds seq_printf(seq, " [%d/%d] [", conf->raid_disks, 96811ce99e6SNeilBrown conf->raid_disks - mddev->degraded); 969ddac7c7eSNeilBrown rcu_read_lock(); 970ddac7c7eSNeilBrown for (i = 0; i < conf->raid_disks; i++) { 971ddac7c7eSNeilBrown mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); 9721da177e4SLinus Torvalds seq_printf(seq, "%s", 973ddac7c7eSNeilBrown rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); 974ddac7c7eSNeilBrown } 975ddac7c7eSNeilBrown rcu_read_unlock(); 9761da177e4SLinus Torvalds seq_printf(seq, "]"); 9771da177e4SLinus Torvalds } 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds static void error(mddev_t *mddev, mdk_rdev_t *rdev) 9811da177e4SLinus Torvalds { 9821da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 9831da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 9841da177e4SLinus Torvalds 9851da177e4SLinus Torvalds /* 9861da177e4SLinus Torvalds * If it is not operational, then we have already marked it as dead 9871da177e4SLinus Torvalds * else if it is the last working disks, ignore the error, let the 9881da177e4SLinus Torvalds * next level up know. 9891da177e4SLinus Torvalds * else mark the drive as failed 9901da177e4SLinus Torvalds */ 991b2d444d7SNeilBrown if (test_bit(In_sync, &rdev->flags) 99211ce99e6SNeilBrown && (conf->raid_disks - mddev->degraded) == 1) 9931da177e4SLinus Torvalds /* 9941da177e4SLinus Torvalds * Don't fail the drive, act as though we were just a 9951da177e4SLinus Torvalds * normal single drive 9961da177e4SLinus Torvalds */ 9971da177e4SLinus Torvalds return; 998c04be0aaSNeilBrown if (test_and_clear_bit(In_sync, &rdev->flags)) { 999c04be0aaSNeilBrown unsigned long flags; 1000c04be0aaSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 10011da177e4SLinus Torvalds mddev->degraded++; 1002dd00a99eSNeilBrown set_bit(Faulty, &rdev->flags); 1003c04be0aaSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 10041da177e4SLinus Torvalds /* 10051da177e4SLinus Torvalds * if recovery is running, make sure it aborts. 10061da177e4SLinus Torvalds */ 10071da177e4SLinus Torvalds set_bit(MD_RECOVERY_ERR, &mddev->recovery); 1008dd00a99eSNeilBrown } else 1009b2d444d7SNeilBrown set_bit(Faulty, &rdev->flags); 1010850b2b42SNeilBrown set_bit(MD_CHANGE_DEVS, &mddev->flags); 10111da177e4SLinus Torvalds printk(KERN_ALERT "raid1: Disk failure on %s, disabling device.\n" 1012d7a420c9SNick Andrew "raid1: Operation continuing on %d devices.\n", 101311ce99e6SNeilBrown bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds static void print_conf(conf_t *conf) 10171da177e4SLinus Torvalds { 10181da177e4SLinus Torvalds int i; 10191da177e4SLinus Torvalds 10201da177e4SLinus Torvalds printk("RAID1 conf printout:\n"); 10211da177e4SLinus Torvalds if (!conf) { 10221da177e4SLinus Torvalds printk("(!conf)\n"); 10231da177e4SLinus Torvalds return; 10241da177e4SLinus Torvalds } 102511ce99e6SNeilBrown printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, 10261da177e4SLinus Torvalds conf->raid_disks); 10271da177e4SLinus Torvalds 1028ddac7c7eSNeilBrown rcu_read_lock(); 10291da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 10301da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1031ddac7c7eSNeilBrown mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); 1032ddac7c7eSNeilBrown if (rdev) 10331da177e4SLinus Torvalds printk(" disk %d, wo:%d, o:%d, dev:%s\n", 1034ddac7c7eSNeilBrown i, !test_bit(In_sync, &rdev->flags), 1035ddac7c7eSNeilBrown !test_bit(Faulty, &rdev->flags), 1036ddac7c7eSNeilBrown bdevname(rdev->bdev,b)); 10371da177e4SLinus Torvalds } 1038ddac7c7eSNeilBrown rcu_read_unlock(); 10391da177e4SLinus Torvalds } 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds static void close_sync(conf_t *conf) 10421da177e4SLinus Torvalds { 104317999be4SNeilBrown wait_barrier(conf); 104417999be4SNeilBrown allow_barrier(conf); 10451da177e4SLinus Torvalds 10461da177e4SLinus Torvalds mempool_destroy(conf->r1buf_pool); 10471da177e4SLinus Torvalds conf->r1buf_pool = NULL; 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 10501da177e4SLinus Torvalds static int raid1_spare_active(mddev_t *mddev) 10511da177e4SLinus Torvalds { 10521da177e4SLinus Torvalds int i; 10531da177e4SLinus Torvalds conf_t *conf = mddev->private; 10541da177e4SLinus Torvalds 10551da177e4SLinus Torvalds /* 10561da177e4SLinus Torvalds * Find all failed disks within the RAID1 configuration 1057ddac7c7eSNeilBrown * and mark them readable. 1058ddac7c7eSNeilBrown * Called under mddev lock, so rcu protection not needed. 10591da177e4SLinus Torvalds */ 10601da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 1061ddac7c7eSNeilBrown mdk_rdev_t *rdev = conf->mirrors[i].rdev; 1062ddac7c7eSNeilBrown if (rdev 1063ddac7c7eSNeilBrown && !test_bit(Faulty, &rdev->flags) 1064c04be0aaSNeilBrown && !test_and_set_bit(In_sync, &rdev->flags)) { 1065c04be0aaSNeilBrown unsigned long flags; 1066c04be0aaSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 10671da177e4SLinus Torvalds mddev->degraded--; 1068c04be0aaSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds 10721da177e4SLinus Torvalds print_conf(conf); 10731da177e4SLinus Torvalds return 0; 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds 10771da177e4SLinus Torvalds static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) 10781da177e4SLinus Torvalds { 10791da177e4SLinus Torvalds conf_t *conf = mddev->private; 10801da177e4SLinus Torvalds int found = 0; 108141158c7eSNeilBrown int mirror = 0; 10821da177e4SLinus Torvalds mirror_info_t *p; 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds for (mirror=0; mirror < mddev->raid_disks; mirror++) 10851da177e4SLinus Torvalds if ( !(p=conf->mirrors+mirror)->rdev) { 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds blk_queue_stack_limits(mddev->queue, 10881da177e4SLinus Torvalds rdev->bdev->bd_disk->queue); 10891da177e4SLinus Torvalds /* as we don't honour merge_bvec_fn, we must never risk 10901da177e4SLinus Torvalds * violating it, so limit ->max_sector to one PAGE, as 10911da177e4SLinus Torvalds * a one page request is never in violation. 10921da177e4SLinus Torvalds */ 10931da177e4SLinus Torvalds if (rdev->bdev->bd_disk->queue->merge_bvec_fn && 10941da177e4SLinus Torvalds mddev->queue->max_sectors > (PAGE_SIZE>>9)) 10951da177e4SLinus Torvalds blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds p->head_position = 0; 10981da177e4SLinus Torvalds rdev->raid_disk = mirror; 10991da177e4SLinus Torvalds found = 1; 11006aea114aSNeilBrown /* As all devices are equivalent, we don't need a full recovery 11016aea114aSNeilBrown * if this was recently any drive of the array 11026aea114aSNeilBrown */ 11036aea114aSNeilBrown if (rdev->saved_raid_disk < 0) 110441158c7eSNeilBrown conf->fullsync = 1; 1105d6065f7bSSuzanne Wood rcu_assign_pointer(p->rdev, rdev); 11061da177e4SLinus Torvalds break; 11071da177e4SLinus Torvalds } 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds print_conf(conf); 11101da177e4SLinus Torvalds return found; 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds 11131da177e4SLinus Torvalds static int raid1_remove_disk(mddev_t *mddev, int number) 11141da177e4SLinus Torvalds { 11151da177e4SLinus Torvalds conf_t *conf = mddev->private; 11161da177e4SLinus Torvalds int err = 0; 11171da177e4SLinus Torvalds mdk_rdev_t *rdev; 11181da177e4SLinus Torvalds mirror_info_t *p = conf->mirrors+ number; 11191da177e4SLinus Torvalds 11201da177e4SLinus Torvalds print_conf(conf); 11211da177e4SLinus Torvalds rdev = p->rdev; 11221da177e4SLinus Torvalds if (rdev) { 1123b2d444d7SNeilBrown if (test_bit(In_sync, &rdev->flags) || 11241da177e4SLinus Torvalds atomic_read(&rdev->nr_pending)) { 11251da177e4SLinus Torvalds err = -EBUSY; 11261da177e4SLinus Torvalds goto abort; 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds p->rdev = NULL; 1129fbd568a3SPaul E. McKenney synchronize_rcu(); 11301da177e4SLinus Torvalds if (atomic_read(&rdev->nr_pending)) { 11311da177e4SLinus Torvalds /* lost the race, try later */ 11321da177e4SLinus Torvalds err = -EBUSY; 11331da177e4SLinus Torvalds p->rdev = rdev; 11341da177e4SLinus Torvalds } 11351da177e4SLinus Torvalds } 11361da177e4SLinus Torvalds abort: 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds print_conf(conf); 11391da177e4SLinus Torvalds return err; 11401da177e4SLinus Torvalds } 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds 11436712ecf8SNeilBrown static void end_sync_read(struct bio *bio, int error) 11441da177e4SLinus Torvalds { 11451da177e4SLinus Torvalds r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); 1146d11c171eSNeilBrown int i; 11471da177e4SLinus Torvalds 1148d11c171eSNeilBrown for (i=r1_bio->mddev->raid_disks; i--; ) 1149d11c171eSNeilBrown if (r1_bio->bios[i] == bio) 1150d11c171eSNeilBrown break; 1151d11c171eSNeilBrown BUG_ON(i < 0); 1152d11c171eSNeilBrown update_head_pos(i, r1_bio); 11531da177e4SLinus Torvalds /* 11541da177e4SLinus Torvalds * we have read a block, now it needs to be re-written, 11551da177e4SLinus Torvalds * or re-read if the read failed. 11561da177e4SLinus Torvalds * We don't do much here, just schedule handling by raid1d 11571da177e4SLinus Torvalds */ 115869382e85SNeilBrown if (test_bit(BIO_UPTODATE, &bio->bi_flags)) 11591da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 1160d11c171eSNeilBrown 1161d11c171eSNeilBrown if (atomic_dec_and_test(&r1_bio->remaining)) 11621da177e4SLinus Torvalds reschedule_retry(r1_bio); 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds 11656712ecf8SNeilBrown static void end_sync_write(struct bio *bio, int error) 11661da177e4SLinus Torvalds { 11671da177e4SLinus Torvalds int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 11681da177e4SLinus Torvalds r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); 11691da177e4SLinus Torvalds mddev_t *mddev = r1_bio->mddev; 11701da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 11711da177e4SLinus Torvalds int i; 11721da177e4SLinus Torvalds int mirror=0; 11731da177e4SLinus Torvalds 11741da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) 11751da177e4SLinus Torvalds if (r1_bio->bios[i] == bio) { 11761da177e4SLinus Torvalds mirror = i; 11771da177e4SLinus Torvalds break; 11781da177e4SLinus Torvalds } 11796b1117d5SNeilBrown if (!uptodate) { 11806b1117d5SNeilBrown int sync_blocks = 0; 11816b1117d5SNeilBrown sector_t s = r1_bio->sector; 11826b1117d5SNeilBrown long sectors_to_go = r1_bio->sectors; 11836b1117d5SNeilBrown /* make sure these bits doesn't get cleared. */ 11846b1117d5SNeilBrown do { 11855e3db645SNeilBrown bitmap_end_sync(mddev->bitmap, s, 11866b1117d5SNeilBrown &sync_blocks, 1); 11876b1117d5SNeilBrown s += sync_blocks; 11886b1117d5SNeilBrown sectors_to_go -= sync_blocks; 11896b1117d5SNeilBrown } while (sectors_to_go > 0); 11901da177e4SLinus Torvalds md_error(mddev, conf->mirrors[mirror].rdev); 11916b1117d5SNeilBrown } 1192e3b9703eSNeilBrown 11931da177e4SLinus Torvalds update_head_pos(mirror, r1_bio); 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds if (atomic_dec_and_test(&r1_bio->remaining)) { 11961da177e4SLinus Torvalds md_done_sync(mddev, r1_bio->sectors, uptodate); 11971da177e4SLinus Torvalds put_buf(r1_bio); 11981da177e4SLinus Torvalds } 11991da177e4SLinus Torvalds } 12001da177e4SLinus Torvalds 12011da177e4SLinus Torvalds static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) 12021da177e4SLinus Torvalds { 12031da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 12041da177e4SLinus Torvalds int i; 12051da177e4SLinus Torvalds int disks = conf->raid_disks; 12061da177e4SLinus Torvalds struct bio *bio, *wbio; 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds bio = r1_bio->bios[r1_bio->read_disk]; 12091da177e4SLinus Torvalds 121069382e85SNeilBrown 1211d11c171eSNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 1212d11c171eSNeilBrown /* We have read all readable devices. If we haven't 1213d11c171eSNeilBrown * got the block, then there is no hope left. 1214d11c171eSNeilBrown * If we have, then we want to do a comparison 1215d11c171eSNeilBrown * and skip the write if everything is the same. 1216d11c171eSNeilBrown * If any blocks failed to read, then we need to 1217d11c171eSNeilBrown * attempt an over-write 12181da177e4SLinus Torvalds */ 1219d11c171eSNeilBrown int primary; 1220d11c171eSNeilBrown if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { 1221d11c171eSNeilBrown for (i=0; i<mddev->raid_disks; i++) 1222d11c171eSNeilBrown if (r1_bio->bios[i]->bi_end_io == end_sync_read) 1223d11c171eSNeilBrown md_error(mddev, conf->mirrors[i].rdev); 1224d11c171eSNeilBrown 1225d11c171eSNeilBrown md_done_sync(mddev, r1_bio->sectors, 1); 1226d11c171eSNeilBrown put_buf(r1_bio); 1227d11c171eSNeilBrown return; 1228d11c171eSNeilBrown } 1229d11c171eSNeilBrown for (primary=0; primary<mddev->raid_disks; primary++) 1230d11c171eSNeilBrown if (r1_bio->bios[primary]->bi_end_io == end_sync_read && 1231d11c171eSNeilBrown test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { 1232d11c171eSNeilBrown r1_bio->bios[primary]->bi_end_io = NULL; 123303c902e1SNeilBrown rdev_dec_pending(conf->mirrors[primary].rdev, mddev); 1234d11c171eSNeilBrown break; 1235d11c171eSNeilBrown } 1236d11c171eSNeilBrown r1_bio->read_disk = primary; 1237d11c171eSNeilBrown for (i=0; i<mddev->raid_disks; i++) 1238ed456662SMike Accetta if (r1_bio->bios[i]->bi_end_io == end_sync_read) { 1239d11c171eSNeilBrown int j; 1240d11c171eSNeilBrown int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); 1241d11c171eSNeilBrown struct bio *pbio = r1_bio->bios[primary]; 1242d11c171eSNeilBrown struct bio *sbio = r1_bio->bios[i]; 1243ed456662SMike Accetta 1244ed456662SMike Accetta if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { 1245ed456662SMike Accetta for (j = vcnt; j-- ; ) { 1246ed456662SMike Accetta struct page *p, *s; 1247ed456662SMike Accetta p = pbio->bi_io_vec[j].bv_page; 1248ed456662SMike Accetta s = sbio->bi_io_vec[j].bv_page; 1249ed456662SMike Accetta if (memcmp(page_address(p), 1250ed456662SMike Accetta page_address(s), 1251d11c171eSNeilBrown PAGE_SIZE)) 1252d11c171eSNeilBrown break; 1253ed456662SMike Accetta } 1254ed456662SMike Accetta } else 1255ed456662SMike Accetta j = 0; 1256d11c171eSNeilBrown if (j >= 0) 1257d11c171eSNeilBrown mddev->resync_mismatches += r1_bio->sectors; 1258cf7a4416SNeilBrown if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) 1259cf7a4416SNeilBrown && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { 1260d11c171eSNeilBrown sbio->bi_end_io = NULL; 126103c902e1SNeilBrown rdev_dec_pending(conf->mirrors[i].rdev, mddev); 126203c902e1SNeilBrown } else { 1263d11c171eSNeilBrown /* fixup the bio for reuse */ 1264d11c171eSNeilBrown sbio->bi_vcnt = vcnt; 1265d11c171eSNeilBrown sbio->bi_size = r1_bio->sectors << 9; 1266d11c171eSNeilBrown sbio->bi_idx = 0; 1267d11c171eSNeilBrown sbio->bi_phys_segments = 0; 1268d11c171eSNeilBrown sbio->bi_hw_segments = 0; 1269d11c171eSNeilBrown sbio->bi_hw_front_size = 0; 1270d11c171eSNeilBrown sbio->bi_hw_back_size = 0; 1271d11c171eSNeilBrown sbio->bi_flags &= ~(BIO_POOL_MASK - 1); 1272d11c171eSNeilBrown sbio->bi_flags |= 1 << BIO_UPTODATE; 1273d11c171eSNeilBrown sbio->bi_next = NULL; 1274d11c171eSNeilBrown sbio->bi_sector = r1_bio->sector + 1275d11c171eSNeilBrown conf->mirrors[i].rdev->data_offset; 1276d11c171eSNeilBrown sbio->bi_bdev = conf->mirrors[i].rdev->bdev; 12773eda22d1SNeilBrown for (j = 0; j < vcnt ; j++) 12783eda22d1SNeilBrown memcpy(page_address(sbio->bi_io_vec[j].bv_page), 12793eda22d1SNeilBrown page_address(pbio->bi_io_vec[j].bv_page), 12803eda22d1SNeilBrown PAGE_SIZE); 12813eda22d1SNeilBrown 1282d11c171eSNeilBrown } 1283d11c171eSNeilBrown } 1284d11c171eSNeilBrown } 12851da177e4SLinus Torvalds if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { 128669382e85SNeilBrown /* ouch - failed to read all of that. 128769382e85SNeilBrown * Try some synchronous reads of other devices to get 128869382e85SNeilBrown * good data, much like with normal read errors. Only 1289ddac7c7eSNeilBrown * read into the pages we already have so we don't 129069382e85SNeilBrown * need to re-issue the read request. 129169382e85SNeilBrown * We don't need to freeze the array, because being in an 129269382e85SNeilBrown * active sync request, there is no normal IO, and 129369382e85SNeilBrown * no overlapping syncs. 12941da177e4SLinus Torvalds */ 129569382e85SNeilBrown sector_t sect = r1_bio->sector; 129669382e85SNeilBrown int sectors = r1_bio->sectors; 129769382e85SNeilBrown int idx = 0; 129869382e85SNeilBrown 129969382e85SNeilBrown while(sectors) { 130069382e85SNeilBrown int s = sectors; 130169382e85SNeilBrown int d = r1_bio->read_disk; 130269382e85SNeilBrown int success = 0; 130369382e85SNeilBrown mdk_rdev_t *rdev; 130469382e85SNeilBrown 130569382e85SNeilBrown if (s > (PAGE_SIZE>>9)) 130669382e85SNeilBrown s = PAGE_SIZE >> 9; 130769382e85SNeilBrown do { 130869382e85SNeilBrown if (r1_bio->bios[d]->bi_end_io == end_sync_read) { 1309ddac7c7eSNeilBrown /* No rcu protection needed here devices 1310ddac7c7eSNeilBrown * can only be removed when no resync is 1311ddac7c7eSNeilBrown * active, and resync is currently active 1312ddac7c7eSNeilBrown */ 131369382e85SNeilBrown rdev = conf->mirrors[d].rdev; 131469382e85SNeilBrown if (sync_page_io(rdev->bdev, 131569382e85SNeilBrown sect + rdev->data_offset, 131669382e85SNeilBrown s<<9, 131769382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 131869382e85SNeilBrown READ)) { 131969382e85SNeilBrown success = 1; 132069382e85SNeilBrown break; 132169382e85SNeilBrown } 132269382e85SNeilBrown } 132369382e85SNeilBrown d++; 132469382e85SNeilBrown if (d == conf->raid_disks) 132569382e85SNeilBrown d = 0; 132669382e85SNeilBrown } while (!success && d != r1_bio->read_disk); 132769382e85SNeilBrown 132869382e85SNeilBrown if (success) { 1329097426f6SNeilBrown int start = d; 133069382e85SNeilBrown /* write it back and re-read */ 133169382e85SNeilBrown set_bit(R1BIO_Uptodate, &r1_bio->state); 133269382e85SNeilBrown while (d != r1_bio->read_disk) { 133369382e85SNeilBrown if (d == 0) 133469382e85SNeilBrown d = conf->raid_disks; 133569382e85SNeilBrown d--; 133669382e85SNeilBrown if (r1_bio->bios[d]->bi_end_io != end_sync_read) 133769382e85SNeilBrown continue; 133869382e85SNeilBrown rdev = conf->mirrors[d].rdev; 13394dbcdc75SNeilBrown atomic_add(s, &rdev->corrected_errors); 134069382e85SNeilBrown if (sync_page_io(rdev->bdev, 134169382e85SNeilBrown sect + rdev->data_offset, 134269382e85SNeilBrown s<<9, 134369382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 1344097426f6SNeilBrown WRITE) == 0) 1345097426f6SNeilBrown md_error(mddev, rdev); 1346097426f6SNeilBrown } 1347097426f6SNeilBrown d = start; 1348097426f6SNeilBrown while (d != r1_bio->read_disk) { 1349097426f6SNeilBrown if (d == 0) 1350097426f6SNeilBrown d = conf->raid_disks; 1351097426f6SNeilBrown d--; 1352097426f6SNeilBrown if (r1_bio->bios[d]->bi_end_io != end_sync_read) 1353097426f6SNeilBrown continue; 1354097426f6SNeilBrown rdev = conf->mirrors[d].rdev; 1355097426f6SNeilBrown if (sync_page_io(rdev->bdev, 135669382e85SNeilBrown sect + rdev->data_offset, 135769382e85SNeilBrown s<<9, 135869382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 1359097426f6SNeilBrown READ) == 0) 136069382e85SNeilBrown md_error(mddev, rdev); 136169382e85SNeilBrown } 136269382e85SNeilBrown } else { 13631da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 136469382e85SNeilBrown /* Cannot read from anywhere, array is toast */ 136569382e85SNeilBrown md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); 13661da177e4SLinus Torvalds printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error" 13671da177e4SLinus Torvalds " for block %llu\n", 13681da177e4SLinus Torvalds bdevname(bio->bi_bdev,b), 13691da177e4SLinus Torvalds (unsigned long long)r1_bio->sector); 13701da177e4SLinus Torvalds md_done_sync(mddev, r1_bio->sectors, 0); 13711da177e4SLinus Torvalds put_buf(r1_bio); 13721da177e4SLinus Torvalds return; 13731da177e4SLinus Torvalds } 137469382e85SNeilBrown sectors -= s; 137569382e85SNeilBrown sect += s; 137669382e85SNeilBrown idx ++; 137769382e85SNeilBrown } 137869382e85SNeilBrown } 1379d11c171eSNeilBrown 1380d11c171eSNeilBrown /* 1381d11c171eSNeilBrown * schedule writes 1382d11c171eSNeilBrown */ 13831da177e4SLinus Torvalds atomic_set(&r1_bio->remaining, 1); 13841da177e4SLinus Torvalds for (i = 0; i < disks ; i++) { 13851da177e4SLinus Torvalds wbio = r1_bio->bios[i]; 13863e198f78SNeilBrown if (wbio->bi_end_io == NULL || 13873e198f78SNeilBrown (wbio->bi_end_io == end_sync_read && 13883e198f78SNeilBrown (i == r1_bio->read_disk || 13893e198f78SNeilBrown !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) 13901da177e4SLinus Torvalds continue; 13911da177e4SLinus Torvalds 13923e198f78SNeilBrown wbio->bi_rw = WRITE; 13933e198f78SNeilBrown wbio->bi_end_io = end_sync_write; 13941da177e4SLinus Torvalds atomic_inc(&r1_bio->remaining); 13951da177e4SLinus Torvalds md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); 1396191ea9b2SNeilBrown 13971da177e4SLinus Torvalds generic_make_request(wbio); 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds if (atomic_dec_and_test(&r1_bio->remaining)) { 1401191ea9b2SNeilBrown /* if we're here, all write(s) have completed, so clean up */ 14021da177e4SLinus Torvalds md_done_sync(mddev, r1_bio->sectors, 1); 14031da177e4SLinus Torvalds put_buf(r1_bio); 14041da177e4SLinus Torvalds } 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds /* 14081da177e4SLinus Torvalds * This is a kernel thread which: 14091da177e4SLinus Torvalds * 14101da177e4SLinus Torvalds * 1. Retries failed read operations on working mirrors. 14111da177e4SLinus Torvalds * 2. Updates the raid superblock when problems encounter. 14121da177e4SLinus Torvalds * 3. Performs writes following reads for array syncronising. 14131da177e4SLinus Torvalds */ 14141da177e4SLinus Torvalds 1415867868fbSNeilBrown static void fix_read_error(conf_t *conf, int read_disk, 1416867868fbSNeilBrown sector_t sect, int sectors) 1417867868fbSNeilBrown { 1418867868fbSNeilBrown mddev_t *mddev = conf->mddev; 1419867868fbSNeilBrown while(sectors) { 1420867868fbSNeilBrown int s = sectors; 1421867868fbSNeilBrown int d = read_disk; 1422867868fbSNeilBrown int success = 0; 1423867868fbSNeilBrown int start; 1424867868fbSNeilBrown mdk_rdev_t *rdev; 1425867868fbSNeilBrown 1426867868fbSNeilBrown if (s > (PAGE_SIZE>>9)) 1427867868fbSNeilBrown s = PAGE_SIZE >> 9; 1428867868fbSNeilBrown 1429867868fbSNeilBrown do { 1430867868fbSNeilBrown /* Note: no rcu protection needed here 1431867868fbSNeilBrown * as this is synchronous in the raid1d thread 1432867868fbSNeilBrown * which is the thread that might remove 1433867868fbSNeilBrown * a device. If raid1d ever becomes multi-threaded.... 1434867868fbSNeilBrown */ 1435867868fbSNeilBrown rdev = conf->mirrors[d].rdev; 1436867868fbSNeilBrown if (rdev && 1437867868fbSNeilBrown test_bit(In_sync, &rdev->flags) && 1438867868fbSNeilBrown sync_page_io(rdev->bdev, 1439867868fbSNeilBrown sect + rdev->data_offset, 1440867868fbSNeilBrown s<<9, 1441867868fbSNeilBrown conf->tmppage, READ)) 1442867868fbSNeilBrown success = 1; 1443867868fbSNeilBrown else { 1444867868fbSNeilBrown d++; 1445867868fbSNeilBrown if (d == conf->raid_disks) 1446867868fbSNeilBrown d = 0; 1447867868fbSNeilBrown } 1448867868fbSNeilBrown } while (!success && d != read_disk); 1449867868fbSNeilBrown 1450867868fbSNeilBrown if (!success) { 1451867868fbSNeilBrown /* Cannot read from anywhere -- bye bye array */ 1452867868fbSNeilBrown md_error(mddev, conf->mirrors[read_disk].rdev); 1453867868fbSNeilBrown break; 1454867868fbSNeilBrown } 1455867868fbSNeilBrown /* write it back and re-read */ 1456867868fbSNeilBrown start = d; 1457867868fbSNeilBrown while (d != read_disk) { 1458867868fbSNeilBrown if (d==0) 1459867868fbSNeilBrown d = conf->raid_disks; 1460867868fbSNeilBrown d--; 1461867868fbSNeilBrown rdev = conf->mirrors[d].rdev; 1462867868fbSNeilBrown if (rdev && 1463867868fbSNeilBrown test_bit(In_sync, &rdev->flags)) { 1464867868fbSNeilBrown if (sync_page_io(rdev->bdev, 1465867868fbSNeilBrown sect + rdev->data_offset, 1466867868fbSNeilBrown s<<9, conf->tmppage, WRITE) 1467867868fbSNeilBrown == 0) 1468867868fbSNeilBrown /* Well, this device is dead */ 1469867868fbSNeilBrown md_error(mddev, rdev); 1470867868fbSNeilBrown } 1471867868fbSNeilBrown } 1472867868fbSNeilBrown d = start; 1473867868fbSNeilBrown while (d != read_disk) { 1474867868fbSNeilBrown char b[BDEVNAME_SIZE]; 1475867868fbSNeilBrown if (d==0) 1476867868fbSNeilBrown d = conf->raid_disks; 1477867868fbSNeilBrown d--; 1478867868fbSNeilBrown rdev = conf->mirrors[d].rdev; 1479867868fbSNeilBrown if (rdev && 1480867868fbSNeilBrown test_bit(In_sync, &rdev->flags)) { 1481867868fbSNeilBrown if (sync_page_io(rdev->bdev, 1482867868fbSNeilBrown sect + rdev->data_offset, 1483867868fbSNeilBrown s<<9, conf->tmppage, READ) 1484867868fbSNeilBrown == 0) 1485867868fbSNeilBrown /* Well, this device is dead */ 1486867868fbSNeilBrown md_error(mddev, rdev); 1487867868fbSNeilBrown else { 1488867868fbSNeilBrown atomic_add(s, &rdev->corrected_errors); 1489867868fbSNeilBrown printk(KERN_INFO 1490867868fbSNeilBrown "raid1:%s: read error corrected " 1491867868fbSNeilBrown "(%d sectors at %llu on %s)\n", 1492867868fbSNeilBrown mdname(mddev), s, 1493969b755aSRandy Dunlap (unsigned long long)(sect + 1494969b755aSRandy Dunlap rdev->data_offset), 1495867868fbSNeilBrown bdevname(rdev->bdev, b)); 1496867868fbSNeilBrown } 1497867868fbSNeilBrown } 1498867868fbSNeilBrown } 1499867868fbSNeilBrown sectors -= s; 1500867868fbSNeilBrown sect += s; 1501867868fbSNeilBrown } 1502867868fbSNeilBrown } 1503867868fbSNeilBrown 15041da177e4SLinus Torvalds static void raid1d(mddev_t *mddev) 15051da177e4SLinus Torvalds { 15061da177e4SLinus Torvalds r1bio_t *r1_bio; 15071da177e4SLinus Torvalds struct bio *bio; 15081da177e4SLinus Torvalds unsigned long flags; 15091da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 15101da177e4SLinus Torvalds struct list_head *head = &conf->retry_list; 15111da177e4SLinus Torvalds int unplug=0; 15121da177e4SLinus Torvalds mdk_rdev_t *rdev; 15131da177e4SLinus Torvalds 15141da177e4SLinus Torvalds md_check_recovery(mddev); 15151da177e4SLinus Torvalds 15161da177e4SLinus Torvalds for (;;) { 15171da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1518a35e63efSNeilBrown 1519a35e63efSNeilBrown unplug += flush_pending_writes(conf); 1520a35e63efSNeilBrown 15211da177e4SLinus Torvalds spin_lock_irqsave(&conf->device_lock, flags); 1522a35e63efSNeilBrown if (list_empty(head)) { 1523191ea9b2SNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 15241da177e4SLinus Torvalds break; 1525a35e63efSNeilBrown } 15261da177e4SLinus Torvalds r1_bio = list_entry(head->prev, r1bio_t, retry_list); 15271da177e4SLinus Torvalds list_del(head->prev); 1528ddaf22abSNeilBrown conf->nr_queued--; 15291da177e4SLinus Torvalds spin_unlock_irqrestore(&conf->device_lock, flags); 15301da177e4SLinus Torvalds 15311da177e4SLinus Torvalds mddev = r1_bio->mddev; 15321da177e4SLinus Torvalds conf = mddev_to_conf(mddev); 15331da177e4SLinus Torvalds if (test_bit(R1BIO_IsSync, &r1_bio->state)) { 15341da177e4SLinus Torvalds sync_request_write(mddev, r1_bio); 15351da177e4SLinus Torvalds unplug = 1; 1536a9701a30SNeilBrown } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) { 1537a9701a30SNeilBrown /* some requests in the r1bio were BIO_RW_BARRIER 1538bea27718SNeilBrown * requests which failed with -EOPNOTSUPP. Hohumm.. 1539a9701a30SNeilBrown * Better resubmit without the barrier. 1540a9701a30SNeilBrown * We know which devices to resubmit for, because 1541a9701a30SNeilBrown * all others have had their bios[] entry cleared. 15425e7dd2abSNeilBrown * We already have a nr_pending reference on these rdevs. 1543a9701a30SNeilBrown */ 1544a9701a30SNeilBrown int i; 1545e3881a68SLars Ellenberg const int do_sync = bio_sync(r1_bio->master_bio); 1546a9701a30SNeilBrown clear_bit(R1BIO_BarrierRetry, &r1_bio->state); 1547a9701a30SNeilBrown clear_bit(R1BIO_Barrier, &r1_bio->state); 1548a9701a30SNeilBrown for (i=0; i < conf->raid_disks; i++) 15492f889129SNeilBrown if (r1_bio->bios[i]) 15502f889129SNeilBrown atomic_inc(&r1_bio->remaining); 15512f889129SNeilBrown for (i=0; i < conf->raid_disks; i++) 1552a9701a30SNeilBrown if (r1_bio->bios[i]) { 1553a9701a30SNeilBrown struct bio_vec *bvec; 1554a9701a30SNeilBrown int j; 1555a9701a30SNeilBrown 1556a9701a30SNeilBrown bio = bio_clone(r1_bio->master_bio, GFP_NOIO); 1557a9701a30SNeilBrown /* copy pages from the failed bio, as 1558a9701a30SNeilBrown * this might be a write-behind device */ 1559a9701a30SNeilBrown __bio_for_each_segment(bvec, bio, j, 0) 1560a9701a30SNeilBrown bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page; 1561a9701a30SNeilBrown bio_put(r1_bio->bios[i]); 1562a9701a30SNeilBrown bio->bi_sector = r1_bio->sector + 1563a9701a30SNeilBrown conf->mirrors[i].rdev->data_offset; 1564a9701a30SNeilBrown bio->bi_bdev = conf->mirrors[i].rdev->bdev; 1565a9701a30SNeilBrown bio->bi_end_io = raid1_end_write_request; 1566e3881a68SLars Ellenberg bio->bi_rw = WRITE | do_sync; 1567a9701a30SNeilBrown bio->bi_private = r1_bio; 1568a9701a30SNeilBrown r1_bio->bios[i] = bio; 1569a9701a30SNeilBrown generic_make_request(bio); 1570a9701a30SNeilBrown } 15711da177e4SLinus Torvalds } else { 15721da177e4SLinus Torvalds int disk; 1573ddaf22abSNeilBrown 1574ddaf22abSNeilBrown /* we got a read error. Maybe the drive is bad. Maybe just 1575ddaf22abSNeilBrown * the block and we can fix it. 1576ddaf22abSNeilBrown * We freeze all other IO, and try reading the block from 1577ddaf22abSNeilBrown * other devices. When we find one, we re-write 1578ddaf22abSNeilBrown * and check it that fixes the read error. 1579ddaf22abSNeilBrown * This is all done synchronously while the array is 1580ddaf22abSNeilBrown * frozen 1581ddaf22abSNeilBrown */ 1582867868fbSNeilBrown if (mddev->ro == 0) { 1583ddaf22abSNeilBrown freeze_array(conf); 1584867868fbSNeilBrown fix_read_error(conf, r1_bio->read_disk, 1585867868fbSNeilBrown r1_bio->sector, 1586867868fbSNeilBrown r1_bio->sectors); 1587ddaf22abSNeilBrown unfreeze_array(conf); 1588867868fbSNeilBrown } 1589ddaf22abSNeilBrown 15901da177e4SLinus Torvalds bio = r1_bio->bios[r1_bio->read_disk]; 15911da177e4SLinus Torvalds if ((disk=read_balance(conf, r1_bio)) == -1) { 15921da177e4SLinus Torvalds printk(KERN_ALERT "raid1: %s: unrecoverable I/O" 15931da177e4SLinus Torvalds " read error for block %llu\n", 15941da177e4SLinus Torvalds bdevname(bio->bi_bdev,b), 15951da177e4SLinus Torvalds (unsigned long long)r1_bio->sector); 15961da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 15971da177e4SLinus Torvalds } else { 1598e3881a68SLars Ellenberg const int do_sync = bio_sync(r1_bio->master_bio); 1599cf30a473SNeilBrown r1_bio->bios[r1_bio->read_disk] = 1600cf30a473SNeilBrown mddev->ro ? IO_BLOCKED : NULL; 16011da177e4SLinus Torvalds r1_bio->read_disk = disk; 16021da177e4SLinus Torvalds bio_put(bio); 16031da177e4SLinus Torvalds bio = bio_clone(r1_bio->master_bio, GFP_NOIO); 16041da177e4SLinus Torvalds r1_bio->bios[r1_bio->read_disk] = bio; 16051da177e4SLinus Torvalds rdev = conf->mirrors[disk].rdev; 16061da177e4SLinus Torvalds if (printk_ratelimit()) 16071da177e4SLinus Torvalds printk(KERN_ERR "raid1: %s: redirecting sector %llu to" 16081da177e4SLinus Torvalds " another mirror\n", 16091da177e4SLinus Torvalds bdevname(rdev->bdev,b), 16101da177e4SLinus Torvalds (unsigned long long)r1_bio->sector); 16111da177e4SLinus Torvalds bio->bi_sector = r1_bio->sector + rdev->data_offset; 16121da177e4SLinus Torvalds bio->bi_bdev = rdev->bdev; 16131da177e4SLinus Torvalds bio->bi_end_io = raid1_end_read_request; 1614e3881a68SLars Ellenberg bio->bi_rw = READ | do_sync; 16151da177e4SLinus Torvalds bio->bi_private = r1_bio; 16161da177e4SLinus Torvalds unplug = 1; 16171da177e4SLinus Torvalds generic_make_request(bio); 16181da177e4SLinus Torvalds } 16191da177e4SLinus Torvalds } 16201da177e4SLinus Torvalds } 16211da177e4SLinus Torvalds if (unplug) 16221da177e4SLinus Torvalds unplug_slaves(mddev); 16231da177e4SLinus Torvalds } 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds static int init_resync(conf_t *conf) 16271da177e4SLinus Torvalds { 16281da177e4SLinus Torvalds int buffs; 16291da177e4SLinus Torvalds 16301da177e4SLinus Torvalds buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; 16319e77c485SEric Sesterhenn BUG_ON(conf->r1buf_pool); 16321da177e4SLinus Torvalds conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, 16331da177e4SLinus Torvalds conf->poolinfo); 16341da177e4SLinus Torvalds if (!conf->r1buf_pool) 16351da177e4SLinus Torvalds return -ENOMEM; 16361da177e4SLinus Torvalds conf->next_resync = 0; 16371da177e4SLinus Torvalds return 0; 16381da177e4SLinus Torvalds } 16391da177e4SLinus Torvalds 16401da177e4SLinus Torvalds /* 16411da177e4SLinus Torvalds * perform a "sync" on one "block" 16421da177e4SLinus Torvalds * 16431da177e4SLinus Torvalds * We need to make sure that no normal I/O request - particularly write 16441da177e4SLinus Torvalds * requests - conflict with active sync requests. 16451da177e4SLinus Torvalds * 16461da177e4SLinus Torvalds * This is achieved by tracking pending requests and a 'barrier' concept 16471da177e4SLinus Torvalds * that can be installed to exclude normal IO requests. 16481da177e4SLinus Torvalds */ 16491da177e4SLinus Torvalds 165057afd89fSNeilBrown static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) 16511da177e4SLinus Torvalds { 16521da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 16531da177e4SLinus Torvalds r1bio_t *r1_bio; 16541da177e4SLinus Torvalds struct bio *bio; 16551da177e4SLinus Torvalds sector_t max_sector, nr_sectors; 16563e198f78SNeilBrown int disk = -1; 16571da177e4SLinus Torvalds int i; 16583e198f78SNeilBrown int wonly = -1; 16593e198f78SNeilBrown int write_targets = 0, read_targets = 0; 1660191ea9b2SNeilBrown int sync_blocks; 1661e3b9703eSNeilBrown int still_degraded = 0; 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds if (!conf->r1buf_pool) 1664191ea9b2SNeilBrown { 1665191ea9b2SNeilBrown /* 1666191ea9b2SNeilBrown printk("sync start - bitmap %p\n", mddev->bitmap); 1667191ea9b2SNeilBrown */ 16681da177e4SLinus Torvalds if (init_resync(conf)) 166957afd89fSNeilBrown return 0; 1670191ea9b2SNeilBrown } 16711da177e4SLinus Torvalds 16721da177e4SLinus Torvalds max_sector = mddev->size << 1; 16731da177e4SLinus Torvalds if (sector_nr >= max_sector) { 1674191ea9b2SNeilBrown /* If we aborted, we need to abort the 1675191ea9b2SNeilBrown * sync on the 'current' bitmap chunk (there will 1676191ea9b2SNeilBrown * only be one in raid1 resync. 1677191ea9b2SNeilBrown * We can find the current addess in mddev->curr_resync 1678191ea9b2SNeilBrown */ 16796a806c51SNeilBrown if (mddev->curr_resync < max_sector) /* aborted */ 16806a806c51SNeilBrown bitmap_end_sync(mddev->bitmap, mddev->curr_resync, 1681191ea9b2SNeilBrown &sync_blocks, 1); 16826a806c51SNeilBrown else /* completed sync */ 1683191ea9b2SNeilBrown conf->fullsync = 0; 16846a806c51SNeilBrown 16856a806c51SNeilBrown bitmap_close_sync(mddev->bitmap); 16861da177e4SLinus Torvalds close_sync(conf); 16871da177e4SLinus Torvalds return 0; 16881da177e4SLinus Torvalds } 16891da177e4SLinus Torvalds 169007d84d10SNeilBrown if (mddev->bitmap == NULL && 169107d84d10SNeilBrown mddev->recovery_cp == MaxSector && 16926394cca5SNeilBrown !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && 169307d84d10SNeilBrown conf->fullsync == 0) { 169407d84d10SNeilBrown *skipped = 1; 169507d84d10SNeilBrown return max_sector - sector_nr; 169607d84d10SNeilBrown } 16976394cca5SNeilBrown /* before building a request, check if we can skip these blocks.. 16986394cca5SNeilBrown * This call the bitmap_start_sync doesn't actually record anything 16996394cca5SNeilBrown */ 1700e3b9703eSNeilBrown if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && 1701e5de485fSNeilBrown !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 1702191ea9b2SNeilBrown /* We can skip this block, and probably several more */ 1703191ea9b2SNeilBrown *skipped = 1; 1704191ea9b2SNeilBrown return sync_blocks; 1705191ea9b2SNeilBrown } 17061da177e4SLinus Torvalds /* 170717999be4SNeilBrown * If there is non-resync activity waiting for a turn, 170817999be4SNeilBrown * and resync is going fast enough, 170917999be4SNeilBrown * then let it though before starting on this new sync request. 17101da177e4SLinus Torvalds */ 171117999be4SNeilBrown if (!go_faster && conf->nr_waiting) 17121da177e4SLinus Torvalds msleep_interruptible(1000); 171317999be4SNeilBrown 1714b47490c9SNeilBrown bitmap_cond_end_sync(mddev->bitmap, sector_nr); 171517999be4SNeilBrown raise_barrier(conf); 171617999be4SNeilBrown 171717999be4SNeilBrown conf->next_resync = sector_nr; 17181da177e4SLinus Torvalds 17191da177e4SLinus Torvalds r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); 17203e198f78SNeilBrown rcu_read_lock(); 17213e198f78SNeilBrown /* 17223e198f78SNeilBrown * If we get a correctably read error during resync or recovery, 17233e198f78SNeilBrown * we might want to read from a different device. So we 17243e198f78SNeilBrown * flag all drives that could conceivably be read from for READ, 17253e198f78SNeilBrown * and any others (which will be non-In_sync devices) for WRITE. 17263e198f78SNeilBrown * If a read fails, we try reading from something else for which READ 17273e198f78SNeilBrown * is OK. 17283e198f78SNeilBrown */ 17291da177e4SLinus Torvalds 17301da177e4SLinus Torvalds r1_bio->mddev = mddev; 17311da177e4SLinus Torvalds r1_bio->sector = sector_nr; 1732191ea9b2SNeilBrown r1_bio->state = 0; 17331da177e4SLinus Torvalds set_bit(R1BIO_IsSync, &r1_bio->state); 17341da177e4SLinus Torvalds 17351da177e4SLinus Torvalds for (i=0; i < conf->raid_disks; i++) { 17363e198f78SNeilBrown mdk_rdev_t *rdev; 17371da177e4SLinus Torvalds bio = r1_bio->bios[i]; 17381da177e4SLinus Torvalds 17391da177e4SLinus Torvalds /* take from bio_init */ 17401da177e4SLinus Torvalds bio->bi_next = NULL; 17411da177e4SLinus Torvalds bio->bi_flags |= 1 << BIO_UPTODATE; 1742802ba064SNeilBrown bio->bi_rw = READ; 17431da177e4SLinus Torvalds bio->bi_vcnt = 0; 17441da177e4SLinus Torvalds bio->bi_idx = 0; 17451da177e4SLinus Torvalds bio->bi_phys_segments = 0; 17461da177e4SLinus Torvalds bio->bi_hw_segments = 0; 17471da177e4SLinus Torvalds bio->bi_size = 0; 17481da177e4SLinus Torvalds bio->bi_end_io = NULL; 17491da177e4SLinus Torvalds bio->bi_private = NULL; 17501da177e4SLinus Torvalds 17513e198f78SNeilBrown rdev = rcu_dereference(conf->mirrors[i].rdev); 17523e198f78SNeilBrown if (rdev == NULL || 17533e198f78SNeilBrown test_bit(Faulty, &rdev->flags)) { 1754e3b9703eSNeilBrown still_degraded = 1; 1755e3b9703eSNeilBrown continue; 17563e198f78SNeilBrown } else if (!test_bit(In_sync, &rdev->flags)) { 17571da177e4SLinus Torvalds bio->bi_rw = WRITE; 17581da177e4SLinus Torvalds bio->bi_end_io = end_sync_write; 17591da177e4SLinus Torvalds write_targets ++; 17603e198f78SNeilBrown } else { 17613e198f78SNeilBrown /* may need to read from here */ 17623e198f78SNeilBrown bio->bi_rw = READ; 17633e198f78SNeilBrown bio->bi_end_io = end_sync_read; 17643e198f78SNeilBrown if (test_bit(WriteMostly, &rdev->flags)) { 17653e198f78SNeilBrown if (wonly < 0) 17663e198f78SNeilBrown wonly = i; 17673e198f78SNeilBrown } else { 17683e198f78SNeilBrown if (disk < 0) 17693e198f78SNeilBrown disk = i; 17703e198f78SNeilBrown } 17713e198f78SNeilBrown read_targets++; 17723e198f78SNeilBrown } 17733e198f78SNeilBrown atomic_inc(&rdev->nr_pending); 17743e198f78SNeilBrown bio->bi_sector = sector_nr + rdev->data_offset; 17753e198f78SNeilBrown bio->bi_bdev = rdev->bdev; 17761da177e4SLinus Torvalds bio->bi_private = r1_bio; 17771da177e4SLinus Torvalds } 17783e198f78SNeilBrown rcu_read_unlock(); 17793e198f78SNeilBrown if (disk < 0) 17803e198f78SNeilBrown disk = wonly; 17813e198f78SNeilBrown r1_bio->read_disk = disk; 1782191ea9b2SNeilBrown 17833e198f78SNeilBrown if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) 17843e198f78SNeilBrown /* extra read targets are also write targets */ 17853e198f78SNeilBrown write_targets += read_targets-1; 17863e198f78SNeilBrown 17873e198f78SNeilBrown if (write_targets == 0 || read_targets == 0) { 17881da177e4SLinus Torvalds /* There is nowhere to write, so all non-sync 17891da177e4SLinus Torvalds * drives must be failed - so we are finished 17901da177e4SLinus Torvalds */ 179157afd89fSNeilBrown sector_t rv = max_sector - sector_nr; 179257afd89fSNeilBrown *skipped = 1; 17931da177e4SLinus Torvalds put_buf(r1_bio); 17941da177e4SLinus Torvalds return rv; 17951da177e4SLinus Torvalds } 17961da177e4SLinus Torvalds 1797c6207277SNeilBrown if (max_sector > mddev->resync_max) 1798c6207277SNeilBrown max_sector = mddev->resync_max; /* Don't do IO beyond here */ 17991da177e4SLinus Torvalds nr_sectors = 0; 1800289e99e8SNeilBrown sync_blocks = 0; 18011da177e4SLinus Torvalds do { 18021da177e4SLinus Torvalds struct page *page; 18031da177e4SLinus Torvalds int len = PAGE_SIZE; 18041da177e4SLinus Torvalds if (sector_nr + (len>>9) > max_sector) 18051da177e4SLinus Torvalds len = (max_sector - sector_nr) << 9; 18061da177e4SLinus Torvalds if (len == 0) 18071da177e4SLinus Torvalds break; 1808ab7a30c7SNeilBrown if (sync_blocks == 0) { 18096a806c51SNeilBrown if (!bitmap_start_sync(mddev->bitmap, sector_nr, 1810e3b9703eSNeilBrown &sync_blocks, still_degraded) && 1811e5de485fSNeilBrown !conf->fullsync && 1812e5de485fSNeilBrown !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 1813191ea9b2SNeilBrown break; 18149e77c485SEric Sesterhenn BUG_ON(sync_blocks < (PAGE_SIZE>>9)); 18156a806c51SNeilBrown if (len > (sync_blocks<<9)) 18166a806c51SNeilBrown len = sync_blocks<<9; 1817ab7a30c7SNeilBrown } 1818191ea9b2SNeilBrown 18191da177e4SLinus Torvalds for (i=0 ; i < conf->raid_disks; i++) { 18201da177e4SLinus Torvalds bio = r1_bio->bios[i]; 18211da177e4SLinus Torvalds if (bio->bi_end_io) { 1822d11c171eSNeilBrown page = bio->bi_io_vec[bio->bi_vcnt].bv_page; 18231da177e4SLinus Torvalds if (bio_add_page(bio, page, len, 0) == 0) { 18241da177e4SLinus Torvalds /* stop here */ 1825d11c171eSNeilBrown bio->bi_io_vec[bio->bi_vcnt].bv_page = page; 18261da177e4SLinus Torvalds while (i > 0) { 18271da177e4SLinus Torvalds i--; 18281da177e4SLinus Torvalds bio = r1_bio->bios[i]; 18296a806c51SNeilBrown if (bio->bi_end_io==NULL) 18306a806c51SNeilBrown continue; 18311da177e4SLinus Torvalds /* remove last page from this bio */ 18321da177e4SLinus Torvalds bio->bi_vcnt--; 18331da177e4SLinus Torvalds bio->bi_size -= len; 18341da177e4SLinus Torvalds bio->bi_flags &= ~(1<< BIO_SEG_VALID); 18351da177e4SLinus Torvalds } 18361da177e4SLinus Torvalds goto bio_full; 18371da177e4SLinus Torvalds } 18381da177e4SLinus Torvalds } 18391da177e4SLinus Torvalds } 18401da177e4SLinus Torvalds nr_sectors += len>>9; 18411da177e4SLinus Torvalds sector_nr += len>>9; 1842191ea9b2SNeilBrown sync_blocks -= (len>>9); 18431da177e4SLinus Torvalds } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); 18441da177e4SLinus Torvalds bio_full: 18451da177e4SLinus Torvalds r1_bio->sectors = nr_sectors; 18461da177e4SLinus Torvalds 1847d11c171eSNeilBrown /* For a user-requested sync, we read all readable devices and do a 1848d11c171eSNeilBrown * compare 1849d11c171eSNeilBrown */ 1850d11c171eSNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 1851d11c171eSNeilBrown atomic_set(&r1_bio->remaining, read_targets); 1852d11c171eSNeilBrown for (i=0; i<conf->raid_disks; i++) { 1853d11c171eSNeilBrown bio = r1_bio->bios[i]; 1854d11c171eSNeilBrown if (bio->bi_end_io == end_sync_read) { 1855ddac7c7eSNeilBrown md_sync_acct(bio->bi_bdev, nr_sectors); 18561da177e4SLinus Torvalds generic_make_request(bio); 1857d11c171eSNeilBrown } 1858d11c171eSNeilBrown } 1859d11c171eSNeilBrown } else { 1860d11c171eSNeilBrown atomic_set(&r1_bio->remaining, 1); 1861d11c171eSNeilBrown bio = r1_bio->bios[r1_bio->read_disk]; 1862ddac7c7eSNeilBrown md_sync_acct(bio->bi_bdev, nr_sectors); 1863d11c171eSNeilBrown generic_make_request(bio); 1864d11c171eSNeilBrown 1865d11c171eSNeilBrown } 18661da177e4SLinus Torvalds return nr_sectors; 18671da177e4SLinus Torvalds } 18681da177e4SLinus Torvalds 18691da177e4SLinus Torvalds static int run(mddev_t *mddev) 18701da177e4SLinus Torvalds { 18711da177e4SLinus Torvalds conf_t *conf; 18721da177e4SLinus Torvalds int i, j, disk_idx; 18731da177e4SLinus Torvalds mirror_info_t *disk; 18741da177e4SLinus Torvalds mdk_rdev_t *rdev; 18751da177e4SLinus Torvalds struct list_head *tmp; 18761da177e4SLinus Torvalds 18771da177e4SLinus Torvalds if (mddev->level != 1) { 18781da177e4SLinus Torvalds printk("raid1: %s: raid level not set to mirroring (%d)\n", 18791da177e4SLinus Torvalds mdname(mddev), mddev->level); 18801da177e4SLinus Torvalds goto out; 18811da177e4SLinus Torvalds } 1882f6705578SNeilBrown if (mddev->reshape_position != MaxSector) { 1883f6705578SNeilBrown printk("raid1: %s: reshape_position set but not supported\n", 1884f6705578SNeilBrown mdname(mddev)); 1885f6705578SNeilBrown goto out; 1886f6705578SNeilBrown } 18871da177e4SLinus Torvalds /* 18881da177e4SLinus Torvalds * copy the already verified devices into our private RAID1 18891da177e4SLinus Torvalds * bookkeeping area. [whatever we allocate in run(), 18901da177e4SLinus Torvalds * should be freed in stop()] 18911da177e4SLinus Torvalds */ 18929ffae0cfSNeilBrown conf = kzalloc(sizeof(conf_t), GFP_KERNEL); 18931da177e4SLinus Torvalds mddev->private = conf; 18941da177e4SLinus Torvalds if (!conf) 18951da177e4SLinus Torvalds goto out_no_mem; 18961da177e4SLinus Torvalds 18979ffae0cfSNeilBrown conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, 18981da177e4SLinus Torvalds GFP_KERNEL); 18991da177e4SLinus Torvalds if (!conf->mirrors) 19001da177e4SLinus Torvalds goto out_no_mem; 19011da177e4SLinus Torvalds 1902ddaf22abSNeilBrown conf->tmppage = alloc_page(GFP_KERNEL); 1903ddaf22abSNeilBrown if (!conf->tmppage) 1904ddaf22abSNeilBrown goto out_no_mem; 1905ddaf22abSNeilBrown 19061da177e4SLinus Torvalds conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL); 19071da177e4SLinus Torvalds if (!conf->poolinfo) 19081da177e4SLinus Torvalds goto out_no_mem; 19091da177e4SLinus Torvalds conf->poolinfo->mddev = mddev; 19101da177e4SLinus Torvalds conf->poolinfo->raid_disks = mddev->raid_disks; 19111da177e4SLinus Torvalds conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 19121da177e4SLinus Torvalds r1bio_pool_free, 19131da177e4SLinus Torvalds conf->poolinfo); 19141da177e4SLinus Torvalds if (!conf->r1bio_pool) 19151da177e4SLinus Torvalds goto out_no_mem; 19161da177e4SLinus Torvalds 1917d089c6afSNeilBrown rdev_for_each(rdev, tmp, mddev) { 19181da177e4SLinus Torvalds disk_idx = rdev->raid_disk; 19191da177e4SLinus Torvalds if (disk_idx >= mddev->raid_disks 19201da177e4SLinus Torvalds || disk_idx < 0) 19211da177e4SLinus Torvalds continue; 19221da177e4SLinus Torvalds disk = conf->mirrors + disk_idx; 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds disk->rdev = rdev; 19251da177e4SLinus Torvalds 19261da177e4SLinus Torvalds blk_queue_stack_limits(mddev->queue, 19271da177e4SLinus Torvalds rdev->bdev->bd_disk->queue); 19281da177e4SLinus Torvalds /* as we don't honour merge_bvec_fn, we must never risk 19291da177e4SLinus Torvalds * violating it, so limit ->max_sector to one PAGE, as 19301da177e4SLinus Torvalds * a one page request is never in violation. 19311da177e4SLinus Torvalds */ 19321da177e4SLinus Torvalds if (rdev->bdev->bd_disk->queue->merge_bvec_fn && 19331da177e4SLinus Torvalds mddev->queue->max_sectors > (PAGE_SIZE>>9)) 19341da177e4SLinus Torvalds blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); 19351da177e4SLinus Torvalds 19361da177e4SLinus Torvalds disk->head_position = 0; 19371da177e4SLinus Torvalds } 19381da177e4SLinus Torvalds conf->raid_disks = mddev->raid_disks; 19391da177e4SLinus Torvalds conf->mddev = mddev; 19401da177e4SLinus Torvalds spin_lock_init(&conf->device_lock); 19411da177e4SLinus Torvalds INIT_LIST_HEAD(&conf->retry_list); 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds spin_lock_init(&conf->resync_lock); 194417999be4SNeilBrown init_waitqueue_head(&conf->wait_barrier); 19451da177e4SLinus Torvalds 1946191ea9b2SNeilBrown bio_list_init(&conf->pending_bio_list); 1947191ea9b2SNeilBrown bio_list_init(&conf->flushing_bio_list); 1948191ea9b2SNeilBrown 19491da177e4SLinus Torvalds 19501da177e4SLinus Torvalds mddev->degraded = 0; 19511da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 19521da177e4SLinus Torvalds 19531da177e4SLinus Torvalds disk = conf->mirrors + i; 19541da177e4SLinus Torvalds 19555fd6c1dcSNeilBrown if (!disk->rdev || 19565fd6c1dcSNeilBrown !test_bit(In_sync, &disk->rdev->flags)) { 19571da177e4SLinus Torvalds disk->head_position = 0; 19581da177e4SLinus Torvalds mddev->degraded++; 1959918f0238SNeilBrown if (disk->rdev) 196017571284SNeilBrown conf->fullsync = 1; 19611da177e4SLinus Torvalds } 19621da177e4SLinus Torvalds } 196311ce99e6SNeilBrown if (mddev->degraded == conf->raid_disks) { 196411ce99e6SNeilBrown printk(KERN_ERR "raid1: no operational mirrors for %s\n", 196511ce99e6SNeilBrown mdname(mddev)); 196611ce99e6SNeilBrown goto out_free_conf; 196711ce99e6SNeilBrown } 196811ce99e6SNeilBrown if (conf->raid_disks - mddev->degraded == 1) 196911ce99e6SNeilBrown mddev->recovery_cp = MaxSector; 19701da177e4SLinus Torvalds 19711da177e4SLinus Torvalds /* 19721da177e4SLinus Torvalds * find the first working one and use it as a starting point 19731da177e4SLinus Torvalds * to read balancing. 19741da177e4SLinus Torvalds */ 19751da177e4SLinus Torvalds for (j = 0; j < conf->raid_disks && 19761da177e4SLinus Torvalds (!conf->mirrors[j].rdev || 1977b2d444d7SNeilBrown !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++) 19781da177e4SLinus Torvalds /* nothing */; 19791da177e4SLinus Torvalds conf->last_used = j; 19801da177e4SLinus Torvalds 19811da177e4SLinus Torvalds 19821da177e4SLinus Torvalds mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1"); 19831da177e4SLinus Torvalds if (!mddev->thread) { 19841da177e4SLinus Torvalds printk(KERN_ERR 19851da177e4SLinus Torvalds "raid1: couldn't allocate thread for %s\n", 19861da177e4SLinus Torvalds mdname(mddev)); 19871da177e4SLinus Torvalds goto out_free_conf; 19881da177e4SLinus Torvalds } 1989191ea9b2SNeilBrown 19901da177e4SLinus Torvalds printk(KERN_INFO 19911da177e4SLinus Torvalds "raid1: raid set %s active with %d out of %d mirrors\n", 19921da177e4SLinus Torvalds mdname(mddev), mddev->raid_disks - mddev->degraded, 19931da177e4SLinus Torvalds mddev->raid_disks); 19941da177e4SLinus Torvalds /* 19951da177e4SLinus Torvalds * Ok, everything is just fine now 19961da177e4SLinus Torvalds */ 19971da177e4SLinus Torvalds mddev->array_size = mddev->size; 19981da177e4SLinus Torvalds 19997a5febe9SNeilBrown mddev->queue->unplug_fn = raid1_unplug; 20000d129228SNeilBrown mddev->queue->backing_dev_info.congested_fn = raid1_congested; 20010d129228SNeilBrown mddev->queue->backing_dev_info.congested_data = mddev; 20027a5febe9SNeilBrown 20031da177e4SLinus Torvalds return 0; 20041da177e4SLinus Torvalds 20051da177e4SLinus Torvalds out_no_mem: 20061da177e4SLinus Torvalds printk(KERN_ERR "raid1: couldn't allocate memory for %s\n", 20071da177e4SLinus Torvalds mdname(mddev)); 20081da177e4SLinus Torvalds 20091da177e4SLinus Torvalds out_free_conf: 20101da177e4SLinus Torvalds if (conf) { 20111da177e4SLinus Torvalds if (conf->r1bio_pool) 20121da177e4SLinus Torvalds mempool_destroy(conf->r1bio_pool); 20131da177e4SLinus Torvalds kfree(conf->mirrors); 20141345b1d8SNeilBrown safe_put_page(conf->tmppage); 20151da177e4SLinus Torvalds kfree(conf->poolinfo); 20161da177e4SLinus Torvalds kfree(conf); 20171da177e4SLinus Torvalds mddev->private = NULL; 20181da177e4SLinus Torvalds } 20191da177e4SLinus Torvalds out: 20201da177e4SLinus Torvalds return -EIO; 20211da177e4SLinus Torvalds } 20221da177e4SLinus Torvalds 20231da177e4SLinus Torvalds static int stop(mddev_t *mddev) 20241da177e4SLinus Torvalds { 20251da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 20264b6d287fSNeilBrown struct bitmap *bitmap = mddev->bitmap; 20274b6d287fSNeilBrown int behind_wait = 0; 20284b6d287fSNeilBrown 20294b6d287fSNeilBrown /* wait for behind writes to complete */ 20304b6d287fSNeilBrown while (bitmap && atomic_read(&bitmap->behind_writes) > 0) { 20314b6d287fSNeilBrown behind_wait++; 20324b6d287fSNeilBrown printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait); 20334b6d287fSNeilBrown set_current_state(TASK_UNINTERRUPTIBLE); 20344b6d287fSNeilBrown schedule_timeout(HZ); /* wait a second */ 20354b6d287fSNeilBrown /* need to kick something here to make sure I/O goes? */ 20364b6d287fSNeilBrown } 20371da177e4SLinus Torvalds 20381da177e4SLinus Torvalds md_unregister_thread(mddev->thread); 20391da177e4SLinus Torvalds mddev->thread = NULL; 20401da177e4SLinus Torvalds blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ 20411da177e4SLinus Torvalds if (conf->r1bio_pool) 20421da177e4SLinus Torvalds mempool_destroy(conf->r1bio_pool); 20431da177e4SLinus Torvalds kfree(conf->mirrors); 20441da177e4SLinus Torvalds kfree(conf->poolinfo); 20451da177e4SLinus Torvalds kfree(conf); 20461da177e4SLinus Torvalds mddev->private = NULL; 20471da177e4SLinus Torvalds return 0; 20481da177e4SLinus Torvalds } 20491da177e4SLinus Torvalds 20501da177e4SLinus Torvalds static int raid1_resize(mddev_t *mddev, sector_t sectors) 20511da177e4SLinus Torvalds { 20521da177e4SLinus Torvalds /* no resync is happening, and there is enough space 20531da177e4SLinus Torvalds * on all devices, so we can resize. 20541da177e4SLinus Torvalds * We need to make sure resync covers any new space. 20551da177e4SLinus Torvalds * If the array is shrinking we should possibly wait until 20561da177e4SLinus Torvalds * any io in the removed space completes, but it hardly seems 20571da177e4SLinus Torvalds * worth it. 20581da177e4SLinus Torvalds */ 20591da177e4SLinus Torvalds mddev->array_size = sectors>>1; 20601da177e4SLinus Torvalds set_capacity(mddev->gendisk, mddev->array_size << 1); 206144ce6294SLinus Torvalds mddev->changed = 1; 20621da177e4SLinus Torvalds if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) { 20631da177e4SLinus Torvalds mddev->recovery_cp = mddev->size << 1; 20641da177e4SLinus Torvalds set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 20651da177e4SLinus Torvalds } 20661da177e4SLinus Torvalds mddev->size = mddev->array_size; 20674b5c7ae8SNeilBrown mddev->resync_max_sectors = sectors; 20681da177e4SLinus Torvalds return 0; 20691da177e4SLinus Torvalds } 20701da177e4SLinus Torvalds 207163c70c4fSNeilBrown static int raid1_reshape(mddev_t *mddev) 20721da177e4SLinus Torvalds { 20731da177e4SLinus Torvalds /* We need to: 20741da177e4SLinus Torvalds * 1/ resize the r1bio_pool 20751da177e4SLinus Torvalds * 2/ resize conf->mirrors 20761da177e4SLinus Torvalds * 20771da177e4SLinus Torvalds * We allocate a new r1bio_pool if we can. 20781da177e4SLinus Torvalds * Then raise a device barrier and wait until all IO stops. 20791da177e4SLinus Torvalds * Then resize conf->mirrors and swap in the new r1bio pool. 20806ea9c07cSNeilBrown * 20816ea9c07cSNeilBrown * At the same time, we "pack" the devices so that all the missing 20826ea9c07cSNeilBrown * devices have the higher raid_disk numbers. 20831da177e4SLinus Torvalds */ 20841da177e4SLinus Torvalds mempool_t *newpool, *oldpool; 20851da177e4SLinus Torvalds struct pool_info *newpoolinfo; 20861da177e4SLinus Torvalds mirror_info_t *newmirrors; 20871da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 208863c70c4fSNeilBrown int cnt, raid_disks; 2089c04be0aaSNeilBrown unsigned long flags; 20906ea9c07cSNeilBrown int d, d2; 20911da177e4SLinus Torvalds 209263c70c4fSNeilBrown /* Cannot change chunk_size, layout, or level */ 209363c70c4fSNeilBrown if (mddev->chunk_size != mddev->new_chunk || 209463c70c4fSNeilBrown mddev->layout != mddev->new_layout || 209563c70c4fSNeilBrown mddev->level != mddev->new_level) { 209663c70c4fSNeilBrown mddev->new_chunk = mddev->chunk_size; 209763c70c4fSNeilBrown mddev->new_layout = mddev->layout; 209863c70c4fSNeilBrown mddev->new_level = mddev->level; 209963c70c4fSNeilBrown return -EINVAL; 210063c70c4fSNeilBrown } 210163c70c4fSNeilBrown 21022a2275d6SNeilBrown md_allow_write(mddev); 21032a2275d6SNeilBrown 210463c70c4fSNeilBrown raid_disks = mddev->raid_disks + mddev->delta_disks; 210563c70c4fSNeilBrown 21066ea9c07cSNeilBrown if (raid_disks < conf->raid_disks) { 21076ea9c07cSNeilBrown cnt=0; 21086ea9c07cSNeilBrown for (d= 0; d < conf->raid_disks; d++) 21091da177e4SLinus Torvalds if (conf->mirrors[d].rdev) 21106ea9c07cSNeilBrown cnt++; 21116ea9c07cSNeilBrown if (cnt > raid_disks) 21121da177e4SLinus Torvalds return -EBUSY; 21136ea9c07cSNeilBrown } 21141da177e4SLinus Torvalds 21151da177e4SLinus Torvalds newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); 21161da177e4SLinus Torvalds if (!newpoolinfo) 21171da177e4SLinus Torvalds return -ENOMEM; 21181da177e4SLinus Torvalds newpoolinfo->mddev = mddev; 21191da177e4SLinus Torvalds newpoolinfo->raid_disks = raid_disks; 21201da177e4SLinus Torvalds 21211da177e4SLinus Torvalds newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 21221da177e4SLinus Torvalds r1bio_pool_free, newpoolinfo); 21231da177e4SLinus Torvalds if (!newpool) { 21241da177e4SLinus Torvalds kfree(newpoolinfo); 21251da177e4SLinus Torvalds return -ENOMEM; 21261da177e4SLinus Torvalds } 21279ffae0cfSNeilBrown newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL); 21281da177e4SLinus Torvalds if (!newmirrors) { 21291da177e4SLinus Torvalds kfree(newpoolinfo); 21301da177e4SLinus Torvalds mempool_destroy(newpool); 21311da177e4SLinus Torvalds return -ENOMEM; 21321da177e4SLinus Torvalds } 21331da177e4SLinus Torvalds 213417999be4SNeilBrown raise_barrier(conf); 21351da177e4SLinus Torvalds 21361da177e4SLinus Torvalds /* ok, everything is stopped */ 21371da177e4SLinus Torvalds oldpool = conf->r1bio_pool; 21381da177e4SLinus Torvalds conf->r1bio_pool = newpool; 21396ea9c07cSNeilBrown 2140a88aa786SNeilBrown for (d = d2 = 0; d < conf->raid_disks; d++) { 2141a88aa786SNeilBrown mdk_rdev_t *rdev = conf->mirrors[d].rdev; 2142a88aa786SNeilBrown if (rdev && rdev->raid_disk != d2) { 2143a88aa786SNeilBrown char nm[20]; 2144a88aa786SNeilBrown sprintf(nm, "rd%d", rdev->raid_disk); 2145a88aa786SNeilBrown sysfs_remove_link(&mddev->kobj, nm); 2146a88aa786SNeilBrown rdev->raid_disk = d2; 2147a88aa786SNeilBrown sprintf(nm, "rd%d", rdev->raid_disk); 2148a88aa786SNeilBrown sysfs_remove_link(&mddev->kobj, nm); 2149a88aa786SNeilBrown if (sysfs_create_link(&mddev->kobj, 2150a88aa786SNeilBrown &rdev->kobj, nm)) 2151a88aa786SNeilBrown printk(KERN_WARNING 2152a88aa786SNeilBrown "md/raid1: cannot register " 2153a88aa786SNeilBrown "%s for %s\n", 2154a88aa786SNeilBrown nm, mdname(mddev)); 2155a88aa786SNeilBrown } 2156a88aa786SNeilBrown if (rdev) 2157a88aa786SNeilBrown newmirrors[d2++].rdev = rdev; 21586ea9c07cSNeilBrown } 21591da177e4SLinus Torvalds kfree(conf->mirrors); 21601da177e4SLinus Torvalds conf->mirrors = newmirrors; 21611da177e4SLinus Torvalds kfree(conf->poolinfo); 21621da177e4SLinus Torvalds conf->poolinfo = newpoolinfo; 21631da177e4SLinus Torvalds 2164c04be0aaSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 21651da177e4SLinus Torvalds mddev->degraded += (raid_disks - conf->raid_disks); 2166c04be0aaSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 21671da177e4SLinus Torvalds conf->raid_disks = mddev->raid_disks = raid_disks; 216863c70c4fSNeilBrown mddev->delta_disks = 0; 21691da177e4SLinus Torvalds 21706ea9c07cSNeilBrown conf->last_used = 0; /* just make sure it is in-range */ 217117999be4SNeilBrown lower_barrier(conf); 21721da177e4SLinus Torvalds 21731da177e4SLinus Torvalds set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 21741da177e4SLinus Torvalds md_wakeup_thread(mddev->thread); 21751da177e4SLinus Torvalds 21761da177e4SLinus Torvalds mempool_destroy(oldpool); 21771da177e4SLinus Torvalds return 0; 21781da177e4SLinus Torvalds } 21791da177e4SLinus Torvalds 2180500af87aSNeilBrown static void raid1_quiesce(mddev_t *mddev, int state) 218136fa3063SNeilBrown { 218236fa3063SNeilBrown conf_t *conf = mddev_to_conf(mddev); 218336fa3063SNeilBrown 218436fa3063SNeilBrown switch(state) { 21859e6603daSNeilBrown case 1: 218617999be4SNeilBrown raise_barrier(conf); 218736fa3063SNeilBrown break; 21889e6603daSNeilBrown case 0: 218917999be4SNeilBrown lower_barrier(conf); 219036fa3063SNeilBrown break; 219136fa3063SNeilBrown } 219236fa3063SNeilBrown } 219336fa3063SNeilBrown 21941da177e4SLinus Torvalds 21952604b703SNeilBrown static struct mdk_personality raid1_personality = 21961da177e4SLinus Torvalds { 21971da177e4SLinus Torvalds .name = "raid1", 21982604b703SNeilBrown .level = 1, 21991da177e4SLinus Torvalds .owner = THIS_MODULE, 22001da177e4SLinus Torvalds .make_request = make_request, 22011da177e4SLinus Torvalds .run = run, 22021da177e4SLinus Torvalds .stop = stop, 22031da177e4SLinus Torvalds .status = status, 22041da177e4SLinus Torvalds .error_handler = error, 22051da177e4SLinus Torvalds .hot_add_disk = raid1_add_disk, 22061da177e4SLinus Torvalds .hot_remove_disk= raid1_remove_disk, 22071da177e4SLinus Torvalds .spare_active = raid1_spare_active, 22081da177e4SLinus Torvalds .sync_request = sync_request, 22091da177e4SLinus Torvalds .resize = raid1_resize, 221063c70c4fSNeilBrown .check_reshape = raid1_reshape, 221136fa3063SNeilBrown .quiesce = raid1_quiesce, 22121da177e4SLinus Torvalds }; 22131da177e4SLinus Torvalds 22141da177e4SLinus Torvalds static int __init raid_init(void) 22151da177e4SLinus Torvalds { 22162604b703SNeilBrown return register_md_personality(&raid1_personality); 22171da177e4SLinus Torvalds } 22181da177e4SLinus Torvalds 22191da177e4SLinus Torvalds static void raid_exit(void) 22201da177e4SLinus Torvalds { 22212604b703SNeilBrown unregister_md_personality(&raid1_personality); 22221da177e4SLinus Torvalds } 22231da177e4SLinus Torvalds 22241da177e4SLinus Torvalds module_init(raid_init); 22251da177e4SLinus Torvalds module_exit(raid_exit); 22261da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 22271da177e4SLinus Torvalds MODULE_ALIAS("md-personality-3"); /* RAID1 */ 2228d9d166c2SNeilBrown MODULE_ALIAS("md-raid1"); 22292604b703SNeilBrown MODULE_ALIAS("md-level-1"); 2230