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 * 121da177e4SLinus Torvalds * Fixes to reconstruction by Jakob �stergaard" <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 2411da177e4SLinus Torvalds bio_endio(bio, bio->bi_size, 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 2581da177e4SLinus Torvalds static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, 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 if (bio->bi_size) 2661da177e4SLinus Torvalds return 1; 2671da177e4SLinus Torvalds 2681da177e4SLinus Torvalds mirror = r1_bio->read_disk; 2691da177e4SLinus Torvalds /* 2701da177e4SLinus Torvalds * this branch is our 'one mirror IO has finished' event handler: 2711da177e4SLinus Torvalds */ 272ddaf22abSNeilBrown update_head_pos(mirror, r1_bio); 273ddaf22abSNeilBrown 274ddaf22abSNeilBrown if (uptodate || conf->working_disks <= 1) { 2751da177e4SLinus Torvalds /* 2761da177e4SLinus Torvalds * Set R1BIO_Uptodate in our master bio, so that 2771da177e4SLinus Torvalds * we will return a good error code for to the higher 2781da177e4SLinus Torvalds * levels even if IO on some other mirrored buffer fails. 2791da177e4SLinus Torvalds * 2801da177e4SLinus Torvalds * The 'master' represents the composite IO operation to 2811da177e4SLinus Torvalds * user-side. So if something waits for IO, then it will 2821da177e4SLinus Torvalds * wait for the 'master' bio. 2831da177e4SLinus Torvalds */ 284220946c9SNeilBrown if (uptodate) 2851da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 2861da177e4SLinus Torvalds 2871da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 288ddaf22abSNeilBrown } else { 2891da177e4SLinus Torvalds /* 2901da177e4SLinus Torvalds * oops, read error: 2911da177e4SLinus Torvalds */ 2921da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 2931da177e4SLinus Torvalds if (printk_ratelimit()) 2941da177e4SLinus Torvalds printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n", 2951da177e4SLinus Torvalds bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector); 2961da177e4SLinus Torvalds reschedule_retry(r1_bio); 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds 2991da177e4SLinus Torvalds rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); 3001da177e4SLinus Torvalds return 0; 3011da177e4SLinus Torvalds } 3021da177e4SLinus Torvalds 3031da177e4SLinus Torvalds static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, 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 if (bio->bi_size) 3121da177e4SLinus Torvalds return 1; 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds for (mirror = 0; mirror < conf->raid_disks; mirror++) 3151da177e4SLinus Torvalds if (r1_bio->bios[mirror] == bio) 3161da177e4SLinus Torvalds break; 3171da177e4SLinus Torvalds 318a9701a30SNeilBrown if (error == -ENOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) { 319a9701a30SNeilBrown set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags); 320a9701a30SNeilBrown set_bit(R1BIO_BarrierRetry, &r1_bio->state); 321a9701a30SNeilBrown r1_bio->mddev->barriers_work = 0; 322a9701a30SNeilBrown } else { 3231da177e4SLinus Torvalds /* 3241da177e4SLinus Torvalds * this branch is our 'one mirror IO has finished' event handler: 3251da177e4SLinus Torvalds */ 326a9701a30SNeilBrown r1_bio->bios[mirror] = NULL; 32704b857f7SNeilBrown to_put = bio; 328191ea9b2SNeilBrown if (!uptodate) { 3291da177e4SLinus Torvalds md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); 330191ea9b2SNeilBrown /* an I/O failed, we can't clear the bitmap */ 331191ea9b2SNeilBrown set_bit(R1BIO_Degraded, &r1_bio->state); 332191ea9b2SNeilBrown } else 3331da177e4SLinus Torvalds /* 3341da177e4SLinus Torvalds * Set R1BIO_Uptodate in our master bio, so that 3351da177e4SLinus Torvalds * we will return a good error code for to the higher 3361da177e4SLinus Torvalds * levels even if IO on some other mirrored buffer fails. 3371da177e4SLinus Torvalds * 3381da177e4SLinus Torvalds * The 'master' represents the composite IO operation to 3391da177e4SLinus Torvalds * user-side. So if something waits for IO, then it will 3401da177e4SLinus Torvalds * wait for the 'master' bio. 3411da177e4SLinus Torvalds */ 3421da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 3431da177e4SLinus Torvalds 3441da177e4SLinus Torvalds update_head_pos(mirror, r1_bio); 3451da177e4SLinus Torvalds 3464b6d287fSNeilBrown if (behind) { 3474b6d287fSNeilBrown if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) 3484b6d287fSNeilBrown atomic_dec(&r1_bio->behind_remaining); 3494b6d287fSNeilBrown 3504b6d287fSNeilBrown /* In behind mode, we ACK the master bio once the I/O has safely 3514b6d287fSNeilBrown * reached all non-writemostly disks. Setting the Returned bit 3524b6d287fSNeilBrown * ensures that this gets done only once -- we don't ever want to 3534b6d287fSNeilBrown * return -EIO here, instead we'll wait */ 3544b6d287fSNeilBrown 3554b6d287fSNeilBrown if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && 3564b6d287fSNeilBrown test_bit(R1BIO_Uptodate, &r1_bio->state)) { 3574b6d287fSNeilBrown /* Maybe we can return now */ 3584b6d287fSNeilBrown if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { 3594b6d287fSNeilBrown struct bio *mbio = r1_bio->master_bio; 3604b6d287fSNeilBrown PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n", 3614b6d287fSNeilBrown (unsigned long long) mbio->bi_sector, 3624b6d287fSNeilBrown (unsigned long long) mbio->bi_sector + 3634b6d287fSNeilBrown (mbio->bi_size >> 9) - 1); 3644b6d287fSNeilBrown bio_endio(mbio, mbio->bi_size, 0); 3654b6d287fSNeilBrown } 3664b6d287fSNeilBrown } 3674b6d287fSNeilBrown } 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)) { 375a9701a30SNeilBrown if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) { 376a9701a30SNeilBrown reschedule_retry(r1_bio); 377a9701a30SNeilBrown /* Don't dec_pending yet, we want to hold 378a9701a30SNeilBrown * the reference over the retry 379a9701a30SNeilBrown */ 38004b857f7SNeilBrown goto out; 381a9701a30SNeilBrown } 3824b6d287fSNeilBrown if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { 3834b6d287fSNeilBrown /* free extra copy of the data pages */ 3844b6d287fSNeilBrown int i = bio->bi_vcnt; 3854b6d287fSNeilBrown while (i--) 3861345b1d8SNeilBrown safe_put_page(bio->bi_io_vec[i].bv_page); 3874b6d287fSNeilBrown } 388191ea9b2SNeilBrown /* clear the bitmap if all writes complete successfully */ 389191ea9b2SNeilBrown bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, 390191ea9b2SNeilBrown r1_bio->sectors, 3914b6d287fSNeilBrown !test_bit(R1BIO_Degraded, &r1_bio->state), 3924b6d287fSNeilBrown behind); 3931da177e4SLinus Torvalds md_write_end(r1_bio->mddev); 3941da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds 3971da177e4SLinus Torvalds rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); 39804b857f7SNeilBrown out: 39904b857f7SNeilBrown if (to_put) 40004b857f7SNeilBrown bio_put(to_put); 40104b857f7SNeilBrown 4021da177e4SLinus Torvalds return 0; 4031da177e4SLinus Torvalds } 4041da177e4SLinus Torvalds 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds /* 4071da177e4SLinus Torvalds * This routine returns the disk from which the requested read should 4081da177e4SLinus Torvalds * be done. There is a per-array 'next expected sequential IO' sector 4091da177e4SLinus Torvalds * number - if this matches on the next IO then we use the last disk. 4101da177e4SLinus Torvalds * There is also a per-disk 'last know head position' sector that is 4111da177e4SLinus Torvalds * maintained from IRQ contexts, both the normal and the resync IO 4121da177e4SLinus Torvalds * completion handlers update this position correctly. If there is no 4131da177e4SLinus Torvalds * perfect sequential match then we pick the disk whose head is closest. 4141da177e4SLinus Torvalds * 4151da177e4SLinus Torvalds * If there are 2 mirrors in the same 2 devices, performance degrades 4161da177e4SLinus Torvalds * because position is mirror, not device based. 4171da177e4SLinus Torvalds * 4181da177e4SLinus Torvalds * The rdev for the device selected will have nr_pending incremented. 4191da177e4SLinus Torvalds */ 4201da177e4SLinus Torvalds static int read_balance(conf_t *conf, r1bio_t *r1_bio) 4211da177e4SLinus Torvalds { 4221da177e4SLinus Torvalds const unsigned long this_sector = r1_bio->sector; 4231da177e4SLinus Torvalds int new_disk = conf->last_used, disk = new_disk; 4248ddf9efeSNeilBrown int wonly_disk = -1; 4251da177e4SLinus Torvalds const int sectors = r1_bio->sectors; 4261da177e4SLinus Torvalds sector_t new_distance, current_distance; 4278ddf9efeSNeilBrown mdk_rdev_t *rdev; 4281da177e4SLinus Torvalds 4291da177e4SLinus Torvalds rcu_read_lock(); 4301da177e4SLinus Torvalds /* 4318ddf9efeSNeilBrown * Check if we can balance. We can balance on the whole 4321da177e4SLinus Torvalds * device if no resync is going on, or below the resync window. 4331da177e4SLinus Torvalds * We take the first readable disk when above the resync window. 4341da177e4SLinus Torvalds */ 4351da177e4SLinus Torvalds retry: 4361da177e4SLinus Torvalds if (conf->mddev->recovery_cp < MaxSector && 4371da177e4SLinus Torvalds (this_sector + sectors >= conf->next_resync)) { 4381da177e4SLinus Torvalds /* Choose the first operation device, for consistancy */ 4391da177e4SLinus Torvalds new_disk = 0; 4401da177e4SLinus Torvalds 441d6065f7bSSuzanne Wood for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); 442cf30a473SNeilBrown r1_bio->bios[new_disk] == IO_BLOCKED || 443b2d444d7SNeilBrown !rdev || !test_bit(In_sync, &rdev->flags) 4448ddf9efeSNeilBrown || test_bit(WriteMostly, &rdev->flags); 445d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) { 4468ddf9efeSNeilBrown 447cf30a473SNeilBrown if (rdev && test_bit(In_sync, &rdev->flags) && 448cf30a473SNeilBrown r1_bio->bios[new_disk] != IO_BLOCKED) 4498ddf9efeSNeilBrown wonly_disk = new_disk; 4508ddf9efeSNeilBrown 4518ddf9efeSNeilBrown if (new_disk == conf->raid_disks - 1) { 4528ddf9efeSNeilBrown new_disk = wonly_disk; 4531da177e4SLinus Torvalds break; 4541da177e4SLinus Torvalds } 4551da177e4SLinus Torvalds } 4561da177e4SLinus Torvalds goto rb_out; 4571da177e4SLinus Torvalds } 4581da177e4SLinus Torvalds 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds /* make sure the disk is operational */ 461d6065f7bSSuzanne Wood for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev); 462cf30a473SNeilBrown r1_bio->bios[new_disk] == IO_BLOCKED || 463b2d444d7SNeilBrown !rdev || !test_bit(In_sync, &rdev->flags) || 4648ddf9efeSNeilBrown test_bit(WriteMostly, &rdev->flags); 465d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) { 4668ddf9efeSNeilBrown 467cf30a473SNeilBrown if (rdev && test_bit(In_sync, &rdev->flags) && 468cf30a473SNeilBrown r1_bio->bios[new_disk] != IO_BLOCKED) 4698ddf9efeSNeilBrown wonly_disk = new_disk; 4708ddf9efeSNeilBrown 4711da177e4SLinus Torvalds if (new_disk <= 0) 4721da177e4SLinus Torvalds new_disk = conf->raid_disks; 4731da177e4SLinus Torvalds new_disk--; 4741da177e4SLinus Torvalds if (new_disk == disk) { 4758ddf9efeSNeilBrown new_disk = wonly_disk; 4768ddf9efeSNeilBrown break; 4778ddf9efeSNeilBrown } 4788ddf9efeSNeilBrown } 4798ddf9efeSNeilBrown 4808ddf9efeSNeilBrown if (new_disk < 0) 4811da177e4SLinus Torvalds goto rb_out; 4828ddf9efeSNeilBrown 4831da177e4SLinus Torvalds disk = new_disk; 4841da177e4SLinus Torvalds /* now disk == new_disk == starting point for search */ 4851da177e4SLinus Torvalds 4861da177e4SLinus Torvalds /* 4871da177e4SLinus Torvalds * Don't change to another disk for sequential reads: 4881da177e4SLinus Torvalds */ 4891da177e4SLinus Torvalds if (conf->next_seq_sect == this_sector) 4901da177e4SLinus Torvalds goto rb_out; 4911da177e4SLinus Torvalds if (this_sector == conf->mirrors[new_disk].head_position) 4921da177e4SLinus Torvalds goto rb_out; 4931da177e4SLinus Torvalds 4941da177e4SLinus Torvalds current_distance = abs(this_sector - conf->mirrors[disk].head_position); 4951da177e4SLinus Torvalds 4961da177e4SLinus Torvalds /* Find the disk whose head is closest */ 4971da177e4SLinus Torvalds 4981da177e4SLinus Torvalds do { 4991da177e4SLinus Torvalds if (disk <= 0) 5001da177e4SLinus Torvalds disk = conf->raid_disks; 5011da177e4SLinus Torvalds disk--; 5021da177e4SLinus Torvalds 503d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[disk].rdev); 5048ddf9efeSNeilBrown 505cf30a473SNeilBrown if (!rdev || r1_bio->bios[disk] == IO_BLOCKED || 506b2d444d7SNeilBrown !test_bit(In_sync, &rdev->flags) || 5078ddf9efeSNeilBrown test_bit(WriteMostly, &rdev->flags)) 5081da177e4SLinus Torvalds continue; 5091da177e4SLinus Torvalds 5101da177e4SLinus Torvalds if (!atomic_read(&rdev->nr_pending)) { 5111da177e4SLinus Torvalds new_disk = disk; 5121da177e4SLinus Torvalds break; 5131da177e4SLinus Torvalds } 5141da177e4SLinus Torvalds new_distance = abs(this_sector - conf->mirrors[disk].head_position); 5151da177e4SLinus Torvalds if (new_distance < current_distance) { 5161da177e4SLinus Torvalds current_distance = new_distance; 5171da177e4SLinus Torvalds new_disk = disk; 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds } while (disk != conf->last_used); 5201da177e4SLinus Torvalds 5211da177e4SLinus Torvalds rb_out: 5221da177e4SLinus Torvalds 5231da177e4SLinus Torvalds 5241da177e4SLinus Torvalds if (new_disk >= 0) { 525d6065f7bSSuzanne Wood rdev = rcu_dereference(conf->mirrors[new_disk].rdev); 5268ddf9efeSNeilBrown if (!rdev) 5278ddf9efeSNeilBrown goto retry; 5288ddf9efeSNeilBrown atomic_inc(&rdev->nr_pending); 529b2d444d7SNeilBrown if (!test_bit(In_sync, &rdev->flags)) { 5301da177e4SLinus Torvalds /* cannot risk returning a device that failed 5311da177e4SLinus Torvalds * before we inc'ed nr_pending 5321da177e4SLinus Torvalds */ 53303c902e1SNeilBrown rdev_dec_pending(rdev, conf->mddev); 5341da177e4SLinus Torvalds goto retry; 5351da177e4SLinus Torvalds } 5368ddf9efeSNeilBrown conf->next_seq_sect = this_sector + sectors; 5378ddf9efeSNeilBrown conf->last_used = new_disk; 5381da177e4SLinus Torvalds } 5391da177e4SLinus Torvalds rcu_read_unlock(); 5401da177e4SLinus Torvalds 5411da177e4SLinus Torvalds return new_disk; 5421da177e4SLinus Torvalds } 5431da177e4SLinus Torvalds 5441da177e4SLinus Torvalds static void unplug_slaves(mddev_t *mddev) 5451da177e4SLinus Torvalds { 5461da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 5471da177e4SLinus Torvalds int i; 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds rcu_read_lock(); 5501da177e4SLinus Torvalds for (i=0; i<mddev->raid_disks; i++) { 551d6065f7bSSuzanne Wood mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); 552b2d444d7SNeilBrown if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { 5531da177e4SLinus Torvalds request_queue_t *r_queue = bdev_get_queue(rdev->bdev); 5541da177e4SLinus Torvalds 5551da177e4SLinus Torvalds atomic_inc(&rdev->nr_pending); 5561da177e4SLinus Torvalds rcu_read_unlock(); 5571da177e4SLinus Torvalds 5581da177e4SLinus Torvalds if (r_queue->unplug_fn) 5591da177e4SLinus Torvalds r_queue->unplug_fn(r_queue); 5601da177e4SLinus Torvalds 5611da177e4SLinus Torvalds rdev_dec_pending(rdev, mddev); 5621da177e4SLinus Torvalds rcu_read_lock(); 5631da177e4SLinus Torvalds } 5641da177e4SLinus Torvalds } 5651da177e4SLinus Torvalds rcu_read_unlock(); 5661da177e4SLinus Torvalds } 5671da177e4SLinus Torvalds 5681da177e4SLinus Torvalds static void raid1_unplug(request_queue_t *q) 5691da177e4SLinus Torvalds { 570191ea9b2SNeilBrown mddev_t *mddev = q->queuedata; 571191ea9b2SNeilBrown 572191ea9b2SNeilBrown unplug_slaves(mddev); 573191ea9b2SNeilBrown md_wakeup_thread(mddev->thread); 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds 5761da177e4SLinus Torvalds static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk, 5771da177e4SLinus Torvalds sector_t *error_sector) 5781da177e4SLinus Torvalds { 5791da177e4SLinus Torvalds mddev_t *mddev = q->queuedata; 5801da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 5811da177e4SLinus Torvalds int i, ret = 0; 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds rcu_read_lock(); 5841da177e4SLinus Torvalds for (i=0; i<mddev->raid_disks && ret == 0; i++) { 585d6065f7bSSuzanne Wood mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev); 586b2d444d7SNeilBrown if (rdev && !test_bit(Faulty, &rdev->flags)) { 5871da177e4SLinus Torvalds struct block_device *bdev = rdev->bdev; 5881da177e4SLinus Torvalds request_queue_t *r_queue = bdev_get_queue(bdev); 5891da177e4SLinus Torvalds 5901da177e4SLinus Torvalds if (!r_queue->issue_flush_fn) 5911da177e4SLinus Torvalds ret = -EOPNOTSUPP; 5921da177e4SLinus Torvalds else { 5931da177e4SLinus Torvalds atomic_inc(&rdev->nr_pending); 5941da177e4SLinus Torvalds rcu_read_unlock(); 5951da177e4SLinus Torvalds ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, 5961da177e4SLinus Torvalds error_sector); 5971da177e4SLinus Torvalds rdev_dec_pending(rdev, mddev); 5981da177e4SLinus Torvalds rcu_read_lock(); 5991da177e4SLinus Torvalds } 6001da177e4SLinus Torvalds } 6011da177e4SLinus Torvalds } 6021da177e4SLinus Torvalds rcu_read_unlock(); 6031da177e4SLinus Torvalds return ret; 6041da177e4SLinus Torvalds } 6051da177e4SLinus Torvalds 60617999be4SNeilBrown /* Barriers.... 60717999be4SNeilBrown * Sometimes we need to suspend IO while we do something else, 60817999be4SNeilBrown * either some resync/recovery, or reconfigure the array. 60917999be4SNeilBrown * To do this we raise a 'barrier'. 61017999be4SNeilBrown * The 'barrier' is a counter that can be raised multiple times 61117999be4SNeilBrown * to count how many activities are happening which preclude 61217999be4SNeilBrown * normal IO. 61317999be4SNeilBrown * We can only raise the barrier if there is no pending IO. 61417999be4SNeilBrown * i.e. if nr_pending == 0. 61517999be4SNeilBrown * We choose only to raise the barrier if no-one is waiting for the 61617999be4SNeilBrown * barrier to go down. This means that as soon as an IO request 61717999be4SNeilBrown * is ready, no other operations which require a barrier will start 61817999be4SNeilBrown * until the IO request has had a chance. 61917999be4SNeilBrown * 62017999be4SNeilBrown * So: regular IO calls 'wait_barrier'. When that returns there 62117999be4SNeilBrown * is no backgroup IO happening, It must arrange to call 62217999be4SNeilBrown * allow_barrier when it has finished its IO. 62317999be4SNeilBrown * backgroup IO calls must call raise_barrier. Once that returns 62417999be4SNeilBrown * there is no normal IO happeing. It must arrange to call 62517999be4SNeilBrown * lower_barrier when the particular background IO completes. 6261da177e4SLinus Torvalds */ 6271da177e4SLinus Torvalds #define RESYNC_DEPTH 32 6281da177e4SLinus Torvalds 62917999be4SNeilBrown static void raise_barrier(conf_t *conf) 6301da177e4SLinus Torvalds { 6311da177e4SLinus Torvalds spin_lock_irq(&conf->resync_lock); 6321da177e4SLinus Torvalds 63317999be4SNeilBrown /* Wait until no block IO is waiting */ 63417999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, 63517999be4SNeilBrown conf->resync_lock, 63617999be4SNeilBrown raid1_unplug(conf->mddev->queue)); 63717999be4SNeilBrown 63817999be4SNeilBrown /* block any new IO from starting */ 63917999be4SNeilBrown conf->barrier++; 64017999be4SNeilBrown 64117999be4SNeilBrown /* No wait for all pending IO to complete */ 64217999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, 64317999be4SNeilBrown !conf->nr_pending && conf->barrier < RESYNC_DEPTH, 64417999be4SNeilBrown conf->resync_lock, 64517999be4SNeilBrown raid1_unplug(conf->mddev->queue)); 64617999be4SNeilBrown 6471da177e4SLinus Torvalds spin_unlock_irq(&conf->resync_lock); 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds 65017999be4SNeilBrown static void lower_barrier(conf_t *conf) 65117999be4SNeilBrown { 65217999be4SNeilBrown unsigned long flags; 65317999be4SNeilBrown spin_lock_irqsave(&conf->resync_lock, flags); 65417999be4SNeilBrown conf->barrier--; 65517999be4SNeilBrown spin_unlock_irqrestore(&conf->resync_lock, flags); 65617999be4SNeilBrown wake_up(&conf->wait_barrier); 65717999be4SNeilBrown } 65817999be4SNeilBrown 65917999be4SNeilBrown static void wait_barrier(conf_t *conf) 66017999be4SNeilBrown { 66117999be4SNeilBrown spin_lock_irq(&conf->resync_lock); 66217999be4SNeilBrown if (conf->barrier) { 66317999be4SNeilBrown conf->nr_waiting++; 66417999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, !conf->barrier, 66517999be4SNeilBrown conf->resync_lock, 66617999be4SNeilBrown raid1_unplug(conf->mddev->queue)); 66717999be4SNeilBrown conf->nr_waiting--; 66817999be4SNeilBrown } 66917999be4SNeilBrown conf->nr_pending++; 67017999be4SNeilBrown spin_unlock_irq(&conf->resync_lock); 67117999be4SNeilBrown } 67217999be4SNeilBrown 67317999be4SNeilBrown static void allow_barrier(conf_t *conf) 67417999be4SNeilBrown { 67517999be4SNeilBrown unsigned long flags; 67617999be4SNeilBrown spin_lock_irqsave(&conf->resync_lock, flags); 67717999be4SNeilBrown conf->nr_pending--; 67817999be4SNeilBrown spin_unlock_irqrestore(&conf->resync_lock, flags); 67917999be4SNeilBrown wake_up(&conf->wait_barrier); 68017999be4SNeilBrown } 68117999be4SNeilBrown 682ddaf22abSNeilBrown static void freeze_array(conf_t *conf) 683ddaf22abSNeilBrown { 684ddaf22abSNeilBrown /* stop syncio and normal IO and wait for everything to 685ddaf22abSNeilBrown * go quite. 686ddaf22abSNeilBrown * We increment barrier and nr_waiting, and then 687ddaf22abSNeilBrown * wait until barrier+nr_pending match nr_queued+2 688ddaf22abSNeilBrown */ 689ddaf22abSNeilBrown spin_lock_irq(&conf->resync_lock); 690ddaf22abSNeilBrown conf->barrier++; 691ddaf22abSNeilBrown conf->nr_waiting++; 692ddaf22abSNeilBrown wait_event_lock_irq(conf->wait_barrier, 693ddaf22abSNeilBrown conf->barrier+conf->nr_pending == conf->nr_queued+2, 694ddaf22abSNeilBrown conf->resync_lock, 695ddaf22abSNeilBrown raid1_unplug(conf->mddev->queue)); 696ddaf22abSNeilBrown spin_unlock_irq(&conf->resync_lock); 697ddaf22abSNeilBrown } 698ddaf22abSNeilBrown static void unfreeze_array(conf_t *conf) 699ddaf22abSNeilBrown { 700ddaf22abSNeilBrown /* reverse the effect of the freeze */ 701ddaf22abSNeilBrown spin_lock_irq(&conf->resync_lock); 702ddaf22abSNeilBrown conf->barrier--; 703ddaf22abSNeilBrown conf->nr_waiting--; 704ddaf22abSNeilBrown wake_up(&conf->wait_barrier); 705ddaf22abSNeilBrown spin_unlock_irq(&conf->resync_lock); 706ddaf22abSNeilBrown } 707ddaf22abSNeilBrown 70817999be4SNeilBrown 7094b6d287fSNeilBrown /* duplicate the data pages for behind I/O */ 7104b6d287fSNeilBrown static struct page **alloc_behind_pages(struct bio *bio) 7114b6d287fSNeilBrown { 7124b6d287fSNeilBrown int i; 7134b6d287fSNeilBrown struct bio_vec *bvec; 7149ffae0cfSNeilBrown struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *), 7154b6d287fSNeilBrown GFP_NOIO); 7164b6d287fSNeilBrown if (unlikely(!pages)) 7174b6d287fSNeilBrown goto do_sync_io; 7184b6d287fSNeilBrown 7194b6d287fSNeilBrown bio_for_each_segment(bvec, bio, i) { 7204b6d287fSNeilBrown pages[i] = alloc_page(GFP_NOIO); 7214b6d287fSNeilBrown if (unlikely(!pages[i])) 7224b6d287fSNeilBrown goto do_sync_io; 7234b6d287fSNeilBrown memcpy(kmap(pages[i]) + bvec->bv_offset, 7244b6d287fSNeilBrown kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); 7254b6d287fSNeilBrown kunmap(pages[i]); 7264b6d287fSNeilBrown kunmap(bvec->bv_page); 7274b6d287fSNeilBrown } 7284b6d287fSNeilBrown 7294b6d287fSNeilBrown return pages; 7304b6d287fSNeilBrown 7314b6d287fSNeilBrown do_sync_io: 7324b6d287fSNeilBrown if (pages) 7334b6d287fSNeilBrown for (i = 0; i < bio->bi_vcnt && pages[i]; i++) 7342d1f3b5dSNeilBrown put_page(pages[i]); 7354b6d287fSNeilBrown kfree(pages); 7364b6d287fSNeilBrown PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); 7374b6d287fSNeilBrown return NULL; 7384b6d287fSNeilBrown } 7394b6d287fSNeilBrown 7401da177e4SLinus Torvalds static int make_request(request_queue_t *q, struct bio * bio) 7411da177e4SLinus Torvalds { 7421da177e4SLinus Torvalds mddev_t *mddev = q->queuedata; 7431da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 7441da177e4SLinus Torvalds mirror_info_t *mirror; 7451da177e4SLinus Torvalds r1bio_t *r1_bio; 7461da177e4SLinus Torvalds struct bio *read_bio; 747191ea9b2SNeilBrown int i, targets = 0, disks; 7481da177e4SLinus Torvalds mdk_rdev_t *rdev; 749191ea9b2SNeilBrown struct bitmap *bitmap = mddev->bitmap; 750191ea9b2SNeilBrown unsigned long flags; 751191ea9b2SNeilBrown struct bio_list bl; 7524b6d287fSNeilBrown struct page **behind_pages = NULL; 753a362357bSJens Axboe const int rw = bio_data_dir(bio); 754a9701a30SNeilBrown int do_barriers; 755191ea9b2SNeilBrown 756a9701a30SNeilBrown if (unlikely(!mddev->barriers_work && bio_barrier(bio))) { 757e5dcdd80SNeilBrown bio_endio(bio, bio->bi_size, -EOPNOTSUPP); 758e5dcdd80SNeilBrown return 0; 759e5dcdd80SNeilBrown } 7601da177e4SLinus Torvalds 7611da177e4SLinus Torvalds /* 7621da177e4SLinus Torvalds * Register the new request and wait if the reconstruction 7631da177e4SLinus Torvalds * thread has put up a bar for new requests. 7641da177e4SLinus Torvalds * Continue immediately if no resync is active currently. 7651da177e4SLinus Torvalds */ 7663d310eb7SNeilBrown md_write_start(mddev, bio); /* wait on superblock update early */ 7673d310eb7SNeilBrown 76817999be4SNeilBrown wait_barrier(conf); 7691da177e4SLinus Torvalds 770a362357bSJens Axboe disk_stat_inc(mddev->gendisk, ios[rw]); 771a362357bSJens Axboe disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio)); 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds /* 7741da177e4SLinus Torvalds * make_request() can abort the operation when READA is being 7751da177e4SLinus Torvalds * used and no empty request is available. 7761da177e4SLinus Torvalds * 7771da177e4SLinus Torvalds */ 7781da177e4SLinus Torvalds r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); 7791da177e4SLinus Torvalds 7801da177e4SLinus Torvalds r1_bio->master_bio = bio; 7811da177e4SLinus Torvalds r1_bio->sectors = bio->bi_size >> 9; 782191ea9b2SNeilBrown r1_bio->state = 0; 7831da177e4SLinus Torvalds r1_bio->mddev = mddev; 7841da177e4SLinus Torvalds r1_bio->sector = bio->bi_sector; 7851da177e4SLinus Torvalds 786a362357bSJens Axboe if (rw == READ) { 7871da177e4SLinus Torvalds /* 7881da177e4SLinus Torvalds * read balancing logic: 7891da177e4SLinus Torvalds */ 7901da177e4SLinus Torvalds int rdisk = read_balance(conf, r1_bio); 7911da177e4SLinus Torvalds 7921da177e4SLinus Torvalds if (rdisk < 0) { 7931da177e4SLinus Torvalds /* couldn't find anywhere to read from */ 7941da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 7951da177e4SLinus Torvalds return 0; 7961da177e4SLinus Torvalds } 7971da177e4SLinus Torvalds mirror = conf->mirrors + rdisk; 7981da177e4SLinus Torvalds 7991da177e4SLinus Torvalds r1_bio->read_disk = rdisk; 8001da177e4SLinus Torvalds 8011da177e4SLinus Torvalds read_bio = bio_clone(bio, GFP_NOIO); 8021da177e4SLinus Torvalds 8031da177e4SLinus Torvalds r1_bio->bios[rdisk] = read_bio; 8041da177e4SLinus Torvalds 8051da177e4SLinus Torvalds read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; 8061da177e4SLinus Torvalds read_bio->bi_bdev = mirror->rdev->bdev; 8071da177e4SLinus Torvalds read_bio->bi_end_io = raid1_end_read_request; 8081da177e4SLinus Torvalds read_bio->bi_rw = READ; 8091da177e4SLinus Torvalds read_bio->bi_private = r1_bio; 8101da177e4SLinus Torvalds 8111da177e4SLinus Torvalds generic_make_request(read_bio); 8121da177e4SLinus Torvalds return 0; 8131da177e4SLinus Torvalds } 8141da177e4SLinus Torvalds 8151da177e4SLinus Torvalds /* 8161da177e4SLinus Torvalds * WRITE: 8171da177e4SLinus Torvalds */ 8181da177e4SLinus Torvalds /* first select target devices under spinlock and 8191da177e4SLinus Torvalds * inc refcount on their rdev. Record them by setting 8201da177e4SLinus Torvalds * bios[x] to bio 8211da177e4SLinus Torvalds */ 8221da177e4SLinus Torvalds disks = conf->raid_disks; 823191ea9b2SNeilBrown #if 0 824191ea9b2SNeilBrown { static int first=1; 825191ea9b2SNeilBrown if (first) printk("First Write sector %llu disks %d\n", 826191ea9b2SNeilBrown (unsigned long long)r1_bio->sector, disks); 827191ea9b2SNeilBrown first = 0; 828191ea9b2SNeilBrown } 829191ea9b2SNeilBrown #endif 8301da177e4SLinus Torvalds rcu_read_lock(); 8311da177e4SLinus Torvalds for (i = 0; i < disks; i++) { 832d6065f7bSSuzanne Wood if ((rdev=rcu_dereference(conf->mirrors[i].rdev)) != NULL && 833b2d444d7SNeilBrown !test_bit(Faulty, &rdev->flags)) { 8341da177e4SLinus Torvalds atomic_inc(&rdev->nr_pending); 835b2d444d7SNeilBrown if (test_bit(Faulty, &rdev->flags)) { 83603c902e1SNeilBrown rdev_dec_pending(rdev, mddev); 8371da177e4SLinus Torvalds r1_bio->bios[i] = NULL; 8381da177e4SLinus Torvalds } else 8391da177e4SLinus Torvalds r1_bio->bios[i] = bio; 840191ea9b2SNeilBrown targets++; 8411da177e4SLinus Torvalds } else 8421da177e4SLinus Torvalds r1_bio->bios[i] = NULL; 8431da177e4SLinus Torvalds } 8441da177e4SLinus Torvalds rcu_read_unlock(); 8451da177e4SLinus Torvalds 8464b6d287fSNeilBrown BUG_ON(targets == 0); /* we never fail the last device */ 8474b6d287fSNeilBrown 848191ea9b2SNeilBrown if (targets < conf->raid_disks) { 849191ea9b2SNeilBrown /* array is degraded, we will not clear the bitmap 850191ea9b2SNeilBrown * on I/O completion (see raid1_end_write_request) */ 851191ea9b2SNeilBrown set_bit(R1BIO_Degraded, &r1_bio->state); 852191ea9b2SNeilBrown } 85306d91a5fSNeilBrown 8544b6d287fSNeilBrown /* do behind I/O ? */ 8554b6d287fSNeilBrown if (bitmap && 8564b6d287fSNeilBrown atomic_read(&bitmap->behind_writes) < bitmap->max_write_behind && 8574b6d287fSNeilBrown (behind_pages = alloc_behind_pages(bio)) != NULL) 8584b6d287fSNeilBrown set_bit(R1BIO_BehindIO, &r1_bio->state); 8594b6d287fSNeilBrown 860191ea9b2SNeilBrown atomic_set(&r1_bio->remaining, 0); 8614b6d287fSNeilBrown atomic_set(&r1_bio->behind_remaining, 0); 862191ea9b2SNeilBrown 86304b857f7SNeilBrown do_barriers = bio_barrier(bio); 864a9701a30SNeilBrown if (do_barriers) 865a9701a30SNeilBrown set_bit(R1BIO_Barrier, &r1_bio->state); 866a9701a30SNeilBrown 867191ea9b2SNeilBrown bio_list_init(&bl); 8681da177e4SLinus Torvalds for (i = 0; i < disks; i++) { 8691da177e4SLinus Torvalds struct bio *mbio; 8701da177e4SLinus Torvalds if (!r1_bio->bios[i]) 8711da177e4SLinus Torvalds continue; 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds mbio = bio_clone(bio, GFP_NOIO); 8741da177e4SLinus Torvalds r1_bio->bios[i] = mbio; 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset; 8771da177e4SLinus Torvalds mbio->bi_bdev = conf->mirrors[i].rdev->bdev; 8781da177e4SLinus Torvalds mbio->bi_end_io = raid1_end_write_request; 879a9701a30SNeilBrown mbio->bi_rw = WRITE | do_barriers; 8801da177e4SLinus Torvalds mbio->bi_private = r1_bio; 8811da177e4SLinus Torvalds 8824b6d287fSNeilBrown if (behind_pages) { 8834b6d287fSNeilBrown struct bio_vec *bvec; 8844b6d287fSNeilBrown int j; 8854b6d287fSNeilBrown 8864b6d287fSNeilBrown /* Yes, I really want the '__' version so that 8874b6d287fSNeilBrown * we clear any unused pointer in the io_vec, rather 8884b6d287fSNeilBrown * than leave them unchanged. This is important 8894b6d287fSNeilBrown * because when we come to free the pages, we won't 8904b6d287fSNeilBrown * know the originial bi_idx, so we just free 8914b6d287fSNeilBrown * them all 8924b6d287fSNeilBrown */ 8934b6d287fSNeilBrown __bio_for_each_segment(bvec, mbio, j, 0) 8944b6d287fSNeilBrown bvec->bv_page = behind_pages[j]; 8954b6d287fSNeilBrown if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) 8964b6d287fSNeilBrown atomic_inc(&r1_bio->behind_remaining); 8974b6d287fSNeilBrown } 8984b6d287fSNeilBrown 8991da177e4SLinus Torvalds atomic_inc(&r1_bio->remaining); 900191ea9b2SNeilBrown 901191ea9b2SNeilBrown bio_list_add(&bl, mbio); 9021da177e4SLinus Torvalds } 9034b6d287fSNeilBrown kfree(behind_pages); /* the behind pages are attached to the bios now */ 9041da177e4SLinus Torvalds 9054b6d287fSNeilBrown bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors, 9064b6d287fSNeilBrown test_bit(R1BIO_BehindIO, &r1_bio->state)); 907191ea9b2SNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 908191ea9b2SNeilBrown bio_list_merge(&conf->pending_bio_list, &bl); 909191ea9b2SNeilBrown bio_list_init(&bl); 910191ea9b2SNeilBrown 911191ea9b2SNeilBrown blk_plug_device(mddev->queue); 912191ea9b2SNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 913191ea9b2SNeilBrown 914191ea9b2SNeilBrown #if 0 915191ea9b2SNeilBrown while ((bio = bio_list_pop(&bl)) != NULL) 916191ea9b2SNeilBrown generic_make_request(bio); 917191ea9b2SNeilBrown #endif 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds return 0; 9201da177e4SLinus Torvalds } 9211da177e4SLinus Torvalds 9221da177e4SLinus Torvalds static void status(struct seq_file *seq, mddev_t *mddev) 9231da177e4SLinus Torvalds { 9241da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 9251da177e4SLinus Torvalds int i; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds seq_printf(seq, " [%d/%d] [", conf->raid_disks, 9281da177e4SLinus Torvalds conf->working_disks); 9291da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) 9301da177e4SLinus Torvalds seq_printf(seq, "%s", 9311da177e4SLinus Torvalds conf->mirrors[i].rdev && 932b2d444d7SNeilBrown test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); 9331da177e4SLinus Torvalds seq_printf(seq, "]"); 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds static void error(mddev_t *mddev, mdk_rdev_t *rdev) 9381da177e4SLinus Torvalds { 9391da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 9401da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 9411da177e4SLinus Torvalds 9421da177e4SLinus Torvalds /* 9431da177e4SLinus Torvalds * If it is not operational, then we have already marked it as dead 9441da177e4SLinus Torvalds * else if it is the last working disks, ignore the error, let the 9451da177e4SLinus Torvalds * next level up know. 9461da177e4SLinus Torvalds * else mark the drive as failed 9471da177e4SLinus Torvalds */ 948b2d444d7SNeilBrown if (test_bit(In_sync, &rdev->flags) 9491da177e4SLinus Torvalds && conf->working_disks == 1) 9501da177e4SLinus Torvalds /* 9511da177e4SLinus Torvalds * Don't fail the drive, act as though we were just a 9521da177e4SLinus Torvalds * normal single drive 9531da177e4SLinus Torvalds */ 9541da177e4SLinus Torvalds return; 955b2d444d7SNeilBrown if (test_bit(In_sync, &rdev->flags)) { 9561da177e4SLinus Torvalds mddev->degraded++; 9571da177e4SLinus Torvalds conf->working_disks--; 9581da177e4SLinus Torvalds /* 9591da177e4SLinus Torvalds * if recovery is running, make sure it aborts. 9601da177e4SLinus Torvalds */ 9611da177e4SLinus Torvalds set_bit(MD_RECOVERY_ERR, &mddev->recovery); 9621da177e4SLinus Torvalds } 963b2d444d7SNeilBrown clear_bit(In_sync, &rdev->flags); 964b2d444d7SNeilBrown set_bit(Faulty, &rdev->flags); 9651da177e4SLinus Torvalds mddev->sb_dirty = 1; 9661da177e4SLinus Torvalds printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n" 9671da177e4SLinus Torvalds " Operation continuing on %d devices\n", 9681da177e4SLinus Torvalds bdevname(rdev->bdev,b), conf->working_disks); 9691da177e4SLinus Torvalds } 9701da177e4SLinus Torvalds 9711da177e4SLinus Torvalds static void print_conf(conf_t *conf) 9721da177e4SLinus Torvalds { 9731da177e4SLinus Torvalds int i; 9741da177e4SLinus Torvalds mirror_info_t *tmp; 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds printk("RAID1 conf printout:\n"); 9771da177e4SLinus Torvalds if (!conf) { 9781da177e4SLinus Torvalds printk("(!conf)\n"); 9791da177e4SLinus Torvalds return; 9801da177e4SLinus Torvalds } 9811da177e4SLinus Torvalds printk(" --- wd:%d rd:%d\n", conf->working_disks, 9821da177e4SLinus Torvalds conf->raid_disks); 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 9851da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 9861da177e4SLinus Torvalds tmp = conf->mirrors + i; 9871da177e4SLinus Torvalds if (tmp->rdev) 9881da177e4SLinus Torvalds printk(" disk %d, wo:%d, o:%d, dev:%s\n", 989b2d444d7SNeilBrown i, !test_bit(In_sync, &tmp->rdev->flags), !test_bit(Faulty, &tmp->rdev->flags), 9901da177e4SLinus Torvalds bdevname(tmp->rdev->bdev,b)); 9911da177e4SLinus Torvalds } 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds 9941da177e4SLinus Torvalds static void close_sync(conf_t *conf) 9951da177e4SLinus Torvalds { 99617999be4SNeilBrown wait_barrier(conf); 99717999be4SNeilBrown allow_barrier(conf); 9981da177e4SLinus Torvalds 9991da177e4SLinus Torvalds mempool_destroy(conf->r1buf_pool); 10001da177e4SLinus Torvalds conf->r1buf_pool = NULL; 10011da177e4SLinus Torvalds } 10021da177e4SLinus Torvalds 10031da177e4SLinus Torvalds static int raid1_spare_active(mddev_t *mddev) 10041da177e4SLinus Torvalds { 10051da177e4SLinus Torvalds int i; 10061da177e4SLinus Torvalds conf_t *conf = mddev->private; 10071da177e4SLinus Torvalds mirror_info_t *tmp; 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds /* 10101da177e4SLinus Torvalds * Find all failed disks within the RAID1 configuration 10111da177e4SLinus Torvalds * and mark them readable 10121da177e4SLinus Torvalds */ 10131da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 10141da177e4SLinus Torvalds tmp = conf->mirrors + i; 10151da177e4SLinus Torvalds if (tmp->rdev 1016b2d444d7SNeilBrown && !test_bit(Faulty, &tmp->rdev->flags) 1017b2d444d7SNeilBrown && !test_bit(In_sync, &tmp->rdev->flags)) { 10181da177e4SLinus Torvalds conf->working_disks++; 10191da177e4SLinus Torvalds mddev->degraded--; 1020b2d444d7SNeilBrown set_bit(In_sync, &tmp->rdev->flags); 10211da177e4SLinus Torvalds } 10221da177e4SLinus Torvalds } 10231da177e4SLinus Torvalds 10241da177e4SLinus Torvalds print_conf(conf); 10251da177e4SLinus Torvalds return 0; 10261da177e4SLinus Torvalds } 10271da177e4SLinus Torvalds 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) 10301da177e4SLinus Torvalds { 10311da177e4SLinus Torvalds conf_t *conf = mddev->private; 10321da177e4SLinus Torvalds int found = 0; 103341158c7eSNeilBrown int mirror = 0; 10341da177e4SLinus Torvalds mirror_info_t *p; 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds for (mirror=0; mirror < mddev->raid_disks; mirror++) 10371da177e4SLinus Torvalds if ( !(p=conf->mirrors+mirror)->rdev) { 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds blk_queue_stack_limits(mddev->queue, 10401da177e4SLinus Torvalds rdev->bdev->bd_disk->queue); 10411da177e4SLinus Torvalds /* as we don't honour merge_bvec_fn, we must never risk 10421da177e4SLinus Torvalds * violating it, so limit ->max_sector to one PAGE, as 10431da177e4SLinus Torvalds * a one page request is never in violation. 10441da177e4SLinus Torvalds */ 10451da177e4SLinus Torvalds if (rdev->bdev->bd_disk->queue->merge_bvec_fn && 10461da177e4SLinus Torvalds mddev->queue->max_sectors > (PAGE_SIZE>>9)) 10471da177e4SLinus Torvalds blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); 10481da177e4SLinus Torvalds 10491da177e4SLinus Torvalds p->head_position = 0; 10501da177e4SLinus Torvalds rdev->raid_disk = mirror; 10511da177e4SLinus Torvalds found = 1; 10526aea114aSNeilBrown /* As all devices are equivalent, we don't need a full recovery 10536aea114aSNeilBrown * if this was recently any drive of the array 10546aea114aSNeilBrown */ 10556aea114aSNeilBrown if (rdev->saved_raid_disk < 0) 105641158c7eSNeilBrown conf->fullsync = 1; 1057d6065f7bSSuzanne Wood rcu_assign_pointer(p->rdev, rdev); 10581da177e4SLinus Torvalds break; 10591da177e4SLinus Torvalds } 10601da177e4SLinus Torvalds 10611da177e4SLinus Torvalds print_conf(conf); 10621da177e4SLinus Torvalds return found; 10631da177e4SLinus Torvalds } 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds static int raid1_remove_disk(mddev_t *mddev, int number) 10661da177e4SLinus Torvalds { 10671da177e4SLinus Torvalds conf_t *conf = mddev->private; 10681da177e4SLinus Torvalds int err = 0; 10691da177e4SLinus Torvalds mdk_rdev_t *rdev; 10701da177e4SLinus Torvalds mirror_info_t *p = conf->mirrors+ number; 10711da177e4SLinus Torvalds 10721da177e4SLinus Torvalds print_conf(conf); 10731da177e4SLinus Torvalds rdev = p->rdev; 10741da177e4SLinus Torvalds if (rdev) { 1075b2d444d7SNeilBrown if (test_bit(In_sync, &rdev->flags) || 10761da177e4SLinus Torvalds atomic_read(&rdev->nr_pending)) { 10771da177e4SLinus Torvalds err = -EBUSY; 10781da177e4SLinus Torvalds goto abort; 10791da177e4SLinus Torvalds } 10801da177e4SLinus Torvalds p->rdev = NULL; 1081fbd568a3SPaul E. McKenney synchronize_rcu(); 10821da177e4SLinus Torvalds if (atomic_read(&rdev->nr_pending)) { 10831da177e4SLinus Torvalds /* lost the race, try later */ 10841da177e4SLinus Torvalds err = -EBUSY; 10851da177e4SLinus Torvalds p->rdev = rdev; 10861da177e4SLinus Torvalds } 10871da177e4SLinus Torvalds } 10881da177e4SLinus Torvalds abort: 10891da177e4SLinus Torvalds 10901da177e4SLinus Torvalds print_conf(conf); 10911da177e4SLinus Torvalds return err; 10921da177e4SLinus Torvalds } 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds 10951da177e4SLinus Torvalds static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error) 10961da177e4SLinus Torvalds { 10971da177e4SLinus Torvalds r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); 1098d11c171eSNeilBrown int i; 10991da177e4SLinus Torvalds 11001da177e4SLinus Torvalds if (bio->bi_size) 11011da177e4SLinus Torvalds return 1; 11021da177e4SLinus Torvalds 1103d11c171eSNeilBrown for (i=r1_bio->mddev->raid_disks; i--; ) 1104d11c171eSNeilBrown if (r1_bio->bios[i] == bio) 1105d11c171eSNeilBrown break; 1106d11c171eSNeilBrown BUG_ON(i < 0); 1107d11c171eSNeilBrown update_head_pos(i, r1_bio); 11081da177e4SLinus Torvalds /* 11091da177e4SLinus Torvalds * we have read a block, now it needs to be re-written, 11101da177e4SLinus Torvalds * or re-read if the read failed. 11111da177e4SLinus Torvalds * We don't do much here, just schedule handling by raid1d 11121da177e4SLinus Torvalds */ 111369382e85SNeilBrown if (test_bit(BIO_UPTODATE, &bio->bi_flags)) 11141da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 1115d11c171eSNeilBrown 1116d11c171eSNeilBrown if (atomic_dec_and_test(&r1_bio->remaining)) 11171da177e4SLinus Torvalds reschedule_retry(r1_bio); 11181da177e4SLinus Torvalds return 0; 11191da177e4SLinus Torvalds } 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error) 11221da177e4SLinus Torvalds { 11231da177e4SLinus Torvalds int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 11241da177e4SLinus Torvalds r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); 11251da177e4SLinus Torvalds mddev_t *mddev = r1_bio->mddev; 11261da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 11271da177e4SLinus Torvalds int i; 11281da177e4SLinus Torvalds int mirror=0; 11291da177e4SLinus Torvalds 11301da177e4SLinus Torvalds if (bio->bi_size) 11311da177e4SLinus Torvalds return 1; 11321da177e4SLinus Torvalds 11331da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) 11341da177e4SLinus Torvalds if (r1_bio->bios[i] == bio) { 11351da177e4SLinus Torvalds mirror = i; 11361da177e4SLinus Torvalds break; 11371da177e4SLinus Torvalds } 11386b1117d5SNeilBrown if (!uptodate) { 11396b1117d5SNeilBrown int sync_blocks = 0; 11406b1117d5SNeilBrown sector_t s = r1_bio->sector; 11416b1117d5SNeilBrown long sectors_to_go = r1_bio->sectors; 11426b1117d5SNeilBrown /* make sure these bits doesn't get cleared. */ 11436b1117d5SNeilBrown do { 11446b1117d5SNeilBrown bitmap_end_sync(mddev->bitmap, r1_bio->sector, 11456b1117d5SNeilBrown &sync_blocks, 1); 11466b1117d5SNeilBrown s += sync_blocks; 11476b1117d5SNeilBrown sectors_to_go -= sync_blocks; 11486b1117d5SNeilBrown } while (sectors_to_go > 0); 11491da177e4SLinus Torvalds md_error(mddev, conf->mirrors[mirror].rdev); 11506b1117d5SNeilBrown } 1151e3b9703eSNeilBrown 11521da177e4SLinus Torvalds update_head_pos(mirror, r1_bio); 11531da177e4SLinus Torvalds 11541da177e4SLinus Torvalds if (atomic_dec_and_test(&r1_bio->remaining)) { 11551da177e4SLinus Torvalds md_done_sync(mddev, r1_bio->sectors, uptodate); 11561da177e4SLinus Torvalds put_buf(r1_bio); 11571da177e4SLinus Torvalds } 11581da177e4SLinus Torvalds return 0; 11591da177e4SLinus Torvalds } 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) 11621da177e4SLinus Torvalds { 11631da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 11641da177e4SLinus Torvalds int i; 11651da177e4SLinus Torvalds int disks = conf->raid_disks; 11661da177e4SLinus Torvalds struct bio *bio, *wbio; 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds bio = r1_bio->bios[r1_bio->read_disk]; 11691da177e4SLinus Torvalds 117069382e85SNeilBrown 1171d11c171eSNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 1172d11c171eSNeilBrown /* We have read all readable devices. If we haven't 1173d11c171eSNeilBrown * got the block, then there is no hope left. 1174d11c171eSNeilBrown * If we have, then we want to do a comparison 1175d11c171eSNeilBrown * and skip the write if everything is the same. 1176d11c171eSNeilBrown * If any blocks failed to read, then we need to 1177d11c171eSNeilBrown * attempt an over-write 11781da177e4SLinus Torvalds */ 1179d11c171eSNeilBrown int primary; 1180d11c171eSNeilBrown if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { 1181d11c171eSNeilBrown for (i=0; i<mddev->raid_disks; i++) 1182d11c171eSNeilBrown if (r1_bio->bios[i]->bi_end_io == end_sync_read) 1183d11c171eSNeilBrown md_error(mddev, conf->mirrors[i].rdev); 1184d11c171eSNeilBrown 1185d11c171eSNeilBrown md_done_sync(mddev, r1_bio->sectors, 1); 1186d11c171eSNeilBrown put_buf(r1_bio); 1187d11c171eSNeilBrown return; 1188d11c171eSNeilBrown } 1189d11c171eSNeilBrown for (primary=0; primary<mddev->raid_disks; primary++) 1190d11c171eSNeilBrown if (r1_bio->bios[primary]->bi_end_io == end_sync_read && 1191d11c171eSNeilBrown test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { 1192d11c171eSNeilBrown r1_bio->bios[primary]->bi_end_io = NULL; 119303c902e1SNeilBrown rdev_dec_pending(conf->mirrors[primary].rdev, mddev); 1194d11c171eSNeilBrown break; 1195d11c171eSNeilBrown } 1196d11c171eSNeilBrown r1_bio->read_disk = primary; 1197d11c171eSNeilBrown for (i=0; i<mddev->raid_disks; i++) 1198d11c171eSNeilBrown if (r1_bio->bios[i]->bi_end_io == end_sync_read && 1199d11c171eSNeilBrown test_bit(BIO_UPTODATE, &r1_bio->bios[i]->bi_flags)) { 1200d11c171eSNeilBrown int j; 1201d11c171eSNeilBrown int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); 1202d11c171eSNeilBrown struct bio *pbio = r1_bio->bios[primary]; 1203d11c171eSNeilBrown struct bio *sbio = r1_bio->bios[i]; 1204d11c171eSNeilBrown for (j = vcnt; j-- ; ) 1205d11c171eSNeilBrown if (memcmp(page_address(pbio->bi_io_vec[j].bv_page), 1206d11c171eSNeilBrown page_address(sbio->bi_io_vec[j].bv_page), 1207d11c171eSNeilBrown PAGE_SIZE)) 1208d11c171eSNeilBrown break; 1209d11c171eSNeilBrown if (j >= 0) 1210d11c171eSNeilBrown mddev->resync_mismatches += r1_bio->sectors; 121103c902e1SNeilBrown if (j < 0 || test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { 1212d11c171eSNeilBrown sbio->bi_end_io = NULL; 121303c902e1SNeilBrown rdev_dec_pending(conf->mirrors[i].rdev, mddev); 121403c902e1SNeilBrown } else { 1215d11c171eSNeilBrown /* fixup the bio for reuse */ 1216d11c171eSNeilBrown sbio->bi_vcnt = vcnt; 1217d11c171eSNeilBrown sbio->bi_size = r1_bio->sectors << 9; 1218d11c171eSNeilBrown sbio->bi_idx = 0; 1219d11c171eSNeilBrown sbio->bi_phys_segments = 0; 1220d11c171eSNeilBrown sbio->bi_hw_segments = 0; 1221d11c171eSNeilBrown sbio->bi_hw_front_size = 0; 1222d11c171eSNeilBrown sbio->bi_hw_back_size = 0; 1223d11c171eSNeilBrown sbio->bi_flags &= ~(BIO_POOL_MASK - 1); 1224d11c171eSNeilBrown sbio->bi_flags |= 1 << BIO_UPTODATE; 1225d11c171eSNeilBrown sbio->bi_next = NULL; 1226d11c171eSNeilBrown sbio->bi_sector = r1_bio->sector + 1227d11c171eSNeilBrown conf->mirrors[i].rdev->data_offset; 1228d11c171eSNeilBrown sbio->bi_bdev = conf->mirrors[i].rdev->bdev; 1229d11c171eSNeilBrown } 1230d11c171eSNeilBrown } 1231d11c171eSNeilBrown } 12321da177e4SLinus Torvalds if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { 123369382e85SNeilBrown /* ouch - failed to read all of that. 123469382e85SNeilBrown * Try some synchronous reads of other devices to get 123569382e85SNeilBrown * good data, much like with normal read errors. Only 123669382e85SNeilBrown * read into the pages we already have so they we don't 123769382e85SNeilBrown * need to re-issue the read request. 123869382e85SNeilBrown * We don't need to freeze the array, because being in an 123969382e85SNeilBrown * active sync request, there is no normal IO, and 124069382e85SNeilBrown * no overlapping syncs. 12411da177e4SLinus Torvalds */ 124269382e85SNeilBrown sector_t sect = r1_bio->sector; 124369382e85SNeilBrown int sectors = r1_bio->sectors; 124469382e85SNeilBrown int idx = 0; 124569382e85SNeilBrown 124669382e85SNeilBrown while(sectors) { 124769382e85SNeilBrown int s = sectors; 124869382e85SNeilBrown int d = r1_bio->read_disk; 124969382e85SNeilBrown int success = 0; 125069382e85SNeilBrown mdk_rdev_t *rdev; 125169382e85SNeilBrown 125269382e85SNeilBrown if (s > (PAGE_SIZE>>9)) 125369382e85SNeilBrown s = PAGE_SIZE >> 9; 125469382e85SNeilBrown do { 125569382e85SNeilBrown if (r1_bio->bios[d]->bi_end_io == end_sync_read) { 125669382e85SNeilBrown rdev = conf->mirrors[d].rdev; 125769382e85SNeilBrown if (sync_page_io(rdev->bdev, 125869382e85SNeilBrown sect + rdev->data_offset, 125969382e85SNeilBrown s<<9, 126069382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 126169382e85SNeilBrown READ)) { 126269382e85SNeilBrown success = 1; 126369382e85SNeilBrown break; 126469382e85SNeilBrown } 126569382e85SNeilBrown } 126669382e85SNeilBrown d++; 126769382e85SNeilBrown if (d == conf->raid_disks) 126869382e85SNeilBrown d = 0; 126969382e85SNeilBrown } while (!success && d != r1_bio->read_disk); 127069382e85SNeilBrown 127169382e85SNeilBrown if (success) { 1272097426f6SNeilBrown int start = d; 127369382e85SNeilBrown /* write it back and re-read */ 127469382e85SNeilBrown set_bit(R1BIO_Uptodate, &r1_bio->state); 127569382e85SNeilBrown while (d != r1_bio->read_disk) { 127669382e85SNeilBrown if (d == 0) 127769382e85SNeilBrown d = conf->raid_disks; 127869382e85SNeilBrown d--; 127969382e85SNeilBrown if (r1_bio->bios[d]->bi_end_io != end_sync_read) 128069382e85SNeilBrown continue; 128169382e85SNeilBrown rdev = conf->mirrors[d].rdev; 12824dbcdc75SNeilBrown atomic_add(s, &rdev->corrected_errors); 128369382e85SNeilBrown if (sync_page_io(rdev->bdev, 128469382e85SNeilBrown sect + rdev->data_offset, 128569382e85SNeilBrown s<<9, 128669382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 1287097426f6SNeilBrown WRITE) == 0) 1288097426f6SNeilBrown md_error(mddev, rdev); 1289097426f6SNeilBrown } 1290097426f6SNeilBrown d = start; 1291097426f6SNeilBrown while (d != r1_bio->read_disk) { 1292097426f6SNeilBrown if (d == 0) 1293097426f6SNeilBrown d = conf->raid_disks; 1294097426f6SNeilBrown d--; 1295097426f6SNeilBrown if (r1_bio->bios[d]->bi_end_io != end_sync_read) 1296097426f6SNeilBrown continue; 1297097426f6SNeilBrown rdev = conf->mirrors[d].rdev; 1298097426f6SNeilBrown if (sync_page_io(rdev->bdev, 129969382e85SNeilBrown sect + rdev->data_offset, 130069382e85SNeilBrown s<<9, 130169382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 1302097426f6SNeilBrown READ) == 0) 130369382e85SNeilBrown md_error(mddev, rdev); 130469382e85SNeilBrown } 130569382e85SNeilBrown } else { 13061da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 130769382e85SNeilBrown /* Cannot read from anywhere, array is toast */ 130869382e85SNeilBrown md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); 13091da177e4SLinus Torvalds printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error" 13101da177e4SLinus Torvalds " for block %llu\n", 13111da177e4SLinus Torvalds bdevname(bio->bi_bdev,b), 13121da177e4SLinus Torvalds (unsigned long long)r1_bio->sector); 13131da177e4SLinus Torvalds md_done_sync(mddev, r1_bio->sectors, 0); 13141da177e4SLinus Torvalds put_buf(r1_bio); 13151da177e4SLinus Torvalds return; 13161da177e4SLinus Torvalds } 131769382e85SNeilBrown sectors -= s; 131869382e85SNeilBrown sect += s; 131969382e85SNeilBrown idx ++; 132069382e85SNeilBrown } 132169382e85SNeilBrown } 1322d11c171eSNeilBrown 1323d11c171eSNeilBrown /* 1324d11c171eSNeilBrown * schedule writes 1325d11c171eSNeilBrown */ 13261da177e4SLinus Torvalds atomic_set(&r1_bio->remaining, 1); 13271da177e4SLinus Torvalds for (i = 0; i < disks ; i++) { 13281da177e4SLinus Torvalds wbio = r1_bio->bios[i]; 13293e198f78SNeilBrown if (wbio->bi_end_io == NULL || 13303e198f78SNeilBrown (wbio->bi_end_io == end_sync_read && 13313e198f78SNeilBrown (i == r1_bio->read_disk || 13323e198f78SNeilBrown !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) 13331da177e4SLinus Torvalds continue; 13341da177e4SLinus Torvalds 13353e198f78SNeilBrown wbio->bi_rw = WRITE; 13363e198f78SNeilBrown wbio->bi_end_io = end_sync_write; 13371da177e4SLinus Torvalds atomic_inc(&r1_bio->remaining); 13381da177e4SLinus Torvalds md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); 1339191ea9b2SNeilBrown 13401da177e4SLinus Torvalds generic_make_request(wbio); 13411da177e4SLinus Torvalds } 13421da177e4SLinus Torvalds 13431da177e4SLinus Torvalds if (atomic_dec_and_test(&r1_bio->remaining)) { 1344191ea9b2SNeilBrown /* if we're here, all write(s) have completed, so clean up */ 13451da177e4SLinus Torvalds md_done_sync(mddev, r1_bio->sectors, 1); 13461da177e4SLinus Torvalds put_buf(r1_bio); 13471da177e4SLinus Torvalds } 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds 13501da177e4SLinus Torvalds /* 13511da177e4SLinus Torvalds * This is a kernel thread which: 13521da177e4SLinus Torvalds * 13531da177e4SLinus Torvalds * 1. Retries failed read operations on working mirrors. 13541da177e4SLinus Torvalds * 2. Updates the raid superblock when problems encounter. 13551da177e4SLinus Torvalds * 3. Performs writes following reads for array syncronising. 13561da177e4SLinus Torvalds */ 13571da177e4SLinus Torvalds 13581da177e4SLinus Torvalds static void raid1d(mddev_t *mddev) 13591da177e4SLinus Torvalds { 13601da177e4SLinus Torvalds r1bio_t *r1_bio; 13611da177e4SLinus Torvalds struct bio *bio; 13621da177e4SLinus Torvalds unsigned long flags; 13631da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 13641da177e4SLinus Torvalds struct list_head *head = &conf->retry_list; 13651da177e4SLinus Torvalds int unplug=0; 13661da177e4SLinus Torvalds mdk_rdev_t *rdev; 13671da177e4SLinus Torvalds 13681da177e4SLinus Torvalds md_check_recovery(mddev); 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds for (;;) { 13711da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 13721da177e4SLinus Torvalds spin_lock_irqsave(&conf->device_lock, flags); 1373191ea9b2SNeilBrown 1374191ea9b2SNeilBrown if (conf->pending_bio_list.head) { 1375191ea9b2SNeilBrown bio = bio_list_get(&conf->pending_bio_list); 1376191ea9b2SNeilBrown blk_remove_plug(mddev->queue); 1377191ea9b2SNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 1378191ea9b2SNeilBrown /* flush any pending bitmap writes to disk before proceeding w/ I/O */ 1379191ea9b2SNeilBrown if (bitmap_unplug(mddev->bitmap) != 0) 1380191ea9b2SNeilBrown printk("%s: bitmap file write failed!\n", mdname(mddev)); 1381191ea9b2SNeilBrown 1382191ea9b2SNeilBrown while (bio) { /* submit pending writes */ 1383191ea9b2SNeilBrown struct bio *next = bio->bi_next; 1384191ea9b2SNeilBrown bio->bi_next = NULL; 1385191ea9b2SNeilBrown generic_make_request(bio); 1386191ea9b2SNeilBrown bio = next; 1387191ea9b2SNeilBrown } 1388191ea9b2SNeilBrown unplug = 1; 1389191ea9b2SNeilBrown 1390191ea9b2SNeilBrown continue; 1391191ea9b2SNeilBrown } 1392191ea9b2SNeilBrown 13931da177e4SLinus Torvalds if (list_empty(head)) 13941da177e4SLinus Torvalds break; 13951da177e4SLinus Torvalds r1_bio = list_entry(head->prev, r1bio_t, retry_list); 13961da177e4SLinus Torvalds list_del(head->prev); 1397ddaf22abSNeilBrown conf->nr_queued--; 13981da177e4SLinus Torvalds spin_unlock_irqrestore(&conf->device_lock, flags); 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds mddev = r1_bio->mddev; 14011da177e4SLinus Torvalds conf = mddev_to_conf(mddev); 14021da177e4SLinus Torvalds if (test_bit(R1BIO_IsSync, &r1_bio->state)) { 14031da177e4SLinus Torvalds sync_request_write(mddev, r1_bio); 14041da177e4SLinus Torvalds unplug = 1; 1405a9701a30SNeilBrown } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) { 1406a9701a30SNeilBrown /* some requests in the r1bio were BIO_RW_BARRIER 1407a9701a30SNeilBrown * requests which failed with -ENOTSUPP. Hohumm.. 1408a9701a30SNeilBrown * Better resubmit without the barrier. 1409a9701a30SNeilBrown * We know which devices to resubmit for, because 1410a9701a30SNeilBrown * all others have had their bios[] entry cleared. 1411a9701a30SNeilBrown */ 1412a9701a30SNeilBrown int i; 1413a9701a30SNeilBrown clear_bit(R1BIO_BarrierRetry, &r1_bio->state); 1414a9701a30SNeilBrown clear_bit(R1BIO_Barrier, &r1_bio->state); 1415a9701a30SNeilBrown for (i=0; i < conf->raid_disks; i++) 14162f889129SNeilBrown if (r1_bio->bios[i]) 14172f889129SNeilBrown atomic_inc(&r1_bio->remaining); 14182f889129SNeilBrown for (i=0; i < conf->raid_disks; i++) 1419a9701a30SNeilBrown if (r1_bio->bios[i]) { 1420a9701a30SNeilBrown struct bio_vec *bvec; 1421a9701a30SNeilBrown int j; 1422a9701a30SNeilBrown 1423a9701a30SNeilBrown bio = bio_clone(r1_bio->master_bio, GFP_NOIO); 1424a9701a30SNeilBrown /* copy pages from the failed bio, as 1425a9701a30SNeilBrown * this might be a write-behind device */ 1426a9701a30SNeilBrown __bio_for_each_segment(bvec, bio, j, 0) 1427a9701a30SNeilBrown bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page; 1428a9701a30SNeilBrown bio_put(r1_bio->bios[i]); 1429a9701a30SNeilBrown bio->bi_sector = r1_bio->sector + 1430a9701a30SNeilBrown conf->mirrors[i].rdev->data_offset; 1431a9701a30SNeilBrown bio->bi_bdev = conf->mirrors[i].rdev->bdev; 1432a9701a30SNeilBrown bio->bi_end_io = raid1_end_write_request; 1433a9701a30SNeilBrown bio->bi_rw = WRITE; 1434a9701a30SNeilBrown bio->bi_private = r1_bio; 1435a9701a30SNeilBrown r1_bio->bios[i] = bio; 1436a9701a30SNeilBrown generic_make_request(bio); 1437a9701a30SNeilBrown } 14381da177e4SLinus Torvalds } else { 14391da177e4SLinus Torvalds int disk; 1440ddaf22abSNeilBrown 1441ddaf22abSNeilBrown /* we got a read error. Maybe the drive is bad. Maybe just 1442ddaf22abSNeilBrown * the block and we can fix it. 1443ddaf22abSNeilBrown * We freeze all other IO, and try reading the block from 1444ddaf22abSNeilBrown * other devices. When we find one, we re-write 1445ddaf22abSNeilBrown * and check it that fixes the read error. 1446ddaf22abSNeilBrown * This is all done synchronously while the array is 1447ddaf22abSNeilBrown * frozen 1448ddaf22abSNeilBrown */ 1449ddaf22abSNeilBrown sector_t sect = r1_bio->sector; 1450ddaf22abSNeilBrown int sectors = r1_bio->sectors; 1451ddaf22abSNeilBrown freeze_array(conf); 1452cf30a473SNeilBrown if (mddev->ro == 0) while(sectors) { 1453ddaf22abSNeilBrown int s = sectors; 1454ddaf22abSNeilBrown int d = r1_bio->read_disk; 1455ddaf22abSNeilBrown int success = 0; 1456ddaf22abSNeilBrown 1457ddaf22abSNeilBrown if (s > (PAGE_SIZE>>9)) 1458ddaf22abSNeilBrown s = PAGE_SIZE >> 9; 1459ddaf22abSNeilBrown 1460ddaf22abSNeilBrown do { 1461ddaf22abSNeilBrown rdev = conf->mirrors[d].rdev; 1462ddaf22abSNeilBrown if (rdev && 1463ddaf22abSNeilBrown test_bit(In_sync, &rdev->flags) && 1464ddaf22abSNeilBrown sync_page_io(rdev->bdev, 1465ddaf22abSNeilBrown sect + rdev->data_offset, 1466ddaf22abSNeilBrown s<<9, 1467ddaf22abSNeilBrown conf->tmppage, READ)) 1468ddaf22abSNeilBrown success = 1; 1469ddaf22abSNeilBrown else { 1470ddaf22abSNeilBrown d++; 1471ddaf22abSNeilBrown if (d == conf->raid_disks) 1472ddaf22abSNeilBrown d = 0; 1473ddaf22abSNeilBrown } 1474ddaf22abSNeilBrown } while (!success && d != r1_bio->read_disk); 1475ddaf22abSNeilBrown 1476ddaf22abSNeilBrown if (success) { 1477ddaf22abSNeilBrown /* write it back and re-read */ 1478097426f6SNeilBrown int start = d; 1479ddaf22abSNeilBrown while (d != r1_bio->read_disk) { 1480ddaf22abSNeilBrown if (d==0) 1481ddaf22abSNeilBrown d = conf->raid_disks; 1482ddaf22abSNeilBrown d--; 1483ddaf22abSNeilBrown rdev = conf->mirrors[d].rdev; 14844dbcdc75SNeilBrown atomic_add(s, &rdev->corrected_errors); 1485ddaf22abSNeilBrown if (rdev && 1486ddaf22abSNeilBrown test_bit(In_sync, &rdev->flags)) { 1487ddaf22abSNeilBrown if (sync_page_io(rdev->bdev, 1488ddaf22abSNeilBrown sect + rdev->data_offset, 1489097426f6SNeilBrown s<<9, conf->tmppage, WRITE) == 0) 1490ddaf22abSNeilBrown /* Well, this device is dead */ 1491ddaf22abSNeilBrown md_error(mddev, rdev); 1492ddaf22abSNeilBrown } 1493ddaf22abSNeilBrown } 1494097426f6SNeilBrown d = start; 1495097426f6SNeilBrown while (d != r1_bio->read_disk) { 1496097426f6SNeilBrown if (d==0) 1497097426f6SNeilBrown d = conf->raid_disks; 1498097426f6SNeilBrown d--; 1499097426f6SNeilBrown rdev = conf->mirrors[d].rdev; 1500097426f6SNeilBrown if (rdev && 1501097426f6SNeilBrown test_bit(In_sync, &rdev->flags)) { 1502097426f6SNeilBrown if (sync_page_io(rdev->bdev, 1503097426f6SNeilBrown sect + rdev->data_offset, 1504097426f6SNeilBrown s<<9, conf->tmppage, READ) == 0) 1505097426f6SNeilBrown /* Well, this device is dead */ 1506097426f6SNeilBrown md_error(mddev, rdev); 1507097426f6SNeilBrown } 1508ddaf22abSNeilBrown } 1509ddaf22abSNeilBrown } else { 1510ddaf22abSNeilBrown /* Cannot read from anywhere -- bye bye array */ 1511ddaf22abSNeilBrown md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); 1512ddaf22abSNeilBrown break; 1513ddaf22abSNeilBrown } 1514ddaf22abSNeilBrown sectors -= s; 1515ddaf22abSNeilBrown sect += s; 1516ddaf22abSNeilBrown } 1517ddaf22abSNeilBrown 1518ddaf22abSNeilBrown unfreeze_array(conf); 1519ddaf22abSNeilBrown 15201da177e4SLinus Torvalds bio = r1_bio->bios[r1_bio->read_disk]; 15211da177e4SLinus Torvalds if ((disk=read_balance(conf, r1_bio)) == -1) { 15221da177e4SLinus Torvalds printk(KERN_ALERT "raid1: %s: unrecoverable I/O" 15231da177e4SLinus Torvalds " read error for block %llu\n", 15241da177e4SLinus Torvalds bdevname(bio->bi_bdev,b), 15251da177e4SLinus Torvalds (unsigned long long)r1_bio->sector); 15261da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 15271da177e4SLinus Torvalds } else { 1528cf30a473SNeilBrown r1_bio->bios[r1_bio->read_disk] = 1529cf30a473SNeilBrown mddev->ro ? IO_BLOCKED : NULL; 15301da177e4SLinus Torvalds r1_bio->read_disk = disk; 15311da177e4SLinus Torvalds bio_put(bio); 15321da177e4SLinus Torvalds bio = bio_clone(r1_bio->master_bio, GFP_NOIO); 15331da177e4SLinus Torvalds r1_bio->bios[r1_bio->read_disk] = bio; 15341da177e4SLinus Torvalds rdev = conf->mirrors[disk].rdev; 15351da177e4SLinus Torvalds if (printk_ratelimit()) 15361da177e4SLinus Torvalds printk(KERN_ERR "raid1: %s: redirecting sector %llu to" 15371da177e4SLinus Torvalds " another mirror\n", 15381da177e4SLinus Torvalds bdevname(rdev->bdev,b), 15391da177e4SLinus Torvalds (unsigned long long)r1_bio->sector); 15401da177e4SLinus Torvalds bio->bi_sector = r1_bio->sector + rdev->data_offset; 15411da177e4SLinus Torvalds bio->bi_bdev = rdev->bdev; 15421da177e4SLinus Torvalds bio->bi_end_io = raid1_end_read_request; 15431da177e4SLinus Torvalds bio->bi_rw = READ; 15441da177e4SLinus Torvalds bio->bi_private = r1_bio; 15451da177e4SLinus Torvalds unplug = 1; 15461da177e4SLinus Torvalds generic_make_request(bio); 15471da177e4SLinus Torvalds } 15481da177e4SLinus Torvalds } 15491da177e4SLinus Torvalds } 15501da177e4SLinus Torvalds spin_unlock_irqrestore(&conf->device_lock, flags); 15511da177e4SLinus Torvalds if (unplug) 15521da177e4SLinus Torvalds unplug_slaves(mddev); 15531da177e4SLinus Torvalds } 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds 15561da177e4SLinus Torvalds static int init_resync(conf_t *conf) 15571da177e4SLinus Torvalds { 15581da177e4SLinus Torvalds int buffs; 15591da177e4SLinus Torvalds 15601da177e4SLinus Torvalds buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; 15619e77c485SEric Sesterhenn BUG_ON(conf->r1buf_pool); 15621da177e4SLinus Torvalds conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, 15631da177e4SLinus Torvalds conf->poolinfo); 15641da177e4SLinus Torvalds if (!conf->r1buf_pool) 15651da177e4SLinus Torvalds return -ENOMEM; 15661da177e4SLinus Torvalds conf->next_resync = 0; 15671da177e4SLinus Torvalds return 0; 15681da177e4SLinus Torvalds } 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds /* 15711da177e4SLinus Torvalds * perform a "sync" on one "block" 15721da177e4SLinus Torvalds * 15731da177e4SLinus Torvalds * We need to make sure that no normal I/O request - particularly write 15741da177e4SLinus Torvalds * requests - conflict with active sync requests. 15751da177e4SLinus Torvalds * 15761da177e4SLinus Torvalds * This is achieved by tracking pending requests and a 'barrier' concept 15771da177e4SLinus Torvalds * that can be installed to exclude normal IO requests. 15781da177e4SLinus Torvalds */ 15791da177e4SLinus Torvalds 158057afd89fSNeilBrown static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) 15811da177e4SLinus Torvalds { 15821da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 15831da177e4SLinus Torvalds r1bio_t *r1_bio; 15841da177e4SLinus Torvalds struct bio *bio; 15851da177e4SLinus Torvalds sector_t max_sector, nr_sectors; 15863e198f78SNeilBrown int disk = -1; 15871da177e4SLinus Torvalds int i; 15883e198f78SNeilBrown int wonly = -1; 15893e198f78SNeilBrown int write_targets = 0, read_targets = 0; 1590191ea9b2SNeilBrown int sync_blocks; 1591e3b9703eSNeilBrown int still_degraded = 0; 15921da177e4SLinus Torvalds 15931da177e4SLinus Torvalds if (!conf->r1buf_pool) 1594191ea9b2SNeilBrown { 1595191ea9b2SNeilBrown /* 1596191ea9b2SNeilBrown printk("sync start - bitmap %p\n", mddev->bitmap); 1597191ea9b2SNeilBrown */ 15981da177e4SLinus Torvalds if (init_resync(conf)) 159957afd89fSNeilBrown return 0; 1600191ea9b2SNeilBrown } 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds max_sector = mddev->size << 1; 16031da177e4SLinus Torvalds if (sector_nr >= max_sector) { 1604191ea9b2SNeilBrown /* If we aborted, we need to abort the 1605191ea9b2SNeilBrown * sync on the 'current' bitmap chunk (there will 1606191ea9b2SNeilBrown * only be one in raid1 resync. 1607191ea9b2SNeilBrown * We can find the current addess in mddev->curr_resync 1608191ea9b2SNeilBrown */ 16096a806c51SNeilBrown if (mddev->curr_resync < max_sector) /* aborted */ 16106a806c51SNeilBrown bitmap_end_sync(mddev->bitmap, mddev->curr_resync, 1611191ea9b2SNeilBrown &sync_blocks, 1); 16126a806c51SNeilBrown else /* completed sync */ 1613191ea9b2SNeilBrown conf->fullsync = 0; 16146a806c51SNeilBrown 16156a806c51SNeilBrown bitmap_close_sync(mddev->bitmap); 16161da177e4SLinus Torvalds close_sync(conf); 16171da177e4SLinus Torvalds return 0; 16181da177e4SLinus Torvalds } 16191da177e4SLinus Torvalds 1620e3b9703eSNeilBrown /* before building a request, check if we can skip these blocks.. 1621e3b9703eSNeilBrown * This call the bitmap_start_sync doesn't actually record anything 1622e3b9703eSNeilBrown */ 1623e3b9703eSNeilBrown if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && 1624e5de485fSNeilBrown !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 1625191ea9b2SNeilBrown /* We can skip this block, and probably several more */ 1626191ea9b2SNeilBrown *skipped = 1; 1627191ea9b2SNeilBrown return sync_blocks; 1628191ea9b2SNeilBrown } 16291da177e4SLinus Torvalds /* 163017999be4SNeilBrown * If there is non-resync activity waiting for a turn, 163117999be4SNeilBrown * and resync is going fast enough, 163217999be4SNeilBrown * then let it though before starting on this new sync request. 16331da177e4SLinus Torvalds */ 163417999be4SNeilBrown if (!go_faster && conf->nr_waiting) 16351da177e4SLinus Torvalds msleep_interruptible(1000); 163617999be4SNeilBrown 163717999be4SNeilBrown raise_barrier(conf); 163817999be4SNeilBrown 163917999be4SNeilBrown conf->next_resync = sector_nr; 16401da177e4SLinus Torvalds 16411da177e4SLinus Torvalds r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); 16423e198f78SNeilBrown rcu_read_lock(); 16433e198f78SNeilBrown /* 16443e198f78SNeilBrown * If we get a correctably read error during resync or recovery, 16453e198f78SNeilBrown * we might want to read from a different device. So we 16463e198f78SNeilBrown * flag all drives that could conceivably be read from for READ, 16473e198f78SNeilBrown * and any others (which will be non-In_sync devices) for WRITE. 16483e198f78SNeilBrown * If a read fails, we try reading from something else for which READ 16493e198f78SNeilBrown * is OK. 16503e198f78SNeilBrown */ 16511da177e4SLinus Torvalds 16521da177e4SLinus Torvalds r1_bio->mddev = mddev; 16531da177e4SLinus Torvalds r1_bio->sector = sector_nr; 1654191ea9b2SNeilBrown r1_bio->state = 0; 16551da177e4SLinus Torvalds set_bit(R1BIO_IsSync, &r1_bio->state); 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds for (i=0; i < conf->raid_disks; i++) { 16583e198f78SNeilBrown mdk_rdev_t *rdev; 16591da177e4SLinus Torvalds bio = r1_bio->bios[i]; 16601da177e4SLinus Torvalds 16611da177e4SLinus Torvalds /* take from bio_init */ 16621da177e4SLinus Torvalds bio->bi_next = NULL; 16631da177e4SLinus Torvalds bio->bi_flags |= 1 << BIO_UPTODATE; 16641da177e4SLinus Torvalds bio->bi_rw = 0; 16651da177e4SLinus Torvalds bio->bi_vcnt = 0; 16661da177e4SLinus Torvalds bio->bi_idx = 0; 16671da177e4SLinus Torvalds bio->bi_phys_segments = 0; 16681da177e4SLinus Torvalds bio->bi_hw_segments = 0; 16691da177e4SLinus Torvalds bio->bi_size = 0; 16701da177e4SLinus Torvalds bio->bi_end_io = NULL; 16711da177e4SLinus Torvalds bio->bi_private = NULL; 16721da177e4SLinus Torvalds 16733e198f78SNeilBrown rdev = rcu_dereference(conf->mirrors[i].rdev); 16743e198f78SNeilBrown if (rdev == NULL || 16753e198f78SNeilBrown test_bit(Faulty, &rdev->flags)) { 1676e3b9703eSNeilBrown still_degraded = 1; 1677e3b9703eSNeilBrown continue; 16783e198f78SNeilBrown } else if (!test_bit(In_sync, &rdev->flags)) { 16791da177e4SLinus Torvalds bio->bi_rw = WRITE; 16801da177e4SLinus Torvalds bio->bi_end_io = end_sync_write; 16811da177e4SLinus Torvalds write_targets ++; 16823e198f78SNeilBrown } else { 16833e198f78SNeilBrown /* may need to read from here */ 16843e198f78SNeilBrown bio->bi_rw = READ; 16853e198f78SNeilBrown bio->bi_end_io = end_sync_read; 16863e198f78SNeilBrown if (test_bit(WriteMostly, &rdev->flags)) { 16873e198f78SNeilBrown if (wonly < 0) 16883e198f78SNeilBrown wonly = i; 16893e198f78SNeilBrown } else { 16903e198f78SNeilBrown if (disk < 0) 16913e198f78SNeilBrown disk = i; 16923e198f78SNeilBrown } 16933e198f78SNeilBrown read_targets++; 16943e198f78SNeilBrown } 16953e198f78SNeilBrown atomic_inc(&rdev->nr_pending); 16963e198f78SNeilBrown bio->bi_sector = sector_nr + rdev->data_offset; 16973e198f78SNeilBrown bio->bi_bdev = rdev->bdev; 16981da177e4SLinus Torvalds bio->bi_private = r1_bio; 16991da177e4SLinus Torvalds } 17003e198f78SNeilBrown rcu_read_unlock(); 17013e198f78SNeilBrown if (disk < 0) 17023e198f78SNeilBrown disk = wonly; 17033e198f78SNeilBrown r1_bio->read_disk = disk; 1704191ea9b2SNeilBrown 17053e198f78SNeilBrown if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) 17063e198f78SNeilBrown /* extra read targets are also write targets */ 17073e198f78SNeilBrown write_targets += read_targets-1; 17083e198f78SNeilBrown 17093e198f78SNeilBrown if (write_targets == 0 || read_targets == 0) { 17101da177e4SLinus Torvalds /* There is nowhere to write, so all non-sync 17111da177e4SLinus Torvalds * drives must be failed - so we are finished 17121da177e4SLinus Torvalds */ 171357afd89fSNeilBrown sector_t rv = max_sector - sector_nr; 171457afd89fSNeilBrown *skipped = 1; 17151da177e4SLinus Torvalds put_buf(r1_bio); 17161da177e4SLinus Torvalds return rv; 17171da177e4SLinus Torvalds } 17181da177e4SLinus Torvalds 17191da177e4SLinus Torvalds nr_sectors = 0; 1720289e99e8SNeilBrown sync_blocks = 0; 17211da177e4SLinus Torvalds do { 17221da177e4SLinus Torvalds struct page *page; 17231da177e4SLinus Torvalds int len = PAGE_SIZE; 17241da177e4SLinus Torvalds if (sector_nr + (len>>9) > max_sector) 17251da177e4SLinus Torvalds len = (max_sector - sector_nr) << 9; 17261da177e4SLinus Torvalds if (len == 0) 17271da177e4SLinus Torvalds break; 1728ab7a30c7SNeilBrown if (sync_blocks == 0) { 17296a806c51SNeilBrown if (!bitmap_start_sync(mddev->bitmap, sector_nr, 1730e3b9703eSNeilBrown &sync_blocks, still_degraded) && 1731e5de485fSNeilBrown !conf->fullsync && 1732e5de485fSNeilBrown !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 1733191ea9b2SNeilBrown break; 17349e77c485SEric Sesterhenn BUG_ON(sync_blocks < (PAGE_SIZE>>9)); 17356a806c51SNeilBrown if (len > (sync_blocks<<9)) 17366a806c51SNeilBrown len = sync_blocks<<9; 1737ab7a30c7SNeilBrown } 1738191ea9b2SNeilBrown 17391da177e4SLinus Torvalds for (i=0 ; i < conf->raid_disks; i++) { 17401da177e4SLinus Torvalds bio = r1_bio->bios[i]; 17411da177e4SLinus Torvalds if (bio->bi_end_io) { 1742d11c171eSNeilBrown page = bio->bi_io_vec[bio->bi_vcnt].bv_page; 17431da177e4SLinus Torvalds if (bio_add_page(bio, page, len, 0) == 0) { 17441da177e4SLinus Torvalds /* stop here */ 1745d11c171eSNeilBrown bio->bi_io_vec[bio->bi_vcnt].bv_page = page; 17461da177e4SLinus Torvalds while (i > 0) { 17471da177e4SLinus Torvalds i--; 17481da177e4SLinus Torvalds bio = r1_bio->bios[i]; 17496a806c51SNeilBrown if (bio->bi_end_io==NULL) 17506a806c51SNeilBrown continue; 17511da177e4SLinus Torvalds /* remove last page from this bio */ 17521da177e4SLinus Torvalds bio->bi_vcnt--; 17531da177e4SLinus Torvalds bio->bi_size -= len; 17541da177e4SLinus Torvalds bio->bi_flags &= ~(1<< BIO_SEG_VALID); 17551da177e4SLinus Torvalds } 17561da177e4SLinus Torvalds goto bio_full; 17571da177e4SLinus Torvalds } 17581da177e4SLinus Torvalds } 17591da177e4SLinus Torvalds } 17601da177e4SLinus Torvalds nr_sectors += len>>9; 17611da177e4SLinus Torvalds sector_nr += len>>9; 1762191ea9b2SNeilBrown sync_blocks -= (len>>9); 17631da177e4SLinus Torvalds } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); 17641da177e4SLinus Torvalds bio_full: 17651da177e4SLinus Torvalds r1_bio->sectors = nr_sectors; 17661da177e4SLinus Torvalds 1767d11c171eSNeilBrown /* For a user-requested sync, we read all readable devices and do a 1768d11c171eSNeilBrown * compare 1769d11c171eSNeilBrown */ 1770d11c171eSNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 1771d11c171eSNeilBrown atomic_set(&r1_bio->remaining, read_targets); 1772d11c171eSNeilBrown for (i=0; i<conf->raid_disks; i++) { 1773d11c171eSNeilBrown bio = r1_bio->bios[i]; 1774d11c171eSNeilBrown if (bio->bi_end_io == end_sync_read) { 1775d11c171eSNeilBrown md_sync_acct(conf->mirrors[i].rdev->bdev, nr_sectors); 17761da177e4SLinus Torvalds generic_make_request(bio); 1777d11c171eSNeilBrown } 1778d11c171eSNeilBrown } 1779d11c171eSNeilBrown } else { 1780d11c171eSNeilBrown atomic_set(&r1_bio->remaining, 1); 1781d11c171eSNeilBrown bio = r1_bio->bios[r1_bio->read_disk]; 1782d11c171eSNeilBrown md_sync_acct(conf->mirrors[r1_bio->read_disk].rdev->bdev, 1783d11c171eSNeilBrown nr_sectors); 1784d11c171eSNeilBrown generic_make_request(bio); 1785d11c171eSNeilBrown 1786d11c171eSNeilBrown } 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds return nr_sectors; 17891da177e4SLinus Torvalds } 17901da177e4SLinus Torvalds 17911da177e4SLinus Torvalds static int run(mddev_t *mddev) 17921da177e4SLinus Torvalds { 17931da177e4SLinus Torvalds conf_t *conf; 17941da177e4SLinus Torvalds int i, j, disk_idx; 17951da177e4SLinus Torvalds mirror_info_t *disk; 17961da177e4SLinus Torvalds mdk_rdev_t *rdev; 17971da177e4SLinus Torvalds struct list_head *tmp; 17981da177e4SLinus Torvalds 17991da177e4SLinus Torvalds if (mddev->level != 1) { 18001da177e4SLinus Torvalds printk("raid1: %s: raid level not set to mirroring (%d)\n", 18011da177e4SLinus Torvalds mdname(mddev), mddev->level); 18021da177e4SLinus Torvalds goto out; 18031da177e4SLinus Torvalds } 1804f6705578SNeilBrown if (mddev->reshape_position != MaxSector) { 1805f6705578SNeilBrown printk("raid1: %s: reshape_position set but not supported\n", 1806f6705578SNeilBrown mdname(mddev)); 1807f6705578SNeilBrown goto out; 1808f6705578SNeilBrown } 18091da177e4SLinus Torvalds /* 18101da177e4SLinus Torvalds * copy the already verified devices into our private RAID1 18111da177e4SLinus Torvalds * bookkeeping area. [whatever we allocate in run(), 18121da177e4SLinus Torvalds * should be freed in stop()] 18131da177e4SLinus Torvalds */ 18149ffae0cfSNeilBrown conf = kzalloc(sizeof(conf_t), GFP_KERNEL); 18151da177e4SLinus Torvalds mddev->private = conf; 18161da177e4SLinus Torvalds if (!conf) 18171da177e4SLinus Torvalds goto out_no_mem; 18181da177e4SLinus Torvalds 18199ffae0cfSNeilBrown conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, 18201da177e4SLinus Torvalds GFP_KERNEL); 18211da177e4SLinus Torvalds if (!conf->mirrors) 18221da177e4SLinus Torvalds goto out_no_mem; 18231da177e4SLinus Torvalds 1824ddaf22abSNeilBrown conf->tmppage = alloc_page(GFP_KERNEL); 1825ddaf22abSNeilBrown if (!conf->tmppage) 1826ddaf22abSNeilBrown goto out_no_mem; 1827ddaf22abSNeilBrown 18281da177e4SLinus Torvalds conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL); 18291da177e4SLinus Torvalds if (!conf->poolinfo) 18301da177e4SLinus Torvalds goto out_no_mem; 18311da177e4SLinus Torvalds conf->poolinfo->mddev = mddev; 18321da177e4SLinus Torvalds conf->poolinfo->raid_disks = mddev->raid_disks; 18331da177e4SLinus Torvalds conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 18341da177e4SLinus Torvalds r1bio_pool_free, 18351da177e4SLinus Torvalds conf->poolinfo); 18361da177e4SLinus Torvalds if (!conf->r1bio_pool) 18371da177e4SLinus Torvalds goto out_no_mem; 18381da177e4SLinus Torvalds 18391da177e4SLinus Torvalds ITERATE_RDEV(mddev, rdev, tmp) { 18401da177e4SLinus Torvalds disk_idx = rdev->raid_disk; 18411da177e4SLinus Torvalds if (disk_idx >= mddev->raid_disks 18421da177e4SLinus Torvalds || disk_idx < 0) 18431da177e4SLinus Torvalds continue; 18441da177e4SLinus Torvalds disk = conf->mirrors + disk_idx; 18451da177e4SLinus Torvalds 18461da177e4SLinus Torvalds disk->rdev = rdev; 18471da177e4SLinus Torvalds 18481da177e4SLinus Torvalds blk_queue_stack_limits(mddev->queue, 18491da177e4SLinus Torvalds rdev->bdev->bd_disk->queue); 18501da177e4SLinus Torvalds /* as we don't honour merge_bvec_fn, we must never risk 18511da177e4SLinus Torvalds * violating it, so limit ->max_sector to one PAGE, as 18521da177e4SLinus Torvalds * a one page request is never in violation. 18531da177e4SLinus Torvalds */ 18541da177e4SLinus Torvalds if (rdev->bdev->bd_disk->queue->merge_bvec_fn && 18551da177e4SLinus Torvalds mddev->queue->max_sectors > (PAGE_SIZE>>9)) 18561da177e4SLinus Torvalds blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); 18571da177e4SLinus Torvalds 18581da177e4SLinus Torvalds disk->head_position = 0; 1859b2d444d7SNeilBrown if (!test_bit(Faulty, &rdev->flags) && test_bit(In_sync, &rdev->flags)) 18601da177e4SLinus Torvalds conf->working_disks++; 18611da177e4SLinus Torvalds } 18621da177e4SLinus Torvalds conf->raid_disks = mddev->raid_disks; 18631da177e4SLinus Torvalds conf->mddev = mddev; 18641da177e4SLinus Torvalds spin_lock_init(&conf->device_lock); 18651da177e4SLinus Torvalds INIT_LIST_HEAD(&conf->retry_list); 18661da177e4SLinus Torvalds if (conf->working_disks == 1) 18671da177e4SLinus Torvalds mddev->recovery_cp = MaxSector; 18681da177e4SLinus Torvalds 18691da177e4SLinus Torvalds spin_lock_init(&conf->resync_lock); 187017999be4SNeilBrown init_waitqueue_head(&conf->wait_barrier); 18711da177e4SLinus Torvalds 1872191ea9b2SNeilBrown bio_list_init(&conf->pending_bio_list); 1873191ea9b2SNeilBrown bio_list_init(&conf->flushing_bio_list); 1874191ea9b2SNeilBrown 18751da177e4SLinus Torvalds if (!conf->working_disks) { 18761da177e4SLinus Torvalds printk(KERN_ERR "raid1: no operational mirrors for %s\n", 18771da177e4SLinus Torvalds mdname(mddev)); 18781da177e4SLinus Torvalds goto out_free_conf; 18791da177e4SLinus Torvalds } 18801da177e4SLinus Torvalds 18811da177e4SLinus Torvalds mddev->degraded = 0; 18821da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 18831da177e4SLinus Torvalds 18841da177e4SLinus Torvalds disk = conf->mirrors + i; 18851da177e4SLinus Torvalds 18861da177e4SLinus Torvalds if (!disk->rdev) { 18871da177e4SLinus Torvalds disk->head_position = 0; 18881da177e4SLinus Torvalds mddev->degraded++; 18891da177e4SLinus Torvalds } 18901da177e4SLinus Torvalds } 18911da177e4SLinus Torvalds 18921da177e4SLinus Torvalds /* 18931da177e4SLinus Torvalds * find the first working one and use it as a starting point 18941da177e4SLinus Torvalds * to read balancing. 18951da177e4SLinus Torvalds */ 18961da177e4SLinus Torvalds for (j = 0; j < conf->raid_disks && 18971da177e4SLinus Torvalds (!conf->mirrors[j].rdev || 1898b2d444d7SNeilBrown !test_bit(In_sync, &conf->mirrors[j].rdev->flags)) ; j++) 18991da177e4SLinus Torvalds /* nothing */; 19001da177e4SLinus Torvalds conf->last_used = j; 19011da177e4SLinus Torvalds 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1"); 19041da177e4SLinus Torvalds if (!mddev->thread) { 19051da177e4SLinus Torvalds printk(KERN_ERR 19061da177e4SLinus Torvalds "raid1: couldn't allocate thread for %s\n", 19071da177e4SLinus Torvalds mdname(mddev)); 19081da177e4SLinus Torvalds goto out_free_conf; 19091da177e4SLinus Torvalds } 1910191ea9b2SNeilBrown 19111da177e4SLinus Torvalds printk(KERN_INFO 19121da177e4SLinus Torvalds "raid1: raid set %s active with %d out of %d mirrors\n", 19131da177e4SLinus Torvalds mdname(mddev), mddev->raid_disks - mddev->degraded, 19141da177e4SLinus Torvalds mddev->raid_disks); 19151da177e4SLinus Torvalds /* 19161da177e4SLinus Torvalds * Ok, everything is just fine now 19171da177e4SLinus Torvalds */ 19181da177e4SLinus Torvalds mddev->array_size = mddev->size; 19191da177e4SLinus Torvalds 19207a5febe9SNeilBrown mddev->queue->unplug_fn = raid1_unplug; 19217a5febe9SNeilBrown mddev->queue->issue_flush_fn = raid1_issue_flush; 19227a5febe9SNeilBrown 19231da177e4SLinus Torvalds return 0; 19241da177e4SLinus Torvalds 19251da177e4SLinus Torvalds out_no_mem: 19261da177e4SLinus Torvalds printk(KERN_ERR "raid1: couldn't allocate memory for %s\n", 19271da177e4SLinus Torvalds mdname(mddev)); 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds out_free_conf: 19301da177e4SLinus Torvalds if (conf) { 19311da177e4SLinus Torvalds if (conf->r1bio_pool) 19321da177e4SLinus Torvalds mempool_destroy(conf->r1bio_pool); 19331da177e4SLinus Torvalds kfree(conf->mirrors); 19341345b1d8SNeilBrown safe_put_page(conf->tmppage); 19351da177e4SLinus Torvalds kfree(conf->poolinfo); 19361da177e4SLinus Torvalds kfree(conf); 19371da177e4SLinus Torvalds mddev->private = NULL; 19381da177e4SLinus Torvalds } 19391da177e4SLinus Torvalds out: 19401da177e4SLinus Torvalds return -EIO; 19411da177e4SLinus Torvalds } 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds static int stop(mddev_t *mddev) 19441da177e4SLinus Torvalds { 19451da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 19464b6d287fSNeilBrown struct bitmap *bitmap = mddev->bitmap; 19474b6d287fSNeilBrown int behind_wait = 0; 19484b6d287fSNeilBrown 19494b6d287fSNeilBrown /* wait for behind writes to complete */ 19504b6d287fSNeilBrown while (bitmap && atomic_read(&bitmap->behind_writes) > 0) { 19514b6d287fSNeilBrown behind_wait++; 19524b6d287fSNeilBrown printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait); 19534b6d287fSNeilBrown set_current_state(TASK_UNINTERRUPTIBLE); 19544b6d287fSNeilBrown schedule_timeout(HZ); /* wait a second */ 19554b6d287fSNeilBrown /* need to kick something here to make sure I/O goes? */ 19564b6d287fSNeilBrown } 19571da177e4SLinus Torvalds 19581da177e4SLinus Torvalds md_unregister_thread(mddev->thread); 19591da177e4SLinus Torvalds mddev->thread = NULL; 19601da177e4SLinus Torvalds blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ 19611da177e4SLinus Torvalds if (conf->r1bio_pool) 19621da177e4SLinus Torvalds mempool_destroy(conf->r1bio_pool); 19631da177e4SLinus Torvalds kfree(conf->mirrors); 19641da177e4SLinus Torvalds kfree(conf->poolinfo); 19651da177e4SLinus Torvalds kfree(conf); 19661da177e4SLinus Torvalds mddev->private = NULL; 19671da177e4SLinus Torvalds return 0; 19681da177e4SLinus Torvalds } 19691da177e4SLinus Torvalds 19701da177e4SLinus Torvalds static int raid1_resize(mddev_t *mddev, sector_t sectors) 19711da177e4SLinus Torvalds { 19721da177e4SLinus Torvalds /* no resync is happening, and there is enough space 19731da177e4SLinus Torvalds * on all devices, so we can resize. 19741da177e4SLinus Torvalds * We need to make sure resync covers any new space. 19751da177e4SLinus Torvalds * If the array is shrinking we should possibly wait until 19761da177e4SLinus Torvalds * any io in the removed space completes, but it hardly seems 19771da177e4SLinus Torvalds * worth it. 19781da177e4SLinus Torvalds */ 19791da177e4SLinus Torvalds mddev->array_size = sectors>>1; 19801da177e4SLinus Torvalds set_capacity(mddev->gendisk, mddev->array_size << 1); 19811da177e4SLinus Torvalds mddev->changed = 1; 19821da177e4SLinus Torvalds if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) { 19831da177e4SLinus Torvalds mddev->recovery_cp = mddev->size << 1; 19841da177e4SLinus Torvalds set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 19851da177e4SLinus Torvalds } 19861da177e4SLinus Torvalds mddev->size = mddev->array_size; 19874b5c7ae8SNeilBrown mddev->resync_max_sectors = sectors; 19881da177e4SLinus Torvalds return 0; 19891da177e4SLinus Torvalds } 19901da177e4SLinus Torvalds 199163c70c4fSNeilBrown static int raid1_reshape(mddev_t *mddev) 19921da177e4SLinus Torvalds { 19931da177e4SLinus Torvalds /* We need to: 19941da177e4SLinus Torvalds * 1/ resize the r1bio_pool 19951da177e4SLinus Torvalds * 2/ resize conf->mirrors 19961da177e4SLinus Torvalds * 19971da177e4SLinus Torvalds * We allocate a new r1bio_pool if we can. 19981da177e4SLinus Torvalds * Then raise a device barrier and wait until all IO stops. 19991da177e4SLinus Torvalds * Then resize conf->mirrors and swap in the new r1bio pool. 20006ea9c07cSNeilBrown * 20016ea9c07cSNeilBrown * At the same time, we "pack" the devices so that all the missing 20026ea9c07cSNeilBrown * devices have the higher raid_disk numbers. 20031da177e4SLinus Torvalds */ 20041da177e4SLinus Torvalds mempool_t *newpool, *oldpool; 20051da177e4SLinus Torvalds struct pool_info *newpoolinfo; 20061da177e4SLinus Torvalds mirror_info_t *newmirrors; 20071da177e4SLinus Torvalds conf_t *conf = mddev_to_conf(mddev); 200863c70c4fSNeilBrown int cnt, raid_disks; 20091da177e4SLinus Torvalds 20106ea9c07cSNeilBrown int d, d2; 20111da177e4SLinus Torvalds 201263c70c4fSNeilBrown /* Cannot change chunk_size, layout, or level */ 201363c70c4fSNeilBrown if (mddev->chunk_size != mddev->new_chunk || 201463c70c4fSNeilBrown mddev->layout != mddev->new_layout || 201563c70c4fSNeilBrown mddev->level != mddev->new_level) { 201663c70c4fSNeilBrown mddev->new_chunk = mddev->chunk_size; 201763c70c4fSNeilBrown mddev->new_layout = mddev->layout; 201863c70c4fSNeilBrown mddev->new_level = mddev->level; 201963c70c4fSNeilBrown return -EINVAL; 202063c70c4fSNeilBrown } 202163c70c4fSNeilBrown 202263c70c4fSNeilBrown raid_disks = mddev->raid_disks + mddev->delta_disks; 202363c70c4fSNeilBrown 20246ea9c07cSNeilBrown if (raid_disks < conf->raid_disks) { 20256ea9c07cSNeilBrown cnt=0; 20266ea9c07cSNeilBrown for (d= 0; d < conf->raid_disks; d++) 20271da177e4SLinus Torvalds if (conf->mirrors[d].rdev) 20286ea9c07cSNeilBrown cnt++; 20296ea9c07cSNeilBrown if (cnt > raid_disks) 20301da177e4SLinus Torvalds return -EBUSY; 20316ea9c07cSNeilBrown } 20321da177e4SLinus Torvalds 20331da177e4SLinus Torvalds newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); 20341da177e4SLinus Torvalds if (!newpoolinfo) 20351da177e4SLinus Torvalds return -ENOMEM; 20361da177e4SLinus Torvalds newpoolinfo->mddev = mddev; 20371da177e4SLinus Torvalds newpoolinfo->raid_disks = raid_disks; 20381da177e4SLinus Torvalds 20391da177e4SLinus Torvalds newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 20401da177e4SLinus Torvalds r1bio_pool_free, newpoolinfo); 20411da177e4SLinus Torvalds if (!newpool) { 20421da177e4SLinus Torvalds kfree(newpoolinfo); 20431da177e4SLinus Torvalds return -ENOMEM; 20441da177e4SLinus Torvalds } 20459ffae0cfSNeilBrown newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL); 20461da177e4SLinus Torvalds if (!newmirrors) { 20471da177e4SLinus Torvalds kfree(newpoolinfo); 20481da177e4SLinus Torvalds mempool_destroy(newpool); 20491da177e4SLinus Torvalds return -ENOMEM; 20501da177e4SLinus Torvalds } 20511da177e4SLinus Torvalds 205217999be4SNeilBrown raise_barrier(conf); 20531da177e4SLinus Torvalds 20541da177e4SLinus Torvalds /* ok, everything is stopped */ 20551da177e4SLinus Torvalds oldpool = conf->r1bio_pool; 20561da177e4SLinus Torvalds conf->r1bio_pool = newpool; 20576ea9c07cSNeilBrown 20586ea9c07cSNeilBrown for (d=d2=0; d < conf->raid_disks; d++) 20596ea9c07cSNeilBrown if (conf->mirrors[d].rdev) { 20606ea9c07cSNeilBrown conf->mirrors[d].rdev->raid_disk = d2; 20616ea9c07cSNeilBrown newmirrors[d2++].rdev = conf->mirrors[d].rdev; 20626ea9c07cSNeilBrown } 20631da177e4SLinus Torvalds kfree(conf->mirrors); 20641da177e4SLinus Torvalds conf->mirrors = newmirrors; 20651da177e4SLinus Torvalds kfree(conf->poolinfo); 20661da177e4SLinus Torvalds conf->poolinfo = newpoolinfo; 20671da177e4SLinus Torvalds 20681da177e4SLinus Torvalds mddev->degraded += (raid_disks - conf->raid_disks); 20691da177e4SLinus Torvalds conf->raid_disks = mddev->raid_disks = raid_disks; 207063c70c4fSNeilBrown mddev->delta_disks = 0; 20711da177e4SLinus Torvalds 20726ea9c07cSNeilBrown conf->last_used = 0; /* just make sure it is in-range */ 207317999be4SNeilBrown lower_barrier(conf); 20741da177e4SLinus Torvalds 20751da177e4SLinus Torvalds set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 20761da177e4SLinus Torvalds md_wakeup_thread(mddev->thread); 20771da177e4SLinus Torvalds 20781da177e4SLinus Torvalds mempool_destroy(oldpool); 20791da177e4SLinus Torvalds return 0; 20801da177e4SLinus Torvalds } 20811da177e4SLinus Torvalds 2082500af87aSNeilBrown static void raid1_quiesce(mddev_t *mddev, int state) 208336fa3063SNeilBrown { 208436fa3063SNeilBrown conf_t *conf = mddev_to_conf(mddev); 208536fa3063SNeilBrown 208636fa3063SNeilBrown switch(state) { 20879e6603daSNeilBrown case 1: 208817999be4SNeilBrown raise_barrier(conf); 208936fa3063SNeilBrown break; 20909e6603daSNeilBrown case 0: 209117999be4SNeilBrown lower_barrier(conf); 209236fa3063SNeilBrown break; 209336fa3063SNeilBrown } 209436fa3063SNeilBrown } 209536fa3063SNeilBrown 20961da177e4SLinus Torvalds 20972604b703SNeilBrown static struct mdk_personality raid1_personality = 20981da177e4SLinus Torvalds { 20991da177e4SLinus Torvalds .name = "raid1", 21002604b703SNeilBrown .level = 1, 21011da177e4SLinus Torvalds .owner = THIS_MODULE, 21021da177e4SLinus Torvalds .make_request = make_request, 21031da177e4SLinus Torvalds .run = run, 21041da177e4SLinus Torvalds .stop = stop, 21051da177e4SLinus Torvalds .status = status, 21061da177e4SLinus Torvalds .error_handler = error, 21071da177e4SLinus Torvalds .hot_add_disk = raid1_add_disk, 21081da177e4SLinus Torvalds .hot_remove_disk= raid1_remove_disk, 21091da177e4SLinus Torvalds .spare_active = raid1_spare_active, 21101da177e4SLinus Torvalds .sync_request = sync_request, 21111da177e4SLinus Torvalds .resize = raid1_resize, 211263c70c4fSNeilBrown .check_reshape = raid1_reshape, 211336fa3063SNeilBrown .quiesce = raid1_quiesce, 21141da177e4SLinus Torvalds }; 21151da177e4SLinus Torvalds 21161da177e4SLinus Torvalds static int __init raid_init(void) 21171da177e4SLinus Torvalds { 21182604b703SNeilBrown return register_md_personality(&raid1_personality); 21191da177e4SLinus Torvalds } 21201da177e4SLinus Torvalds 21211da177e4SLinus Torvalds static void raid_exit(void) 21221da177e4SLinus Torvalds { 21232604b703SNeilBrown unregister_md_personality(&raid1_personality); 21241da177e4SLinus Torvalds } 21251da177e4SLinus Torvalds 21261da177e4SLinus Torvalds module_init(raid_init); 21271da177e4SLinus Torvalds module_exit(raid_exit); 21281da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 21291da177e4SLinus Torvalds MODULE_ALIAS("md-personality-3"); /* RAID1 */ 2130d9d166c2SNeilBrown MODULE_ALIAS("md-raid1"); 21312604b703SNeilBrown MODULE_ALIAS("md-level-1"); 2132