11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * raid1.c : Multiple Devices driver for Linux 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * RAID-1 management functions. 91da177e4SLinus Torvalds * 101da177e4SLinus Torvalds * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 111da177e4SLinus Torvalds * 1296de0e25SJan Engelhardt * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> 131da177e4SLinus Torvalds * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> 141da177e4SLinus Torvalds * 15191ea9b2SNeilBrown * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support 16191ea9b2SNeilBrown * bitmapped intelligence in resync: 17191ea9b2SNeilBrown * 18191ea9b2SNeilBrown * - bitmap marked during normal i/o 19191ea9b2SNeilBrown * - bitmap used to skip nondirty blocks during sync 20191ea9b2SNeilBrown * 21191ea9b2SNeilBrown * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: 22191ea9b2SNeilBrown * - persistent bitmap code 23191ea9b2SNeilBrown * 241da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or modify 251da177e4SLinus Torvalds * it under the terms of the GNU General Public License as published by 261da177e4SLinus Torvalds * the Free Software Foundation; either version 2, or (at your option) 271da177e4SLinus Torvalds * any later version. 281da177e4SLinus Torvalds * 291da177e4SLinus Torvalds * You should have received a copy of the GNU General Public License 301da177e4SLinus Torvalds * (for example /usr/src/linux/COPYING); if not, write to the Free 311da177e4SLinus Torvalds * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 321da177e4SLinus Torvalds */ 331da177e4SLinus Torvalds 345a0e3ad6STejun Heo #include <linux/slab.h> 3525570727SStephen Rothwell #include <linux/delay.h> 36bff61975SNeilBrown #include <linux/blkdev.h> 37056075c7SPaul Gortmaker #include <linux/module.h> 38bff61975SNeilBrown #include <linux/seq_file.h> 398bda470eSChristian Dietrich #include <linux/ratelimit.h> 4043b2e5d8SNeilBrown #include "md.h" 41ef740c37SChristoph Hellwig #include "raid1.h" 42ef740c37SChristoph Hellwig #include "bitmap.h" 43191ea9b2SNeilBrown 441da177e4SLinus Torvalds /* 451da177e4SLinus Torvalds * Number of guaranteed r1bios in case of extreme VM load: 461da177e4SLinus Torvalds */ 471da177e4SLinus Torvalds #define NR_RAID1_BIOS 256 481da177e4SLinus Torvalds 49473e87ceSJonathan Brassow /* when we get a read error on a read-only array, we redirect to another 50473e87ceSJonathan Brassow * device without failing the first device, or trying to over-write to 51473e87ceSJonathan Brassow * correct the read error. To keep track of bad blocks on a per-bio 52473e87ceSJonathan Brassow * level, we store IO_BLOCKED in the appropriate 'bios' pointer 53473e87ceSJonathan Brassow */ 54473e87ceSJonathan Brassow #define IO_BLOCKED ((struct bio *)1) 55473e87ceSJonathan Brassow /* When we successfully write to a known bad-block, we need to remove the 56473e87ceSJonathan Brassow * bad-block marking which must be done from process context. So we record 57473e87ceSJonathan Brassow * the success by setting devs[n].bio to IO_MADE_GOOD 58473e87ceSJonathan Brassow */ 59473e87ceSJonathan Brassow #define IO_MADE_GOOD ((struct bio *)2) 60473e87ceSJonathan Brassow 61473e87ceSJonathan Brassow #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) 62473e87ceSJonathan Brassow 6334db0cd6SNeilBrown /* When there are this many requests queue to be written by 6434db0cd6SNeilBrown * the raid1 thread, we become 'congested' to provide back-pressure 6534db0cd6SNeilBrown * for writeback. 6634db0cd6SNeilBrown */ 6734db0cd6SNeilBrown static int max_queued_requests = 1024; 681da177e4SLinus Torvalds 6979ef3a8aSmajianpeng static void allow_barrier(struct r1conf *conf, sector_t start_next_window, 7079ef3a8aSmajianpeng sector_t bi_sector); 71e8096360SNeilBrown static void lower_barrier(struct r1conf *conf); 721da177e4SLinus Torvalds 73dd0fc66fSAl Viro static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) 741da177e4SLinus Torvalds { 751da177e4SLinus Torvalds struct pool_info *pi = data; 769f2c9d12SNeilBrown int size = offsetof(struct r1bio, bios[pi->raid_disks]); 771da177e4SLinus Torvalds 781da177e4SLinus Torvalds /* allocate a r1bio with room for raid_disks entries in the bios array */ 797eaceaccSJens Axboe return kzalloc(size, gfp_flags); 801da177e4SLinus Torvalds } 811da177e4SLinus Torvalds 821da177e4SLinus Torvalds static void r1bio_pool_free(void *r1_bio, void *data) 831da177e4SLinus Torvalds { 841da177e4SLinus Torvalds kfree(r1_bio); 851da177e4SLinus Torvalds } 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds #define RESYNC_BLOCK_SIZE (64*1024) 888e005f7cSmajianpeng #define RESYNC_DEPTH 32 891da177e4SLinus Torvalds #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) 901da177e4SLinus Torvalds #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) 918e005f7cSmajianpeng #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) 928e005f7cSmajianpeng #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) 938e005f7cSmajianpeng #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) 941da177e4SLinus Torvalds 95dd0fc66fSAl Viro static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) 961da177e4SLinus Torvalds { 971da177e4SLinus Torvalds struct pool_info *pi = data; 989f2c9d12SNeilBrown struct r1bio *r1_bio; 991da177e4SLinus Torvalds struct bio *bio; 1001da177e4SLinus Torvalds int i, j; 1011da177e4SLinus Torvalds 1021da177e4SLinus Torvalds r1_bio = r1bio_pool_alloc(gfp_flags, pi); 1037eaceaccSJens Axboe if (!r1_bio) 1041da177e4SLinus Torvalds return NULL; 1051da177e4SLinus Torvalds 1061da177e4SLinus Torvalds /* 1071da177e4SLinus Torvalds * Allocate bios : 1 for reading, n-1 for writing 1081da177e4SLinus Torvalds */ 1091da177e4SLinus Torvalds for (j = pi->raid_disks ; j-- ; ) { 1106746557fSNeilBrown bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); 1111da177e4SLinus Torvalds if (!bio) 1121da177e4SLinus Torvalds goto out_free_bio; 1131da177e4SLinus Torvalds r1_bio->bios[j] = bio; 1141da177e4SLinus Torvalds } 1151da177e4SLinus Torvalds /* 1161da177e4SLinus Torvalds * Allocate RESYNC_PAGES data pages and attach them to 117d11c171eSNeilBrown * the first bio. 118d11c171eSNeilBrown * If this is a user-requested check/repair, allocate 119d11c171eSNeilBrown * RESYNC_PAGES for each bio. 1201da177e4SLinus Torvalds */ 121d11c171eSNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) 122d11c171eSNeilBrown j = pi->raid_disks; 123d11c171eSNeilBrown else 124d11c171eSNeilBrown j = 1; 125d11c171eSNeilBrown while(j--) { 126d11c171eSNeilBrown bio = r1_bio->bios[j]; 127a0787606SKent Overstreet bio->bi_vcnt = RESYNC_PAGES; 1281da177e4SLinus Torvalds 129a0787606SKent Overstreet if (bio_alloc_pages(bio, gfp_flags)) 130a0787606SKent Overstreet goto out_free_bio; 131d11c171eSNeilBrown } 132d11c171eSNeilBrown /* If not user-requests, copy the page pointers to all bios */ 133d11c171eSNeilBrown if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { 134d11c171eSNeilBrown for (i=0; i<RESYNC_PAGES ; i++) 135d11c171eSNeilBrown for (j=1; j<pi->raid_disks; j++) 136d11c171eSNeilBrown r1_bio->bios[j]->bi_io_vec[i].bv_page = 137d11c171eSNeilBrown r1_bio->bios[0]->bi_io_vec[i].bv_page; 138d11c171eSNeilBrown } 1391da177e4SLinus Torvalds 1401da177e4SLinus Torvalds r1_bio->master_bio = NULL; 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds return r1_bio; 1431da177e4SLinus Torvalds 1441da177e4SLinus Torvalds out_free_bio: 1451da177e4SLinus Torvalds while (++j < pi->raid_disks) 1461da177e4SLinus Torvalds bio_put(r1_bio->bios[j]); 1471da177e4SLinus Torvalds r1bio_pool_free(r1_bio, data); 1481da177e4SLinus Torvalds return NULL; 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 1511da177e4SLinus Torvalds static void r1buf_pool_free(void *__r1_bio, void *data) 1521da177e4SLinus Torvalds { 1531da177e4SLinus Torvalds struct pool_info *pi = data; 154d11c171eSNeilBrown int i,j; 1559f2c9d12SNeilBrown struct r1bio *r1bio = __r1_bio; 1561da177e4SLinus Torvalds 157d11c171eSNeilBrown for (i = 0; i < RESYNC_PAGES; i++) 158d11c171eSNeilBrown for (j = pi->raid_disks; j-- ;) { 159d11c171eSNeilBrown if (j == 0 || 160d11c171eSNeilBrown r1bio->bios[j]->bi_io_vec[i].bv_page != 161d11c171eSNeilBrown r1bio->bios[0]->bi_io_vec[i].bv_page) 1621345b1d8SNeilBrown safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds for (i=0 ; i < pi->raid_disks; i++) 1651da177e4SLinus Torvalds bio_put(r1bio->bios[i]); 1661da177e4SLinus Torvalds 1671da177e4SLinus Torvalds r1bio_pool_free(r1bio, data); 1681da177e4SLinus Torvalds } 1691da177e4SLinus Torvalds 170e8096360SNeilBrown static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) 1711da177e4SLinus Torvalds { 1721da177e4SLinus Torvalds int i; 1731da177e4SLinus Torvalds 1748f19ccb2SNeilBrown for (i = 0; i < conf->raid_disks * 2; i++) { 1751da177e4SLinus Torvalds struct bio **bio = r1_bio->bios + i; 1764367af55SNeilBrown if (!BIO_SPECIAL(*bio)) 1771da177e4SLinus Torvalds bio_put(*bio); 1781da177e4SLinus Torvalds *bio = NULL; 1791da177e4SLinus Torvalds } 1801da177e4SLinus Torvalds } 1811da177e4SLinus Torvalds 1829f2c9d12SNeilBrown static void free_r1bio(struct r1bio *r1_bio) 1831da177e4SLinus Torvalds { 184e8096360SNeilBrown struct r1conf *conf = r1_bio->mddev->private; 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds put_all_bios(conf, r1_bio); 1871da177e4SLinus Torvalds mempool_free(r1_bio, conf->r1bio_pool); 1881da177e4SLinus Torvalds } 1891da177e4SLinus Torvalds 1909f2c9d12SNeilBrown static void put_buf(struct r1bio *r1_bio) 1911da177e4SLinus Torvalds { 192e8096360SNeilBrown struct r1conf *conf = r1_bio->mddev->private; 1933e198f78SNeilBrown int i; 1943e198f78SNeilBrown 1958f19ccb2SNeilBrown for (i = 0; i < conf->raid_disks * 2; i++) { 1963e198f78SNeilBrown struct bio *bio = r1_bio->bios[i]; 1973e198f78SNeilBrown if (bio->bi_end_io) 1983e198f78SNeilBrown rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); 1993e198f78SNeilBrown } 2001da177e4SLinus Torvalds 2011da177e4SLinus Torvalds mempool_free(r1_bio, conf->r1buf_pool); 2021da177e4SLinus Torvalds 20317999be4SNeilBrown lower_barrier(conf); 2041da177e4SLinus Torvalds } 2051da177e4SLinus Torvalds 2069f2c9d12SNeilBrown static void reschedule_retry(struct r1bio *r1_bio) 2071da177e4SLinus Torvalds { 2081da177e4SLinus Torvalds unsigned long flags; 209fd01b88cSNeilBrown struct mddev *mddev = r1_bio->mddev; 210e8096360SNeilBrown struct r1conf *conf = mddev->private; 2111da177e4SLinus Torvalds 2121da177e4SLinus Torvalds spin_lock_irqsave(&conf->device_lock, flags); 2131da177e4SLinus Torvalds list_add(&r1_bio->retry_list, &conf->retry_list); 214ddaf22abSNeilBrown conf->nr_queued ++; 2151da177e4SLinus Torvalds spin_unlock_irqrestore(&conf->device_lock, flags); 2161da177e4SLinus Torvalds 21717999be4SNeilBrown wake_up(&conf->wait_barrier); 2181da177e4SLinus Torvalds md_wakeup_thread(mddev->thread); 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds 2211da177e4SLinus Torvalds /* 2221da177e4SLinus Torvalds * raid_end_bio_io() is called when we have finished servicing a mirrored 2231da177e4SLinus Torvalds * operation and are ready to return a success/failure code to the buffer 2241da177e4SLinus Torvalds * cache layer. 2251da177e4SLinus Torvalds */ 2269f2c9d12SNeilBrown static void call_bio_endio(struct r1bio *r1_bio) 227d2eb35acSNeilBrown { 228d2eb35acSNeilBrown struct bio *bio = r1_bio->master_bio; 229d2eb35acSNeilBrown int done; 230e8096360SNeilBrown struct r1conf *conf = r1_bio->mddev->private; 23179ef3a8aSmajianpeng sector_t start_next_window = r1_bio->start_next_window; 2324f024f37SKent Overstreet sector_t bi_sector = bio->bi_iter.bi_sector; 233d2eb35acSNeilBrown 234d2eb35acSNeilBrown if (bio->bi_phys_segments) { 235d2eb35acSNeilBrown unsigned long flags; 236d2eb35acSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 237d2eb35acSNeilBrown bio->bi_phys_segments--; 238d2eb35acSNeilBrown done = (bio->bi_phys_segments == 0); 239d2eb35acSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 24079ef3a8aSmajianpeng /* 24179ef3a8aSmajianpeng * make_request() might be waiting for 24279ef3a8aSmajianpeng * bi_phys_segments to decrease 24379ef3a8aSmajianpeng */ 24479ef3a8aSmajianpeng wake_up(&conf->wait_barrier); 245d2eb35acSNeilBrown } else 246d2eb35acSNeilBrown done = 1; 247d2eb35acSNeilBrown 248d2eb35acSNeilBrown if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) 249d2eb35acSNeilBrown clear_bit(BIO_UPTODATE, &bio->bi_flags); 250d2eb35acSNeilBrown if (done) { 251d2eb35acSNeilBrown bio_endio(bio, 0); 252d2eb35acSNeilBrown /* 253d2eb35acSNeilBrown * Wake up any possible resync thread that waits for the device 254d2eb35acSNeilBrown * to go idle. 255d2eb35acSNeilBrown */ 25679ef3a8aSmajianpeng allow_barrier(conf, start_next_window, bi_sector); 257d2eb35acSNeilBrown } 258d2eb35acSNeilBrown } 259d2eb35acSNeilBrown 2609f2c9d12SNeilBrown static void raid_end_bio_io(struct r1bio *r1_bio) 2611da177e4SLinus Torvalds { 2621da177e4SLinus Torvalds struct bio *bio = r1_bio->master_bio; 2631da177e4SLinus Torvalds 2644b6d287fSNeilBrown /* if nobody has done the final endio yet, do it now */ 2654b6d287fSNeilBrown if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { 26636a4e1feSNeilBrown pr_debug("raid1: sync end %s on sectors %llu-%llu\n", 2674b6d287fSNeilBrown (bio_data_dir(bio) == WRITE) ? "write" : "read", 2684f024f37SKent Overstreet (unsigned long long) bio->bi_iter.bi_sector, 2694f024f37SKent Overstreet (unsigned long long) bio_end_sector(bio) - 1); 2704b6d287fSNeilBrown 271d2eb35acSNeilBrown call_bio_endio(r1_bio); 2724b6d287fSNeilBrown } 2731da177e4SLinus Torvalds free_r1bio(r1_bio); 2741da177e4SLinus Torvalds } 2751da177e4SLinus Torvalds 2761da177e4SLinus Torvalds /* 2771da177e4SLinus Torvalds * Update disk head position estimator based on IRQ completion info. 2781da177e4SLinus Torvalds */ 2799f2c9d12SNeilBrown static inline void update_head_pos(int disk, struct r1bio *r1_bio) 2801da177e4SLinus Torvalds { 281e8096360SNeilBrown struct r1conf *conf = r1_bio->mddev->private; 2821da177e4SLinus Torvalds 2831da177e4SLinus Torvalds conf->mirrors[disk].head_position = 2841da177e4SLinus Torvalds r1_bio->sector + (r1_bio->sectors); 2851da177e4SLinus Torvalds } 2861da177e4SLinus Torvalds 287ba3ae3beSNamhyung Kim /* 288ba3ae3beSNamhyung Kim * Find the disk number which triggered given bio 289ba3ae3beSNamhyung Kim */ 2909f2c9d12SNeilBrown static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) 291ba3ae3beSNamhyung Kim { 292ba3ae3beSNamhyung Kim int mirror; 29330194636SNeilBrown struct r1conf *conf = r1_bio->mddev->private; 29430194636SNeilBrown int raid_disks = conf->raid_disks; 295ba3ae3beSNamhyung Kim 2968f19ccb2SNeilBrown for (mirror = 0; mirror < raid_disks * 2; mirror++) 297ba3ae3beSNamhyung Kim if (r1_bio->bios[mirror] == bio) 298ba3ae3beSNamhyung Kim break; 299ba3ae3beSNamhyung Kim 3008f19ccb2SNeilBrown BUG_ON(mirror == raid_disks * 2); 301ba3ae3beSNamhyung Kim update_head_pos(mirror, r1_bio); 302ba3ae3beSNamhyung Kim 303ba3ae3beSNamhyung Kim return mirror; 304ba3ae3beSNamhyung Kim } 305ba3ae3beSNamhyung Kim 3066712ecf8SNeilBrown static void raid1_end_read_request(struct bio *bio, int error) 3071da177e4SLinus Torvalds { 3081da177e4SLinus Torvalds int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 3099f2c9d12SNeilBrown struct r1bio *r1_bio = bio->bi_private; 3101da177e4SLinus Torvalds int mirror; 311e8096360SNeilBrown struct r1conf *conf = r1_bio->mddev->private; 3121da177e4SLinus Torvalds 3131da177e4SLinus Torvalds mirror = r1_bio->read_disk; 3141da177e4SLinus Torvalds /* 3151da177e4SLinus Torvalds * this branch is our 'one mirror IO has finished' event handler: 3161da177e4SLinus Torvalds */ 317ddaf22abSNeilBrown update_head_pos(mirror, r1_bio); 318ddaf22abSNeilBrown 319220946c9SNeilBrown if (uptodate) 3201da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 321dd00a99eSNeilBrown else { 322dd00a99eSNeilBrown /* If all other devices have failed, we want to return 323dd00a99eSNeilBrown * the error upwards rather than fail the last device. 324dd00a99eSNeilBrown * Here we redefine "uptodate" to mean "Don't want to retry" 325dd00a99eSNeilBrown */ 326dd00a99eSNeilBrown unsigned long flags; 327dd00a99eSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 328dd00a99eSNeilBrown if (r1_bio->mddev->degraded == conf->raid_disks || 329dd00a99eSNeilBrown (r1_bio->mddev->degraded == conf->raid_disks-1 && 330dd00a99eSNeilBrown !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) 331dd00a99eSNeilBrown uptodate = 1; 332dd00a99eSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 333dd00a99eSNeilBrown } 3341da177e4SLinus Torvalds 3357ad4d4a6SNeilBrown if (uptodate) { 3361da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 3377ad4d4a6SNeilBrown rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); 3387ad4d4a6SNeilBrown } else { 3391da177e4SLinus Torvalds /* 3401da177e4SLinus Torvalds * oops, read error: 3411da177e4SLinus Torvalds */ 3421da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 3438bda470eSChristian Dietrich printk_ratelimited( 3448bda470eSChristian Dietrich KERN_ERR "md/raid1:%s: %s: " 3458bda470eSChristian Dietrich "rescheduling sector %llu\n", 3469dd1e2faSNeilBrown mdname(conf->mddev), 3478bda470eSChristian Dietrich bdevname(conf->mirrors[mirror].rdev->bdev, 3488bda470eSChristian Dietrich b), 3498bda470eSChristian Dietrich (unsigned long long)r1_bio->sector); 350d2eb35acSNeilBrown set_bit(R1BIO_ReadError, &r1_bio->state); 3511da177e4SLinus Torvalds reschedule_retry(r1_bio); 3527ad4d4a6SNeilBrown /* don't drop the reference on read_disk yet */ 3531da177e4SLinus Torvalds } 3541da177e4SLinus Torvalds } 3551da177e4SLinus Torvalds 3569f2c9d12SNeilBrown static void close_write(struct r1bio *r1_bio) 3574e78064fSNeilBrown { 3584e78064fSNeilBrown /* it really is the end of this request */ 3594e78064fSNeilBrown if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { 3604e78064fSNeilBrown /* free extra copy of the data pages */ 361af6d7b76SNeilBrown int i = r1_bio->behind_page_count; 3624e78064fSNeilBrown while (i--) 3632ca68f5eSNeilBrown safe_put_page(r1_bio->behind_bvecs[i].bv_page); 3642ca68f5eSNeilBrown kfree(r1_bio->behind_bvecs); 3652ca68f5eSNeilBrown r1_bio->behind_bvecs = NULL; 3664e78064fSNeilBrown } 3674e78064fSNeilBrown /* clear the bitmap if all writes complete successfully */ 3684e78064fSNeilBrown bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, 3694e78064fSNeilBrown r1_bio->sectors, 3704e78064fSNeilBrown !test_bit(R1BIO_Degraded, &r1_bio->state), 371af6d7b76SNeilBrown test_bit(R1BIO_BehindIO, &r1_bio->state)); 3724e78064fSNeilBrown md_write_end(r1_bio->mddev); 373cd5ff9a1SNeilBrown } 374cd5ff9a1SNeilBrown 3759f2c9d12SNeilBrown static void r1_bio_write_done(struct r1bio *r1_bio) 376cd5ff9a1SNeilBrown { 377cd5ff9a1SNeilBrown if (!atomic_dec_and_test(&r1_bio->remaining)) 378cd5ff9a1SNeilBrown return; 379cd5ff9a1SNeilBrown 380cd5ff9a1SNeilBrown if (test_bit(R1BIO_WriteError, &r1_bio->state)) 381cd5ff9a1SNeilBrown reschedule_retry(r1_bio); 382cd5ff9a1SNeilBrown else { 383cd5ff9a1SNeilBrown close_write(r1_bio); 3844367af55SNeilBrown if (test_bit(R1BIO_MadeGood, &r1_bio->state)) 3854367af55SNeilBrown reschedule_retry(r1_bio); 3864367af55SNeilBrown else 3874e78064fSNeilBrown raid_end_bio_io(r1_bio); 3884e78064fSNeilBrown } 3894e78064fSNeilBrown } 3904e78064fSNeilBrown 3916712ecf8SNeilBrown static void raid1_end_write_request(struct bio *bio, int error) 3921da177e4SLinus Torvalds { 3931da177e4SLinus Torvalds int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 3949f2c9d12SNeilBrown struct r1bio *r1_bio = bio->bi_private; 395a9701a30SNeilBrown int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); 396e8096360SNeilBrown struct r1conf *conf = r1_bio->mddev->private; 39704b857f7SNeilBrown struct bio *to_put = NULL; 3981da177e4SLinus Torvalds 399ba3ae3beSNamhyung Kim mirror = find_bio_disk(r1_bio, bio); 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds /* 402e9c7469bSTejun Heo * 'one mirror IO has finished' event handler: 4031da177e4SLinus Torvalds */ 404191ea9b2SNeilBrown if (!uptodate) { 405cd5ff9a1SNeilBrown set_bit(WriteErrorSeen, 406cd5ff9a1SNeilBrown &conf->mirrors[mirror].rdev->flags); 40719d67169SNeilBrown if (!test_and_set_bit(WantReplacement, 40819d67169SNeilBrown &conf->mirrors[mirror].rdev->flags)) 40919d67169SNeilBrown set_bit(MD_RECOVERY_NEEDED, & 41019d67169SNeilBrown conf->mddev->recovery); 41119d67169SNeilBrown 412cd5ff9a1SNeilBrown set_bit(R1BIO_WriteError, &r1_bio->state); 4134367af55SNeilBrown } else { 4141da177e4SLinus Torvalds /* 415e9c7469bSTejun Heo * Set R1BIO_Uptodate in our master bio, so that we 416e9c7469bSTejun Heo * will return a good error code for to the higher 417e9c7469bSTejun Heo * levels even if IO on some other mirrored buffer 418e9c7469bSTejun Heo * fails. 4191da177e4SLinus Torvalds * 420e9c7469bSTejun Heo * The 'master' represents the composite IO operation 421e9c7469bSTejun Heo * to user-side. So if something waits for IO, then it 422e9c7469bSTejun Heo * will wait for the 'master' bio. 4231da177e4SLinus Torvalds */ 4244367af55SNeilBrown sector_t first_bad; 4254367af55SNeilBrown int bad_sectors; 4264367af55SNeilBrown 427cd5ff9a1SNeilBrown r1_bio->bios[mirror] = NULL; 428cd5ff9a1SNeilBrown to_put = bio; 4293056e3aeSAlex Lyakas /* 4303056e3aeSAlex Lyakas * Do not set R1BIO_Uptodate if the current device is 4313056e3aeSAlex Lyakas * rebuilding or Faulty. This is because we cannot use 4323056e3aeSAlex Lyakas * such device for properly reading the data back (we could 4333056e3aeSAlex Lyakas * potentially use it, if the current write would have felt 4343056e3aeSAlex Lyakas * before rdev->recovery_offset, but for simplicity we don't 4353056e3aeSAlex Lyakas * check this here. 4363056e3aeSAlex Lyakas */ 4373056e3aeSAlex Lyakas if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) && 4383056e3aeSAlex Lyakas !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)) 4391da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 4401da177e4SLinus Torvalds 4414367af55SNeilBrown /* Maybe we can clear some bad blocks. */ 4424367af55SNeilBrown if (is_badblock(conf->mirrors[mirror].rdev, 4434367af55SNeilBrown r1_bio->sector, r1_bio->sectors, 4444367af55SNeilBrown &first_bad, &bad_sectors)) { 4454367af55SNeilBrown r1_bio->bios[mirror] = IO_MADE_GOOD; 4464367af55SNeilBrown set_bit(R1BIO_MadeGood, &r1_bio->state); 4474367af55SNeilBrown } 4484367af55SNeilBrown } 4494367af55SNeilBrown 4504b6d287fSNeilBrown if (behind) { 4514b6d287fSNeilBrown if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) 4524b6d287fSNeilBrown atomic_dec(&r1_bio->behind_remaining); 4534b6d287fSNeilBrown 454e9c7469bSTejun Heo /* 455e9c7469bSTejun Heo * In behind mode, we ACK the master bio once the I/O 456e9c7469bSTejun Heo * has safely reached all non-writemostly 457e9c7469bSTejun Heo * disks. Setting the Returned bit ensures that this 458e9c7469bSTejun Heo * gets done only once -- we don't ever want to return 459e9c7469bSTejun Heo * -EIO here, instead we'll wait 460e9c7469bSTejun Heo */ 4614b6d287fSNeilBrown if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && 4624b6d287fSNeilBrown test_bit(R1BIO_Uptodate, &r1_bio->state)) { 4634b6d287fSNeilBrown /* Maybe we can return now */ 4644b6d287fSNeilBrown if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { 4654b6d287fSNeilBrown struct bio *mbio = r1_bio->master_bio; 46636a4e1feSNeilBrown pr_debug("raid1: behind end write sectors" 46736a4e1feSNeilBrown " %llu-%llu\n", 4684f024f37SKent Overstreet (unsigned long long) mbio->bi_iter.bi_sector, 4694f024f37SKent Overstreet (unsigned long long) bio_end_sector(mbio) - 1); 470d2eb35acSNeilBrown call_bio_endio(r1_bio); 4714b6d287fSNeilBrown } 4724b6d287fSNeilBrown } 4734b6d287fSNeilBrown } 4744367af55SNeilBrown if (r1_bio->bios[mirror] == NULL) 4754367af55SNeilBrown rdev_dec_pending(conf->mirrors[mirror].rdev, 4764367af55SNeilBrown conf->mddev); 477e9c7469bSTejun Heo 4781da177e4SLinus Torvalds /* 4791da177e4SLinus Torvalds * Let's see if all mirrored write operations have finished 4801da177e4SLinus Torvalds * already. 4811da177e4SLinus Torvalds */ 482af6d7b76SNeilBrown r1_bio_write_done(r1_bio); 483c70810b3SNeilBrown 48404b857f7SNeilBrown if (to_put) 48504b857f7SNeilBrown bio_put(to_put); 4861da177e4SLinus Torvalds } 4871da177e4SLinus Torvalds 4881da177e4SLinus Torvalds 4891da177e4SLinus Torvalds /* 4901da177e4SLinus Torvalds * This routine returns the disk from which the requested read should 4911da177e4SLinus Torvalds * be done. There is a per-array 'next expected sequential IO' sector 4921da177e4SLinus Torvalds * number - if this matches on the next IO then we use the last disk. 4931da177e4SLinus Torvalds * There is also a per-disk 'last know head position' sector that is 4941da177e4SLinus Torvalds * maintained from IRQ contexts, both the normal and the resync IO 4951da177e4SLinus Torvalds * completion handlers update this position correctly. If there is no 4961da177e4SLinus Torvalds * perfect sequential match then we pick the disk whose head is closest. 4971da177e4SLinus Torvalds * 4981da177e4SLinus Torvalds * If there are 2 mirrors in the same 2 devices, performance degrades 4991da177e4SLinus Torvalds * because position is mirror, not device based. 5001da177e4SLinus Torvalds * 5011da177e4SLinus Torvalds * The rdev for the device selected will have nr_pending incremented. 5021da177e4SLinus Torvalds */ 503e8096360SNeilBrown static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) 5041da177e4SLinus Torvalds { 505af3a2cd6SNeilBrown const sector_t this_sector = r1_bio->sector; 506d2eb35acSNeilBrown int sectors; 507d2eb35acSNeilBrown int best_good_sectors; 5089dedf603SShaohua Li int best_disk, best_dist_disk, best_pending_disk; 5099dedf603SShaohua Li int has_nonrot_disk; 510be4d3280SShaohua Li int disk; 51176073054SNeilBrown sector_t best_dist; 5129dedf603SShaohua Li unsigned int min_pending; 5133cb03002SNeilBrown struct md_rdev *rdev; 514f3ac8bf7SNeilBrown int choose_first; 51512cee5a8SShaohua Li int choose_next_idle; 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds rcu_read_lock(); 5181da177e4SLinus Torvalds /* 5198ddf9efeSNeilBrown * Check if we can balance. We can balance on the whole 5201da177e4SLinus Torvalds * device if no resync is going on, or below the resync window. 5211da177e4SLinus Torvalds * We take the first readable disk when above the resync window. 5221da177e4SLinus Torvalds */ 5231da177e4SLinus Torvalds retry: 524d2eb35acSNeilBrown sectors = r1_bio->sectors; 52576073054SNeilBrown best_disk = -1; 5269dedf603SShaohua Li best_dist_disk = -1; 52776073054SNeilBrown best_dist = MaxSector; 5289dedf603SShaohua Li best_pending_disk = -1; 5299dedf603SShaohua Li min_pending = UINT_MAX; 530d2eb35acSNeilBrown best_good_sectors = 0; 5319dedf603SShaohua Li has_nonrot_disk = 0; 53212cee5a8SShaohua Li choose_next_idle = 0; 533d2eb35acSNeilBrown 5341da177e4SLinus Torvalds if (conf->mddev->recovery_cp < MaxSector && 535be4d3280SShaohua Li (this_sector + sectors >= conf->next_resync)) 536f3ac8bf7SNeilBrown choose_first = 1; 537be4d3280SShaohua Li else 538f3ac8bf7SNeilBrown choose_first = 0; 5391da177e4SLinus Torvalds 540be4d3280SShaohua Li for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { 54176073054SNeilBrown sector_t dist; 542d2eb35acSNeilBrown sector_t first_bad; 543d2eb35acSNeilBrown int bad_sectors; 5449dedf603SShaohua Li unsigned int pending; 54512cee5a8SShaohua Li bool nonrot; 546d2eb35acSNeilBrown 547f3ac8bf7SNeilBrown rdev = rcu_dereference(conf->mirrors[disk].rdev); 548f3ac8bf7SNeilBrown if (r1_bio->bios[disk] == IO_BLOCKED 549f3ac8bf7SNeilBrown || rdev == NULL 5506b740b8dSNeilBrown || test_bit(Unmerged, &rdev->flags) 55176073054SNeilBrown || test_bit(Faulty, &rdev->flags)) 552f3ac8bf7SNeilBrown continue; 55376073054SNeilBrown if (!test_bit(In_sync, &rdev->flags) && 55476073054SNeilBrown rdev->recovery_offset < this_sector + sectors) 55576073054SNeilBrown continue; 55676073054SNeilBrown if (test_bit(WriteMostly, &rdev->flags)) { 55776073054SNeilBrown /* Don't balance among write-mostly, just 55876073054SNeilBrown * use the first as a last resort */ 559307729c8SNeilBrown if (best_disk < 0) { 560307729c8SNeilBrown if (is_badblock(rdev, this_sector, sectors, 561307729c8SNeilBrown &first_bad, &bad_sectors)) { 562307729c8SNeilBrown if (first_bad < this_sector) 563307729c8SNeilBrown /* Cannot use this */ 564307729c8SNeilBrown continue; 565307729c8SNeilBrown best_good_sectors = first_bad - this_sector; 566307729c8SNeilBrown } else 567307729c8SNeilBrown best_good_sectors = sectors; 56876073054SNeilBrown best_disk = disk; 569307729c8SNeilBrown } 57076073054SNeilBrown continue; 5718ddf9efeSNeilBrown } 57276073054SNeilBrown /* This is a reasonable device to use. It might 57376073054SNeilBrown * even be best. 5741da177e4SLinus Torvalds */ 575d2eb35acSNeilBrown if (is_badblock(rdev, this_sector, sectors, 576d2eb35acSNeilBrown &first_bad, &bad_sectors)) { 577d2eb35acSNeilBrown if (best_dist < MaxSector) 578d2eb35acSNeilBrown /* already have a better device */ 579d2eb35acSNeilBrown continue; 580d2eb35acSNeilBrown if (first_bad <= this_sector) { 581d2eb35acSNeilBrown /* cannot read here. If this is the 'primary' 582d2eb35acSNeilBrown * device, then we must not read beyond 583d2eb35acSNeilBrown * bad_sectors from another device.. 584d2eb35acSNeilBrown */ 585d2eb35acSNeilBrown bad_sectors -= (this_sector - first_bad); 586d2eb35acSNeilBrown if (choose_first && sectors > bad_sectors) 587d2eb35acSNeilBrown sectors = bad_sectors; 588d2eb35acSNeilBrown if (best_good_sectors > sectors) 589d2eb35acSNeilBrown best_good_sectors = sectors; 590d2eb35acSNeilBrown 591d2eb35acSNeilBrown } else { 592d2eb35acSNeilBrown sector_t good_sectors = first_bad - this_sector; 593d2eb35acSNeilBrown if (good_sectors > best_good_sectors) { 594d2eb35acSNeilBrown best_good_sectors = good_sectors; 595d2eb35acSNeilBrown best_disk = disk; 596d2eb35acSNeilBrown } 597d2eb35acSNeilBrown if (choose_first) 598d2eb35acSNeilBrown break; 599d2eb35acSNeilBrown } 600d2eb35acSNeilBrown continue; 601d2eb35acSNeilBrown } else 602d2eb35acSNeilBrown best_good_sectors = sectors; 603d2eb35acSNeilBrown 60412cee5a8SShaohua Li nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); 60512cee5a8SShaohua Li has_nonrot_disk |= nonrot; 6069dedf603SShaohua Li pending = atomic_read(&rdev->nr_pending); 60776073054SNeilBrown dist = abs(this_sector - conf->mirrors[disk].head_position); 60812cee5a8SShaohua Li if (choose_first) { 60976073054SNeilBrown best_disk = disk; 6101da177e4SLinus Torvalds break; 6111da177e4SLinus Torvalds } 61212cee5a8SShaohua Li /* Don't change to another disk for sequential reads */ 61312cee5a8SShaohua Li if (conf->mirrors[disk].next_seq_sect == this_sector 61412cee5a8SShaohua Li || dist == 0) { 61512cee5a8SShaohua Li int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; 61612cee5a8SShaohua Li struct raid1_info *mirror = &conf->mirrors[disk]; 61712cee5a8SShaohua Li 61812cee5a8SShaohua Li best_disk = disk; 61912cee5a8SShaohua Li /* 62012cee5a8SShaohua Li * If buffered sequential IO size exceeds optimal 62112cee5a8SShaohua Li * iosize, check if there is idle disk. If yes, choose 62212cee5a8SShaohua Li * the idle disk. read_balance could already choose an 62312cee5a8SShaohua Li * idle disk before noticing it's a sequential IO in 62412cee5a8SShaohua Li * this disk. This doesn't matter because this disk 62512cee5a8SShaohua Li * will idle, next time it will be utilized after the 62612cee5a8SShaohua Li * first disk has IO size exceeds optimal iosize. In 62712cee5a8SShaohua Li * this way, iosize of the first disk will be optimal 62812cee5a8SShaohua Li * iosize at least. iosize of the second disk might be 62912cee5a8SShaohua Li * small, but not a big deal since when the second disk 63012cee5a8SShaohua Li * starts IO, the first disk is likely still busy. 63112cee5a8SShaohua Li */ 63212cee5a8SShaohua Li if (nonrot && opt_iosize > 0 && 63312cee5a8SShaohua Li mirror->seq_start != MaxSector && 63412cee5a8SShaohua Li mirror->next_seq_sect > opt_iosize && 63512cee5a8SShaohua Li mirror->next_seq_sect - opt_iosize >= 63612cee5a8SShaohua Li mirror->seq_start) { 63712cee5a8SShaohua Li choose_next_idle = 1; 63812cee5a8SShaohua Li continue; 63912cee5a8SShaohua Li } 64012cee5a8SShaohua Li break; 64112cee5a8SShaohua Li } 64212cee5a8SShaohua Li /* If device is idle, use it */ 64312cee5a8SShaohua Li if (pending == 0) { 64412cee5a8SShaohua Li best_disk = disk; 64512cee5a8SShaohua Li break; 64612cee5a8SShaohua Li } 64712cee5a8SShaohua Li 64812cee5a8SShaohua Li if (choose_next_idle) 64912cee5a8SShaohua Li continue; 6509dedf603SShaohua Li 6519dedf603SShaohua Li if (min_pending > pending) { 6529dedf603SShaohua Li min_pending = pending; 6539dedf603SShaohua Li best_pending_disk = disk; 6549dedf603SShaohua Li } 6559dedf603SShaohua Li 65676073054SNeilBrown if (dist < best_dist) { 65776073054SNeilBrown best_dist = dist; 6589dedf603SShaohua Li best_dist_disk = disk; 6591da177e4SLinus Torvalds } 660f3ac8bf7SNeilBrown } 6611da177e4SLinus Torvalds 6629dedf603SShaohua Li /* 6639dedf603SShaohua Li * If all disks are rotational, choose the closest disk. If any disk is 6649dedf603SShaohua Li * non-rotational, choose the disk with less pending request even the 6659dedf603SShaohua Li * disk is rotational, which might/might not be optimal for raids with 6669dedf603SShaohua Li * mixed ratation/non-rotational disks depending on workload. 6679dedf603SShaohua Li */ 6689dedf603SShaohua Li if (best_disk == -1) { 6699dedf603SShaohua Li if (has_nonrot_disk) 6709dedf603SShaohua Li best_disk = best_pending_disk; 6719dedf603SShaohua Li else 6729dedf603SShaohua Li best_disk = best_dist_disk; 6739dedf603SShaohua Li } 6749dedf603SShaohua Li 67576073054SNeilBrown if (best_disk >= 0) { 67676073054SNeilBrown rdev = rcu_dereference(conf->mirrors[best_disk].rdev); 6778ddf9efeSNeilBrown if (!rdev) 6788ddf9efeSNeilBrown goto retry; 6798ddf9efeSNeilBrown atomic_inc(&rdev->nr_pending); 68076073054SNeilBrown if (test_bit(Faulty, &rdev->flags)) { 6811da177e4SLinus Torvalds /* cannot risk returning a device that failed 6821da177e4SLinus Torvalds * before we inc'ed nr_pending 6831da177e4SLinus Torvalds */ 68403c902e1SNeilBrown rdev_dec_pending(rdev, conf->mddev); 6851da177e4SLinus Torvalds goto retry; 6861da177e4SLinus Torvalds } 687d2eb35acSNeilBrown sectors = best_good_sectors; 68812cee5a8SShaohua Li 68912cee5a8SShaohua Li if (conf->mirrors[best_disk].next_seq_sect != this_sector) 69012cee5a8SShaohua Li conf->mirrors[best_disk].seq_start = this_sector; 69112cee5a8SShaohua Li 692be4d3280SShaohua Li conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; 6931da177e4SLinus Torvalds } 6941da177e4SLinus Torvalds rcu_read_unlock(); 695d2eb35acSNeilBrown *max_sectors = sectors; 6961da177e4SLinus Torvalds 69776073054SNeilBrown return best_disk; 6981da177e4SLinus Torvalds } 6991da177e4SLinus Torvalds 7006b740b8dSNeilBrown static int raid1_mergeable_bvec(struct request_queue *q, 7016b740b8dSNeilBrown struct bvec_merge_data *bvm, 7026b740b8dSNeilBrown struct bio_vec *biovec) 7036b740b8dSNeilBrown { 7046b740b8dSNeilBrown struct mddev *mddev = q->queuedata; 7056b740b8dSNeilBrown struct r1conf *conf = mddev->private; 7066b740b8dSNeilBrown sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); 7076b740b8dSNeilBrown int max = biovec->bv_len; 7086b740b8dSNeilBrown 7096b740b8dSNeilBrown if (mddev->merge_check_needed) { 7106b740b8dSNeilBrown int disk; 7116b740b8dSNeilBrown rcu_read_lock(); 7126b740b8dSNeilBrown for (disk = 0; disk < conf->raid_disks * 2; disk++) { 7136b740b8dSNeilBrown struct md_rdev *rdev = rcu_dereference( 7146b740b8dSNeilBrown conf->mirrors[disk].rdev); 7156b740b8dSNeilBrown if (rdev && !test_bit(Faulty, &rdev->flags)) { 7166b740b8dSNeilBrown struct request_queue *q = 7176b740b8dSNeilBrown bdev_get_queue(rdev->bdev); 7186b740b8dSNeilBrown if (q->merge_bvec_fn) { 7196b740b8dSNeilBrown bvm->bi_sector = sector + 7206b740b8dSNeilBrown rdev->data_offset; 7216b740b8dSNeilBrown bvm->bi_bdev = rdev->bdev; 7226b740b8dSNeilBrown max = min(max, q->merge_bvec_fn( 7236b740b8dSNeilBrown q, bvm, biovec)); 7246b740b8dSNeilBrown } 7256b740b8dSNeilBrown } 7266b740b8dSNeilBrown } 7276b740b8dSNeilBrown rcu_read_unlock(); 7286b740b8dSNeilBrown } 7296b740b8dSNeilBrown return max; 7306b740b8dSNeilBrown 7316b740b8dSNeilBrown } 7326b740b8dSNeilBrown 733fd01b88cSNeilBrown int md_raid1_congested(struct mddev *mddev, int bits) 7340d129228SNeilBrown { 735e8096360SNeilBrown struct r1conf *conf = mddev->private; 7360d129228SNeilBrown int i, ret = 0; 7370d129228SNeilBrown 73834db0cd6SNeilBrown if ((bits & (1 << BDI_async_congested)) && 73934db0cd6SNeilBrown conf->pending_count >= max_queued_requests) 74034db0cd6SNeilBrown return 1; 74134db0cd6SNeilBrown 7420d129228SNeilBrown rcu_read_lock(); 743f53e29fcSNeilBrown for (i = 0; i < conf->raid_disks * 2; i++) { 7443cb03002SNeilBrown struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 7450d129228SNeilBrown if (rdev && !test_bit(Faulty, &rdev->flags)) { 746165125e1SJens Axboe struct request_queue *q = bdev_get_queue(rdev->bdev); 7470d129228SNeilBrown 7481ed7242eSJonathan Brassow BUG_ON(!q); 7491ed7242eSJonathan Brassow 7500d129228SNeilBrown /* Note the '|| 1' - when read_balance prefers 7510d129228SNeilBrown * non-congested targets, it can be removed 7520d129228SNeilBrown */ 75391a9e99dSAlexander Beregalov if ((bits & (1<<BDI_async_congested)) || 1) 7540d129228SNeilBrown ret |= bdi_congested(&q->backing_dev_info, bits); 7550d129228SNeilBrown else 7560d129228SNeilBrown ret &= bdi_congested(&q->backing_dev_info, bits); 7570d129228SNeilBrown } 7580d129228SNeilBrown } 7590d129228SNeilBrown rcu_read_unlock(); 7600d129228SNeilBrown return ret; 7610d129228SNeilBrown } 7621ed7242eSJonathan Brassow EXPORT_SYMBOL_GPL(md_raid1_congested); 7630d129228SNeilBrown 7641ed7242eSJonathan Brassow static int raid1_congested(void *data, int bits) 7651ed7242eSJonathan Brassow { 766fd01b88cSNeilBrown struct mddev *mddev = data; 7671ed7242eSJonathan Brassow 7681ed7242eSJonathan Brassow return mddev_congested(mddev, bits) || 7691ed7242eSJonathan Brassow md_raid1_congested(mddev, bits); 7701ed7242eSJonathan Brassow } 7710d129228SNeilBrown 772e8096360SNeilBrown static void flush_pending_writes(struct r1conf *conf) 773a35e63efSNeilBrown { 774a35e63efSNeilBrown /* Any writes that have been queued but are awaiting 775a35e63efSNeilBrown * bitmap updates get flushed here. 776a35e63efSNeilBrown */ 777a35e63efSNeilBrown spin_lock_irq(&conf->device_lock); 778a35e63efSNeilBrown 779a35e63efSNeilBrown if (conf->pending_bio_list.head) { 780a35e63efSNeilBrown struct bio *bio; 781a35e63efSNeilBrown bio = bio_list_get(&conf->pending_bio_list); 78234db0cd6SNeilBrown conf->pending_count = 0; 783a35e63efSNeilBrown spin_unlock_irq(&conf->device_lock); 784a35e63efSNeilBrown /* flush any pending bitmap writes to 785a35e63efSNeilBrown * disk before proceeding w/ I/O */ 786a35e63efSNeilBrown bitmap_unplug(conf->mddev->bitmap); 78734db0cd6SNeilBrown wake_up(&conf->wait_barrier); 788a35e63efSNeilBrown 789a35e63efSNeilBrown while (bio) { /* submit pending writes */ 790a35e63efSNeilBrown struct bio *next = bio->bi_next; 791a35e63efSNeilBrown bio->bi_next = NULL; 7922ff8cc2cSShaohua Li if (unlikely((bio->bi_rw & REQ_DISCARD) && 7932ff8cc2cSShaohua Li !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) 7942ff8cc2cSShaohua Li /* Just ignore it */ 7952ff8cc2cSShaohua Li bio_endio(bio, 0); 7962ff8cc2cSShaohua Li else 797a35e63efSNeilBrown generic_make_request(bio); 798a35e63efSNeilBrown bio = next; 799a35e63efSNeilBrown } 800a35e63efSNeilBrown } else 801a35e63efSNeilBrown spin_unlock_irq(&conf->device_lock); 8027eaceaccSJens Axboe } 8037eaceaccSJens Axboe 80417999be4SNeilBrown /* Barriers.... 80517999be4SNeilBrown * Sometimes we need to suspend IO while we do something else, 80617999be4SNeilBrown * either some resync/recovery, or reconfigure the array. 80717999be4SNeilBrown * To do this we raise a 'barrier'. 80817999be4SNeilBrown * The 'barrier' is a counter that can be raised multiple times 80917999be4SNeilBrown * to count how many activities are happening which preclude 81017999be4SNeilBrown * normal IO. 81117999be4SNeilBrown * We can only raise the barrier if there is no pending IO. 81217999be4SNeilBrown * i.e. if nr_pending == 0. 81317999be4SNeilBrown * We choose only to raise the barrier if no-one is waiting for the 81417999be4SNeilBrown * barrier to go down. This means that as soon as an IO request 81517999be4SNeilBrown * is ready, no other operations which require a barrier will start 81617999be4SNeilBrown * until the IO request has had a chance. 81717999be4SNeilBrown * 81817999be4SNeilBrown * So: regular IO calls 'wait_barrier'. When that returns there 81917999be4SNeilBrown * is no backgroup IO happening, It must arrange to call 82017999be4SNeilBrown * allow_barrier when it has finished its IO. 82117999be4SNeilBrown * backgroup IO calls must call raise_barrier. Once that returns 82217999be4SNeilBrown * there is no normal IO happeing. It must arrange to call 82317999be4SNeilBrown * lower_barrier when the particular background IO completes. 8241da177e4SLinus Torvalds */ 825e8096360SNeilBrown static void raise_barrier(struct r1conf *conf) 8261da177e4SLinus Torvalds { 8271da177e4SLinus Torvalds spin_lock_irq(&conf->resync_lock); 8281da177e4SLinus Torvalds 82917999be4SNeilBrown /* Wait until no block IO is waiting */ 83017999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, 831eed8c02eSLukas Czerner conf->resync_lock); 83217999be4SNeilBrown 83317999be4SNeilBrown /* block any new IO from starting */ 83417999be4SNeilBrown conf->barrier++; 83517999be4SNeilBrown 83679ef3a8aSmajianpeng /* For these conditions we must wait: 83779ef3a8aSmajianpeng * A: while the array is in frozen state 83879ef3a8aSmajianpeng * B: while barrier >= RESYNC_DEPTH, meaning resync reach 83979ef3a8aSmajianpeng * the max count which allowed. 84079ef3a8aSmajianpeng * C: next_resync + RESYNC_SECTORS > start_next_window, meaning 84179ef3a8aSmajianpeng * next resync will reach to the window which normal bios are 84279ef3a8aSmajianpeng * handling. 84379ef3a8aSmajianpeng */ 84417999be4SNeilBrown wait_event_lock_irq(conf->wait_barrier, 845b364e3d0Smajianpeng !conf->array_frozen && 84679ef3a8aSmajianpeng conf->barrier < RESYNC_DEPTH && 84779ef3a8aSmajianpeng (conf->start_next_window >= 84879ef3a8aSmajianpeng conf->next_resync + RESYNC_SECTORS), 849eed8c02eSLukas Czerner conf->resync_lock); 85017999be4SNeilBrown 8511da177e4SLinus Torvalds spin_unlock_irq(&conf->resync_lock); 8521da177e4SLinus Torvalds } 8531da177e4SLinus Torvalds 854e8096360SNeilBrown static void lower_barrier(struct r1conf *conf) 85517999be4SNeilBrown { 85617999be4SNeilBrown unsigned long flags; 857709ae487SNeilBrown BUG_ON(conf->barrier <= 0); 85817999be4SNeilBrown spin_lock_irqsave(&conf->resync_lock, flags); 85917999be4SNeilBrown conf->barrier--; 86017999be4SNeilBrown spin_unlock_irqrestore(&conf->resync_lock, flags); 86117999be4SNeilBrown wake_up(&conf->wait_barrier); 86217999be4SNeilBrown } 86317999be4SNeilBrown 86479ef3a8aSmajianpeng static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) 86517999be4SNeilBrown { 86679ef3a8aSmajianpeng bool wait = false; 86779ef3a8aSmajianpeng 86879ef3a8aSmajianpeng if (conf->array_frozen || !bio) 86979ef3a8aSmajianpeng wait = true; 87079ef3a8aSmajianpeng else if (conf->barrier && bio_data_dir(bio) == WRITE) { 87179ef3a8aSmajianpeng if (conf->next_resync < RESYNC_WINDOW_SECTORS) 87279ef3a8aSmajianpeng wait = true; 87379ef3a8aSmajianpeng else if ((conf->next_resync - RESYNC_WINDOW_SECTORS 87479ef3a8aSmajianpeng >= bio_end_sector(bio)) || 87579ef3a8aSmajianpeng (conf->next_resync + NEXT_NORMALIO_DISTANCE 8764f024f37SKent Overstreet <= bio->bi_iter.bi_sector)) 87779ef3a8aSmajianpeng wait = false; 87879ef3a8aSmajianpeng else 87979ef3a8aSmajianpeng wait = true; 88079ef3a8aSmajianpeng } 88179ef3a8aSmajianpeng 88279ef3a8aSmajianpeng return wait; 88379ef3a8aSmajianpeng } 88479ef3a8aSmajianpeng 88579ef3a8aSmajianpeng static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) 88679ef3a8aSmajianpeng { 88779ef3a8aSmajianpeng sector_t sector = 0; 88879ef3a8aSmajianpeng 88917999be4SNeilBrown spin_lock_irq(&conf->resync_lock); 89079ef3a8aSmajianpeng if (need_to_wait_for_sync(conf, bio)) { 89117999be4SNeilBrown conf->nr_waiting++; 892d6b42dcbSNeilBrown /* Wait for the barrier to drop. 893d6b42dcbSNeilBrown * However if there are already pending 894d6b42dcbSNeilBrown * requests (preventing the barrier from 895d6b42dcbSNeilBrown * rising completely), and the 896d6b42dcbSNeilBrown * pre-process bio queue isn't empty, 897d6b42dcbSNeilBrown * then don't wait, as we need to empty 898d6b42dcbSNeilBrown * that queue to get the nr_pending 899d6b42dcbSNeilBrown * count down. 900d6b42dcbSNeilBrown */ 901d6b42dcbSNeilBrown wait_event_lock_irq(conf->wait_barrier, 902b364e3d0Smajianpeng !conf->array_frozen && 903b364e3d0Smajianpeng (!conf->barrier || 90479ef3a8aSmajianpeng ((conf->start_next_window < 90579ef3a8aSmajianpeng conf->next_resync + RESYNC_SECTORS) && 906d6b42dcbSNeilBrown current->bio_list && 907b364e3d0Smajianpeng !bio_list_empty(current->bio_list))), 908eed8c02eSLukas Czerner conf->resync_lock); 90917999be4SNeilBrown conf->nr_waiting--; 91017999be4SNeilBrown } 91179ef3a8aSmajianpeng 91279ef3a8aSmajianpeng if (bio && bio_data_dir(bio) == WRITE) { 91379ef3a8aSmajianpeng if (conf->next_resync + NEXT_NORMALIO_DISTANCE 9144f024f37SKent Overstreet <= bio->bi_iter.bi_sector) { 91579ef3a8aSmajianpeng if (conf->start_next_window == MaxSector) 91679ef3a8aSmajianpeng conf->start_next_window = 91779ef3a8aSmajianpeng conf->next_resync + 91879ef3a8aSmajianpeng NEXT_NORMALIO_DISTANCE; 91979ef3a8aSmajianpeng 92079ef3a8aSmajianpeng if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) 9214f024f37SKent Overstreet <= bio->bi_iter.bi_sector) 92279ef3a8aSmajianpeng conf->next_window_requests++; 92379ef3a8aSmajianpeng else 92479ef3a8aSmajianpeng conf->current_window_requests++; 92579ef3a8aSmajianpeng } 9264f024f37SKent Overstreet if (bio->bi_iter.bi_sector >= conf->start_next_window) 92779ef3a8aSmajianpeng sector = conf->start_next_window; 92817999be4SNeilBrown } 92917999be4SNeilBrown 93079ef3a8aSmajianpeng conf->nr_pending++; 93179ef3a8aSmajianpeng spin_unlock_irq(&conf->resync_lock); 93279ef3a8aSmajianpeng return sector; 93379ef3a8aSmajianpeng } 93479ef3a8aSmajianpeng 93579ef3a8aSmajianpeng static void allow_barrier(struct r1conf *conf, sector_t start_next_window, 93679ef3a8aSmajianpeng sector_t bi_sector) 93717999be4SNeilBrown { 93817999be4SNeilBrown unsigned long flags; 93979ef3a8aSmajianpeng 94017999be4SNeilBrown spin_lock_irqsave(&conf->resync_lock, flags); 94117999be4SNeilBrown conf->nr_pending--; 94279ef3a8aSmajianpeng if (start_next_window) { 94379ef3a8aSmajianpeng if (start_next_window == conf->start_next_window) { 94479ef3a8aSmajianpeng if (conf->start_next_window + NEXT_NORMALIO_DISTANCE 94579ef3a8aSmajianpeng <= bi_sector) 94679ef3a8aSmajianpeng conf->next_window_requests--; 94779ef3a8aSmajianpeng else 94879ef3a8aSmajianpeng conf->current_window_requests--; 94979ef3a8aSmajianpeng } else 95079ef3a8aSmajianpeng conf->current_window_requests--; 95179ef3a8aSmajianpeng 95279ef3a8aSmajianpeng if (!conf->current_window_requests) { 95379ef3a8aSmajianpeng if (conf->next_window_requests) { 95479ef3a8aSmajianpeng conf->current_window_requests = 95579ef3a8aSmajianpeng conf->next_window_requests; 95679ef3a8aSmajianpeng conf->next_window_requests = 0; 95779ef3a8aSmajianpeng conf->start_next_window += 95879ef3a8aSmajianpeng NEXT_NORMALIO_DISTANCE; 95979ef3a8aSmajianpeng } else 96079ef3a8aSmajianpeng conf->start_next_window = MaxSector; 96179ef3a8aSmajianpeng } 96279ef3a8aSmajianpeng } 96317999be4SNeilBrown spin_unlock_irqrestore(&conf->resync_lock, flags); 96417999be4SNeilBrown wake_up(&conf->wait_barrier); 96517999be4SNeilBrown } 96617999be4SNeilBrown 967e2d59925SNeilBrown static void freeze_array(struct r1conf *conf, int extra) 968ddaf22abSNeilBrown { 969ddaf22abSNeilBrown /* stop syncio and normal IO and wait for everything to 970ddaf22abSNeilBrown * go quite. 971b364e3d0Smajianpeng * We wait until nr_pending match nr_queued+extra 9721c830532SNeilBrown * This is called in the context of one normal IO request 9731c830532SNeilBrown * that has failed. Thus any sync request that might be pending 9741c830532SNeilBrown * will be blocked by nr_pending, and we need to wait for 9751c830532SNeilBrown * pending IO requests to complete or be queued for re-try. 976e2d59925SNeilBrown * Thus the number queued (nr_queued) plus this request (extra) 9771c830532SNeilBrown * must match the number of pending IOs (nr_pending) before 9781c830532SNeilBrown * we continue. 979ddaf22abSNeilBrown */ 980ddaf22abSNeilBrown spin_lock_irq(&conf->resync_lock); 981b364e3d0Smajianpeng conf->array_frozen = 1; 982eed8c02eSLukas Czerner wait_event_lock_irq_cmd(conf->wait_barrier, 983e2d59925SNeilBrown conf->nr_pending == conf->nr_queued+extra, 984ddaf22abSNeilBrown conf->resync_lock, 985c3b328acSNeilBrown flush_pending_writes(conf)); 986ddaf22abSNeilBrown spin_unlock_irq(&conf->resync_lock); 987ddaf22abSNeilBrown } 988e8096360SNeilBrown static void unfreeze_array(struct r1conf *conf) 989ddaf22abSNeilBrown { 990ddaf22abSNeilBrown /* reverse the effect of the freeze */ 991ddaf22abSNeilBrown spin_lock_irq(&conf->resync_lock); 992b364e3d0Smajianpeng conf->array_frozen = 0; 993ddaf22abSNeilBrown wake_up(&conf->wait_barrier); 994ddaf22abSNeilBrown spin_unlock_irq(&conf->resync_lock); 995ddaf22abSNeilBrown } 996ddaf22abSNeilBrown 99717999be4SNeilBrown 9984e78064fSNeilBrown /* duplicate the data pages for behind I/O 9994e78064fSNeilBrown */ 10009f2c9d12SNeilBrown static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) 10014b6d287fSNeilBrown { 10024b6d287fSNeilBrown int i; 10034b6d287fSNeilBrown struct bio_vec *bvec; 10042ca68f5eSNeilBrown struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), 10054b6d287fSNeilBrown GFP_NOIO); 10062ca68f5eSNeilBrown if (unlikely(!bvecs)) 1007af6d7b76SNeilBrown return; 10084b6d287fSNeilBrown 1009cb34e057SKent Overstreet bio_for_each_segment_all(bvec, bio, i) { 10102ca68f5eSNeilBrown bvecs[i] = *bvec; 10112ca68f5eSNeilBrown bvecs[i].bv_page = alloc_page(GFP_NOIO); 10122ca68f5eSNeilBrown if (unlikely(!bvecs[i].bv_page)) 10134b6d287fSNeilBrown goto do_sync_io; 10142ca68f5eSNeilBrown memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, 10154b6d287fSNeilBrown kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); 10162ca68f5eSNeilBrown kunmap(bvecs[i].bv_page); 10174b6d287fSNeilBrown kunmap(bvec->bv_page); 10184b6d287fSNeilBrown } 10192ca68f5eSNeilBrown r1_bio->behind_bvecs = bvecs; 1020af6d7b76SNeilBrown r1_bio->behind_page_count = bio->bi_vcnt; 1021af6d7b76SNeilBrown set_bit(R1BIO_BehindIO, &r1_bio->state); 1022af6d7b76SNeilBrown return; 10234b6d287fSNeilBrown 10244b6d287fSNeilBrown do_sync_io: 1025af6d7b76SNeilBrown for (i = 0; i < bio->bi_vcnt; i++) 10262ca68f5eSNeilBrown if (bvecs[i].bv_page) 10272ca68f5eSNeilBrown put_page(bvecs[i].bv_page); 10282ca68f5eSNeilBrown kfree(bvecs); 10294f024f37SKent Overstreet pr_debug("%dB behind alloc failed, doing sync I/O\n", 10304f024f37SKent Overstreet bio->bi_iter.bi_size); 10314b6d287fSNeilBrown } 10324b6d287fSNeilBrown 1033f54a9d0eSNeilBrown struct raid1_plug_cb { 1034f54a9d0eSNeilBrown struct blk_plug_cb cb; 1035f54a9d0eSNeilBrown struct bio_list pending; 1036f54a9d0eSNeilBrown int pending_cnt; 1037f54a9d0eSNeilBrown }; 1038f54a9d0eSNeilBrown 1039f54a9d0eSNeilBrown static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) 1040f54a9d0eSNeilBrown { 1041f54a9d0eSNeilBrown struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, 1042f54a9d0eSNeilBrown cb); 1043f54a9d0eSNeilBrown struct mddev *mddev = plug->cb.data; 1044f54a9d0eSNeilBrown struct r1conf *conf = mddev->private; 1045f54a9d0eSNeilBrown struct bio *bio; 1046f54a9d0eSNeilBrown 1047874807a8SNeilBrown if (from_schedule || current->bio_list) { 1048f54a9d0eSNeilBrown spin_lock_irq(&conf->device_lock); 1049f54a9d0eSNeilBrown bio_list_merge(&conf->pending_bio_list, &plug->pending); 1050f54a9d0eSNeilBrown conf->pending_count += plug->pending_cnt; 1051f54a9d0eSNeilBrown spin_unlock_irq(&conf->device_lock); 1052ee0b0244SNeilBrown wake_up(&conf->wait_barrier); 1053f54a9d0eSNeilBrown md_wakeup_thread(mddev->thread); 1054f54a9d0eSNeilBrown kfree(plug); 1055f54a9d0eSNeilBrown return; 1056f54a9d0eSNeilBrown } 1057f54a9d0eSNeilBrown 1058f54a9d0eSNeilBrown /* we aren't scheduling, so we can do the write-out directly. */ 1059f54a9d0eSNeilBrown bio = bio_list_get(&plug->pending); 1060f54a9d0eSNeilBrown bitmap_unplug(mddev->bitmap); 1061f54a9d0eSNeilBrown wake_up(&conf->wait_barrier); 1062f54a9d0eSNeilBrown 1063f54a9d0eSNeilBrown while (bio) { /* submit pending writes */ 1064f54a9d0eSNeilBrown struct bio *next = bio->bi_next; 1065f54a9d0eSNeilBrown bio->bi_next = NULL; 106632f9f570SShaohua Li if (unlikely((bio->bi_rw & REQ_DISCARD) && 106732f9f570SShaohua Li !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) 106832f9f570SShaohua Li /* Just ignore it */ 106932f9f570SShaohua Li bio_endio(bio, 0); 107032f9f570SShaohua Li else 1071f54a9d0eSNeilBrown generic_make_request(bio); 1072f54a9d0eSNeilBrown bio = next; 1073f54a9d0eSNeilBrown } 1074f54a9d0eSNeilBrown kfree(plug); 1075f54a9d0eSNeilBrown } 1076f54a9d0eSNeilBrown 1077b4fdcb02SLinus Torvalds static void make_request(struct mddev *mddev, struct bio * bio) 10781da177e4SLinus Torvalds { 1079e8096360SNeilBrown struct r1conf *conf = mddev->private; 10800eaf822cSJonathan Brassow struct raid1_info *mirror; 10819f2c9d12SNeilBrown struct r1bio *r1_bio; 10821da177e4SLinus Torvalds struct bio *read_bio; 10831f68f0c4SNeilBrown int i, disks; 108484255d10SNeilBrown struct bitmap *bitmap; 1085191ea9b2SNeilBrown unsigned long flags; 1086a362357bSJens Axboe const int rw = bio_data_dir(bio); 10872c7d46ecSNeilBrown const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); 1088e9c7469bSTejun Heo const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); 10892ff8cc2cSShaohua Li const unsigned long do_discard = (bio->bi_rw 10902ff8cc2cSShaohua Li & (REQ_DISCARD | REQ_SECURE)); 1091c8dc9c65SJoe Lawrence const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME); 10923cb03002SNeilBrown struct md_rdev *blocked_rdev; 1093f54a9d0eSNeilBrown struct blk_plug_cb *cb; 1094f54a9d0eSNeilBrown struct raid1_plug_cb *plug = NULL; 10951f68f0c4SNeilBrown int first_clone; 10961f68f0c4SNeilBrown int sectors_handled; 10971f68f0c4SNeilBrown int max_sectors; 109879ef3a8aSmajianpeng sector_t start_next_window; 1099191ea9b2SNeilBrown 11001da177e4SLinus Torvalds /* 11011da177e4SLinus Torvalds * Register the new request and wait if the reconstruction 11021da177e4SLinus Torvalds * thread has put up a bar for new requests. 11031da177e4SLinus Torvalds * Continue immediately if no resync is active currently. 11041da177e4SLinus Torvalds */ 110562de608dSNeilBrown 11063d310eb7SNeilBrown md_write_start(mddev, bio); /* wait on superblock update early */ 11073d310eb7SNeilBrown 11086eef4b21SNeilBrown if (bio_data_dir(bio) == WRITE && 1109f73a1c7dSKent Overstreet bio_end_sector(bio) > mddev->suspend_lo && 11104f024f37SKent Overstreet bio->bi_iter.bi_sector < mddev->suspend_hi) { 11116eef4b21SNeilBrown /* As the suspend_* range is controlled by 11126eef4b21SNeilBrown * userspace, we want an interruptible 11136eef4b21SNeilBrown * wait. 11146eef4b21SNeilBrown */ 11156eef4b21SNeilBrown DEFINE_WAIT(w); 11166eef4b21SNeilBrown for (;;) { 11176eef4b21SNeilBrown flush_signals(current); 11186eef4b21SNeilBrown prepare_to_wait(&conf->wait_barrier, 11196eef4b21SNeilBrown &w, TASK_INTERRUPTIBLE); 1120f73a1c7dSKent Overstreet if (bio_end_sector(bio) <= mddev->suspend_lo || 11214f024f37SKent Overstreet bio->bi_iter.bi_sector >= mddev->suspend_hi) 11226eef4b21SNeilBrown break; 11236eef4b21SNeilBrown schedule(); 11246eef4b21SNeilBrown } 11256eef4b21SNeilBrown finish_wait(&conf->wait_barrier, &w); 11266eef4b21SNeilBrown } 112762de608dSNeilBrown 112879ef3a8aSmajianpeng start_next_window = wait_barrier(conf, bio); 11291da177e4SLinus Torvalds 113084255d10SNeilBrown bitmap = mddev->bitmap; 113184255d10SNeilBrown 11321da177e4SLinus Torvalds /* 11331da177e4SLinus Torvalds * make_request() can abort the operation when READA is being 11341da177e4SLinus Torvalds * used and no empty request is available. 11351da177e4SLinus Torvalds * 11361da177e4SLinus Torvalds */ 11371da177e4SLinus Torvalds r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); 11381da177e4SLinus Torvalds 11391da177e4SLinus Torvalds r1_bio->master_bio = bio; 1140aa8b57aaSKent Overstreet r1_bio->sectors = bio_sectors(bio); 1141191ea9b2SNeilBrown r1_bio->state = 0; 11421da177e4SLinus Torvalds r1_bio->mddev = mddev; 11434f024f37SKent Overstreet r1_bio->sector = bio->bi_iter.bi_sector; 11441da177e4SLinus Torvalds 1145d2eb35acSNeilBrown /* We might need to issue multiple reads to different 1146d2eb35acSNeilBrown * devices if there are bad blocks around, so we keep 1147d2eb35acSNeilBrown * track of the number of reads in bio->bi_phys_segments. 1148d2eb35acSNeilBrown * If this is 0, there is only one r1_bio and no locking 1149d2eb35acSNeilBrown * will be needed when requests complete. If it is 1150d2eb35acSNeilBrown * non-zero, then it is the number of not-completed requests. 1151d2eb35acSNeilBrown */ 1152d2eb35acSNeilBrown bio->bi_phys_segments = 0; 1153d2eb35acSNeilBrown clear_bit(BIO_SEG_VALID, &bio->bi_flags); 1154d2eb35acSNeilBrown 1155a362357bSJens Axboe if (rw == READ) { 11561da177e4SLinus Torvalds /* 11571da177e4SLinus Torvalds * read balancing logic: 11581da177e4SLinus Torvalds */ 1159d2eb35acSNeilBrown int rdisk; 1160d2eb35acSNeilBrown 1161d2eb35acSNeilBrown read_again: 1162d2eb35acSNeilBrown rdisk = read_balance(conf, r1_bio, &max_sectors); 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds if (rdisk < 0) { 11651da177e4SLinus Torvalds /* couldn't find anywhere to read from */ 11661da177e4SLinus Torvalds raid_end_bio_io(r1_bio); 11675a7bbad2SChristoph Hellwig return; 11681da177e4SLinus Torvalds } 11691da177e4SLinus Torvalds mirror = conf->mirrors + rdisk; 11701da177e4SLinus Torvalds 1171e555190dSNeilBrown if (test_bit(WriteMostly, &mirror->rdev->flags) && 1172e555190dSNeilBrown bitmap) { 1173e555190dSNeilBrown /* Reading from a write-mostly device must 1174e555190dSNeilBrown * take care not to over-take any writes 1175e555190dSNeilBrown * that are 'behind' 1176e555190dSNeilBrown */ 1177e555190dSNeilBrown wait_event(bitmap->behind_wait, 1178e555190dSNeilBrown atomic_read(&bitmap->behind_writes) == 0); 1179e555190dSNeilBrown } 11801da177e4SLinus Torvalds r1_bio->read_disk = rdisk; 11811da177e4SLinus Torvalds 1182a167f663SNeilBrown read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); 11834f024f37SKent Overstreet bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, 1184d2eb35acSNeilBrown max_sectors); 11851da177e4SLinus Torvalds 11861da177e4SLinus Torvalds r1_bio->bios[rdisk] = read_bio; 11871da177e4SLinus Torvalds 11884f024f37SKent Overstreet read_bio->bi_iter.bi_sector = r1_bio->sector + 11894f024f37SKent Overstreet mirror->rdev->data_offset; 11901da177e4SLinus Torvalds read_bio->bi_bdev = mirror->rdev->bdev; 11911da177e4SLinus Torvalds read_bio->bi_end_io = raid1_end_read_request; 11927b6d91daSChristoph Hellwig read_bio->bi_rw = READ | do_sync; 11931da177e4SLinus Torvalds read_bio->bi_private = r1_bio; 11941da177e4SLinus Torvalds 1195d2eb35acSNeilBrown if (max_sectors < r1_bio->sectors) { 1196d2eb35acSNeilBrown /* could not read all from this device, so we will 1197d2eb35acSNeilBrown * need another r1_bio. 1198d2eb35acSNeilBrown */ 1199d2eb35acSNeilBrown 1200d2eb35acSNeilBrown sectors_handled = (r1_bio->sector + max_sectors 12014f024f37SKent Overstreet - bio->bi_iter.bi_sector); 1202d2eb35acSNeilBrown r1_bio->sectors = max_sectors; 1203d2eb35acSNeilBrown spin_lock_irq(&conf->device_lock); 1204d2eb35acSNeilBrown if (bio->bi_phys_segments == 0) 1205d2eb35acSNeilBrown bio->bi_phys_segments = 2; 1206d2eb35acSNeilBrown else 1207d2eb35acSNeilBrown bio->bi_phys_segments++; 1208d2eb35acSNeilBrown spin_unlock_irq(&conf->device_lock); 1209d2eb35acSNeilBrown /* Cannot call generic_make_request directly 1210d2eb35acSNeilBrown * as that will be queued in __make_request 1211d2eb35acSNeilBrown * and subsequent mempool_alloc might block waiting 1212d2eb35acSNeilBrown * for it. So hand bio over to raid1d. 1213d2eb35acSNeilBrown */ 1214d2eb35acSNeilBrown reschedule_retry(r1_bio); 1215d2eb35acSNeilBrown 1216d2eb35acSNeilBrown r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); 1217d2eb35acSNeilBrown 1218d2eb35acSNeilBrown r1_bio->master_bio = bio; 1219aa8b57aaSKent Overstreet r1_bio->sectors = bio_sectors(bio) - sectors_handled; 1220d2eb35acSNeilBrown r1_bio->state = 0; 1221d2eb35acSNeilBrown r1_bio->mddev = mddev; 12224f024f37SKent Overstreet r1_bio->sector = bio->bi_iter.bi_sector + 12234f024f37SKent Overstreet sectors_handled; 1224d2eb35acSNeilBrown goto read_again; 1225d2eb35acSNeilBrown } else 12261da177e4SLinus Torvalds generic_make_request(read_bio); 12275a7bbad2SChristoph Hellwig return; 12281da177e4SLinus Torvalds } 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds /* 12311da177e4SLinus Torvalds * WRITE: 12321da177e4SLinus Torvalds */ 123334db0cd6SNeilBrown if (conf->pending_count >= max_queued_requests) { 123434db0cd6SNeilBrown md_wakeup_thread(mddev->thread); 123534db0cd6SNeilBrown wait_event(conf->wait_barrier, 123634db0cd6SNeilBrown conf->pending_count < max_queued_requests); 123734db0cd6SNeilBrown } 12381f68f0c4SNeilBrown /* first select target devices under rcu_lock and 12391da177e4SLinus Torvalds * inc refcount on their rdev. Record them by setting 12401da177e4SLinus Torvalds * bios[x] to bio 12411f68f0c4SNeilBrown * If there are known/acknowledged bad blocks on any device on 12421f68f0c4SNeilBrown * which we have seen a write error, we want to avoid writing those 12431f68f0c4SNeilBrown * blocks. 12441f68f0c4SNeilBrown * This potentially requires several writes to write around 12451f68f0c4SNeilBrown * the bad blocks. Each set of writes gets it's own r1bio 12461f68f0c4SNeilBrown * with a set of bios attached. 12471da177e4SLinus Torvalds */ 1248c3b328acSNeilBrown 12498f19ccb2SNeilBrown disks = conf->raid_disks * 2; 12506bfe0b49SDan Williams retry_write: 125179ef3a8aSmajianpeng r1_bio->start_next_window = start_next_window; 12526bfe0b49SDan Williams blocked_rdev = NULL; 12531da177e4SLinus Torvalds rcu_read_lock(); 12541f68f0c4SNeilBrown max_sectors = r1_bio->sectors; 12551da177e4SLinus Torvalds for (i = 0; i < disks; i++) { 12563cb03002SNeilBrown struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 12576bfe0b49SDan Williams if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { 12586bfe0b49SDan Williams atomic_inc(&rdev->nr_pending); 12596bfe0b49SDan Williams blocked_rdev = rdev; 12606bfe0b49SDan Williams break; 12616bfe0b49SDan Williams } 12621da177e4SLinus Torvalds r1_bio->bios[i] = NULL; 12636b740b8dSNeilBrown if (!rdev || test_bit(Faulty, &rdev->flags) 12646b740b8dSNeilBrown || test_bit(Unmerged, &rdev->flags)) { 12658f19ccb2SNeilBrown if (i < conf->raid_disks) 12661f68f0c4SNeilBrown set_bit(R1BIO_Degraded, &r1_bio->state); 12671f68f0c4SNeilBrown continue; 1268964147d5SNeilBrown } 12691f68f0c4SNeilBrown 12701f68f0c4SNeilBrown atomic_inc(&rdev->nr_pending); 12711f68f0c4SNeilBrown if (test_bit(WriteErrorSeen, &rdev->flags)) { 12721f68f0c4SNeilBrown sector_t first_bad; 12731f68f0c4SNeilBrown int bad_sectors; 12741f68f0c4SNeilBrown int is_bad; 12751f68f0c4SNeilBrown 12761f68f0c4SNeilBrown is_bad = is_badblock(rdev, r1_bio->sector, 12771f68f0c4SNeilBrown max_sectors, 12781f68f0c4SNeilBrown &first_bad, &bad_sectors); 12791f68f0c4SNeilBrown if (is_bad < 0) { 12801f68f0c4SNeilBrown /* mustn't write here until the bad block is 12811f68f0c4SNeilBrown * acknowledged*/ 12821f68f0c4SNeilBrown set_bit(BlockedBadBlocks, &rdev->flags); 12831f68f0c4SNeilBrown blocked_rdev = rdev; 12841f68f0c4SNeilBrown break; 12851f68f0c4SNeilBrown } 12861f68f0c4SNeilBrown if (is_bad && first_bad <= r1_bio->sector) { 12871f68f0c4SNeilBrown /* Cannot write here at all */ 12881f68f0c4SNeilBrown bad_sectors -= (r1_bio->sector - first_bad); 12891f68f0c4SNeilBrown if (bad_sectors < max_sectors) 12901f68f0c4SNeilBrown /* mustn't write more than bad_sectors 12911f68f0c4SNeilBrown * to other devices yet 12921f68f0c4SNeilBrown */ 12931f68f0c4SNeilBrown max_sectors = bad_sectors; 12941f68f0c4SNeilBrown rdev_dec_pending(rdev, mddev); 12951f68f0c4SNeilBrown /* We don't set R1BIO_Degraded as that 12961f68f0c4SNeilBrown * only applies if the disk is 12971f68f0c4SNeilBrown * missing, so it might be re-added, 12981f68f0c4SNeilBrown * and we want to know to recover this 12991f68f0c4SNeilBrown * chunk. 13001f68f0c4SNeilBrown * In this case the device is here, 13011f68f0c4SNeilBrown * and the fact that this chunk is not 13021f68f0c4SNeilBrown * in-sync is recorded in the bad 13031f68f0c4SNeilBrown * block log 13041f68f0c4SNeilBrown */ 13051f68f0c4SNeilBrown continue; 13061f68f0c4SNeilBrown } 13071f68f0c4SNeilBrown if (is_bad) { 13081f68f0c4SNeilBrown int good_sectors = first_bad - r1_bio->sector; 13091f68f0c4SNeilBrown if (good_sectors < max_sectors) 13101f68f0c4SNeilBrown max_sectors = good_sectors; 13111f68f0c4SNeilBrown } 13121f68f0c4SNeilBrown } 13131f68f0c4SNeilBrown r1_bio->bios[i] = bio; 13141da177e4SLinus Torvalds } 13151da177e4SLinus Torvalds rcu_read_unlock(); 13161da177e4SLinus Torvalds 13176bfe0b49SDan Williams if (unlikely(blocked_rdev)) { 13186bfe0b49SDan Williams /* Wait for this device to become unblocked */ 13196bfe0b49SDan Williams int j; 132079ef3a8aSmajianpeng sector_t old = start_next_window; 13216bfe0b49SDan Williams 13226bfe0b49SDan Williams for (j = 0; j < i; j++) 13236bfe0b49SDan Williams if (r1_bio->bios[j]) 13246bfe0b49SDan Williams rdev_dec_pending(conf->mirrors[j].rdev, mddev); 13251f68f0c4SNeilBrown r1_bio->state = 0; 13264f024f37SKent Overstreet allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); 13276bfe0b49SDan Williams md_wait_for_blocked_rdev(blocked_rdev, mddev); 132879ef3a8aSmajianpeng start_next_window = wait_barrier(conf, bio); 132979ef3a8aSmajianpeng /* 133079ef3a8aSmajianpeng * We must make sure the multi r1bios of bio have 133179ef3a8aSmajianpeng * the same value of bi_phys_segments 133279ef3a8aSmajianpeng */ 133379ef3a8aSmajianpeng if (bio->bi_phys_segments && old && 133479ef3a8aSmajianpeng old != start_next_window) 133579ef3a8aSmajianpeng /* Wait for the former r1bio(s) to complete */ 133679ef3a8aSmajianpeng wait_event(conf->wait_barrier, 133779ef3a8aSmajianpeng bio->bi_phys_segments == 1); 13386bfe0b49SDan Williams goto retry_write; 13396bfe0b49SDan Williams } 13406bfe0b49SDan Williams 13411f68f0c4SNeilBrown if (max_sectors < r1_bio->sectors) { 13421f68f0c4SNeilBrown /* We are splitting this write into multiple parts, so 13431f68f0c4SNeilBrown * we need to prepare for allocating another r1_bio. 13441f68f0c4SNeilBrown */ 13451f68f0c4SNeilBrown r1_bio->sectors = max_sectors; 13461f68f0c4SNeilBrown spin_lock_irq(&conf->device_lock); 13471f68f0c4SNeilBrown if (bio->bi_phys_segments == 0) 13481f68f0c4SNeilBrown bio->bi_phys_segments = 2; 13491f68f0c4SNeilBrown else 13501f68f0c4SNeilBrown bio->bi_phys_segments++; 13511f68f0c4SNeilBrown spin_unlock_irq(&conf->device_lock); 1352191ea9b2SNeilBrown } 13534f024f37SKent Overstreet sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; 13544b6d287fSNeilBrown 13554e78064fSNeilBrown atomic_set(&r1_bio->remaining, 1); 13564b6d287fSNeilBrown atomic_set(&r1_bio->behind_remaining, 0); 1357191ea9b2SNeilBrown 13581f68f0c4SNeilBrown first_clone = 1; 13591da177e4SLinus Torvalds for (i = 0; i < disks; i++) { 13601da177e4SLinus Torvalds struct bio *mbio; 13611da177e4SLinus Torvalds if (!r1_bio->bios[i]) 13621da177e4SLinus Torvalds continue; 13631da177e4SLinus Torvalds 1364a167f663SNeilBrown mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); 13654f024f37SKent Overstreet bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors); 13661da177e4SLinus Torvalds 13671f68f0c4SNeilBrown if (first_clone) { 13681f68f0c4SNeilBrown /* do behind I/O ? 13691f68f0c4SNeilBrown * Not if there are too many, or cannot 13701f68f0c4SNeilBrown * allocate memory, or a reader on WriteMostly 13711f68f0c4SNeilBrown * is waiting for behind writes to flush */ 13721f68f0c4SNeilBrown if (bitmap && 13731f68f0c4SNeilBrown (atomic_read(&bitmap->behind_writes) 13741f68f0c4SNeilBrown < mddev->bitmap_info.max_write_behind) && 13751f68f0c4SNeilBrown !waitqueue_active(&bitmap->behind_wait)) 13761f68f0c4SNeilBrown alloc_behind_pages(mbio, r1_bio); 13771da177e4SLinus Torvalds 13781f68f0c4SNeilBrown bitmap_startwrite(bitmap, r1_bio->sector, 13791f68f0c4SNeilBrown r1_bio->sectors, 13801f68f0c4SNeilBrown test_bit(R1BIO_BehindIO, 13811f68f0c4SNeilBrown &r1_bio->state)); 13821f68f0c4SNeilBrown first_clone = 0; 13831f68f0c4SNeilBrown } 13842ca68f5eSNeilBrown if (r1_bio->behind_bvecs) { 13854b6d287fSNeilBrown struct bio_vec *bvec; 13864b6d287fSNeilBrown int j; 13874b6d287fSNeilBrown 1388cb34e057SKent Overstreet /* 1389cb34e057SKent Overstreet * We trimmed the bio, so _all is legit 13904b6d287fSNeilBrown */ 1391d74c6d51SKent Overstreet bio_for_each_segment_all(bvec, mbio, j) 13922ca68f5eSNeilBrown bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; 13934b6d287fSNeilBrown if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) 13944b6d287fSNeilBrown atomic_inc(&r1_bio->behind_remaining); 13954b6d287fSNeilBrown } 13964b6d287fSNeilBrown 13971f68f0c4SNeilBrown r1_bio->bios[i] = mbio; 13981f68f0c4SNeilBrown 13994f024f37SKent Overstreet mbio->bi_iter.bi_sector = (r1_bio->sector + 14001f68f0c4SNeilBrown conf->mirrors[i].rdev->data_offset); 14011f68f0c4SNeilBrown mbio->bi_bdev = conf->mirrors[i].rdev->bdev; 14021f68f0c4SNeilBrown mbio->bi_end_io = raid1_end_write_request; 1403c8dc9c65SJoe Lawrence mbio->bi_rw = 1404c8dc9c65SJoe Lawrence WRITE | do_flush_fua | do_sync | do_discard | do_same; 14051f68f0c4SNeilBrown mbio->bi_private = r1_bio; 14061f68f0c4SNeilBrown 14071da177e4SLinus Torvalds atomic_inc(&r1_bio->remaining); 1408f54a9d0eSNeilBrown 1409f54a9d0eSNeilBrown cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); 1410f54a9d0eSNeilBrown if (cb) 1411f54a9d0eSNeilBrown plug = container_of(cb, struct raid1_plug_cb, cb); 1412f54a9d0eSNeilBrown else 1413f54a9d0eSNeilBrown plug = NULL; 1414191ea9b2SNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 1415f54a9d0eSNeilBrown if (plug) { 1416f54a9d0eSNeilBrown bio_list_add(&plug->pending, mbio); 1417f54a9d0eSNeilBrown plug->pending_cnt++; 1418f54a9d0eSNeilBrown } else { 14194e78064fSNeilBrown bio_list_add(&conf->pending_bio_list, mbio); 142034db0cd6SNeilBrown conf->pending_count++; 1421f54a9d0eSNeilBrown } 1422191ea9b2SNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 1423f54a9d0eSNeilBrown if (!plug) 1424b357f04aSNeilBrown md_wakeup_thread(mddev->thread); 14254e78064fSNeilBrown } 1426079fa166SNeilBrown /* Mustn't call r1_bio_write_done before this next test, 1427079fa166SNeilBrown * as it could result in the bio being freed. 1428079fa166SNeilBrown */ 1429aa8b57aaSKent Overstreet if (sectors_handled < bio_sectors(bio)) { 1430079fa166SNeilBrown r1_bio_write_done(r1_bio); 14311f68f0c4SNeilBrown /* We need another r1_bio. It has already been counted 14321f68f0c4SNeilBrown * in bio->bi_phys_segments 14331f68f0c4SNeilBrown */ 14341f68f0c4SNeilBrown r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); 14351f68f0c4SNeilBrown r1_bio->master_bio = bio; 1436aa8b57aaSKent Overstreet r1_bio->sectors = bio_sectors(bio) - sectors_handled; 14371f68f0c4SNeilBrown r1_bio->state = 0; 14381f68f0c4SNeilBrown r1_bio->mddev = mddev; 14394f024f37SKent Overstreet r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; 14401f68f0c4SNeilBrown goto retry_write; 14411f68f0c4SNeilBrown } 14421f68f0c4SNeilBrown 1443079fa166SNeilBrown r1_bio_write_done(r1_bio); 1444079fa166SNeilBrown 1445079fa166SNeilBrown /* In case raid1d snuck in to freeze_array */ 1446079fa166SNeilBrown wake_up(&conf->wait_barrier); 14471da177e4SLinus Torvalds } 14481da177e4SLinus Torvalds 1449fd01b88cSNeilBrown static void status(struct seq_file *seq, struct mddev *mddev) 14501da177e4SLinus Torvalds { 1451e8096360SNeilBrown struct r1conf *conf = mddev->private; 14521da177e4SLinus Torvalds int i; 14531da177e4SLinus Torvalds 14541da177e4SLinus Torvalds seq_printf(seq, " [%d/%d] [", conf->raid_disks, 145511ce99e6SNeilBrown conf->raid_disks - mddev->degraded); 1456ddac7c7eSNeilBrown rcu_read_lock(); 1457ddac7c7eSNeilBrown for (i = 0; i < conf->raid_disks; i++) { 14583cb03002SNeilBrown struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 14591da177e4SLinus Torvalds seq_printf(seq, "%s", 1460ddac7c7eSNeilBrown rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); 1461ddac7c7eSNeilBrown } 1462ddac7c7eSNeilBrown rcu_read_unlock(); 14631da177e4SLinus Torvalds seq_printf(seq, "]"); 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds 14661da177e4SLinus Torvalds 1467fd01b88cSNeilBrown static void error(struct mddev *mddev, struct md_rdev *rdev) 14681da177e4SLinus Torvalds { 14691da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1470e8096360SNeilBrown struct r1conf *conf = mddev->private; 14711da177e4SLinus Torvalds 14721da177e4SLinus Torvalds /* 14731da177e4SLinus Torvalds * If it is not operational, then we have already marked it as dead 14741da177e4SLinus Torvalds * else if it is the last working disks, ignore the error, let the 14751da177e4SLinus Torvalds * next level up know. 14761da177e4SLinus Torvalds * else mark the drive as failed 14771da177e4SLinus Torvalds */ 1478b2d444d7SNeilBrown if (test_bit(In_sync, &rdev->flags) 14794044ba58SNeilBrown && (conf->raid_disks - mddev->degraded) == 1) { 14801da177e4SLinus Torvalds /* 14811da177e4SLinus Torvalds * Don't fail the drive, act as though we were just a 14824044ba58SNeilBrown * normal single drive. 14834044ba58SNeilBrown * However don't try a recovery from this drive as 14844044ba58SNeilBrown * it is very likely to fail. 14851da177e4SLinus Torvalds */ 14865389042fSNeilBrown conf->recovery_disabled = mddev->recovery_disabled; 14871da177e4SLinus Torvalds return; 14884044ba58SNeilBrown } 1489de393cdeSNeilBrown set_bit(Blocked, &rdev->flags); 1490c04be0aaSNeilBrown if (test_and_clear_bit(In_sync, &rdev->flags)) { 1491c04be0aaSNeilBrown unsigned long flags; 1492c04be0aaSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 14931da177e4SLinus Torvalds mddev->degraded++; 1494dd00a99eSNeilBrown set_bit(Faulty, &rdev->flags); 1495c04be0aaSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 14961da177e4SLinus Torvalds /* 14971da177e4SLinus Torvalds * if recovery is running, make sure it aborts. 14981da177e4SLinus Torvalds */ 1499dfc70645SNeilBrown set_bit(MD_RECOVERY_INTR, &mddev->recovery); 1500dd00a99eSNeilBrown } else 1501b2d444d7SNeilBrown set_bit(Faulty, &rdev->flags); 1502850b2b42SNeilBrown set_bit(MD_CHANGE_DEVS, &mddev->flags); 1503067032bcSJoe Perches printk(KERN_ALERT 1504067032bcSJoe Perches "md/raid1:%s: Disk failure on %s, disabling device.\n" 1505067032bcSJoe Perches "md/raid1:%s: Operation continuing on %d devices.\n", 15069dd1e2faSNeilBrown mdname(mddev), bdevname(rdev->bdev, b), 15079dd1e2faSNeilBrown mdname(mddev), conf->raid_disks - mddev->degraded); 15081da177e4SLinus Torvalds } 15091da177e4SLinus Torvalds 1510e8096360SNeilBrown static void print_conf(struct r1conf *conf) 15111da177e4SLinus Torvalds { 15121da177e4SLinus Torvalds int i; 15131da177e4SLinus Torvalds 15149dd1e2faSNeilBrown printk(KERN_DEBUG "RAID1 conf printout:\n"); 15151da177e4SLinus Torvalds if (!conf) { 15169dd1e2faSNeilBrown printk(KERN_DEBUG "(!conf)\n"); 15171da177e4SLinus Torvalds return; 15181da177e4SLinus Torvalds } 15199dd1e2faSNeilBrown printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, 15201da177e4SLinus Torvalds conf->raid_disks); 15211da177e4SLinus Torvalds 1522ddac7c7eSNeilBrown rcu_read_lock(); 15231da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 15241da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 15253cb03002SNeilBrown struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); 1526ddac7c7eSNeilBrown if (rdev) 15279dd1e2faSNeilBrown printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", 1528ddac7c7eSNeilBrown i, !test_bit(In_sync, &rdev->flags), 1529ddac7c7eSNeilBrown !test_bit(Faulty, &rdev->flags), 1530ddac7c7eSNeilBrown bdevname(rdev->bdev,b)); 15311da177e4SLinus Torvalds } 1532ddac7c7eSNeilBrown rcu_read_unlock(); 15331da177e4SLinus Torvalds } 15341da177e4SLinus Torvalds 1535e8096360SNeilBrown static void close_sync(struct r1conf *conf) 15361da177e4SLinus Torvalds { 153779ef3a8aSmajianpeng wait_barrier(conf, NULL); 153879ef3a8aSmajianpeng allow_barrier(conf, 0, 0); 15391da177e4SLinus Torvalds 15401da177e4SLinus Torvalds mempool_destroy(conf->r1buf_pool); 15411da177e4SLinus Torvalds conf->r1buf_pool = NULL; 154279ef3a8aSmajianpeng 154379ef3a8aSmajianpeng conf->next_resync = 0; 154479ef3a8aSmajianpeng conf->start_next_window = MaxSector; 15451da177e4SLinus Torvalds } 15461da177e4SLinus Torvalds 1547fd01b88cSNeilBrown static int raid1_spare_active(struct mddev *mddev) 15481da177e4SLinus Torvalds { 15491da177e4SLinus Torvalds int i; 1550e8096360SNeilBrown struct r1conf *conf = mddev->private; 15516b965620SNeilBrown int count = 0; 15526b965620SNeilBrown unsigned long flags; 15531da177e4SLinus Torvalds 15541da177e4SLinus Torvalds /* 15551da177e4SLinus Torvalds * Find all failed disks within the RAID1 configuration 1556ddac7c7eSNeilBrown * and mark them readable. 1557ddac7c7eSNeilBrown * Called under mddev lock, so rcu protection not needed. 15581da177e4SLinus Torvalds */ 15591da177e4SLinus Torvalds for (i = 0; i < conf->raid_disks; i++) { 15603cb03002SNeilBrown struct md_rdev *rdev = conf->mirrors[i].rdev; 15618c7a2c2bSNeilBrown struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; 15628c7a2c2bSNeilBrown if (repl 15638c7a2c2bSNeilBrown && repl->recovery_offset == MaxSector 15648c7a2c2bSNeilBrown && !test_bit(Faulty, &repl->flags) 15658c7a2c2bSNeilBrown && !test_and_set_bit(In_sync, &repl->flags)) { 15668c7a2c2bSNeilBrown /* replacement has just become active */ 15678c7a2c2bSNeilBrown if (!rdev || 15688c7a2c2bSNeilBrown !test_and_clear_bit(In_sync, &rdev->flags)) 15698c7a2c2bSNeilBrown count++; 15708c7a2c2bSNeilBrown if (rdev) { 15718c7a2c2bSNeilBrown /* Replaced device not technically 15728c7a2c2bSNeilBrown * faulty, but we need to be sure 15738c7a2c2bSNeilBrown * it gets removed and never re-added 15748c7a2c2bSNeilBrown */ 15758c7a2c2bSNeilBrown set_bit(Faulty, &rdev->flags); 15768c7a2c2bSNeilBrown sysfs_notify_dirent_safe( 15778c7a2c2bSNeilBrown rdev->sysfs_state); 15788c7a2c2bSNeilBrown } 15798c7a2c2bSNeilBrown } 1580ddac7c7eSNeilBrown if (rdev 158161e4947cSLukasz Dorau && rdev->recovery_offset == MaxSector 1582ddac7c7eSNeilBrown && !test_bit(Faulty, &rdev->flags) 1583c04be0aaSNeilBrown && !test_and_set_bit(In_sync, &rdev->flags)) { 15846b965620SNeilBrown count++; 1585654e8b5aSJonathan Brassow sysfs_notify_dirent_safe(rdev->sysfs_state); 15861da177e4SLinus Torvalds } 15871da177e4SLinus Torvalds } 15886b965620SNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 15896b965620SNeilBrown mddev->degraded -= count; 15906b965620SNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 15911da177e4SLinus Torvalds 15921da177e4SLinus Torvalds print_conf(conf); 15936b965620SNeilBrown return count; 15941da177e4SLinus Torvalds } 15951da177e4SLinus Torvalds 15961da177e4SLinus Torvalds 1597fd01b88cSNeilBrown static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) 15981da177e4SLinus Torvalds { 1599e8096360SNeilBrown struct r1conf *conf = mddev->private; 1600199050eaSNeil Brown int err = -EEXIST; 160141158c7eSNeilBrown int mirror = 0; 16020eaf822cSJonathan Brassow struct raid1_info *p; 16036c2fce2eSNeil Brown int first = 0; 160430194636SNeilBrown int last = conf->raid_disks - 1; 16056b740b8dSNeilBrown struct request_queue *q = bdev_get_queue(rdev->bdev); 16061da177e4SLinus Torvalds 16075389042fSNeilBrown if (mddev->recovery_disabled == conf->recovery_disabled) 16085389042fSNeilBrown return -EBUSY; 16095389042fSNeilBrown 16106c2fce2eSNeil Brown if (rdev->raid_disk >= 0) 16116c2fce2eSNeil Brown first = last = rdev->raid_disk; 16126c2fce2eSNeil Brown 16136b740b8dSNeilBrown if (q->merge_bvec_fn) { 16146b740b8dSNeilBrown set_bit(Unmerged, &rdev->flags); 16156b740b8dSNeilBrown mddev->merge_check_needed = 1; 16166b740b8dSNeilBrown } 16176b740b8dSNeilBrown 16187ef449d1SNeilBrown for (mirror = first; mirror <= last; mirror++) { 16197ef449d1SNeilBrown p = conf->mirrors+mirror; 16207ef449d1SNeilBrown if (!p->rdev) { 16211da177e4SLinus Torvalds 16229092c02dSJonathan Brassow if (mddev->gendisk) 16238f6c2e4bSMartin K. Petersen disk_stack_limits(mddev->gendisk, rdev->bdev, 16248f6c2e4bSMartin K. Petersen rdev->data_offset << 9); 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds p->head_position = 0; 16271da177e4SLinus Torvalds rdev->raid_disk = mirror; 1628199050eaSNeil Brown err = 0; 16296aea114aSNeilBrown /* As all devices are equivalent, we don't need a full recovery 16306aea114aSNeilBrown * if this was recently any drive of the array 16316aea114aSNeilBrown */ 16326aea114aSNeilBrown if (rdev->saved_raid_disk < 0) 163341158c7eSNeilBrown conf->fullsync = 1; 1634d6065f7bSSuzanne Wood rcu_assign_pointer(p->rdev, rdev); 16351da177e4SLinus Torvalds break; 16361da177e4SLinus Torvalds } 16377ef449d1SNeilBrown if (test_bit(WantReplacement, &p->rdev->flags) && 16387ef449d1SNeilBrown p[conf->raid_disks].rdev == NULL) { 16397ef449d1SNeilBrown /* Add this device as a replacement */ 16407ef449d1SNeilBrown clear_bit(In_sync, &rdev->flags); 16417ef449d1SNeilBrown set_bit(Replacement, &rdev->flags); 16427ef449d1SNeilBrown rdev->raid_disk = mirror; 16437ef449d1SNeilBrown err = 0; 16447ef449d1SNeilBrown conf->fullsync = 1; 16457ef449d1SNeilBrown rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); 16467ef449d1SNeilBrown break; 16477ef449d1SNeilBrown } 16487ef449d1SNeilBrown } 16496b740b8dSNeilBrown if (err == 0 && test_bit(Unmerged, &rdev->flags)) { 16506b740b8dSNeilBrown /* Some requests might not have seen this new 16516b740b8dSNeilBrown * merge_bvec_fn. We must wait for them to complete 16526b740b8dSNeilBrown * before merging the device fully. 16536b740b8dSNeilBrown * First we make sure any code which has tested 16546b740b8dSNeilBrown * our function has submitted the request, then 16556b740b8dSNeilBrown * we wait for all outstanding requests to complete. 16566b740b8dSNeilBrown */ 16576b740b8dSNeilBrown synchronize_sched(); 1658e2d59925SNeilBrown freeze_array(conf, 0); 1659e2d59925SNeilBrown unfreeze_array(conf); 16606b740b8dSNeilBrown clear_bit(Unmerged, &rdev->flags); 16616b740b8dSNeilBrown } 1662ac5e7113SAndre Noll md_integrity_add_rdev(rdev, mddev); 16639092c02dSJonathan Brassow if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) 16642ff8cc2cSShaohua Li queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); 16651da177e4SLinus Torvalds print_conf(conf); 1666199050eaSNeil Brown return err; 16671da177e4SLinus Torvalds } 16681da177e4SLinus Torvalds 1669b8321b68SNeilBrown static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) 16701da177e4SLinus Torvalds { 1671e8096360SNeilBrown struct r1conf *conf = mddev->private; 16721da177e4SLinus Torvalds int err = 0; 1673b8321b68SNeilBrown int number = rdev->raid_disk; 16740eaf822cSJonathan Brassow struct raid1_info *p = conf->mirrors + number; 16751da177e4SLinus Torvalds 1676b014f14cSNeilBrown if (rdev != p->rdev) 1677b014f14cSNeilBrown p = conf->mirrors + conf->raid_disks + number; 1678b014f14cSNeilBrown 16791da177e4SLinus Torvalds print_conf(conf); 1680b8321b68SNeilBrown if (rdev == p->rdev) { 1681b2d444d7SNeilBrown if (test_bit(In_sync, &rdev->flags) || 16821da177e4SLinus Torvalds atomic_read(&rdev->nr_pending)) { 16831da177e4SLinus Torvalds err = -EBUSY; 16841da177e4SLinus Torvalds goto abort; 16851da177e4SLinus Torvalds } 1686046abeedSNeilBrown /* Only remove non-faulty devices if recovery 1687dfc70645SNeilBrown * is not possible. 1688dfc70645SNeilBrown */ 1689dfc70645SNeilBrown if (!test_bit(Faulty, &rdev->flags) && 16905389042fSNeilBrown mddev->recovery_disabled != conf->recovery_disabled && 1691dfc70645SNeilBrown mddev->degraded < conf->raid_disks) { 1692dfc70645SNeilBrown err = -EBUSY; 1693dfc70645SNeilBrown goto abort; 1694dfc70645SNeilBrown } 16951da177e4SLinus Torvalds p->rdev = NULL; 1696fbd568a3SPaul E. McKenney synchronize_rcu(); 16971da177e4SLinus Torvalds if (atomic_read(&rdev->nr_pending)) { 16981da177e4SLinus Torvalds /* lost the race, try later */ 16991da177e4SLinus Torvalds err = -EBUSY; 17001da177e4SLinus Torvalds p->rdev = rdev; 1701ac5e7113SAndre Noll goto abort; 17028c7a2c2bSNeilBrown } else if (conf->mirrors[conf->raid_disks + number].rdev) { 17038c7a2c2bSNeilBrown /* We just removed a device that is being replaced. 17048c7a2c2bSNeilBrown * Move down the replacement. We drain all IO before 17058c7a2c2bSNeilBrown * doing this to avoid confusion. 17068c7a2c2bSNeilBrown */ 17078c7a2c2bSNeilBrown struct md_rdev *repl = 17088c7a2c2bSNeilBrown conf->mirrors[conf->raid_disks + number].rdev; 1709e2d59925SNeilBrown freeze_array(conf, 0); 17108c7a2c2bSNeilBrown clear_bit(Replacement, &repl->flags); 17118c7a2c2bSNeilBrown p->rdev = repl; 17128c7a2c2bSNeilBrown conf->mirrors[conf->raid_disks + number].rdev = NULL; 1713e2d59925SNeilBrown unfreeze_array(conf); 1714b014f14cSNeilBrown clear_bit(WantReplacement, &rdev->flags); 17158c7a2c2bSNeilBrown } else 17168c7a2c2bSNeilBrown clear_bit(WantReplacement, &rdev->flags); 1717a91a2785SMartin K. Petersen err = md_integrity_register(mddev); 17181da177e4SLinus Torvalds } 17191da177e4SLinus Torvalds abort: 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds print_conf(conf); 17221da177e4SLinus Torvalds return err; 17231da177e4SLinus Torvalds } 17241da177e4SLinus Torvalds 17251da177e4SLinus Torvalds 17266712ecf8SNeilBrown static void end_sync_read(struct bio *bio, int error) 17271da177e4SLinus Torvalds { 17289f2c9d12SNeilBrown struct r1bio *r1_bio = bio->bi_private; 17291da177e4SLinus Torvalds 17300fc280f6SNeilBrown update_head_pos(r1_bio->read_disk, r1_bio); 1731ba3ae3beSNamhyung Kim 17321da177e4SLinus Torvalds /* 17331da177e4SLinus Torvalds * we have read a block, now it needs to be re-written, 17341da177e4SLinus Torvalds * or re-read if the read failed. 17351da177e4SLinus Torvalds * We don't do much here, just schedule handling by raid1d 17361da177e4SLinus Torvalds */ 173769382e85SNeilBrown if (test_bit(BIO_UPTODATE, &bio->bi_flags)) 17381da177e4SLinus Torvalds set_bit(R1BIO_Uptodate, &r1_bio->state); 1739d11c171eSNeilBrown 1740d11c171eSNeilBrown if (atomic_dec_and_test(&r1_bio->remaining)) 17411da177e4SLinus Torvalds reschedule_retry(r1_bio); 17421da177e4SLinus Torvalds } 17431da177e4SLinus Torvalds 17446712ecf8SNeilBrown static void end_sync_write(struct bio *bio, int error) 17451da177e4SLinus Torvalds { 17461da177e4SLinus Torvalds int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); 17479f2c9d12SNeilBrown struct r1bio *r1_bio = bio->bi_private; 1748fd01b88cSNeilBrown struct mddev *mddev = r1_bio->mddev; 1749e8096360SNeilBrown struct r1conf *conf = mddev->private; 17501da177e4SLinus Torvalds int mirror=0; 17514367af55SNeilBrown sector_t first_bad; 17524367af55SNeilBrown int bad_sectors; 17531da177e4SLinus Torvalds 1754ba3ae3beSNamhyung Kim mirror = find_bio_disk(r1_bio, bio); 1755ba3ae3beSNamhyung Kim 17566b1117d5SNeilBrown if (!uptodate) { 175757dab0bdSNeilBrown sector_t sync_blocks = 0; 17586b1117d5SNeilBrown sector_t s = r1_bio->sector; 17596b1117d5SNeilBrown long sectors_to_go = r1_bio->sectors; 17606b1117d5SNeilBrown /* make sure these bits doesn't get cleared. */ 17616b1117d5SNeilBrown do { 17625e3db645SNeilBrown bitmap_end_sync(mddev->bitmap, s, 17636b1117d5SNeilBrown &sync_blocks, 1); 17646b1117d5SNeilBrown s += sync_blocks; 17656b1117d5SNeilBrown sectors_to_go -= sync_blocks; 17666b1117d5SNeilBrown } while (sectors_to_go > 0); 1767d8f05d29SNeilBrown set_bit(WriteErrorSeen, 1768d8f05d29SNeilBrown &conf->mirrors[mirror].rdev->flags); 176919d67169SNeilBrown if (!test_and_set_bit(WantReplacement, 177019d67169SNeilBrown &conf->mirrors[mirror].rdev->flags)) 177119d67169SNeilBrown set_bit(MD_RECOVERY_NEEDED, & 177219d67169SNeilBrown mddev->recovery); 1773d8f05d29SNeilBrown set_bit(R1BIO_WriteError, &r1_bio->state); 17744367af55SNeilBrown } else if (is_badblock(conf->mirrors[mirror].rdev, 17754367af55SNeilBrown r1_bio->sector, 17764367af55SNeilBrown r1_bio->sectors, 17773a9f28a5SNeilBrown &first_bad, &bad_sectors) && 17783a9f28a5SNeilBrown !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, 17793a9f28a5SNeilBrown r1_bio->sector, 17803a9f28a5SNeilBrown r1_bio->sectors, 17813a9f28a5SNeilBrown &first_bad, &bad_sectors) 17823a9f28a5SNeilBrown ) 17834367af55SNeilBrown set_bit(R1BIO_MadeGood, &r1_bio->state); 1784e3b9703eSNeilBrown 17851da177e4SLinus Torvalds if (atomic_dec_and_test(&r1_bio->remaining)) { 17864367af55SNeilBrown int s = r1_bio->sectors; 1787d8f05d29SNeilBrown if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 1788d8f05d29SNeilBrown test_bit(R1BIO_WriteError, &r1_bio->state)) 17894367af55SNeilBrown reschedule_retry(r1_bio); 17904367af55SNeilBrown else { 17911da177e4SLinus Torvalds put_buf(r1_bio); 179273d5c38aSNeilBrown md_done_sync(mddev, s, uptodate); 17931da177e4SLinus Torvalds } 17941da177e4SLinus Torvalds } 17954367af55SNeilBrown } 17961da177e4SLinus Torvalds 17973cb03002SNeilBrown static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, 1798d8f05d29SNeilBrown int sectors, struct page *page, int rw) 1799d8f05d29SNeilBrown { 1800d8f05d29SNeilBrown if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) 1801d8f05d29SNeilBrown /* success */ 1802d8f05d29SNeilBrown return 1; 180319d67169SNeilBrown if (rw == WRITE) { 1804d8f05d29SNeilBrown set_bit(WriteErrorSeen, &rdev->flags); 180519d67169SNeilBrown if (!test_and_set_bit(WantReplacement, 180619d67169SNeilBrown &rdev->flags)) 180719d67169SNeilBrown set_bit(MD_RECOVERY_NEEDED, & 180819d67169SNeilBrown rdev->mddev->recovery); 180919d67169SNeilBrown } 1810d8f05d29SNeilBrown /* need to record an error - either for the block or the device */ 1811d8f05d29SNeilBrown if (!rdev_set_badblocks(rdev, sector, sectors, 0)) 1812d8f05d29SNeilBrown md_error(rdev->mddev, rdev); 1813d8f05d29SNeilBrown return 0; 1814d8f05d29SNeilBrown } 1815d8f05d29SNeilBrown 18169f2c9d12SNeilBrown static int fix_sync_read_error(struct r1bio *r1_bio) 18171da177e4SLinus Torvalds { 1818a68e5870SNeilBrown /* Try some synchronous reads of other devices to get 181969382e85SNeilBrown * good data, much like with normal read errors. Only 1820ddac7c7eSNeilBrown * read into the pages we already have so we don't 182169382e85SNeilBrown * need to re-issue the read request. 182269382e85SNeilBrown * We don't need to freeze the array, because being in an 182369382e85SNeilBrown * active sync request, there is no normal IO, and 182469382e85SNeilBrown * no overlapping syncs. 182506f60385SNeilBrown * We don't need to check is_badblock() again as we 182606f60385SNeilBrown * made sure that anything with a bad block in range 182706f60385SNeilBrown * will have bi_end_io clear. 18281da177e4SLinus Torvalds */ 1829fd01b88cSNeilBrown struct mddev *mddev = r1_bio->mddev; 1830e8096360SNeilBrown struct r1conf *conf = mddev->private; 1831a68e5870SNeilBrown struct bio *bio = r1_bio->bios[r1_bio->read_disk]; 183269382e85SNeilBrown sector_t sect = r1_bio->sector; 183369382e85SNeilBrown int sectors = r1_bio->sectors; 183469382e85SNeilBrown int idx = 0; 183569382e85SNeilBrown 183669382e85SNeilBrown while(sectors) { 183769382e85SNeilBrown int s = sectors; 183869382e85SNeilBrown int d = r1_bio->read_disk; 183969382e85SNeilBrown int success = 0; 18403cb03002SNeilBrown struct md_rdev *rdev; 184178d7f5f7SNeilBrown int start; 184269382e85SNeilBrown 184369382e85SNeilBrown if (s > (PAGE_SIZE>>9)) 184469382e85SNeilBrown s = PAGE_SIZE >> 9; 184569382e85SNeilBrown do { 184669382e85SNeilBrown if (r1_bio->bios[d]->bi_end_io == end_sync_read) { 1847ddac7c7eSNeilBrown /* No rcu protection needed here devices 1848ddac7c7eSNeilBrown * can only be removed when no resync is 1849ddac7c7eSNeilBrown * active, and resync is currently active 1850ddac7c7eSNeilBrown */ 185169382e85SNeilBrown rdev = conf->mirrors[d].rdev; 18529d3d8011SNamhyung Kim if (sync_page_io(rdev, sect, s<<9, 185369382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 1854ccebd4c4SJonathan Brassow READ, false)) { 185569382e85SNeilBrown success = 1; 185669382e85SNeilBrown break; 185769382e85SNeilBrown } 185869382e85SNeilBrown } 185969382e85SNeilBrown d++; 18608f19ccb2SNeilBrown if (d == conf->raid_disks * 2) 186169382e85SNeilBrown d = 0; 186269382e85SNeilBrown } while (!success && d != r1_bio->read_disk); 186369382e85SNeilBrown 186478d7f5f7SNeilBrown if (!success) { 186578d7f5f7SNeilBrown char b[BDEVNAME_SIZE]; 18663a9f28a5SNeilBrown int abort = 0; 18673a9f28a5SNeilBrown /* Cannot read from anywhere, this block is lost. 18683a9f28a5SNeilBrown * Record a bad block on each device. If that doesn't 18693a9f28a5SNeilBrown * work just disable and interrupt the recovery. 18703a9f28a5SNeilBrown * Don't fail devices as that won't really help. 18713a9f28a5SNeilBrown */ 187278d7f5f7SNeilBrown printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" 187378d7f5f7SNeilBrown " for block %llu\n", 187478d7f5f7SNeilBrown mdname(mddev), 187578d7f5f7SNeilBrown bdevname(bio->bi_bdev, b), 187678d7f5f7SNeilBrown (unsigned long long)r1_bio->sector); 18778f19ccb2SNeilBrown for (d = 0; d < conf->raid_disks * 2; d++) { 18783a9f28a5SNeilBrown rdev = conf->mirrors[d].rdev; 18793a9f28a5SNeilBrown if (!rdev || test_bit(Faulty, &rdev->flags)) 18803a9f28a5SNeilBrown continue; 18813a9f28a5SNeilBrown if (!rdev_set_badblocks(rdev, sect, s, 0)) 18823a9f28a5SNeilBrown abort = 1; 18833a9f28a5SNeilBrown } 18843a9f28a5SNeilBrown if (abort) { 1885d890fa2bSNeilBrown conf->recovery_disabled = 1886d890fa2bSNeilBrown mddev->recovery_disabled; 18873a9f28a5SNeilBrown set_bit(MD_RECOVERY_INTR, &mddev->recovery); 188878d7f5f7SNeilBrown md_done_sync(mddev, r1_bio->sectors, 0); 188978d7f5f7SNeilBrown put_buf(r1_bio); 189078d7f5f7SNeilBrown return 0; 189178d7f5f7SNeilBrown } 18923a9f28a5SNeilBrown /* Try next page */ 18933a9f28a5SNeilBrown sectors -= s; 18943a9f28a5SNeilBrown sect += s; 18953a9f28a5SNeilBrown idx++; 18963a9f28a5SNeilBrown continue; 18973a9f28a5SNeilBrown } 189878d7f5f7SNeilBrown 189978d7f5f7SNeilBrown start = d; 190069382e85SNeilBrown /* write it back and re-read */ 190169382e85SNeilBrown while (d != r1_bio->read_disk) { 190269382e85SNeilBrown if (d == 0) 19038f19ccb2SNeilBrown d = conf->raid_disks * 2; 190469382e85SNeilBrown d--; 190569382e85SNeilBrown if (r1_bio->bios[d]->bi_end_io != end_sync_read) 190669382e85SNeilBrown continue; 190769382e85SNeilBrown rdev = conf->mirrors[d].rdev; 1908d8f05d29SNeilBrown if (r1_sync_page_io(rdev, sect, s, 190969382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 1910d8f05d29SNeilBrown WRITE) == 0) { 191178d7f5f7SNeilBrown r1_bio->bios[d]->bi_end_io = NULL; 191278d7f5f7SNeilBrown rdev_dec_pending(rdev, mddev); 19139d3d8011SNamhyung Kim } 1914097426f6SNeilBrown } 1915097426f6SNeilBrown d = start; 1916097426f6SNeilBrown while (d != r1_bio->read_disk) { 1917097426f6SNeilBrown if (d == 0) 19188f19ccb2SNeilBrown d = conf->raid_disks * 2; 1919097426f6SNeilBrown d--; 1920097426f6SNeilBrown if (r1_bio->bios[d]->bi_end_io != end_sync_read) 1921097426f6SNeilBrown continue; 1922097426f6SNeilBrown rdev = conf->mirrors[d].rdev; 1923d8f05d29SNeilBrown if (r1_sync_page_io(rdev, sect, s, 192469382e85SNeilBrown bio->bi_io_vec[idx].bv_page, 1925d8f05d29SNeilBrown READ) != 0) 19269d3d8011SNamhyung Kim atomic_add(s, &rdev->corrected_errors); 192769382e85SNeilBrown } 192869382e85SNeilBrown sectors -= s; 192969382e85SNeilBrown sect += s; 193069382e85SNeilBrown idx ++; 193169382e85SNeilBrown } 193278d7f5f7SNeilBrown set_bit(R1BIO_Uptodate, &r1_bio->state); 19337ca78d57SNeilBrown set_bit(BIO_UPTODATE, &bio->bi_flags); 1934a68e5870SNeilBrown return 1; 193569382e85SNeilBrown } 1936d11c171eSNeilBrown 19379f2c9d12SNeilBrown static int process_checks(struct r1bio *r1_bio) 1938a68e5870SNeilBrown { 1939a68e5870SNeilBrown /* We have read all readable devices. If we haven't 1940a68e5870SNeilBrown * got the block, then there is no hope left. 1941a68e5870SNeilBrown * If we have, then we want to do a comparison 1942a68e5870SNeilBrown * and skip the write if everything is the same. 1943a68e5870SNeilBrown * If any blocks failed to read, then we need to 1944a68e5870SNeilBrown * attempt an over-write 1945a68e5870SNeilBrown */ 1946fd01b88cSNeilBrown struct mddev *mddev = r1_bio->mddev; 1947e8096360SNeilBrown struct r1conf *conf = mddev->private; 1948a68e5870SNeilBrown int primary; 1949a68e5870SNeilBrown int i; 1950f4380a91Smajianpeng int vcnt; 1951a68e5870SNeilBrown 195230bc9b53SNeilBrown /* Fix variable parts of all bios */ 195330bc9b53SNeilBrown vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); 195430bc9b53SNeilBrown for (i = 0; i < conf->raid_disks * 2; i++) { 195530bc9b53SNeilBrown int j; 195630bc9b53SNeilBrown int size; 195730bc9b53SNeilBrown struct bio *b = r1_bio->bios[i]; 195830bc9b53SNeilBrown if (b->bi_end_io != end_sync_read) 195930bc9b53SNeilBrown continue; 196030bc9b53SNeilBrown /* fixup the bio for reuse */ 196130bc9b53SNeilBrown bio_reset(b); 196230bc9b53SNeilBrown b->bi_vcnt = vcnt; 19634f024f37SKent Overstreet b->bi_iter.bi_size = r1_bio->sectors << 9; 19644f024f37SKent Overstreet b->bi_iter.bi_sector = r1_bio->sector + 196530bc9b53SNeilBrown conf->mirrors[i].rdev->data_offset; 196630bc9b53SNeilBrown b->bi_bdev = conf->mirrors[i].rdev->bdev; 196730bc9b53SNeilBrown b->bi_end_io = end_sync_read; 196830bc9b53SNeilBrown b->bi_private = r1_bio; 196930bc9b53SNeilBrown 19704f024f37SKent Overstreet size = b->bi_iter.bi_size; 197130bc9b53SNeilBrown for (j = 0; j < vcnt ; j++) { 197230bc9b53SNeilBrown struct bio_vec *bi; 197330bc9b53SNeilBrown bi = &b->bi_io_vec[j]; 197430bc9b53SNeilBrown bi->bv_offset = 0; 197530bc9b53SNeilBrown if (size > PAGE_SIZE) 197630bc9b53SNeilBrown bi->bv_len = PAGE_SIZE; 197730bc9b53SNeilBrown else 197830bc9b53SNeilBrown bi->bv_len = size; 197930bc9b53SNeilBrown size -= PAGE_SIZE; 198030bc9b53SNeilBrown } 198130bc9b53SNeilBrown } 19828f19ccb2SNeilBrown for (primary = 0; primary < conf->raid_disks * 2; primary++) 1983a68e5870SNeilBrown if (r1_bio->bios[primary]->bi_end_io == end_sync_read && 1984a68e5870SNeilBrown test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { 1985a68e5870SNeilBrown r1_bio->bios[primary]->bi_end_io = NULL; 1986a68e5870SNeilBrown rdev_dec_pending(conf->mirrors[primary].rdev, mddev); 1987a68e5870SNeilBrown break; 1988a68e5870SNeilBrown } 1989a68e5870SNeilBrown r1_bio->read_disk = primary; 19908f19ccb2SNeilBrown for (i = 0; i < conf->raid_disks * 2; i++) { 1991a68e5870SNeilBrown int j; 1992a68e5870SNeilBrown struct bio *pbio = r1_bio->bios[primary]; 1993a68e5870SNeilBrown struct bio *sbio = r1_bio->bios[i]; 199478d7f5f7SNeilBrown 19952aabaa65SKent Overstreet if (sbio->bi_end_io != end_sync_read) 199678d7f5f7SNeilBrown continue; 1997a68e5870SNeilBrown 1998a68e5870SNeilBrown if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { 1999a68e5870SNeilBrown for (j = vcnt; j-- ; ) { 2000a68e5870SNeilBrown struct page *p, *s; 2001a68e5870SNeilBrown p = pbio->bi_io_vec[j].bv_page; 2002a68e5870SNeilBrown s = sbio->bi_io_vec[j].bv_page; 2003a68e5870SNeilBrown if (memcmp(page_address(p), 2004a68e5870SNeilBrown page_address(s), 20055020ad7dSNeilBrown sbio->bi_io_vec[j].bv_len)) 2006a68e5870SNeilBrown break; 2007a68e5870SNeilBrown } 2008a68e5870SNeilBrown } else 2009a68e5870SNeilBrown j = 0; 2010a68e5870SNeilBrown if (j >= 0) 20117f7583d4SJianpeng Ma atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); 2012a68e5870SNeilBrown if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) 2013a68e5870SNeilBrown && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { 201478d7f5f7SNeilBrown /* No need to write to this device. */ 2015a68e5870SNeilBrown sbio->bi_end_io = NULL; 2016a68e5870SNeilBrown rdev_dec_pending(conf->mirrors[i].rdev, mddev); 201778d7f5f7SNeilBrown continue; 201878d7f5f7SNeilBrown } 2019d3b45c2aSKent Overstreet 2020d3b45c2aSKent Overstreet bio_copy_data(sbio, pbio); 2021a68e5870SNeilBrown } 2022a68e5870SNeilBrown return 0; 2023a68e5870SNeilBrown } 2024a68e5870SNeilBrown 20259f2c9d12SNeilBrown static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) 2026a68e5870SNeilBrown { 2027e8096360SNeilBrown struct r1conf *conf = mddev->private; 2028a68e5870SNeilBrown int i; 20298f19ccb2SNeilBrown int disks = conf->raid_disks * 2; 2030a68e5870SNeilBrown struct bio *bio, *wbio; 2031a68e5870SNeilBrown 2032a68e5870SNeilBrown bio = r1_bio->bios[r1_bio->read_disk]; 2033a68e5870SNeilBrown 2034a68e5870SNeilBrown if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) 2035a68e5870SNeilBrown /* ouch - failed to read all of that. */ 2036a68e5870SNeilBrown if (!fix_sync_read_error(r1_bio)) 2037a68e5870SNeilBrown return; 20387ca78d57SNeilBrown 20397ca78d57SNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 20407ca78d57SNeilBrown if (process_checks(r1_bio) < 0) 20417ca78d57SNeilBrown return; 2042d11c171eSNeilBrown /* 2043d11c171eSNeilBrown * schedule writes 2044d11c171eSNeilBrown */ 20451da177e4SLinus Torvalds atomic_set(&r1_bio->remaining, 1); 20461da177e4SLinus Torvalds for (i = 0; i < disks ; i++) { 20471da177e4SLinus Torvalds wbio = r1_bio->bios[i]; 20483e198f78SNeilBrown if (wbio->bi_end_io == NULL || 20493e198f78SNeilBrown (wbio->bi_end_io == end_sync_read && 20503e198f78SNeilBrown (i == r1_bio->read_disk || 20513e198f78SNeilBrown !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) 20521da177e4SLinus Torvalds continue; 20531da177e4SLinus Torvalds 20543e198f78SNeilBrown wbio->bi_rw = WRITE; 20553e198f78SNeilBrown wbio->bi_end_io = end_sync_write; 20561da177e4SLinus Torvalds atomic_inc(&r1_bio->remaining); 2057aa8b57aaSKent Overstreet md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); 2058191ea9b2SNeilBrown 20591da177e4SLinus Torvalds generic_make_request(wbio); 20601da177e4SLinus Torvalds } 20611da177e4SLinus Torvalds 20621da177e4SLinus Torvalds if (atomic_dec_and_test(&r1_bio->remaining)) { 2063191ea9b2SNeilBrown /* if we're here, all write(s) have completed, so clean up */ 206458e94ae1SNeilBrown int s = r1_bio->sectors; 206558e94ae1SNeilBrown if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 206658e94ae1SNeilBrown test_bit(R1BIO_WriteError, &r1_bio->state)) 206758e94ae1SNeilBrown reschedule_retry(r1_bio); 206858e94ae1SNeilBrown else { 20691da177e4SLinus Torvalds put_buf(r1_bio); 207058e94ae1SNeilBrown md_done_sync(mddev, s, 1); 207158e94ae1SNeilBrown } 20721da177e4SLinus Torvalds } 20731da177e4SLinus Torvalds } 20741da177e4SLinus Torvalds 20751da177e4SLinus Torvalds /* 20761da177e4SLinus Torvalds * This is a kernel thread which: 20771da177e4SLinus Torvalds * 20781da177e4SLinus Torvalds * 1. Retries failed read operations on working mirrors. 20791da177e4SLinus Torvalds * 2. Updates the raid superblock when problems encounter. 2080d2eb35acSNeilBrown * 3. Performs writes following reads for array synchronising. 20811da177e4SLinus Torvalds */ 20821da177e4SLinus Torvalds 2083e8096360SNeilBrown static void fix_read_error(struct r1conf *conf, int read_disk, 2084867868fbSNeilBrown sector_t sect, int sectors) 2085867868fbSNeilBrown { 2086fd01b88cSNeilBrown struct mddev *mddev = conf->mddev; 2087867868fbSNeilBrown while(sectors) { 2088867868fbSNeilBrown int s = sectors; 2089867868fbSNeilBrown int d = read_disk; 2090867868fbSNeilBrown int success = 0; 2091867868fbSNeilBrown int start; 20923cb03002SNeilBrown struct md_rdev *rdev; 2093867868fbSNeilBrown 2094867868fbSNeilBrown if (s > (PAGE_SIZE>>9)) 2095867868fbSNeilBrown s = PAGE_SIZE >> 9; 2096867868fbSNeilBrown 2097867868fbSNeilBrown do { 2098867868fbSNeilBrown /* Note: no rcu protection needed here 2099867868fbSNeilBrown * as this is synchronous in the raid1d thread 2100867868fbSNeilBrown * which is the thread that might remove 2101867868fbSNeilBrown * a device. If raid1d ever becomes multi-threaded.... 2102867868fbSNeilBrown */ 2103d2eb35acSNeilBrown sector_t first_bad; 2104d2eb35acSNeilBrown int bad_sectors; 2105d2eb35acSNeilBrown 2106867868fbSNeilBrown rdev = conf->mirrors[d].rdev; 2107867868fbSNeilBrown if (rdev && 2108da8840a7Smajianpeng (test_bit(In_sync, &rdev->flags) || 2109da8840a7Smajianpeng (!test_bit(Faulty, &rdev->flags) && 2110da8840a7Smajianpeng rdev->recovery_offset >= sect + s)) && 2111d2eb35acSNeilBrown is_badblock(rdev, sect, s, 2112d2eb35acSNeilBrown &first_bad, &bad_sectors) == 0 && 2113ccebd4c4SJonathan Brassow sync_page_io(rdev, sect, s<<9, 2114ccebd4c4SJonathan Brassow conf->tmppage, READ, false)) 2115867868fbSNeilBrown success = 1; 2116867868fbSNeilBrown else { 2117867868fbSNeilBrown d++; 21188f19ccb2SNeilBrown if (d == conf->raid_disks * 2) 2119867868fbSNeilBrown d = 0; 2120867868fbSNeilBrown } 2121867868fbSNeilBrown } while (!success && d != read_disk); 2122867868fbSNeilBrown 2123867868fbSNeilBrown if (!success) { 2124d8f05d29SNeilBrown /* Cannot read from anywhere - mark it bad */ 21253cb03002SNeilBrown struct md_rdev *rdev = conf->mirrors[read_disk].rdev; 2126d8f05d29SNeilBrown if (!rdev_set_badblocks(rdev, sect, s, 0)) 2127d8f05d29SNeilBrown md_error(mddev, rdev); 2128867868fbSNeilBrown break; 2129867868fbSNeilBrown } 2130867868fbSNeilBrown /* write it back and re-read */ 2131867868fbSNeilBrown start = d; 2132867868fbSNeilBrown while (d != read_disk) { 2133867868fbSNeilBrown if (d==0) 21348f19ccb2SNeilBrown d = conf->raid_disks * 2; 2135867868fbSNeilBrown d--; 2136867868fbSNeilBrown rdev = conf->mirrors[d].rdev; 2137867868fbSNeilBrown if (rdev && 2138d8f05d29SNeilBrown test_bit(In_sync, &rdev->flags)) 2139d8f05d29SNeilBrown r1_sync_page_io(rdev, sect, s, 2140d8f05d29SNeilBrown conf->tmppage, WRITE); 2141867868fbSNeilBrown } 2142867868fbSNeilBrown d = start; 2143867868fbSNeilBrown while (d != read_disk) { 2144867868fbSNeilBrown char b[BDEVNAME_SIZE]; 2145867868fbSNeilBrown if (d==0) 21468f19ccb2SNeilBrown d = conf->raid_disks * 2; 2147867868fbSNeilBrown d--; 2148867868fbSNeilBrown rdev = conf->mirrors[d].rdev; 2149867868fbSNeilBrown if (rdev && 2150867868fbSNeilBrown test_bit(In_sync, &rdev->flags)) { 2151d8f05d29SNeilBrown if (r1_sync_page_io(rdev, sect, s, 2152d8f05d29SNeilBrown conf->tmppage, READ)) { 2153867868fbSNeilBrown atomic_add(s, &rdev->corrected_errors); 2154867868fbSNeilBrown printk(KERN_INFO 21559dd1e2faSNeilBrown "md/raid1:%s: read error corrected " 2156867868fbSNeilBrown "(%d sectors at %llu on %s)\n", 2157867868fbSNeilBrown mdname(mddev), s, 2158969b755aSRandy Dunlap (unsigned long long)(sect + 2159969b755aSRandy Dunlap rdev->data_offset), 2160867868fbSNeilBrown bdevname(rdev->bdev, b)); 2161867868fbSNeilBrown } 2162867868fbSNeilBrown } 2163867868fbSNeilBrown } 2164867868fbSNeilBrown sectors -= s; 2165867868fbSNeilBrown sect += s; 2166867868fbSNeilBrown } 2167867868fbSNeilBrown } 2168867868fbSNeilBrown 21699f2c9d12SNeilBrown static int narrow_write_error(struct r1bio *r1_bio, int i) 2170cd5ff9a1SNeilBrown { 2171fd01b88cSNeilBrown struct mddev *mddev = r1_bio->mddev; 2172e8096360SNeilBrown struct r1conf *conf = mddev->private; 21733cb03002SNeilBrown struct md_rdev *rdev = conf->mirrors[i].rdev; 2174cd5ff9a1SNeilBrown 2175cd5ff9a1SNeilBrown /* bio has the data to be written to device 'i' where 2176cd5ff9a1SNeilBrown * we just recently had a write error. 2177cd5ff9a1SNeilBrown * We repeatedly clone the bio and trim down to one block, 2178cd5ff9a1SNeilBrown * then try the write. Where the write fails we record 2179cd5ff9a1SNeilBrown * a bad block. 2180cd5ff9a1SNeilBrown * It is conceivable that the bio doesn't exactly align with 2181cd5ff9a1SNeilBrown * blocks. We must handle this somehow. 2182cd5ff9a1SNeilBrown * 2183cd5ff9a1SNeilBrown * We currently own a reference on the rdev. 2184cd5ff9a1SNeilBrown */ 2185cd5ff9a1SNeilBrown 2186cd5ff9a1SNeilBrown int block_sectors; 2187cd5ff9a1SNeilBrown sector_t sector; 2188cd5ff9a1SNeilBrown int sectors; 2189cd5ff9a1SNeilBrown int sect_to_write = r1_bio->sectors; 2190cd5ff9a1SNeilBrown int ok = 1; 2191cd5ff9a1SNeilBrown 2192cd5ff9a1SNeilBrown if (rdev->badblocks.shift < 0) 2193cd5ff9a1SNeilBrown return 0; 2194cd5ff9a1SNeilBrown 2195cd5ff9a1SNeilBrown block_sectors = 1 << rdev->badblocks.shift; 2196cd5ff9a1SNeilBrown sector = r1_bio->sector; 2197cd5ff9a1SNeilBrown sectors = ((sector + block_sectors) 2198cd5ff9a1SNeilBrown & ~(sector_t)(block_sectors - 1)) 2199cd5ff9a1SNeilBrown - sector; 2200cd5ff9a1SNeilBrown 2201cd5ff9a1SNeilBrown while (sect_to_write) { 2202cd5ff9a1SNeilBrown struct bio *wbio; 2203cd5ff9a1SNeilBrown if (sectors > sect_to_write) 2204cd5ff9a1SNeilBrown sectors = sect_to_write; 2205cd5ff9a1SNeilBrown /* Write at 'sector' for 'sectors'*/ 2206cd5ff9a1SNeilBrown 2207b783863fSKent Overstreet if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { 2208b783863fSKent Overstreet unsigned vcnt = r1_bio->behind_page_count; 2209b783863fSKent Overstreet struct bio_vec *vec = r1_bio->behind_bvecs; 2210b783863fSKent Overstreet 2211b783863fSKent Overstreet while (!vec->bv_page) { 2212b783863fSKent Overstreet vec++; 2213b783863fSKent Overstreet vcnt--; 2214b783863fSKent Overstreet } 2215b783863fSKent Overstreet 2216cd5ff9a1SNeilBrown wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); 2217cd5ff9a1SNeilBrown memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); 2218b783863fSKent Overstreet 2219cd5ff9a1SNeilBrown wbio->bi_vcnt = vcnt; 2220b783863fSKent Overstreet } else { 2221b783863fSKent Overstreet wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); 2222b783863fSKent Overstreet } 2223b783863fSKent Overstreet 2224b783863fSKent Overstreet wbio->bi_rw = WRITE; 22254f024f37SKent Overstreet wbio->bi_iter.bi_sector = r1_bio->sector; 22264f024f37SKent Overstreet wbio->bi_iter.bi_size = r1_bio->sectors << 9; 2227cd5ff9a1SNeilBrown 22286678d83fSKent Overstreet bio_trim(wbio, sector - r1_bio->sector, sectors); 22294f024f37SKent Overstreet wbio->bi_iter.bi_sector += rdev->data_offset; 2230cd5ff9a1SNeilBrown wbio->bi_bdev = rdev->bdev; 2231cd5ff9a1SNeilBrown if (submit_bio_wait(WRITE, wbio) == 0) 2232cd5ff9a1SNeilBrown /* failure! */ 2233cd5ff9a1SNeilBrown ok = rdev_set_badblocks(rdev, sector, 2234cd5ff9a1SNeilBrown sectors, 0) 2235cd5ff9a1SNeilBrown && ok; 2236cd5ff9a1SNeilBrown 2237cd5ff9a1SNeilBrown bio_put(wbio); 2238cd5ff9a1SNeilBrown sect_to_write -= sectors; 2239cd5ff9a1SNeilBrown sector += sectors; 2240cd5ff9a1SNeilBrown sectors = block_sectors; 2241cd5ff9a1SNeilBrown } 2242cd5ff9a1SNeilBrown return ok; 2243cd5ff9a1SNeilBrown } 2244cd5ff9a1SNeilBrown 2245e8096360SNeilBrown static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) 224662096bceSNeilBrown { 224762096bceSNeilBrown int m; 224862096bceSNeilBrown int s = r1_bio->sectors; 22498f19ccb2SNeilBrown for (m = 0; m < conf->raid_disks * 2 ; m++) { 22503cb03002SNeilBrown struct md_rdev *rdev = conf->mirrors[m].rdev; 225162096bceSNeilBrown struct bio *bio = r1_bio->bios[m]; 225262096bceSNeilBrown if (bio->bi_end_io == NULL) 225362096bceSNeilBrown continue; 225462096bceSNeilBrown if (test_bit(BIO_UPTODATE, &bio->bi_flags) && 225562096bceSNeilBrown test_bit(R1BIO_MadeGood, &r1_bio->state)) { 2256c6563a8cSNeilBrown rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); 225762096bceSNeilBrown } 225862096bceSNeilBrown if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && 225962096bceSNeilBrown test_bit(R1BIO_WriteError, &r1_bio->state)) { 226062096bceSNeilBrown if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) 226162096bceSNeilBrown md_error(conf->mddev, rdev); 226262096bceSNeilBrown } 226362096bceSNeilBrown } 226462096bceSNeilBrown put_buf(r1_bio); 226562096bceSNeilBrown md_done_sync(conf->mddev, s, 1); 226662096bceSNeilBrown } 226762096bceSNeilBrown 2268e8096360SNeilBrown static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) 226962096bceSNeilBrown { 227062096bceSNeilBrown int m; 22718f19ccb2SNeilBrown for (m = 0; m < conf->raid_disks * 2 ; m++) 227262096bceSNeilBrown if (r1_bio->bios[m] == IO_MADE_GOOD) { 22733cb03002SNeilBrown struct md_rdev *rdev = conf->mirrors[m].rdev; 227462096bceSNeilBrown rdev_clear_badblocks(rdev, 227562096bceSNeilBrown r1_bio->sector, 2276c6563a8cSNeilBrown r1_bio->sectors, 0); 227762096bceSNeilBrown rdev_dec_pending(rdev, conf->mddev); 227862096bceSNeilBrown } else if (r1_bio->bios[m] != NULL) { 227962096bceSNeilBrown /* This drive got a write error. We need to 228062096bceSNeilBrown * narrow down and record precise write 228162096bceSNeilBrown * errors. 228262096bceSNeilBrown */ 228362096bceSNeilBrown if (!narrow_write_error(r1_bio, m)) { 228462096bceSNeilBrown md_error(conf->mddev, 228562096bceSNeilBrown conf->mirrors[m].rdev); 228662096bceSNeilBrown /* an I/O failed, we can't clear the bitmap */ 228762096bceSNeilBrown set_bit(R1BIO_Degraded, &r1_bio->state); 228862096bceSNeilBrown } 228962096bceSNeilBrown rdev_dec_pending(conf->mirrors[m].rdev, 229062096bceSNeilBrown conf->mddev); 229162096bceSNeilBrown } 229262096bceSNeilBrown if (test_bit(R1BIO_WriteError, &r1_bio->state)) 229362096bceSNeilBrown close_write(r1_bio); 229462096bceSNeilBrown raid_end_bio_io(r1_bio); 229562096bceSNeilBrown } 229662096bceSNeilBrown 2297e8096360SNeilBrown static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) 229862096bceSNeilBrown { 229962096bceSNeilBrown int disk; 230062096bceSNeilBrown int max_sectors; 2301fd01b88cSNeilBrown struct mddev *mddev = conf->mddev; 230262096bceSNeilBrown struct bio *bio; 230362096bceSNeilBrown char b[BDEVNAME_SIZE]; 23043cb03002SNeilBrown struct md_rdev *rdev; 230562096bceSNeilBrown 230662096bceSNeilBrown clear_bit(R1BIO_ReadError, &r1_bio->state); 230762096bceSNeilBrown /* we got a read error. Maybe the drive is bad. Maybe just 230862096bceSNeilBrown * the block and we can fix it. 230962096bceSNeilBrown * We freeze all other IO, and try reading the block from 231062096bceSNeilBrown * other devices. When we find one, we re-write 231162096bceSNeilBrown * and check it that fixes the read error. 231262096bceSNeilBrown * This is all done synchronously while the array is 231362096bceSNeilBrown * frozen 231462096bceSNeilBrown */ 231562096bceSNeilBrown if (mddev->ro == 0) { 2316e2d59925SNeilBrown freeze_array(conf, 1); 231762096bceSNeilBrown fix_read_error(conf, r1_bio->read_disk, 231862096bceSNeilBrown r1_bio->sector, r1_bio->sectors); 231962096bceSNeilBrown unfreeze_array(conf); 232062096bceSNeilBrown } else 232162096bceSNeilBrown md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); 23227ad4d4a6SNeilBrown rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); 232362096bceSNeilBrown 232462096bceSNeilBrown bio = r1_bio->bios[r1_bio->read_disk]; 232562096bceSNeilBrown bdevname(bio->bi_bdev, b); 232662096bceSNeilBrown read_more: 232762096bceSNeilBrown disk = read_balance(conf, r1_bio, &max_sectors); 232862096bceSNeilBrown if (disk == -1) { 232962096bceSNeilBrown printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" 233062096bceSNeilBrown " read error for block %llu\n", 233162096bceSNeilBrown mdname(mddev), b, (unsigned long long)r1_bio->sector); 233262096bceSNeilBrown raid_end_bio_io(r1_bio); 233362096bceSNeilBrown } else { 233462096bceSNeilBrown const unsigned long do_sync 233562096bceSNeilBrown = r1_bio->master_bio->bi_rw & REQ_SYNC; 233662096bceSNeilBrown if (bio) { 233762096bceSNeilBrown r1_bio->bios[r1_bio->read_disk] = 233862096bceSNeilBrown mddev->ro ? IO_BLOCKED : NULL; 233962096bceSNeilBrown bio_put(bio); 234062096bceSNeilBrown } 234162096bceSNeilBrown r1_bio->read_disk = disk; 234262096bceSNeilBrown bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); 23434f024f37SKent Overstreet bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, 23444f024f37SKent Overstreet max_sectors); 234562096bceSNeilBrown r1_bio->bios[r1_bio->read_disk] = bio; 234662096bceSNeilBrown rdev = conf->mirrors[disk].rdev; 234762096bceSNeilBrown printk_ratelimited(KERN_ERR 234862096bceSNeilBrown "md/raid1:%s: redirecting sector %llu" 234962096bceSNeilBrown " to other mirror: %s\n", 235062096bceSNeilBrown mdname(mddev), 235162096bceSNeilBrown (unsigned long long)r1_bio->sector, 235262096bceSNeilBrown bdevname(rdev->bdev, b)); 23534f024f37SKent Overstreet bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; 235462096bceSNeilBrown bio->bi_bdev = rdev->bdev; 235562096bceSNeilBrown bio->bi_end_io = raid1_end_read_request; 235662096bceSNeilBrown bio->bi_rw = READ | do_sync; 235762096bceSNeilBrown bio->bi_private = r1_bio; 235862096bceSNeilBrown if (max_sectors < r1_bio->sectors) { 235962096bceSNeilBrown /* Drat - have to split this up more */ 236062096bceSNeilBrown struct bio *mbio = r1_bio->master_bio; 236162096bceSNeilBrown int sectors_handled = (r1_bio->sector + max_sectors 23624f024f37SKent Overstreet - mbio->bi_iter.bi_sector); 236362096bceSNeilBrown r1_bio->sectors = max_sectors; 236462096bceSNeilBrown spin_lock_irq(&conf->device_lock); 236562096bceSNeilBrown if (mbio->bi_phys_segments == 0) 236662096bceSNeilBrown mbio->bi_phys_segments = 2; 236762096bceSNeilBrown else 236862096bceSNeilBrown mbio->bi_phys_segments++; 236962096bceSNeilBrown spin_unlock_irq(&conf->device_lock); 237062096bceSNeilBrown generic_make_request(bio); 237162096bceSNeilBrown bio = NULL; 237262096bceSNeilBrown 237362096bceSNeilBrown r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); 237462096bceSNeilBrown 237562096bceSNeilBrown r1_bio->master_bio = mbio; 2376aa8b57aaSKent Overstreet r1_bio->sectors = bio_sectors(mbio) - sectors_handled; 237762096bceSNeilBrown r1_bio->state = 0; 237862096bceSNeilBrown set_bit(R1BIO_ReadError, &r1_bio->state); 237962096bceSNeilBrown r1_bio->mddev = mddev; 23804f024f37SKent Overstreet r1_bio->sector = mbio->bi_iter.bi_sector + 23814f024f37SKent Overstreet sectors_handled; 238262096bceSNeilBrown 238362096bceSNeilBrown goto read_more; 238462096bceSNeilBrown } else 238562096bceSNeilBrown generic_make_request(bio); 238662096bceSNeilBrown } 238762096bceSNeilBrown } 238862096bceSNeilBrown 23894ed8731dSShaohua Li static void raid1d(struct md_thread *thread) 23901da177e4SLinus Torvalds { 23914ed8731dSShaohua Li struct mddev *mddev = thread->mddev; 23929f2c9d12SNeilBrown struct r1bio *r1_bio; 23931da177e4SLinus Torvalds unsigned long flags; 2394e8096360SNeilBrown struct r1conf *conf = mddev->private; 23951da177e4SLinus Torvalds struct list_head *head = &conf->retry_list; 2396e1dfa0a2SNeilBrown struct blk_plug plug; 23971da177e4SLinus Torvalds 23981da177e4SLinus Torvalds md_check_recovery(mddev); 23991da177e4SLinus Torvalds 2400e1dfa0a2SNeilBrown blk_start_plug(&plug); 24011da177e4SLinus Torvalds for (;;) { 2402a35e63efSNeilBrown 24037eaceaccSJens Axboe flush_pending_writes(conf); 2404a35e63efSNeilBrown 24051da177e4SLinus Torvalds spin_lock_irqsave(&conf->device_lock, flags); 2406a35e63efSNeilBrown if (list_empty(head)) { 2407191ea9b2SNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 24081da177e4SLinus Torvalds break; 2409a35e63efSNeilBrown } 24109f2c9d12SNeilBrown r1_bio = list_entry(head->prev, struct r1bio, retry_list); 24111da177e4SLinus Torvalds list_del(head->prev); 2412ddaf22abSNeilBrown conf->nr_queued--; 24131da177e4SLinus Torvalds spin_unlock_irqrestore(&conf->device_lock, flags); 24141da177e4SLinus Torvalds 24151da177e4SLinus Torvalds mddev = r1_bio->mddev; 2416070ec55dSNeilBrown conf = mddev->private; 24174367af55SNeilBrown if (test_bit(R1BIO_IsSync, &r1_bio->state)) { 2418d8f05d29SNeilBrown if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 241962096bceSNeilBrown test_bit(R1BIO_WriteError, &r1_bio->state)) 242062096bceSNeilBrown handle_sync_write_finished(conf, r1_bio); 242162096bceSNeilBrown else 24221da177e4SLinus Torvalds sync_request_write(mddev, r1_bio); 2423cd5ff9a1SNeilBrown } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || 242462096bceSNeilBrown test_bit(R1BIO_WriteError, &r1_bio->state)) 242562096bceSNeilBrown handle_write_finished(conf, r1_bio); 242662096bceSNeilBrown else if (test_bit(R1BIO_ReadError, &r1_bio->state)) 242762096bceSNeilBrown handle_read_error(conf, r1_bio); 2428d2eb35acSNeilBrown else 2429d2eb35acSNeilBrown /* just a partial read to be scheduled from separate 2430d2eb35acSNeilBrown * context 2431d2eb35acSNeilBrown */ 2432d2eb35acSNeilBrown generic_make_request(r1_bio->bios[r1_bio->read_disk]); 243362096bceSNeilBrown 24341d9d5241SNeilBrown cond_resched(); 2435de393cdeSNeilBrown if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) 2436de393cdeSNeilBrown md_check_recovery(mddev); 24371da177e4SLinus Torvalds } 2438e1dfa0a2SNeilBrown blk_finish_plug(&plug); 24391da177e4SLinus Torvalds } 24401da177e4SLinus Torvalds 24411da177e4SLinus Torvalds 2442e8096360SNeilBrown static int init_resync(struct r1conf *conf) 24431da177e4SLinus Torvalds { 24441da177e4SLinus Torvalds int buffs; 24451da177e4SLinus Torvalds 24461da177e4SLinus Torvalds buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; 24479e77c485SEric Sesterhenn BUG_ON(conf->r1buf_pool); 24481da177e4SLinus Torvalds conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, 24491da177e4SLinus Torvalds conf->poolinfo); 24501da177e4SLinus Torvalds if (!conf->r1buf_pool) 24511da177e4SLinus Torvalds return -ENOMEM; 24521da177e4SLinus Torvalds conf->next_resync = 0; 24531da177e4SLinus Torvalds return 0; 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds 24561da177e4SLinus Torvalds /* 24571da177e4SLinus Torvalds * perform a "sync" on one "block" 24581da177e4SLinus Torvalds * 24591da177e4SLinus Torvalds * We need to make sure that no normal I/O request - particularly write 24601da177e4SLinus Torvalds * requests - conflict with active sync requests. 24611da177e4SLinus Torvalds * 24621da177e4SLinus Torvalds * This is achieved by tracking pending requests and a 'barrier' concept 24631da177e4SLinus Torvalds * that can be installed to exclude normal IO requests. 24641da177e4SLinus Torvalds */ 24651da177e4SLinus Torvalds 2466fd01b88cSNeilBrown static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) 24671da177e4SLinus Torvalds { 2468e8096360SNeilBrown struct r1conf *conf = mddev->private; 24699f2c9d12SNeilBrown struct r1bio *r1_bio; 24701da177e4SLinus Torvalds struct bio *bio; 24711da177e4SLinus Torvalds sector_t max_sector, nr_sectors; 24723e198f78SNeilBrown int disk = -1; 24731da177e4SLinus Torvalds int i; 24743e198f78SNeilBrown int wonly = -1; 24753e198f78SNeilBrown int write_targets = 0, read_targets = 0; 247657dab0bdSNeilBrown sector_t sync_blocks; 2477e3b9703eSNeilBrown int still_degraded = 0; 247806f60385SNeilBrown int good_sectors = RESYNC_SECTORS; 247906f60385SNeilBrown int min_bad = 0; /* number of sectors that are bad in all devices */ 24801da177e4SLinus Torvalds 24811da177e4SLinus Torvalds if (!conf->r1buf_pool) 24821da177e4SLinus Torvalds if (init_resync(conf)) 248357afd89fSNeilBrown return 0; 24841da177e4SLinus Torvalds 248558c0fed4SAndre Noll max_sector = mddev->dev_sectors; 24861da177e4SLinus Torvalds if (sector_nr >= max_sector) { 2487191ea9b2SNeilBrown /* If we aborted, we need to abort the 2488191ea9b2SNeilBrown * sync on the 'current' bitmap chunk (there will 2489191ea9b2SNeilBrown * only be one in raid1 resync. 2490191ea9b2SNeilBrown * We can find the current addess in mddev->curr_resync 2491191ea9b2SNeilBrown */ 24926a806c51SNeilBrown if (mddev->curr_resync < max_sector) /* aborted */ 24936a806c51SNeilBrown bitmap_end_sync(mddev->bitmap, mddev->curr_resync, 2494191ea9b2SNeilBrown &sync_blocks, 1); 24956a806c51SNeilBrown else /* completed sync */ 2496191ea9b2SNeilBrown conf->fullsync = 0; 24976a806c51SNeilBrown 24986a806c51SNeilBrown bitmap_close_sync(mddev->bitmap); 24991da177e4SLinus Torvalds close_sync(conf); 25001da177e4SLinus Torvalds return 0; 25011da177e4SLinus Torvalds } 25021da177e4SLinus Torvalds 250307d84d10SNeilBrown if (mddev->bitmap == NULL && 250407d84d10SNeilBrown mddev->recovery_cp == MaxSector && 25056394cca5SNeilBrown !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && 250607d84d10SNeilBrown conf->fullsync == 0) { 250707d84d10SNeilBrown *skipped = 1; 250807d84d10SNeilBrown return max_sector - sector_nr; 250907d84d10SNeilBrown } 25106394cca5SNeilBrown /* before building a request, check if we can skip these blocks.. 25116394cca5SNeilBrown * This call the bitmap_start_sync doesn't actually record anything 25126394cca5SNeilBrown */ 2513e3b9703eSNeilBrown if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && 2514e5de485fSNeilBrown !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 2515191ea9b2SNeilBrown /* We can skip this block, and probably several more */ 2516191ea9b2SNeilBrown *skipped = 1; 2517191ea9b2SNeilBrown return sync_blocks; 2518191ea9b2SNeilBrown } 25191da177e4SLinus Torvalds /* 252017999be4SNeilBrown * If there is non-resync activity waiting for a turn, 252117999be4SNeilBrown * and resync is going fast enough, 252217999be4SNeilBrown * then let it though before starting on this new sync request. 25231da177e4SLinus Torvalds */ 252417999be4SNeilBrown if (!go_faster && conf->nr_waiting) 25251da177e4SLinus Torvalds msleep_interruptible(1000); 252617999be4SNeilBrown 2527b47490c9SNeilBrown bitmap_cond_end_sync(mddev->bitmap, sector_nr); 25281c4588e9SNeilBrown r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); 252917999be4SNeilBrown raise_barrier(conf); 253017999be4SNeilBrown 253117999be4SNeilBrown conf->next_resync = sector_nr; 25321da177e4SLinus Torvalds 25333e198f78SNeilBrown rcu_read_lock(); 25343e198f78SNeilBrown /* 25353e198f78SNeilBrown * If we get a correctably read error during resync or recovery, 25363e198f78SNeilBrown * we might want to read from a different device. So we 25373e198f78SNeilBrown * flag all drives that could conceivably be read from for READ, 25383e198f78SNeilBrown * and any others (which will be non-In_sync devices) for WRITE. 25393e198f78SNeilBrown * If a read fails, we try reading from something else for which READ 25403e198f78SNeilBrown * is OK. 25413e198f78SNeilBrown */ 25421da177e4SLinus Torvalds 25431da177e4SLinus Torvalds r1_bio->mddev = mddev; 25441da177e4SLinus Torvalds r1_bio->sector = sector_nr; 2545191ea9b2SNeilBrown r1_bio->state = 0; 25461da177e4SLinus Torvalds set_bit(R1BIO_IsSync, &r1_bio->state); 25471da177e4SLinus Torvalds 25488f19ccb2SNeilBrown for (i = 0; i < conf->raid_disks * 2; i++) { 25493cb03002SNeilBrown struct md_rdev *rdev; 25501da177e4SLinus Torvalds bio = r1_bio->bios[i]; 25512aabaa65SKent Overstreet bio_reset(bio); 25521da177e4SLinus Torvalds 25533e198f78SNeilBrown rdev = rcu_dereference(conf->mirrors[i].rdev); 25543e198f78SNeilBrown if (rdev == NULL || 25553e198f78SNeilBrown test_bit(Faulty, &rdev->flags)) { 25568f19ccb2SNeilBrown if (i < conf->raid_disks) 2557e3b9703eSNeilBrown still_degraded = 1; 25583e198f78SNeilBrown } else if (!test_bit(In_sync, &rdev->flags)) { 25591da177e4SLinus Torvalds bio->bi_rw = WRITE; 25601da177e4SLinus Torvalds bio->bi_end_io = end_sync_write; 25611da177e4SLinus Torvalds write_targets ++; 25623e198f78SNeilBrown } else { 25633e198f78SNeilBrown /* may need to read from here */ 256406f60385SNeilBrown sector_t first_bad = MaxSector; 256506f60385SNeilBrown int bad_sectors; 256606f60385SNeilBrown 256706f60385SNeilBrown if (is_badblock(rdev, sector_nr, good_sectors, 256806f60385SNeilBrown &first_bad, &bad_sectors)) { 256906f60385SNeilBrown if (first_bad > sector_nr) 257006f60385SNeilBrown good_sectors = first_bad - sector_nr; 257106f60385SNeilBrown else { 257206f60385SNeilBrown bad_sectors -= (sector_nr - first_bad); 257306f60385SNeilBrown if (min_bad == 0 || 257406f60385SNeilBrown min_bad > bad_sectors) 257506f60385SNeilBrown min_bad = bad_sectors; 257606f60385SNeilBrown } 257706f60385SNeilBrown } 257806f60385SNeilBrown if (sector_nr < first_bad) { 25793e198f78SNeilBrown if (test_bit(WriteMostly, &rdev->flags)) { 25803e198f78SNeilBrown if (wonly < 0) 25813e198f78SNeilBrown wonly = i; 25823e198f78SNeilBrown } else { 25833e198f78SNeilBrown if (disk < 0) 25843e198f78SNeilBrown disk = i; 25853e198f78SNeilBrown } 258606f60385SNeilBrown bio->bi_rw = READ; 258706f60385SNeilBrown bio->bi_end_io = end_sync_read; 25883e198f78SNeilBrown read_targets++; 2589d57368afSAlexander Lyakas } else if (!test_bit(WriteErrorSeen, &rdev->flags) && 2590d57368afSAlexander Lyakas test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && 2591d57368afSAlexander Lyakas !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { 2592d57368afSAlexander Lyakas /* 2593d57368afSAlexander Lyakas * The device is suitable for reading (InSync), 2594d57368afSAlexander Lyakas * but has bad block(s) here. Let's try to correct them, 2595d57368afSAlexander Lyakas * if we are doing resync or repair. Otherwise, leave 2596d57368afSAlexander Lyakas * this device alone for this sync request. 2597d57368afSAlexander Lyakas */ 2598d57368afSAlexander Lyakas bio->bi_rw = WRITE; 2599d57368afSAlexander Lyakas bio->bi_end_io = end_sync_write; 2600d57368afSAlexander Lyakas write_targets++; 26013e198f78SNeilBrown } 260206f60385SNeilBrown } 260306f60385SNeilBrown if (bio->bi_end_io) { 26043e198f78SNeilBrown atomic_inc(&rdev->nr_pending); 26054f024f37SKent Overstreet bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; 26063e198f78SNeilBrown bio->bi_bdev = rdev->bdev; 26071da177e4SLinus Torvalds bio->bi_private = r1_bio; 26081da177e4SLinus Torvalds } 260906f60385SNeilBrown } 26103e198f78SNeilBrown rcu_read_unlock(); 26113e198f78SNeilBrown if (disk < 0) 26123e198f78SNeilBrown disk = wonly; 26133e198f78SNeilBrown r1_bio->read_disk = disk; 2614191ea9b2SNeilBrown 261506f60385SNeilBrown if (read_targets == 0 && min_bad > 0) { 261606f60385SNeilBrown /* These sectors are bad on all InSync devices, so we 261706f60385SNeilBrown * need to mark them bad on all write targets 261806f60385SNeilBrown */ 261906f60385SNeilBrown int ok = 1; 26208f19ccb2SNeilBrown for (i = 0 ; i < conf->raid_disks * 2 ; i++) 262106f60385SNeilBrown if (r1_bio->bios[i]->bi_end_io == end_sync_write) { 2622a42f9d83Smajianpeng struct md_rdev *rdev = conf->mirrors[i].rdev; 262306f60385SNeilBrown ok = rdev_set_badblocks(rdev, sector_nr, 262406f60385SNeilBrown min_bad, 0 262506f60385SNeilBrown ) && ok; 262606f60385SNeilBrown } 262706f60385SNeilBrown set_bit(MD_CHANGE_DEVS, &mddev->flags); 262806f60385SNeilBrown *skipped = 1; 262906f60385SNeilBrown put_buf(r1_bio); 263006f60385SNeilBrown 263106f60385SNeilBrown if (!ok) { 263206f60385SNeilBrown /* Cannot record the badblocks, so need to 263306f60385SNeilBrown * abort the resync. 263406f60385SNeilBrown * If there are multiple read targets, could just 263506f60385SNeilBrown * fail the really bad ones ??? 263606f60385SNeilBrown */ 263706f60385SNeilBrown conf->recovery_disabled = mddev->recovery_disabled; 263806f60385SNeilBrown set_bit(MD_RECOVERY_INTR, &mddev->recovery); 263906f60385SNeilBrown return 0; 264006f60385SNeilBrown } else 264106f60385SNeilBrown return min_bad; 264206f60385SNeilBrown 264306f60385SNeilBrown } 264406f60385SNeilBrown if (min_bad > 0 && min_bad < good_sectors) { 264506f60385SNeilBrown /* only resync enough to reach the next bad->good 264606f60385SNeilBrown * transition */ 264706f60385SNeilBrown good_sectors = min_bad; 264806f60385SNeilBrown } 264906f60385SNeilBrown 26503e198f78SNeilBrown if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) 26513e198f78SNeilBrown /* extra read targets are also write targets */ 26523e198f78SNeilBrown write_targets += read_targets-1; 26533e198f78SNeilBrown 26543e198f78SNeilBrown if (write_targets == 0 || read_targets == 0) { 26551da177e4SLinus Torvalds /* There is nowhere to write, so all non-sync 26561da177e4SLinus Torvalds * drives must be failed - so we are finished 26571da177e4SLinus Torvalds */ 2658b7219ccbSNeilBrown sector_t rv; 2659b7219ccbSNeilBrown if (min_bad > 0) 2660b7219ccbSNeilBrown max_sector = sector_nr + min_bad; 2661b7219ccbSNeilBrown rv = max_sector - sector_nr; 266257afd89fSNeilBrown *skipped = 1; 26631da177e4SLinus Torvalds put_buf(r1_bio); 26641da177e4SLinus Torvalds return rv; 26651da177e4SLinus Torvalds } 26661da177e4SLinus Torvalds 2667c6207277SNeilBrown if (max_sector > mddev->resync_max) 2668c6207277SNeilBrown max_sector = mddev->resync_max; /* Don't do IO beyond here */ 266906f60385SNeilBrown if (max_sector > sector_nr + good_sectors) 267006f60385SNeilBrown max_sector = sector_nr + good_sectors; 26711da177e4SLinus Torvalds nr_sectors = 0; 2672289e99e8SNeilBrown sync_blocks = 0; 26731da177e4SLinus Torvalds do { 26741da177e4SLinus Torvalds struct page *page; 26751da177e4SLinus Torvalds int len = PAGE_SIZE; 26761da177e4SLinus Torvalds if (sector_nr + (len>>9) > max_sector) 26771da177e4SLinus Torvalds len = (max_sector - sector_nr) << 9; 26781da177e4SLinus Torvalds if (len == 0) 26791da177e4SLinus Torvalds break; 2680ab7a30c7SNeilBrown if (sync_blocks == 0) { 26816a806c51SNeilBrown if (!bitmap_start_sync(mddev->bitmap, sector_nr, 2682e3b9703eSNeilBrown &sync_blocks, still_degraded) && 2683e5de485fSNeilBrown !conf->fullsync && 2684e5de485fSNeilBrown !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) 2685191ea9b2SNeilBrown break; 26869e77c485SEric Sesterhenn BUG_ON(sync_blocks < (PAGE_SIZE>>9)); 26877571ae88SNeilBrown if ((len >> 9) > sync_blocks) 26886a806c51SNeilBrown len = sync_blocks<<9; 2689ab7a30c7SNeilBrown } 2690191ea9b2SNeilBrown 26918f19ccb2SNeilBrown for (i = 0 ; i < conf->raid_disks * 2; i++) { 26921da177e4SLinus Torvalds bio = r1_bio->bios[i]; 26931da177e4SLinus Torvalds if (bio->bi_end_io) { 2694d11c171eSNeilBrown page = bio->bi_io_vec[bio->bi_vcnt].bv_page; 26951da177e4SLinus Torvalds if (bio_add_page(bio, page, len, 0) == 0) { 26961da177e4SLinus Torvalds /* stop here */ 2697d11c171eSNeilBrown bio->bi_io_vec[bio->bi_vcnt].bv_page = page; 26981da177e4SLinus Torvalds while (i > 0) { 26991da177e4SLinus Torvalds i--; 27001da177e4SLinus Torvalds bio = r1_bio->bios[i]; 27016a806c51SNeilBrown if (bio->bi_end_io==NULL) 27026a806c51SNeilBrown continue; 27031da177e4SLinus Torvalds /* remove last page from this bio */ 27041da177e4SLinus Torvalds bio->bi_vcnt--; 27054f024f37SKent Overstreet bio->bi_iter.bi_size -= len; 27061da177e4SLinus Torvalds bio->bi_flags &= ~(1<< BIO_SEG_VALID); 27071da177e4SLinus Torvalds } 27081da177e4SLinus Torvalds goto bio_full; 27091da177e4SLinus Torvalds } 27101da177e4SLinus Torvalds } 27111da177e4SLinus Torvalds } 27121da177e4SLinus Torvalds nr_sectors += len>>9; 27131da177e4SLinus Torvalds sector_nr += len>>9; 2714191ea9b2SNeilBrown sync_blocks -= (len>>9); 27151da177e4SLinus Torvalds } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); 27161da177e4SLinus Torvalds bio_full: 27171da177e4SLinus Torvalds r1_bio->sectors = nr_sectors; 27181da177e4SLinus Torvalds 2719d11c171eSNeilBrown /* For a user-requested sync, we read all readable devices and do a 2720d11c171eSNeilBrown * compare 2721d11c171eSNeilBrown */ 2722d11c171eSNeilBrown if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { 2723d11c171eSNeilBrown atomic_set(&r1_bio->remaining, read_targets); 27242d4f4f33SNeilBrown for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { 2725d11c171eSNeilBrown bio = r1_bio->bios[i]; 2726d11c171eSNeilBrown if (bio->bi_end_io == end_sync_read) { 27272d4f4f33SNeilBrown read_targets--; 2728ddac7c7eSNeilBrown md_sync_acct(bio->bi_bdev, nr_sectors); 27291da177e4SLinus Torvalds generic_make_request(bio); 2730d11c171eSNeilBrown } 2731d11c171eSNeilBrown } 2732d11c171eSNeilBrown } else { 2733d11c171eSNeilBrown atomic_set(&r1_bio->remaining, 1); 2734d11c171eSNeilBrown bio = r1_bio->bios[r1_bio->read_disk]; 2735ddac7c7eSNeilBrown md_sync_acct(bio->bi_bdev, nr_sectors); 2736d11c171eSNeilBrown generic_make_request(bio); 2737d11c171eSNeilBrown 2738d11c171eSNeilBrown } 27391da177e4SLinus Torvalds return nr_sectors; 27401da177e4SLinus Torvalds } 27411da177e4SLinus Torvalds 2742fd01b88cSNeilBrown static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) 274380c3a6ceSDan Williams { 274480c3a6ceSDan Williams if (sectors) 274580c3a6ceSDan Williams return sectors; 274680c3a6ceSDan Williams 274780c3a6ceSDan Williams return mddev->dev_sectors; 274880c3a6ceSDan Williams } 274980c3a6ceSDan Williams 2750e8096360SNeilBrown static struct r1conf *setup_conf(struct mddev *mddev) 27511da177e4SLinus Torvalds { 2752e8096360SNeilBrown struct r1conf *conf; 2753709ae487SNeilBrown int i; 27540eaf822cSJonathan Brassow struct raid1_info *disk; 27553cb03002SNeilBrown struct md_rdev *rdev; 2756709ae487SNeilBrown int err = -ENOMEM; 27571da177e4SLinus Torvalds 2758e8096360SNeilBrown conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); 27591da177e4SLinus Torvalds if (!conf) 2760709ae487SNeilBrown goto abort; 27611da177e4SLinus Torvalds 27620eaf822cSJonathan Brassow conf->mirrors = kzalloc(sizeof(struct raid1_info) 27638f19ccb2SNeilBrown * mddev->raid_disks * 2, 27641da177e4SLinus Torvalds GFP_KERNEL); 27651da177e4SLinus Torvalds if (!conf->mirrors) 2766709ae487SNeilBrown goto abort; 27671da177e4SLinus Torvalds 2768ddaf22abSNeilBrown conf->tmppage = alloc_page(GFP_KERNEL); 2769ddaf22abSNeilBrown if (!conf->tmppage) 2770709ae487SNeilBrown goto abort; 2771ddaf22abSNeilBrown 2772709ae487SNeilBrown conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); 27731da177e4SLinus Torvalds if (!conf->poolinfo) 2774709ae487SNeilBrown goto abort; 27758f19ccb2SNeilBrown conf->poolinfo->raid_disks = mddev->raid_disks * 2; 27761da177e4SLinus Torvalds conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 27771da177e4SLinus Torvalds r1bio_pool_free, 27781da177e4SLinus Torvalds conf->poolinfo); 27791da177e4SLinus Torvalds if (!conf->r1bio_pool) 2780709ae487SNeilBrown goto abort; 2781709ae487SNeilBrown 2782ed9bfdf1SNeilBrown conf->poolinfo->mddev = mddev; 27831da177e4SLinus Torvalds 2784c19d5798SNeilBrown err = -EINVAL; 2785e7e72bf6SNeil Brown spin_lock_init(&conf->device_lock); 2786dafb20faSNeilBrown rdev_for_each(rdev, mddev) { 2787aba336bdSNeilBrown struct request_queue *q; 2788709ae487SNeilBrown int disk_idx = rdev->raid_disk; 27891da177e4SLinus Torvalds if (disk_idx >= mddev->raid_disks 27901da177e4SLinus Torvalds || disk_idx < 0) 27911da177e4SLinus Torvalds continue; 2792c19d5798SNeilBrown if (test_bit(Replacement, &rdev->flags)) 279302b898f2SNeilBrown disk = conf->mirrors + mddev->raid_disks + disk_idx; 2794c19d5798SNeilBrown else 27951da177e4SLinus Torvalds disk = conf->mirrors + disk_idx; 27961da177e4SLinus Torvalds 2797c19d5798SNeilBrown if (disk->rdev) 2798c19d5798SNeilBrown goto abort; 27991da177e4SLinus Torvalds disk->rdev = rdev; 2800aba336bdSNeilBrown q = bdev_get_queue(rdev->bdev); 2801aba336bdSNeilBrown if (q->merge_bvec_fn) 2802aba336bdSNeilBrown mddev->merge_check_needed = 1; 28031da177e4SLinus Torvalds 28041da177e4SLinus Torvalds disk->head_position = 0; 280512cee5a8SShaohua Li disk->seq_start = MaxSector; 28061da177e4SLinus Torvalds } 28071da177e4SLinus Torvalds conf->raid_disks = mddev->raid_disks; 28081da177e4SLinus Torvalds conf->mddev = mddev; 28091da177e4SLinus Torvalds INIT_LIST_HEAD(&conf->retry_list); 28101da177e4SLinus Torvalds 28111da177e4SLinus Torvalds spin_lock_init(&conf->resync_lock); 281217999be4SNeilBrown init_waitqueue_head(&conf->wait_barrier); 28131da177e4SLinus Torvalds 2814191ea9b2SNeilBrown bio_list_init(&conf->pending_bio_list); 281534db0cd6SNeilBrown conf->pending_count = 0; 2816d890fa2bSNeilBrown conf->recovery_disabled = mddev->recovery_disabled - 1; 2817191ea9b2SNeilBrown 281879ef3a8aSmajianpeng conf->start_next_window = MaxSector; 281979ef3a8aSmajianpeng conf->current_window_requests = conf->next_window_requests = 0; 282079ef3a8aSmajianpeng 2821c19d5798SNeilBrown err = -EIO; 28228f19ccb2SNeilBrown for (i = 0; i < conf->raid_disks * 2; i++) { 28231da177e4SLinus Torvalds 28241da177e4SLinus Torvalds disk = conf->mirrors + i; 28251da177e4SLinus Torvalds 2826c19d5798SNeilBrown if (i < conf->raid_disks && 2827c19d5798SNeilBrown disk[conf->raid_disks].rdev) { 2828c19d5798SNeilBrown /* This slot has a replacement. */ 2829c19d5798SNeilBrown if (!disk->rdev) { 2830c19d5798SNeilBrown /* No original, just make the replacement 2831c19d5798SNeilBrown * a recovering spare 2832c19d5798SNeilBrown */ 2833c19d5798SNeilBrown disk->rdev = 2834c19d5798SNeilBrown disk[conf->raid_disks].rdev; 2835c19d5798SNeilBrown disk[conf->raid_disks].rdev = NULL; 2836c19d5798SNeilBrown } else if (!test_bit(In_sync, &disk->rdev->flags)) 2837c19d5798SNeilBrown /* Original is not in_sync - bad */ 2838c19d5798SNeilBrown goto abort; 2839c19d5798SNeilBrown } 2840c19d5798SNeilBrown 28415fd6c1dcSNeilBrown if (!disk->rdev || 28425fd6c1dcSNeilBrown !test_bit(In_sync, &disk->rdev->flags)) { 28431da177e4SLinus Torvalds disk->head_position = 0; 28444f0a5e01SJonathan Brassow if (disk->rdev && 28454f0a5e01SJonathan Brassow (disk->rdev->saved_raid_disk < 0)) 284617571284SNeilBrown conf->fullsync = 1; 2847be4d3280SShaohua Li } 28481da177e4SLinus Torvalds } 2849709ae487SNeilBrown 2850709ae487SNeilBrown err = -ENOMEM; 28510232605dSNeilBrown conf->thread = md_register_thread(raid1d, mddev, "raid1"); 2852709ae487SNeilBrown if (!conf->thread) { 28531da177e4SLinus Torvalds printk(KERN_ERR 28549dd1e2faSNeilBrown "md/raid1:%s: couldn't allocate thread\n", 28551da177e4SLinus Torvalds mdname(mddev)); 2856709ae487SNeilBrown goto abort; 28571da177e4SLinus Torvalds } 2858191ea9b2SNeilBrown 2859709ae487SNeilBrown return conf; 2860709ae487SNeilBrown 2861709ae487SNeilBrown abort: 2862709ae487SNeilBrown if (conf) { 2863709ae487SNeilBrown if (conf->r1bio_pool) 2864709ae487SNeilBrown mempool_destroy(conf->r1bio_pool); 2865709ae487SNeilBrown kfree(conf->mirrors); 2866709ae487SNeilBrown safe_put_page(conf->tmppage); 2867709ae487SNeilBrown kfree(conf->poolinfo); 2868709ae487SNeilBrown kfree(conf); 2869709ae487SNeilBrown } 2870709ae487SNeilBrown return ERR_PTR(err); 2871709ae487SNeilBrown } 2872709ae487SNeilBrown 28735220ea1eSmajianpeng static int stop(struct mddev *mddev); 2874fd01b88cSNeilBrown static int run(struct mddev *mddev) 2875709ae487SNeilBrown { 2876e8096360SNeilBrown struct r1conf *conf; 2877709ae487SNeilBrown int i; 28783cb03002SNeilBrown struct md_rdev *rdev; 28795220ea1eSmajianpeng int ret; 28802ff8cc2cSShaohua Li bool discard_supported = false; 2881709ae487SNeilBrown 2882709ae487SNeilBrown if (mddev->level != 1) { 28839dd1e2faSNeilBrown printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", 2884709ae487SNeilBrown mdname(mddev), mddev->level); 2885709ae487SNeilBrown return -EIO; 2886709ae487SNeilBrown } 2887709ae487SNeilBrown if (mddev->reshape_position != MaxSector) { 28889dd1e2faSNeilBrown printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", 2889709ae487SNeilBrown mdname(mddev)); 2890709ae487SNeilBrown return -EIO; 2891709ae487SNeilBrown } 2892709ae487SNeilBrown /* 2893709ae487SNeilBrown * copy the already verified devices into our private RAID1 2894709ae487SNeilBrown * bookkeeping area. [whatever we allocate in run(), 2895709ae487SNeilBrown * should be freed in stop()] 2896709ae487SNeilBrown */ 2897709ae487SNeilBrown if (mddev->private == NULL) 2898709ae487SNeilBrown conf = setup_conf(mddev); 2899709ae487SNeilBrown else 2900709ae487SNeilBrown conf = mddev->private; 2901709ae487SNeilBrown 2902709ae487SNeilBrown if (IS_ERR(conf)) 2903709ae487SNeilBrown return PTR_ERR(conf); 2904709ae487SNeilBrown 2905c8dc9c65SJoe Lawrence if (mddev->queue) 29065026d7a9SH. Peter Anvin blk_queue_max_write_same_sectors(mddev->queue, 0); 29075026d7a9SH. Peter Anvin 2908dafb20faSNeilBrown rdev_for_each(rdev, mddev) { 29091ed7242eSJonathan Brassow if (!mddev->gendisk) 29101ed7242eSJonathan Brassow continue; 2911709ae487SNeilBrown disk_stack_limits(mddev->gendisk, rdev->bdev, 2912709ae487SNeilBrown rdev->data_offset << 9); 29132ff8cc2cSShaohua Li if (blk_queue_discard(bdev_get_queue(rdev->bdev))) 29142ff8cc2cSShaohua Li discard_supported = true; 2915709ae487SNeilBrown } 2916709ae487SNeilBrown 2917709ae487SNeilBrown mddev->degraded = 0; 2918709ae487SNeilBrown for (i=0; i < conf->raid_disks; i++) 2919709ae487SNeilBrown if (conf->mirrors[i].rdev == NULL || 2920709ae487SNeilBrown !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || 2921709ae487SNeilBrown test_bit(Faulty, &conf->mirrors[i].rdev->flags)) 2922709ae487SNeilBrown mddev->degraded++; 2923709ae487SNeilBrown 2924709ae487SNeilBrown if (conf->raid_disks - mddev->degraded == 1) 2925709ae487SNeilBrown mddev->recovery_cp = MaxSector; 2926709ae487SNeilBrown 29278c6ac868SAndre Noll if (mddev->recovery_cp != MaxSector) 29289dd1e2faSNeilBrown printk(KERN_NOTICE "md/raid1:%s: not clean" 29298c6ac868SAndre Noll " -- starting background reconstruction\n", 29308c6ac868SAndre Noll mdname(mddev)); 29311da177e4SLinus Torvalds printk(KERN_INFO 29329dd1e2faSNeilBrown "md/raid1:%s: active with %d out of %d mirrors\n", 29331da177e4SLinus Torvalds mdname(mddev), mddev->raid_disks - mddev->degraded, 29341da177e4SLinus Torvalds mddev->raid_disks); 2935709ae487SNeilBrown 29361da177e4SLinus Torvalds /* 29371da177e4SLinus Torvalds * Ok, everything is just fine now 29381da177e4SLinus Torvalds */ 2939709ae487SNeilBrown mddev->thread = conf->thread; 2940709ae487SNeilBrown conf->thread = NULL; 2941709ae487SNeilBrown mddev->private = conf; 2942709ae487SNeilBrown 29431f403624SDan Williams md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); 29441da177e4SLinus Torvalds 29451ed7242eSJonathan Brassow if (mddev->queue) { 29460d129228SNeilBrown mddev->queue->backing_dev_info.congested_fn = raid1_congested; 29470d129228SNeilBrown mddev->queue->backing_dev_info.congested_data = mddev; 29486b740b8dSNeilBrown blk_queue_merge_bvec(mddev->queue, raid1_mergeable_bvec); 29492ff8cc2cSShaohua Li 29502ff8cc2cSShaohua Li if (discard_supported) 29512ff8cc2cSShaohua Li queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, 29522ff8cc2cSShaohua Li mddev->queue); 29532ff8cc2cSShaohua Li else 29542ff8cc2cSShaohua Li queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, 29552ff8cc2cSShaohua Li mddev->queue); 29561ed7242eSJonathan Brassow } 29575220ea1eSmajianpeng 29585220ea1eSmajianpeng ret = md_integrity_register(mddev); 29595220ea1eSmajianpeng if (ret) 29605220ea1eSmajianpeng stop(mddev); 29615220ea1eSmajianpeng return ret; 29621da177e4SLinus Torvalds } 29631da177e4SLinus Torvalds 2964fd01b88cSNeilBrown static int stop(struct mddev *mddev) 29651da177e4SLinus Torvalds { 2966e8096360SNeilBrown struct r1conf *conf = mddev->private; 29674b6d287fSNeilBrown struct bitmap *bitmap = mddev->bitmap; 29684b6d287fSNeilBrown 29694b6d287fSNeilBrown /* wait for behind writes to complete */ 2970e555190dSNeilBrown if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { 29719dd1e2faSNeilBrown printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n", 29729dd1e2faSNeilBrown mdname(mddev)); 29734b6d287fSNeilBrown /* need to kick something here to make sure I/O goes? */ 2974e555190dSNeilBrown wait_event(bitmap->behind_wait, 2975e555190dSNeilBrown atomic_read(&bitmap->behind_writes) == 0); 29764b6d287fSNeilBrown } 29771da177e4SLinus Torvalds 297807169fd4Smajianpeng freeze_array(conf, 0); 297907169fd4Smajianpeng unfreeze_array(conf); 2980409c57f3SNeilBrown 298101f96c0aSNeilBrown md_unregister_thread(&mddev->thread); 29821da177e4SLinus Torvalds if (conf->r1bio_pool) 29831da177e4SLinus Torvalds mempool_destroy(conf->r1bio_pool); 29841da177e4SLinus Torvalds kfree(conf->mirrors); 29850fea7ed8SHirokazu Takahashi safe_put_page(conf->tmppage); 29861da177e4SLinus Torvalds kfree(conf->poolinfo); 29871da177e4SLinus Torvalds kfree(conf); 29881da177e4SLinus Torvalds mddev->private = NULL; 29891da177e4SLinus Torvalds return 0; 29901da177e4SLinus Torvalds } 29911da177e4SLinus Torvalds 2992fd01b88cSNeilBrown static int raid1_resize(struct mddev *mddev, sector_t sectors) 29931da177e4SLinus Torvalds { 29941da177e4SLinus Torvalds /* no resync is happening, and there is enough space 29951da177e4SLinus Torvalds * on all devices, so we can resize. 29961da177e4SLinus Torvalds * We need to make sure resync covers any new space. 29971da177e4SLinus Torvalds * If the array is shrinking we should possibly wait until 29981da177e4SLinus Torvalds * any io in the removed space completes, but it hardly seems 29991da177e4SLinus Torvalds * worth it. 30001da177e4SLinus Torvalds */ 3001a4a6125aSNeilBrown sector_t newsize = raid1_size(mddev, sectors, 0); 3002a4a6125aSNeilBrown if (mddev->external_size && 3003a4a6125aSNeilBrown mddev->array_sectors > newsize) 3004b522adcdSDan Williams return -EINVAL; 3005a4a6125aSNeilBrown if (mddev->bitmap) { 3006a4a6125aSNeilBrown int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); 3007a4a6125aSNeilBrown if (ret) 3008a4a6125aSNeilBrown return ret; 3009a4a6125aSNeilBrown } 3010a4a6125aSNeilBrown md_set_array_sectors(mddev, newsize); 3011f233ea5cSAndre Noll set_capacity(mddev->gendisk, mddev->array_sectors); 3012449aad3eSNeilBrown revalidate_disk(mddev->gendisk); 3013b522adcdSDan Williams if (sectors > mddev->dev_sectors && 3014b098636cSNeilBrown mddev->recovery_cp > mddev->dev_sectors) { 301558c0fed4SAndre Noll mddev->recovery_cp = mddev->dev_sectors; 30161da177e4SLinus Torvalds set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 30171da177e4SLinus Torvalds } 3018b522adcdSDan Williams mddev->dev_sectors = sectors; 30194b5c7ae8SNeilBrown mddev->resync_max_sectors = sectors; 30201da177e4SLinus Torvalds return 0; 30211da177e4SLinus Torvalds } 30221da177e4SLinus Torvalds 3023fd01b88cSNeilBrown static int raid1_reshape(struct mddev *mddev) 30241da177e4SLinus Torvalds { 30251da177e4SLinus Torvalds /* We need to: 30261da177e4SLinus Torvalds * 1/ resize the r1bio_pool 30271da177e4SLinus Torvalds * 2/ resize conf->mirrors 30281da177e4SLinus Torvalds * 30291da177e4SLinus Torvalds * We allocate a new r1bio_pool if we can. 30301da177e4SLinus Torvalds * Then raise a device barrier and wait until all IO stops. 30311da177e4SLinus Torvalds * Then resize conf->mirrors and swap in the new r1bio pool. 30326ea9c07cSNeilBrown * 30336ea9c07cSNeilBrown * At the same time, we "pack" the devices so that all the missing 30346ea9c07cSNeilBrown * devices have the higher raid_disk numbers. 30351da177e4SLinus Torvalds */ 30361da177e4SLinus Torvalds mempool_t *newpool, *oldpool; 30371da177e4SLinus Torvalds struct pool_info *newpoolinfo; 30380eaf822cSJonathan Brassow struct raid1_info *newmirrors; 3039e8096360SNeilBrown struct r1conf *conf = mddev->private; 304063c70c4fSNeilBrown int cnt, raid_disks; 3041c04be0aaSNeilBrown unsigned long flags; 3042b5470dc5SDan Williams int d, d2, err; 30431da177e4SLinus Torvalds 304463c70c4fSNeilBrown /* Cannot change chunk_size, layout, or level */ 3045664e7c41SAndre Noll if (mddev->chunk_sectors != mddev->new_chunk_sectors || 304663c70c4fSNeilBrown mddev->layout != mddev->new_layout || 304763c70c4fSNeilBrown mddev->level != mddev->new_level) { 3048664e7c41SAndre Noll mddev->new_chunk_sectors = mddev->chunk_sectors; 304963c70c4fSNeilBrown mddev->new_layout = mddev->layout; 305063c70c4fSNeilBrown mddev->new_level = mddev->level; 305163c70c4fSNeilBrown return -EINVAL; 305263c70c4fSNeilBrown } 305363c70c4fSNeilBrown 3054b5470dc5SDan Williams err = md_allow_write(mddev); 3055b5470dc5SDan Williams if (err) 3056b5470dc5SDan Williams return err; 30572a2275d6SNeilBrown 305863c70c4fSNeilBrown raid_disks = mddev->raid_disks + mddev->delta_disks; 305963c70c4fSNeilBrown 30606ea9c07cSNeilBrown if (raid_disks < conf->raid_disks) { 30616ea9c07cSNeilBrown cnt=0; 30626ea9c07cSNeilBrown for (d= 0; d < conf->raid_disks; d++) 30631da177e4SLinus Torvalds if (conf->mirrors[d].rdev) 30646ea9c07cSNeilBrown cnt++; 30656ea9c07cSNeilBrown if (cnt > raid_disks) 30661da177e4SLinus Torvalds return -EBUSY; 30676ea9c07cSNeilBrown } 30681da177e4SLinus Torvalds 30691da177e4SLinus Torvalds newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); 30701da177e4SLinus Torvalds if (!newpoolinfo) 30711da177e4SLinus Torvalds return -ENOMEM; 30721da177e4SLinus Torvalds newpoolinfo->mddev = mddev; 30738f19ccb2SNeilBrown newpoolinfo->raid_disks = raid_disks * 2; 30741da177e4SLinus Torvalds 30751da177e4SLinus Torvalds newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, 30761da177e4SLinus Torvalds r1bio_pool_free, newpoolinfo); 30771da177e4SLinus Torvalds if (!newpool) { 30781da177e4SLinus Torvalds kfree(newpoolinfo); 30791da177e4SLinus Torvalds return -ENOMEM; 30801da177e4SLinus Torvalds } 30810eaf822cSJonathan Brassow newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, 30828f19ccb2SNeilBrown GFP_KERNEL); 30831da177e4SLinus Torvalds if (!newmirrors) { 30841da177e4SLinus Torvalds kfree(newpoolinfo); 30851da177e4SLinus Torvalds mempool_destroy(newpool); 30861da177e4SLinus Torvalds return -ENOMEM; 30871da177e4SLinus Torvalds } 30881da177e4SLinus Torvalds 3089e2d59925SNeilBrown freeze_array(conf, 0); 30901da177e4SLinus Torvalds 30911da177e4SLinus Torvalds /* ok, everything is stopped */ 30921da177e4SLinus Torvalds oldpool = conf->r1bio_pool; 30931da177e4SLinus Torvalds conf->r1bio_pool = newpool; 30946ea9c07cSNeilBrown 3095a88aa786SNeilBrown for (d = d2 = 0; d < conf->raid_disks; d++) { 30963cb03002SNeilBrown struct md_rdev *rdev = conf->mirrors[d].rdev; 3097a88aa786SNeilBrown if (rdev && rdev->raid_disk != d2) { 309836fad858SNamhyung Kim sysfs_unlink_rdev(mddev, rdev); 3099a88aa786SNeilBrown rdev->raid_disk = d2; 310036fad858SNamhyung Kim sysfs_unlink_rdev(mddev, rdev); 310136fad858SNamhyung Kim if (sysfs_link_rdev(mddev, rdev)) 3102a88aa786SNeilBrown printk(KERN_WARNING 310336fad858SNamhyung Kim "md/raid1:%s: cannot register rd%d\n", 310436fad858SNamhyung Kim mdname(mddev), rdev->raid_disk); 3105a88aa786SNeilBrown } 3106a88aa786SNeilBrown if (rdev) 3107a88aa786SNeilBrown newmirrors[d2++].rdev = rdev; 31086ea9c07cSNeilBrown } 31091da177e4SLinus Torvalds kfree(conf->mirrors); 31101da177e4SLinus Torvalds conf->mirrors = newmirrors; 31111da177e4SLinus Torvalds kfree(conf->poolinfo); 31121da177e4SLinus Torvalds conf->poolinfo = newpoolinfo; 31131da177e4SLinus Torvalds 3114c04be0aaSNeilBrown spin_lock_irqsave(&conf->device_lock, flags); 31151da177e4SLinus Torvalds mddev->degraded += (raid_disks - conf->raid_disks); 3116c04be0aaSNeilBrown spin_unlock_irqrestore(&conf->device_lock, flags); 31171da177e4SLinus Torvalds conf->raid_disks = mddev->raid_disks = raid_disks; 311863c70c4fSNeilBrown mddev->delta_disks = 0; 31191da177e4SLinus Torvalds 3120e2d59925SNeilBrown unfreeze_array(conf); 31211da177e4SLinus Torvalds 31221da177e4SLinus Torvalds set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); 31231da177e4SLinus Torvalds md_wakeup_thread(mddev->thread); 31241da177e4SLinus Torvalds 31251da177e4SLinus Torvalds mempool_destroy(oldpool); 31261da177e4SLinus Torvalds return 0; 31271da177e4SLinus Torvalds } 31281da177e4SLinus Torvalds 3129fd01b88cSNeilBrown static void raid1_quiesce(struct mddev *mddev, int state) 313036fa3063SNeilBrown { 3131e8096360SNeilBrown struct r1conf *conf = mddev->private; 313236fa3063SNeilBrown 313336fa3063SNeilBrown switch(state) { 31346eef4b21SNeilBrown case 2: /* wake for suspend */ 31356eef4b21SNeilBrown wake_up(&conf->wait_barrier); 31366eef4b21SNeilBrown break; 31379e6603daSNeilBrown case 1: 313807169fd4Smajianpeng freeze_array(conf, 0); 313936fa3063SNeilBrown break; 31409e6603daSNeilBrown case 0: 314107169fd4Smajianpeng unfreeze_array(conf); 314236fa3063SNeilBrown break; 314336fa3063SNeilBrown } 314436fa3063SNeilBrown } 314536fa3063SNeilBrown 3146fd01b88cSNeilBrown static void *raid1_takeover(struct mddev *mddev) 3147709ae487SNeilBrown { 3148709ae487SNeilBrown /* raid1 can take over: 3149709ae487SNeilBrown * raid5 with 2 devices, any layout or chunk size 3150709ae487SNeilBrown */ 3151709ae487SNeilBrown if (mddev->level == 5 && mddev->raid_disks == 2) { 3152e8096360SNeilBrown struct r1conf *conf; 3153709ae487SNeilBrown mddev->new_level = 1; 3154709ae487SNeilBrown mddev->new_layout = 0; 3155709ae487SNeilBrown mddev->new_chunk_sectors = 0; 3156709ae487SNeilBrown conf = setup_conf(mddev); 3157709ae487SNeilBrown if (!IS_ERR(conf)) 315807169fd4Smajianpeng /* Array must appear to be quiesced */ 315907169fd4Smajianpeng conf->array_frozen = 1; 3160709ae487SNeilBrown return conf; 3161709ae487SNeilBrown } 3162709ae487SNeilBrown return ERR_PTR(-EINVAL); 3163709ae487SNeilBrown } 31641da177e4SLinus Torvalds 316584fc4b56SNeilBrown static struct md_personality raid1_personality = 31661da177e4SLinus Torvalds { 31671da177e4SLinus Torvalds .name = "raid1", 31682604b703SNeilBrown .level = 1, 31691da177e4SLinus Torvalds .owner = THIS_MODULE, 31701da177e4SLinus Torvalds .make_request = make_request, 31711da177e4SLinus Torvalds .run = run, 31721da177e4SLinus Torvalds .stop = stop, 31731da177e4SLinus Torvalds .status = status, 31741da177e4SLinus Torvalds .error_handler = error, 31751da177e4SLinus Torvalds .hot_add_disk = raid1_add_disk, 31761da177e4SLinus Torvalds .hot_remove_disk= raid1_remove_disk, 31771da177e4SLinus Torvalds .spare_active = raid1_spare_active, 31781da177e4SLinus Torvalds .sync_request = sync_request, 31791da177e4SLinus Torvalds .resize = raid1_resize, 318080c3a6ceSDan Williams .size = raid1_size, 318163c70c4fSNeilBrown .check_reshape = raid1_reshape, 318236fa3063SNeilBrown .quiesce = raid1_quiesce, 3183709ae487SNeilBrown .takeover = raid1_takeover, 31841da177e4SLinus Torvalds }; 31851da177e4SLinus Torvalds 31861da177e4SLinus Torvalds static int __init raid_init(void) 31871da177e4SLinus Torvalds { 31882604b703SNeilBrown return register_md_personality(&raid1_personality); 31891da177e4SLinus Torvalds } 31901da177e4SLinus Torvalds 31911da177e4SLinus Torvalds static void raid_exit(void) 31921da177e4SLinus Torvalds { 31932604b703SNeilBrown unregister_md_personality(&raid1_personality); 31941da177e4SLinus Torvalds } 31951da177e4SLinus Torvalds 31961da177e4SLinus Torvalds module_init(raid_init); 31971da177e4SLinus Torvalds module_exit(raid_exit); 31981da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 31990efb9e61SNeilBrown MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); 32001da177e4SLinus Torvalds MODULE_ALIAS("md-personality-3"); /* RAID1 */ 3201d9d166c2SNeilBrown MODULE_ALIAS("md-raid1"); 32022604b703SNeilBrown MODULE_ALIAS("md-level-1"); 320334db0cd6SNeilBrown 320434db0cd6SNeilBrown module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); 3205