1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * linear.c : Multiple Devices driver for Linux Copyright (C) 1994-96 Marc 4 * ZYNGIER <zyngier@ufr-info-p7.ibp.fr> or <maz@gloups.fdn.fr> 5 */ 6 7 #include <linux/blkdev.h> 8 #include <linux/raid/md_u.h> 9 #include <linux/seq_file.h> 10 #include <linux/module.h> 11 #include <linux/slab.h> 12 #include <trace/events/block.h> 13 #include "md.h" 14 15 struct dev_info { 16 struct md_rdev *rdev; 17 sector_t end_sector; 18 }; 19 20 struct linear_conf { 21 struct rcu_head rcu; 22 sector_t array_sectors; 23 /* a copy of mddev->raid_disks */ 24 int raid_disks; 25 struct dev_info disks[] __counted_by(raid_disks); 26 }; 27 28 /* 29 * find which device holds a particular offset 30 */ 31 static inline struct dev_info *which_dev(struct mddev *mddev, sector_t sector) 32 { 33 int lo, mid, hi; 34 struct linear_conf *conf; 35 36 lo = 0; 37 hi = mddev->raid_disks - 1; 38 conf = mddev->private; 39 40 /* 41 * Binary Search 42 */ 43 44 while (hi > lo) { 45 46 mid = (hi + lo) / 2; 47 if (sector < conf->disks[mid].end_sector) 48 hi = mid; 49 else 50 lo = mid + 1; 51 } 52 53 return conf->disks + lo; 54 } 55 56 static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks) 57 { 58 struct linear_conf *conf; 59 sector_t array_sectors; 60 61 conf = mddev->private; 62 WARN_ONCE(sectors || raid_disks, 63 "%s does not support generic reshape\n", __func__); 64 array_sectors = conf->array_sectors; 65 66 return array_sectors; 67 } 68 69 static int linear_set_limits(struct mddev *mddev) 70 { 71 struct queue_limits lim; 72 int err; 73 74 md_init_stacking_limits(&lim); 75 lim.max_hw_sectors = mddev->chunk_sectors; 76 lim.max_write_zeroes_sectors = mddev->chunk_sectors; 77 lim.io_min = mddev->chunk_sectors << 9; 78 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY); 79 if (err) 80 return err; 81 82 return queue_limits_set(mddev->gendisk->queue, &lim); 83 } 84 85 static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks) 86 { 87 struct linear_conf *conf; 88 struct md_rdev *rdev; 89 int ret = -EINVAL; 90 int cnt; 91 int i; 92 93 conf = kzalloc(struct_size(conf, disks, raid_disks), GFP_KERNEL); 94 if (!conf) 95 return ERR_PTR(-ENOMEM); 96 97 /* 98 * conf->raid_disks is copy of mddev->raid_disks. The reason to 99 * keep a copy of mddev->raid_disks in struct linear_conf is, 100 * mddev->raid_disks may not be consistent with pointers number of 101 * conf->disks[] when it is updated in linear_add() and used to 102 * iterate old conf->disks[] earray in linear_congested(). 103 * Here conf->raid_disks is always consitent with number of 104 * pointers in conf->disks[] array, and mddev->private is updated 105 * with rcu_assign_pointer() in linear_addr(), such race can be 106 * avoided. 107 */ 108 conf->raid_disks = raid_disks; 109 110 cnt = 0; 111 conf->array_sectors = 0; 112 113 rdev_for_each(rdev, mddev) { 114 int j = rdev->raid_disk; 115 struct dev_info *disk = conf->disks + j; 116 sector_t sectors; 117 118 if (j < 0 || j >= raid_disks || disk->rdev) { 119 pr_warn("md/linear:%s: disk numbering problem. Aborting!\n", 120 mdname(mddev)); 121 goto out; 122 } 123 124 disk->rdev = rdev; 125 if (mddev->chunk_sectors) { 126 sectors = rdev->sectors; 127 sector_div(sectors, mddev->chunk_sectors); 128 rdev->sectors = sectors * mddev->chunk_sectors; 129 } 130 131 conf->array_sectors += rdev->sectors; 132 cnt++; 133 } 134 if (cnt != raid_disks) { 135 pr_warn("md/linear:%s: not enough drives present. Aborting!\n", 136 mdname(mddev)); 137 goto out; 138 } 139 140 /* 141 * Here we calculate the device offsets. 142 */ 143 conf->disks[0].end_sector = conf->disks[0].rdev->sectors; 144 145 for (i = 1; i < raid_disks; i++) 146 conf->disks[i].end_sector = 147 conf->disks[i-1].end_sector + 148 conf->disks[i].rdev->sectors; 149 150 if (!mddev_is_dm(mddev)) { 151 ret = linear_set_limits(mddev); 152 if (ret) 153 goto out; 154 } 155 156 return conf; 157 158 out: 159 kfree(conf); 160 return ERR_PTR(ret); 161 } 162 163 static int linear_run(struct mddev *mddev) 164 { 165 struct linear_conf *conf; 166 int ret; 167 168 if (md_check_no_bitmap(mddev)) 169 return -EINVAL; 170 171 conf = linear_conf(mddev, mddev->raid_disks); 172 if (IS_ERR(conf)) 173 return PTR_ERR(conf); 174 175 mddev->private = conf; 176 md_set_array_sectors(mddev, linear_size(mddev, 0, 0)); 177 178 ret = md_integrity_register(mddev); 179 if (ret) { 180 kfree(conf); 181 mddev->private = NULL; 182 } 183 return ret; 184 } 185 186 static int linear_add(struct mddev *mddev, struct md_rdev *rdev) 187 { 188 /* Adding a drive to a linear array allows the array to grow. 189 * It is permitted if the new drive has a matching superblock 190 * already on it, with raid_disk equal to raid_disks. 191 * It is achieved by creating a new linear_private_data structure 192 * and swapping it in in-place of the current one. 193 * The current one is never freed until the array is stopped. 194 * This avoids races. 195 */ 196 struct linear_conf *newconf, *oldconf; 197 198 if (rdev->saved_raid_disk != mddev->raid_disks) 199 return -EINVAL; 200 201 rdev->raid_disk = rdev->saved_raid_disk; 202 rdev->saved_raid_disk = -1; 203 204 newconf = linear_conf(mddev, mddev->raid_disks + 1); 205 if (IS_ERR(newconf)) 206 return PTR_ERR(newconf); 207 208 /* newconf->raid_disks already keeps a copy of * the increased 209 * value of mddev->raid_disks, WARN_ONCE() is just used to make 210 * sure of this. It is possible that oldconf is still referenced 211 * in linear_congested(), therefore kfree_rcu() is used to free 212 * oldconf until no one uses it anymore. 213 */ 214 oldconf = rcu_dereference_protected(mddev->private, 215 lockdep_is_held(&mddev->reconfig_mutex)); 216 mddev->raid_disks++; 217 WARN_ONCE(mddev->raid_disks != newconf->raid_disks, 218 "copied raid_disks doesn't match mddev->raid_disks"); 219 rcu_assign_pointer(mddev->private, newconf); 220 md_set_array_sectors(mddev, linear_size(mddev, 0, 0)); 221 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors); 222 kfree_rcu(oldconf, rcu); 223 return 0; 224 } 225 226 static void linear_free(struct mddev *mddev, void *priv) 227 { 228 struct linear_conf *conf = priv; 229 230 kfree(conf); 231 } 232 233 static bool linear_make_request(struct mddev *mddev, struct bio *bio) 234 { 235 struct dev_info *tmp_dev; 236 sector_t start_sector, end_sector, data_offset; 237 sector_t bio_sector = bio->bi_iter.bi_sector; 238 239 if (unlikely(bio->bi_opf & REQ_PREFLUSH) 240 && md_flush_request(mddev, bio)) 241 return true; 242 243 tmp_dev = which_dev(mddev, bio_sector); 244 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors; 245 end_sector = tmp_dev->end_sector; 246 data_offset = tmp_dev->rdev->data_offset; 247 248 if (unlikely(bio_sector >= end_sector || 249 bio_sector < start_sector)) 250 goto out_of_bounds; 251 252 if (unlikely(is_rdev_broken(tmp_dev->rdev))) { 253 md_error(mddev, tmp_dev->rdev); 254 bio_io_error(bio); 255 return true; 256 } 257 258 if (unlikely(bio_end_sector(bio) > end_sector)) { 259 /* This bio crosses a device boundary, so we have to split it */ 260 struct bio *split = bio_split(bio, end_sector - bio_sector, 261 GFP_NOIO, &mddev->bio_set); 262 263 if (IS_ERR(split)) { 264 bio->bi_status = errno_to_blk_status(PTR_ERR(split)); 265 bio_endio(bio); 266 return true; 267 } 268 269 bio_chain(split, bio); 270 submit_bio_noacct(bio); 271 bio = split; 272 } 273 274 md_account_bio(mddev, &bio); 275 bio_set_dev(bio, tmp_dev->rdev->bdev); 276 bio->bi_iter.bi_sector = bio->bi_iter.bi_sector - 277 start_sector + data_offset; 278 279 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && 280 !bdev_max_discard_sectors(bio->bi_bdev))) { 281 /* Just ignore it */ 282 bio_endio(bio); 283 } else { 284 if (mddev->gendisk) 285 trace_block_bio_remap(bio, disk_devt(mddev->gendisk), 286 bio_sector); 287 mddev_check_write_zeroes(mddev, bio); 288 submit_bio_noacct(bio); 289 } 290 return true; 291 292 out_of_bounds: 293 pr_err("md/linear:%s: make_request: Sector %llu out of bounds on dev %pg: %llu sectors, offset %llu\n", 294 mdname(mddev), 295 (unsigned long long)bio->bi_iter.bi_sector, 296 tmp_dev->rdev->bdev, 297 (unsigned long long)tmp_dev->rdev->sectors, 298 (unsigned long long)start_sector); 299 bio_io_error(bio); 300 return true; 301 } 302 303 static void linear_status(struct seq_file *seq, struct mddev *mddev) 304 { 305 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2); 306 } 307 308 static void linear_error(struct mddev *mddev, struct md_rdev *rdev) 309 { 310 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) { 311 char *md_name = mdname(mddev); 312 313 pr_crit("md/linear%s: Disk failure on %pg detected, failing array.\n", 314 md_name, rdev->bdev); 315 } 316 } 317 318 static void linear_quiesce(struct mddev *mddev, int state) 319 { 320 } 321 322 static struct md_personality linear_personality = { 323 .name = "linear", 324 .level = LEVEL_LINEAR, 325 .owner = THIS_MODULE, 326 .make_request = linear_make_request, 327 .run = linear_run, 328 .free = linear_free, 329 .status = linear_status, 330 .hot_add_disk = linear_add, 331 .size = linear_size, 332 .quiesce = linear_quiesce, 333 .error_handler = linear_error, 334 }; 335 336 static int __init linear_init(void) 337 { 338 return register_md_personality(&linear_personality); 339 } 340 341 static void linear_exit(void) 342 { 343 unregister_md_personality(&linear_personality); 344 } 345 346 module_init(linear_init); 347 module_exit(linear_exit); 348 MODULE_LICENSE("GPL"); 349 MODULE_DESCRIPTION("Linear device concatenation personality for MD (deprecated)"); 350 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/ 351 MODULE_ALIAS("md-linear"); 352 MODULE_ALIAS("md-level--1"); 353