| 1 | |
| 2 | |
| 3 | * raid10.c : Multiple Devices driver for Linux
|
| 4 | *
|
| 5 | * Copyright (C) 2000-2004 Neil Brown
|
| 6 | *
|
| 7 | * RAID-10 support for md.
|
| 8 | *
|
| 9 | * Base on code in raid1.c. See raid1.c for further copyright information.
|
| 10 |
|
| 11 |
|
| 12 | #include <linux/slab.h>
|
| 13 | #include <linux/delay.h>
|
| 14 | #include <linux/blkdev.h>
|
| 15 | #include <linux/module.h>
|
| 16 | #include <linux/seq_file.h>
|
| 17 | #include <linux/ratelimit.h>
|
| 18 | #include <linux/kthread.h>
|
| 19 | #include <linux/raid/md_p.h>
|
| 20 | #include <trace/events/block.h>
|
| 21 | #include "md.h"
|
| 22 |
|
| 23 | #define RAID_1_10_NAME "raid10"
|
| 24 | #include "raid10.h"
|
| 25 | #include "raid0.h"
|
| 26 | #include "md-bitmap.h"
|
| 27 | #include "md-cluster.h"
|
| 28 |
|
| 29 | |
| 30 | * RAID10 provides a combination of RAID0 and RAID1 functionality.
|
| 31 | * The layout of data is defined by
|
| 32 | * chunk_size
|
| 33 | * raid_disks
|
| 34 | * near_copies (stored in low byte of layout)
|
| 35 | * far_copies (stored in second byte of layout)
|
| 36 | * far_offset (stored in bit 16 of layout )
|
| 37 | * use_far_sets (stored in bit 17 of layout )
|
| 38 | * use_far_sets_bugfixed (stored in bit 18 of layout )
|
| 39 | *
|
| 40 | * The data to be stored is divided into chunks using chunksize. Each device
|
| 41 | * is divided into far_copies sections. In each section, chunks are laid out
|
| 42 | * in a style similar to raid0, but near_copies copies of each chunk is stored
|
| 43 | * (each on a different drive). The starting device for each section is offset
|
| 44 | * near_copies from the starting device of the previous section. Thus there
|
| 45 | * are (near_copies * far_copies) of each chunk, and each is on a different
|
| 46 | * drive. near_copies and far_copies must be at least one, and their product
|
| 47 | * is at most raid_disks.
|
| 48 | *
|
| 49 | * If far_offset is true, then the far_copies are handled a bit differently.
|
| 50 | * The copies are still in different stripes, but instead of being very far
|
| 51 | * apart on disk, there are adjacent stripes.
|
| 52 | *
|
| 53 | * The far and offset algorithms are handled slightly differently if
|
| 54 | * 'use_far_sets' is true. In this case, the array's devices are grouped into
|
| 55 | * sets that are (near_copies * far_copies) in size. The far copied stripes
|
| 56 | * are still shifted by 'near_copies' devices, but this shifting stays confined
|
| 57 | * to the set rather than the entire array. This is done to improve the number
|
| 58 | * of device combinations that can fail without causing the array to fail.
|
| 59 | * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk
|
| 60 | * on a device):
|
| 61 | * A B C D A B C D E
|
| 62 | * ... ...
|
| 63 | * D A B C E A B C D
|
| 64 | * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s):
|
| 65 | * [A B] [C D] [A B] [C D E]
|
| 66 | * |...| |...| |...| | ... |
|
| 67 | * [B A] [D C] [B A] [E C D]
|
| 68 |
|
| 69 |
|
| 70 | static void allow_barrier(struct r10conf *conf);
|
| 71 | static void lower_barrier(struct r10conf *conf);
|
| 72 | static int _enough(struct r10conf *conf, int previous, int ignore);
|
| 73 | static int enough(struct r10conf *conf, int ignore);
|
| 74 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
|
| 75 | int *skipped);
|
| 76 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio);
|
| 77 | static void end_reshape_write(struct bio *bio);
|
| 78 | static void end_reshape(struct r10conf *conf);
|
| 79 |
|
| 80 | #include "raid1-10.c"
|
| 81 |
|
| 82 | #define NULL_CMD
|
| 83 | #define cmd_before(conf, cmd) \
|
| 84 | do { \
|
| 85 | write_sequnlock_irq(&(conf)->resync_lock); \
|
| 86 | cmd; \
|
| 87 | } while (0)
|
| 88 | #define cmd_after(conf) write_seqlock_irq(&(conf)->resync_lock)
|
| 89 |
|
| 90 | #define wait_event_barrier_cmd(conf, cond, cmd) \
|
| 91 | wait_event_cmd((conf)->wait_barrier, cond, cmd_before(conf, cmd), \
|
| 92 | cmd_after(conf))
|
| 93 |
|
| 94 | #define wait_event_barrier(conf, cond) \
|
| 95 | wait_event_barrier_cmd(conf, cond, NULL_CMD)
|
| 96 |
|
| 97 | |
| 98 | * for resync bio, r10bio pointer can be retrieved from the per-bio
|
| 99 | * 'struct resync_pages'.
|
| 100 |
|
| 101 | static inline struct r10bio *get_resync_r10bio(struct bio *bio)
|
| 102 | {
|
| 103 | return get_resync_pages(bio)->raid_bio;
|
| 104 | }
|
| 105 |
|
| 106 | static inline int calc_r10bio_size(unsigned int raid_disks)
|
| 107 | {
|
| 108 | return offsetof(struct r10bio, devs[raid_disks]);
|
| 109 | }
|
| 110 |
|
| 111 | static mempool_t *create_r10bio_pool(unsigned int raid_disks)
|
| 112 | {
|
| 113 | int size = calc_r10bio_size(raid_disks);
|
| 114 |
|
| 115 | return mempool_create_kmalloc_pool(NR_RAID_BIOS, size);
|
| 116 | }
|
| 117 |
|
| 118 | static struct r10bio *alloc_r10bio(unsigned int raid_disks, gfp_t gfp_flags)
|
| 119 | {
|
| 120 | int size = calc_r10bio_size(raid_disks);
|
| 121 |
|
| 122 |
|
| 123 | return kzalloc(size, gfp_flags);
|
| 124 | }
|
| 125 |
|
| 126 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
|
| 127 |
|
| 128 | #define RESYNC_WINDOW (1024*1024)
|
| 129 |
|
| 130 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
|
| 131 | #define CLUSTER_RESYNC_WINDOW (32 * RESYNC_WINDOW)
|
| 132 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
|
| 133 |
|
| 134 | |
| 135 | * When performing a resync, we need to read and compare, so
|
| 136 | * we need as many pages are there are copies.
|
| 137 | * When performing a recovery, we need 2 bios, one for read,
|
| 138 | * one for write (we recover only one drive per r10buf)
|
| 139 | *
|
| 140 |
|
| 141 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data)
|
| 142 | {
|
| 143 | struct r10conf *conf = data;
|
| 144 | struct r10bio *r10_bio;
|
| 145 | struct bio *bio;
|
| 146 | int j;
|
| 147 | int nalloc, nalloc_rp;
|
| 148 | struct resync_pages *rps;
|
| 149 |
|
| 150 | r10_bio = alloc_r10bio(conf->geo.raid_disks, gfp_flags);
|
| 151 | if (!r10_bio)
|
| 152 | return NULL;
|
| 153 |
|
| 154 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) ||
|
| 155 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery))
|
| 156 | nalloc = conf->copies;
|
| 157 | else
|
| 158 | nalloc = 2;
|
| 159 |
|
| 160 |
|
| 161 | if (!conf->have_replacement)
|
| 162 | nalloc_rp = nalloc;
|
| 163 | else
|
| 164 | nalloc_rp = nalloc * 2;
|
| 165 | rps = kmalloc_objs(struct resync_pages, nalloc_rp, gfp_flags);
|
| 166 | if (!rps)
|
| 167 | goto out_free_r10bio;
|
| 168 |
|
| 169 | |
| 170 | * Allocate bios.
|
| 171 |
|
| 172 | for (j = nalloc ; j-- ; ) {
|
| 173 | bio = bio_kmalloc(RESYNC_PAGES, gfp_flags);
|
| 174 | if (!bio)
|
| 175 | goto out_free_bio;
|
| 176 | bio_init_inline(bio, NULL, RESYNC_PAGES, 0);
|
| 177 | r10_bio->devs[j].bio = bio;
|
| 178 | if (!conf->have_replacement)
|
| 179 | continue;
|
| 180 | bio = bio_kmalloc(RESYNC_PAGES, gfp_flags);
|
| 181 | if (!bio)
|
| 182 | goto out_free_bio;
|
| 183 | bio_init_inline(bio, NULL, RESYNC_PAGES, 0);
|
| 184 | r10_bio->devs[j].repl_bio = bio;
|
| 185 | }
|
| 186 | |
| 187 | * Allocate RESYNC_PAGES data pages and attach them
|
| 188 | * where needed.
|
| 189 |
|
| 190 | for (j = 0; j < nalloc; j++) {
|
| 191 | struct bio *rbio = r10_bio->devs[j].repl_bio;
|
| 192 | struct resync_pages *rp, *rp_repl;
|
| 193 |
|
| 194 | rp = &rps[j];
|
| 195 | if (rbio)
|
| 196 | rp_repl = &rps[nalloc + j];
|
| 197 |
|
| 198 | bio = r10_bio->devs[j].bio;
|
| 199 |
|
| 200 | if (!j || test_bit(MD_RECOVERY_SYNC,
|
| 201 | &conf->mddev->recovery)) {
|
| 202 | if (resync_alloc_pages(rp, gfp_flags))
|
| 203 | goto out_free_pages;
|
| 204 | } else {
|
| 205 | memcpy(rp, &rps[0], sizeof(*rp));
|
| 206 | resync_get_all_pages(rp);
|
| 207 | }
|
| 208 |
|
| 209 | rp->raid_bio = r10_bio;
|
| 210 | bio->bi_private = rp;
|
| 211 | if (rbio) {
|
| 212 | memcpy(rp_repl, rp, sizeof(*rp));
|
| 213 | rbio->bi_private = rp_repl;
|
| 214 | }
|
| 215 | }
|
| 216 |
|
| 217 | return r10_bio;
|
| 218 |
|
| 219 | out_free_pages:
|
| 220 | while (--j >= 0)
|
| 221 | resync_free_pages(&rps[j]);
|
| 222 |
|
| 223 | j = 0;
|
| 224 | out_free_bio:
|
| 225 | for ( ; j < nalloc; j++) {
|
| 226 | if (r10_bio->devs[j].bio)
|
| 227 | bio_uninit(r10_bio->devs[j].bio);
|
| 228 | kfree(r10_bio->devs[j].bio);
|
| 229 | if (r10_bio->devs[j].repl_bio)
|
| 230 | bio_uninit(r10_bio->devs[j].repl_bio);
|
| 231 | kfree(r10_bio->devs[j].repl_bio);
|
| 232 | }
|
| 233 | kfree(rps);
|
| 234 | out_free_r10bio:
|
| 235 | rbio_pool_free(r10_bio, conf);
|
| 236 | return NULL;
|
| 237 | }
|
| 238 |
|
| 239 | static void r10buf_pool_free(void *__r10_bio, void *data)
|
| 240 | {
|
| 241 | struct r10conf *conf = data;
|
| 242 | struct r10bio *r10bio = __r10_bio;
|
| 243 | int j;
|
| 244 | struct resync_pages *rp = NULL;
|
| 245 |
|
| 246 | for (j = conf->copies; j--; ) {
|
| 247 | struct bio *bio = r10bio->devs[j].bio;
|
| 248 |
|
| 249 | if (bio) {
|
| 250 | rp = get_resync_pages(bio);
|
| 251 | resync_free_pages(rp);
|
| 252 | bio_uninit(bio);
|
| 253 | kfree(bio);
|
| 254 | }
|
| 255 |
|
| 256 | bio = r10bio->devs[j].repl_bio;
|
| 257 | if (bio) {
|
| 258 | bio_uninit(bio);
|
| 259 | kfree(bio);
|
| 260 | }
|
| 261 | }
|
| 262 |
|
| 263 |
|
| 264 | kfree(rp);
|
| 265 |
|
| 266 | rbio_pool_free(r10bio, conf);
|
| 267 | }
|
| 268 |
|
| 269 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio)
|
| 270 | {
|
| 271 | int i;
|
| 272 |
|
| 273 | for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 274 | struct bio **bio = & r10_bio->devs[i].bio;
|
| 275 | if (!BIO_SPECIAL(*bio))
|
| 276 | bio_put(*bio);
|
| 277 | *bio = NULL;
|
| 278 | bio = &r10_bio->devs[i].repl_bio;
|
| 279 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio))
|
| 280 | bio_put(*bio);
|
| 281 | *bio = NULL;
|
| 282 | }
|
| 283 | }
|
| 284 |
|
| 285 | static void free_r10bio(struct r10bio *r10_bio)
|
| 286 | {
|
| 287 | struct r10conf *conf = r10_bio->mddev->private;
|
| 288 |
|
| 289 | put_all_bios(conf, r10_bio);
|
| 290 | mempool_free(r10_bio, conf->r10bio_pool);
|
| 291 | }
|
| 292 |
|
| 293 | static void put_buf(struct r10bio *r10_bio)
|
| 294 | {
|
| 295 | struct r10conf *conf = r10_bio->mddev->private;
|
| 296 |
|
| 297 | mempool_free(r10_bio, &conf->r10buf_pool);
|
| 298 |
|
| 299 | lower_barrier(conf);
|
| 300 | }
|
| 301 |
|
| 302 | static void wake_up_barrier(struct r10conf *conf)
|
| 303 | {
|
| 304 | if (wq_has_sleeper(&conf->wait_barrier))
|
| 305 | wake_up(&conf->wait_barrier);
|
| 306 | }
|
| 307 |
|
| 308 | static void reschedule_retry(struct r10bio *r10_bio)
|
| 309 | {
|
| 310 | unsigned long flags;
|
| 311 | struct mddev *mddev = r10_bio->mddev;
|
| 312 | struct r10conf *conf = mddev->private;
|
| 313 |
|
| 314 | spin_lock_irqsave(&conf->device_lock, flags);
|
| 315 | list_add(&r10_bio->retry_list, &conf->retry_list);
|
| 316 | conf->nr_queued ++;
|
| 317 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 318 |
|
| 319 |
|
| 320 | wake_up(&conf->wait_barrier);
|
| 321 |
|
| 322 | md_wakeup_thread(mddev->thread);
|
| 323 | }
|
| 324 |
|
| 325 | |
| 326 | * raid_end_bio_io() is called when we have finished servicing a mirrored
|
| 327 | * operation and are ready to return a success/failure code to the buffer
|
| 328 | * cache layer.
|
| 329 |
|
| 330 | static void raid_end_bio_io(struct r10bio *r10_bio)
|
| 331 | {
|
| 332 | struct bio *bio = r10_bio->master_bio;
|
| 333 | struct r10conf *conf = r10_bio->mddev->private;
|
| 334 | bool returned = true;
|
| 335 |
|
| 336 | if (!test_and_set_bit(R10BIO_Returned, &r10_bio->state)) {
|
| 337 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
|
| 338 | bio->bi_status = BLK_STS_IOERR;
|
| 339 | returned = false;
|
| 340 | }
|
| 341 |
|
| 342 | free_r10bio(r10_bio);
|
| 343 |
|
| 344 | if (!returned)
|
| 345 | bio_endio(bio);
|
| 346 |
|
| 347 | |
| 348 | * Wake up any possible resync thread that waits for the device
|
| 349 | * to go idle.
|
| 350 |
|
| 351 | allow_barrier(conf);
|
| 352 | }
|
| 353 |
|
| 354 | |
| 355 | * Update disk head position estimator based on IRQ completion info.
|
| 356 |
|
| 357 | static inline void update_head_pos(int slot, struct r10bio *r10_bio)
|
| 358 | {
|
| 359 | struct r10conf *conf = r10_bio->mddev->private;
|
| 360 |
|
| 361 | conf->mirrors[r10_bio->devs[slot].devnum].head_position =
|
| 362 | r10_bio->devs[slot].addr + (r10_bio->sectors);
|
| 363 | }
|
| 364 |
|
| 365 | |
| 366 | * Find the disk number which triggered given bio
|
| 367 |
|
| 368 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio,
|
| 369 | struct bio *bio, int *slotp, int *replp)
|
| 370 | {
|
| 371 | int slot;
|
| 372 | int repl = 0;
|
| 373 |
|
| 374 | for (slot = 0; slot < conf->geo.raid_disks; slot++) {
|
| 375 | if (r10_bio->devs[slot].bio == bio)
|
| 376 | break;
|
| 377 | if (r10_bio->devs[slot].repl_bio == bio) {
|
| 378 | repl = 1;
|
| 379 | break;
|
| 380 | }
|
| 381 | }
|
| 382 |
|
| 383 | update_head_pos(slot, r10_bio);
|
| 384 |
|
| 385 | if (slotp)
|
| 386 | *slotp = slot;
|
| 387 | if (replp)
|
| 388 | *replp = repl;
|
| 389 | return r10_bio->devs[slot].devnum;
|
| 390 | }
|
| 391 |
|
| 392 | static void raid10_end_read_request(struct bio *bio)
|
| 393 | {
|
| 394 | int uptodate = !bio->bi_status;
|
| 395 | struct r10bio *r10_bio = bio->bi_private;
|
| 396 | int slot;
|
| 397 | struct md_rdev *rdev;
|
| 398 | struct r10conf *conf = r10_bio->mddev->private;
|
| 399 |
|
| 400 | slot = r10_bio->read_slot;
|
| 401 | rdev = r10_bio->devs[slot].rdev;
|
| 402 | |
| 403 | * this branch is our 'one mirror IO has finished' event handler:
|
| 404 |
|
| 405 | update_head_pos(slot, r10_bio);
|
| 406 |
|
| 407 | if (uptodate) {
|
| 408 | |
| 409 | * Set R10BIO_Uptodate in our master bio, so that
|
| 410 | * we will return a good error code to the higher
|
| 411 | * levels even if IO on some other mirrored buffer fails.
|
| 412 | *
|
| 413 | * The 'master' represents the composite IO operation to
|
| 414 | * user-side. So if something waits for IO, then it will
|
| 415 | * wait for the 'master' bio.
|
| 416 |
|
| 417 | set_bit(R10BIO_Uptodate, &r10_bio->state);
|
| 418 | } else if (!raid1_should_handle_error(bio)) {
|
| 419 | uptodate = 1;
|
| 420 | } else {
|
| 421 | |
| 422 | * failed, we want to return the error upwards rather
|
| 423 | * than fail the last device. Here we redefine
|
| 424 | * "uptodate" to mean "Don't want to retry"
|
| 425 |
|
| 426 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state),
|
| 427 | rdev->raid_disk))
|
| 428 | uptodate = 1;
|
| 429 | }
|
| 430 | if (uptodate) {
|
| 431 | raid_end_bio_io(r10_bio);
|
| 432 | rdev_dec_pending(rdev, conf->mddev);
|
| 433 | } else {
|
| 434 | |
| 435 | * oops, read error - keep the refcount on the rdev
|
| 436 |
|
| 437 | pr_err_ratelimited("md/raid10:%s: %pg: rescheduling sector %llu\n",
|
| 438 | mdname(conf->mddev),
|
| 439 | rdev->bdev,
|
| 440 | (unsigned long long)r10_bio->sector);
|
| 441 | set_bit(R10BIO_ReadError, &r10_bio->state);
|
| 442 | reschedule_retry(r10_bio);
|
| 443 | }
|
| 444 | }
|
| 445 |
|
| 446 | static void close_write(struct r10bio *r10_bio)
|
| 447 | {
|
| 448 | struct mddev *mddev = r10_bio->mddev;
|
| 449 |
|
| 450 | md_write_end(mddev);
|
| 451 | }
|
| 452 |
|
| 453 | static void one_write_done(struct r10bio *r10_bio)
|
| 454 | {
|
| 455 | if (atomic_dec_and_test(&r10_bio->remaining)) {
|
| 456 | if (test_bit(R10BIO_WriteError, &r10_bio->state))
|
| 457 | reschedule_retry(r10_bio);
|
| 458 | else {
|
| 459 | close_write(r10_bio);
|
| 460 | if (test_bit(R10BIO_MadeGood, &r10_bio->state))
|
| 461 | reschedule_retry(r10_bio);
|
| 462 | else
|
| 463 | raid_end_bio_io(r10_bio);
|
| 464 | }
|
| 465 | }
|
| 466 | }
|
| 467 |
|
| 468 | static void raid10_end_write_request(struct bio *bio)
|
| 469 | {
|
| 470 | struct r10bio *r10_bio = bio->bi_private;
|
| 471 | int dev;
|
| 472 | int dec_rdev = 1;
|
| 473 | struct r10conf *conf = r10_bio->mddev->private;
|
| 474 | int slot, repl;
|
| 475 | struct md_rdev *rdev = NULL;
|
| 476 | struct bio *to_put = NULL;
|
| 477 | bool ignore_error = !raid1_should_handle_error(bio) ||
|
| 478 | (bio->bi_status && bio_op(bio) == REQ_OP_DISCARD);
|
| 479 |
|
| 480 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
|
| 481 |
|
| 482 | if (repl)
|
| 483 | rdev = conf->mirrors[dev].replacement;
|
| 484 | if (!rdev) {
|
| 485 | smp_rmb();
|
| 486 | repl = 0;
|
| 487 | rdev = conf->mirrors[dev].rdev;
|
| 488 | }
|
| 489 | |
| 490 | * this branch is our 'one mirror IO has finished' event handler:
|
| 491 |
|
| 492 | if (bio->bi_status && !ignore_error) {
|
| 493 | if (repl)
|
| 494 | |
| 495 | * just fail it.
|
| 496 |
|
| 497 | md_error(rdev->mddev, rdev);
|
| 498 | else {
|
| 499 | set_bit(WriteErrorSeen, &rdev->flags);
|
| 500 | if (!test_and_set_bit(WantReplacement, &rdev->flags))
|
| 501 | set_bit(MD_RECOVERY_NEEDED,
|
| 502 | &rdev->mddev->recovery);
|
| 503 |
|
| 504 | dec_rdev = 0;
|
| 505 | if (test_bit(FailFast, &rdev->flags) &&
|
| 506 | (bio->bi_opf & MD_FAILFAST)) {
|
| 507 | md_error(rdev->mddev, rdev);
|
| 508 | }
|
| 509 |
|
| 510 | |
| 511 | * When the device is faulty, it is not necessary to
|
| 512 | * handle write error.
|
| 513 |
|
| 514 | if (!test_bit(Faulty, &rdev->flags))
|
| 515 | set_bit(R10BIO_WriteError, &r10_bio->state);
|
| 516 | else {
|
| 517 |
|
| 518 | r10_bio->devs[slot].bio = NULL;
|
| 519 | to_put = bio;
|
| 520 | dec_rdev = 1;
|
| 521 | }
|
| 522 | }
|
| 523 | } else {
|
| 524 | |
| 525 | * Set R10BIO_Uptodate in our master bio, so that
|
| 526 | * we will return a good error code for to the higher
|
| 527 | * levels even if IO on some other mirrored buffer fails.
|
| 528 | *
|
| 529 | * The 'master' represents the composite IO operation to
|
| 530 | * user-side. So if something waits for IO, then it will
|
| 531 | * wait for the 'master' bio.
|
| 532 | *
|
| 533 | * Do not set R10BIO_Uptodate if the current device is
|
| 534 | * rebuilding or Faulty. This is because we cannot use
|
| 535 | * such device for properly reading the data back (we could
|
| 536 | * potentially use it, if the current write would have felt
|
| 537 | * before rdev->recovery_offset, but for simplicity we don't
|
| 538 | * check this here.
|
| 539 |
|
| 540 | if (test_bit(In_sync, &rdev->flags) &&
|
| 541 | !test_bit(Faulty, &rdev->flags))
|
| 542 | set_bit(R10BIO_Uptodate, &r10_bio->state);
|
| 543 |
|
| 544 |
|
| 545 | if (rdev_has_badblock(rdev, r10_bio->devs[slot].addr,
|
| 546 | r10_bio->sectors) &&
|
| 547 | !ignore_error) {
|
| 548 | bio_put(bio);
|
| 549 | if (repl)
|
| 550 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD;
|
| 551 | else
|
| 552 | r10_bio->devs[slot].bio = IO_MADE_GOOD;
|
| 553 | dec_rdev = 0;
|
| 554 | set_bit(R10BIO_MadeGood, &r10_bio->state);
|
| 555 | }
|
| 556 | }
|
| 557 |
|
| 558 | |
| 559 | *
|
| 560 | * Let's see if all mirrored write operations have finished
|
| 561 | * already.
|
| 562 |
|
| 563 | one_write_done(r10_bio);
|
| 564 | if (dec_rdev)
|
| 565 | rdev_dec_pending(rdev, conf->mddev);
|
| 566 | if (to_put)
|
| 567 | bio_put(to_put);
|
| 568 | }
|
| 569 |
|
| 570 | |
| 571 | * RAID10 layout manager
|
| 572 | * As well as the chunksize and raid_disks count, there are two
|
| 573 | * parameters: near_copies and far_copies.
|
| 574 | * near_copies * far_copies must be <= raid_disks.
|
| 575 | * Normally one of these will be 1.
|
| 576 | * If both are 1, we get raid0.
|
| 577 | * If near_copies == raid_disks, we get raid1.
|
| 578 | *
|
| 579 | * Chunks are laid out in raid0 style with near_copies copies of the
|
| 580 | * first chunk, followed by near_copies copies of the next chunk and
|
| 581 | * so on.
|
| 582 | * If far_copies > 1, then after 1/far_copies of the array has been assigned
|
| 583 | * as described above, we start again with a device offset of near_copies.
|
| 584 | * So we effectively have another copy of the whole array further down all
|
| 585 | * the drives, but with blocks on different drives.
|
| 586 | * With this layout, and block is never stored twice on the one device.
|
| 587 | *
|
| 588 | * raid10_find_phys finds the sector offset of a given virtual sector
|
| 589 | * on each device that it is on.
|
| 590 | *
|
| 591 | * raid10_find_virt does the reverse mapping, from a device and a
|
| 592 | * sector offset to a virtual address
|
| 593 |
|
| 594 |
|
| 595 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio)
|
| 596 | {
|
| 597 | int n,f;
|
| 598 | sector_t sector;
|
| 599 | sector_t chunk;
|
| 600 | sector_t stripe;
|
| 601 | int dev;
|
| 602 | int slot = 0;
|
| 603 | int last_far_set_start, last_far_set_size;
|
| 604 |
|
| 605 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
|
| 606 | last_far_set_start *= geo->far_set_size;
|
| 607 |
|
| 608 | last_far_set_size = geo->far_set_size;
|
| 609 | last_far_set_size += (geo->raid_disks % geo->far_set_size);
|
| 610 |
|
| 611 |
|
| 612 | chunk = r10bio->sector >> geo->chunk_shift;
|
| 613 | sector = r10bio->sector & geo->chunk_mask;
|
| 614 |
|
| 615 | chunk *= geo->near_copies;
|
| 616 | stripe = chunk;
|
| 617 | dev = sector_div(stripe, geo->raid_disks);
|
| 618 | if (geo->far_offset)
|
| 619 | stripe *= geo->far_copies;
|
| 620 |
|
| 621 | sector += stripe << geo->chunk_shift;
|
| 622 |
|
| 623 |
|
| 624 | for (n = 0; n < geo->near_copies; n++) {
|
| 625 | int d = dev;
|
| 626 | int set;
|
| 627 | sector_t s = sector;
|
| 628 | r10bio->devs[slot].devnum = d;
|
| 629 | r10bio->devs[slot].addr = s;
|
| 630 | slot++;
|
| 631 |
|
| 632 | for (f = 1; f < geo->far_copies; f++) {
|
| 633 | set = d / geo->far_set_size;
|
| 634 | d += geo->near_copies;
|
| 635 |
|
| 636 | if ((geo->raid_disks % geo->far_set_size) &&
|
| 637 | (d > last_far_set_start)) {
|
| 638 | d -= last_far_set_start;
|
| 639 | d %= last_far_set_size;
|
| 640 | d += last_far_set_start;
|
| 641 | } else {
|
| 642 | d %= geo->far_set_size;
|
| 643 | d += geo->far_set_size * set;
|
| 644 | }
|
| 645 | s += geo->stride;
|
| 646 | r10bio->devs[slot].devnum = d;
|
| 647 | r10bio->devs[slot].addr = s;
|
| 648 | slot++;
|
| 649 | }
|
| 650 | dev++;
|
| 651 | if (dev >= geo->raid_disks) {
|
| 652 | dev = 0;
|
| 653 | sector += (geo->chunk_mask + 1);
|
| 654 | }
|
| 655 | }
|
| 656 | }
|
| 657 |
|
| 658 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio)
|
| 659 | {
|
| 660 | struct geom *geo = &conf->geo;
|
| 661 |
|
| 662 | if (conf->reshape_progress != MaxSector &&
|
| 663 | ((r10bio->sector >= conf->reshape_progress) !=
|
| 664 | conf->mddev->reshape_backwards)) {
|
| 665 | set_bit(R10BIO_Previous, &r10bio->state);
|
| 666 | geo = &conf->prev;
|
| 667 | } else
|
| 668 | clear_bit(R10BIO_Previous, &r10bio->state);
|
| 669 |
|
| 670 | __raid10_find_phys(geo, r10bio);
|
| 671 | }
|
| 672 |
|
| 673 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev)
|
| 674 | {
|
| 675 | sector_t offset, chunk, vchunk;
|
| 676 | |
| 677 | * or recovery, so reshape isn't happening
|
| 678 |
|
| 679 | struct geom *geo = &conf->geo;
|
| 680 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size;
|
| 681 | int far_set_size = geo->far_set_size;
|
| 682 | int last_far_set_start;
|
| 683 |
|
| 684 | if (geo->raid_disks % geo->far_set_size) {
|
| 685 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1;
|
| 686 | last_far_set_start *= geo->far_set_size;
|
| 687 |
|
| 688 | if (dev >= last_far_set_start) {
|
| 689 | far_set_size = geo->far_set_size;
|
| 690 | far_set_size += (geo->raid_disks % geo->far_set_size);
|
| 691 | far_set_start = last_far_set_start;
|
| 692 | }
|
| 693 | }
|
| 694 |
|
| 695 | offset = sector & geo->chunk_mask;
|
| 696 | if (geo->far_offset) {
|
| 697 | int fc;
|
| 698 | chunk = sector >> geo->chunk_shift;
|
| 699 | fc = sector_div(chunk, geo->far_copies);
|
| 700 | dev -= fc * geo->near_copies;
|
| 701 | if (dev < far_set_start)
|
| 702 | dev += far_set_size;
|
| 703 | } else {
|
| 704 | while (sector >= geo->stride) {
|
| 705 | sector -= geo->stride;
|
| 706 | if (dev < (geo->near_copies + far_set_start))
|
| 707 | dev += far_set_size - geo->near_copies;
|
| 708 | else
|
| 709 | dev -= geo->near_copies;
|
| 710 | }
|
| 711 | chunk = sector >> geo->chunk_shift;
|
| 712 | }
|
| 713 | vchunk = chunk * geo->raid_disks + dev;
|
| 714 | sector_div(vchunk, geo->near_copies);
|
| 715 | return (vchunk << geo->chunk_shift) + offset;
|
| 716 | }
|
| 717 |
|
| 718 | |
| 719 | * This routine returns the disk from which the requested read should
|
| 720 | * be done. There is a per-array 'next expected sequential IO' sector
|
| 721 | * number - if this matches on the next IO then we use the last disk.
|
| 722 | * There is also a per-disk 'last know head position' sector that is
|
| 723 | * maintained from IRQ contexts, both the normal and the resync IO
|
| 724 | * completion handlers update this position correctly. If there is no
|
| 725 | * perfect sequential match then we pick the disk whose head is closest.
|
| 726 | *
|
| 727 | * If there are 2 mirrors in the same 2 devices, performance degrades
|
| 728 | * because position is mirror, not device based.
|
| 729 | *
|
| 730 | * The rdev for the device selected will have nr_pending incremented.
|
| 731 |
|
| 732 |
|
| 733 | |
| 734 | * FIXME: possibly should rethink readbalancing and do it differently
|
| 735 | * depending on near_copies / far_copies geometry.
|
| 736 |
|
| 737 | static struct md_rdev *read_balance(struct r10conf *conf,
|
| 738 | struct r10bio *r10_bio,
|
| 739 | int *max_sectors)
|
| 740 | {
|
| 741 | const sector_t this_sector = r10_bio->sector;
|
| 742 | int disk, slot;
|
| 743 | int sectors = r10_bio->sectors;
|
| 744 | int best_good_sectors;
|
| 745 | sector_t new_distance, best_dist;
|
| 746 | struct md_rdev *best_dist_rdev, *best_pending_rdev, *rdev = NULL;
|
| 747 | int do_balance;
|
| 748 | int best_dist_slot, best_pending_slot;
|
| 749 | bool has_nonrot_disk = false;
|
| 750 | unsigned int min_pending;
|
| 751 | struct geom *geo = &conf->geo;
|
| 752 |
|
| 753 | raid10_find_phys(conf, r10_bio);
|
| 754 | best_dist_slot = -1;
|
| 755 | min_pending = UINT_MAX;
|
| 756 | best_dist_rdev = NULL;
|
| 757 | best_pending_rdev = NULL;
|
| 758 | best_dist = MaxSector;
|
| 759 | best_good_sectors = 0;
|
| 760 | do_balance = 1;
|
| 761 | clear_bit(R10BIO_FailFast, &r10_bio->state);
|
| 762 |
|
| 763 | if (raid1_should_read_first(conf->mddev, this_sector, sectors))
|
| 764 | do_balance = 0;
|
| 765 |
|
| 766 | for (slot = 0; slot < conf->copies ; slot++) {
|
| 767 | sector_t first_bad;
|
| 768 | sector_t bad_sectors;
|
| 769 | sector_t dev_sector;
|
| 770 | unsigned int pending;
|
| 771 | bool nonrot;
|
| 772 |
|
| 773 | if (r10_bio->devs[slot].bio == IO_BLOCKED)
|
| 774 | continue;
|
| 775 | disk = r10_bio->devs[slot].devnum;
|
| 776 | rdev = conf->mirrors[disk].replacement;
|
| 777 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) ||
|
| 778 | r10_bio->devs[slot].addr + sectors >
|
| 779 | rdev->recovery_offset)
|
| 780 | rdev = conf->mirrors[disk].rdev;
|
| 781 | if (rdev == NULL ||
|
| 782 | test_bit(Faulty, &rdev->flags))
|
| 783 | continue;
|
| 784 | if (!test_bit(In_sync, &rdev->flags) &&
|
| 785 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset)
|
| 786 | continue;
|
| 787 |
|
| 788 | dev_sector = r10_bio->devs[slot].addr;
|
| 789 | if (is_badblock(rdev, dev_sector, sectors,
|
| 790 | &first_bad, &bad_sectors)) {
|
| 791 | if (best_dist < MaxSector)
|
| 792 |
|
| 793 | continue;
|
| 794 | if (first_bad <= dev_sector) {
|
| 795 | |
| 796 | * 'primary' device, then we must not read
|
| 797 | * beyond 'bad_sectors' from another device.
|
| 798 |
|
| 799 | bad_sectors -= (dev_sector - first_bad);
|
| 800 | if (!do_balance && sectors > bad_sectors)
|
| 801 | sectors = bad_sectors;
|
| 802 | if (best_good_sectors > sectors)
|
| 803 | best_good_sectors = sectors;
|
| 804 | } else {
|
| 805 | sector_t good_sectors =
|
| 806 | first_bad - dev_sector;
|
| 807 | if (good_sectors > best_good_sectors) {
|
| 808 | best_good_sectors = good_sectors;
|
| 809 | best_dist_slot = slot;
|
| 810 | best_dist_rdev = rdev;
|
| 811 | }
|
| 812 | if (!do_balance)
|
| 813 |
|
| 814 | break;
|
| 815 | }
|
| 816 | continue;
|
| 817 | } else
|
| 818 | best_good_sectors = sectors;
|
| 819 |
|
| 820 | if (!do_balance)
|
| 821 | break;
|
| 822 |
|
| 823 | nonrot = !bdev_rot(rdev->bdev);
|
| 824 | has_nonrot_disk |= nonrot;
|
| 825 | pending = atomic_read(&rdev->nr_pending);
|
| 826 | if (min_pending > pending && nonrot) {
|
| 827 | min_pending = pending;
|
| 828 | best_pending_slot = slot;
|
| 829 | best_pending_rdev = rdev;
|
| 830 | }
|
| 831 |
|
| 832 | if (best_dist_slot >= 0)
|
| 833 |
|
| 834 | set_bit(R10BIO_FailFast, &r10_bio->state);
|
| 835 | |
| 836 | * sequential read speed for 'far copies' arrays. So only
|
| 837 | * keep it for 'near' arrays, and review those later.
|
| 838 |
|
| 839 | if (geo->near_copies > 1 && !pending)
|
| 840 | new_distance = 0;
|
| 841 |
|
| 842 |
|
| 843 | else if (geo->far_copies > 1)
|
| 844 | new_distance = r10_bio->devs[slot].addr;
|
| 845 | else
|
| 846 | new_distance = abs(r10_bio->devs[slot].addr -
|
| 847 | conf->mirrors[disk].head_position);
|
| 848 |
|
| 849 | if (new_distance < best_dist) {
|
| 850 | best_dist = new_distance;
|
| 851 | best_dist_slot = slot;
|
| 852 | best_dist_rdev = rdev;
|
| 853 | }
|
| 854 | }
|
| 855 | if (slot >= conf->copies) {
|
| 856 | if (has_nonrot_disk) {
|
| 857 | slot = best_pending_slot;
|
| 858 | rdev = best_pending_rdev;
|
| 859 | } else {
|
| 860 | slot = best_dist_slot;
|
| 861 | rdev = best_dist_rdev;
|
| 862 | }
|
| 863 | }
|
| 864 |
|
| 865 | if (slot >= 0) {
|
| 866 | atomic_inc(&rdev->nr_pending);
|
| 867 | r10_bio->read_slot = slot;
|
| 868 | } else
|
| 869 | rdev = NULL;
|
| 870 | *max_sectors = best_good_sectors;
|
| 871 |
|
| 872 | return rdev;
|
| 873 | }
|
| 874 |
|
| 875 | static void flush_pending_writes(struct r10conf *conf)
|
| 876 | {
|
| 877 | |
| 878 | * bitmap updates get flushed here.
|
| 879 |
|
| 880 | spin_lock_irq(&conf->device_lock);
|
| 881 |
|
| 882 | if (conf->pending_bio_list.head) {
|
| 883 | struct blk_plug plug;
|
| 884 | struct bio *bio;
|
| 885 |
|
| 886 | bio = bio_list_get(&conf->pending_bio_list);
|
| 887 | spin_unlock_irq(&conf->device_lock);
|
| 888 |
|
| 889 | |
| 890 | * As this is called in a wait_event() loop (see freeze_array),
|
| 891 | * current->state might be TASK_UNINTERRUPTIBLE which will
|
| 892 | * cause a warning when we prepare to wait again. As it is
|
| 893 | * rare that this path is taken, it is perfectly safe to force
|
| 894 | * us to go around the wait_event() loop again, so the warning
|
| 895 | * is a false-positive. Silence the warning by resetting
|
| 896 | * thread state
|
| 897 |
|
| 898 | __set_current_state(TASK_RUNNING);
|
| 899 |
|
| 900 | blk_start_plug(&plug);
|
| 901 | raid1_prepare_flush_writes(conf->mddev);
|
| 902 | wake_up(&conf->wait_barrier);
|
| 903 |
|
| 904 | while (bio) {
|
| 905 | struct bio *next = bio->bi_next;
|
| 906 |
|
| 907 | raid1_submit_write(bio);
|
| 908 | bio = next;
|
| 909 | cond_resched();
|
| 910 | }
|
| 911 | blk_finish_plug(&plug);
|
| 912 | } else
|
| 913 | spin_unlock_irq(&conf->device_lock);
|
| 914 | }
|
| 915 |
|
| 916 | |
| 917 | * Sometimes we need to suspend IO while we do something else,
|
| 918 | * either some resync/recovery, or reconfigure the array.
|
| 919 | * To do this we raise a 'barrier'.
|
| 920 | * The 'barrier' is a counter that can be raised multiple times
|
| 921 | * to count how many activities are happening which preclude
|
| 922 | * normal IO.
|
| 923 | * We can only raise the barrier if there is no pending IO.
|
| 924 | * i.e. if nr_pending == 0.
|
| 925 | * We choose only to raise the barrier if no-one is waiting for the
|
| 926 | * barrier to go down. This means that as soon as an IO request
|
| 927 | * is ready, no other operations which require a barrier will start
|
| 928 | * until the IO request has had a chance.
|
| 929 | *
|
| 930 | * So: regular IO calls 'wait_barrier'. When that returns there
|
| 931 | * is no backgroup IO happening, It must arrange to call
|
| 932 | * allow_barrier when it has finished its IO.
|
| 933 | * backgroup IO calls must call raise_barrier. Once that returns
|
| 934 | * there is no normal IO happeing. It must arrange to call
|
| 935 | * lower_barrier when the particular background IO completes.
|
| 936 |
|
| 937 |
|
| 938 | static void raise_barrier(struct r10conf *conf, int force)
|
| 939 | {
|
| 940 | write_seqlock_irq(&conf->resync_lock);
|
| 941 |
|
| 942 | if (WARN_ON_ONCE(force && !conf->barrier))
|
| 943 | force = false;
|
| 944 |
|
| 945 |
|
| 946 | wait_event_barrier(conf, force || !conf->nr_waiting);
|
| 947 |
|
| 948 |
|
| 949 | WRITE_ONCE(conf->barrier, conf->barrier + 1);
|
| 950 |
|
| 951 |
|
| 952 | wait_event_barrier(conf, !atomic_read(&conf->nr_pending) &&
|
| 953 | conf->barrier < RESYNC_DEPTH);
|
| 954 |
|
| 955 | write_sequnlock_irq(&conf->resync_lock);
|
| 956 | }
|
| 957 |
|
| 958 | static void lower_barrier(struct r10conf *conf)
|
| 959 | {
|
| 960 | unsigned long flags;
|
| 961 |
|
| 962 | write_seqlock_irqsave(&conf->resync_lock, flags);
|
| 963 | WRITE_ONCE(conf->barrier, conf->barrier - 1);
|
| 964 | write_sequnlock_irqrestore(&conf->resync_lock, flags);
|
| 965 | wake_up(&conf->wait_barrier);
|
| 966 | }
|
| 967 |
|
| 968 | static bool stop_waiting_barrier(struct r10conf *conf)
|
| 969 | {
|
| 970 | struct bio_list *bio_list = current->bio_list;
|
| 971 | struct md_thread *thread;
|
| 972 |
|
| 973 |
|
| 974 | if (!conf->barrier)
|
| 975 | return true;
|
| 976 |
|
| 977 | |
| 978 | * If there are already pending requests (preventing the barrier from
|
| 979 | * rising completely), and the pre-process bio queue isn't empty, then
|
| 980 | * don't wait, as we need to empty that queue to get the nr_pending
|
| 981 | * count down.
|
| 982 |
|
| 983 | if (atomic_read(&conf->nr_pending) && bio_list &&
|
| 984 | (!bio_list_empty(&bio_list[0]) || !bio_list_empty(&bio_list[1])))
|
| 985 | return true;
|
| 986 |
|
| 987 |
|
| 988 | thread = rcu_dereference_protected(conf->mddev->thread, true);
|
| 989 | |
| 990 | * move on if io is issued from raid10d(), nr_pending is not released
|
| 991 | * from original io(see handle_read_error()). All raise barrier is
|
| 992 | * blocked until this io is done.
|
| 993 |
|
| 994 | if (thread->tsk == current) {
|
| 995 | WARN_ON_ONCE(atomic_read(&conf->nr_pending) == 0);
|
| 996 | return true;
|
| 997 | }
|
| 998 |
|
| 999 | return false;
|
| 1000 | }
|
| 1001 |
|
| 1002 | static bool wait_barrier_nolock(struct r10conf *conf)
|
| 1003 | {
|
| 1004 | unsigned int seq = read_seqbegin(&conf->resync_lock);
|
| 1005 |
|
| 1006 | if (READ_ONCE(conf->barrier))
|
| 1007 | return false;
|
| 1008 |
|
| 1009 | atomic_inc(&conf->nr_pending);
|
| 1010 | if (!read_seqretry(&conf->resync_lock, seq))
|
| 1011 | return true;
|
| 1012 |
|
| 1013 | if (atomic_dec_and_test(&conf->nr_pending))
|
| 1014 | wake_up_barrier(conf);
|
| 1015 |
|
| 1016 | return false;
|
| 1017 | }
|
| 1018 |
|
| 1019 | static void wait_barrier(struct r10conf *conf)
|
| 1020 | {
|
| 1021 | if (wait_barrier_nolock(conf))
|
| 1022 | return;
|
| 1023 |
|
| 1024 | write_seqlock_irq(&conf->resync_lock);
|
| 1025 | if (conf->barrier) {
|
| 1026 | conf->nr_waiting++;
|
| 1027 | mddev_add_trace_msg(conf->mddev, "raid10 wait barrier");
|
| 1028 | wait_event_barrier(conf, stop_waiting_barrier(conf));
|
| 1029 | conf->nr_waiting--;
|
| 1030 | if (!conf->nr_waiting)
|
| 1031 | wake_up(&conf->wait_barrier);
|
| 1032 | }
|
| 1033 | atomic_inc(&conf->nr_pending);
|
| 1034 | write_sequnlock_irq(&conf->resync_lock);
|
| 1035 | }
|
| 1036 |
|
| 1037 | static void allow_barrier(struct r10conf *conf)
|
| 1038 | {
|
| 1039 | if ((atomic_dec_and_test(&conf->nr_pending)) ||
|
| 1040 | (conf->array_freeze_pending))
|
| 1041 | wake_up_barrier(conf);
|
| 1042 | }
|
| 1043 |
|
| 1044 | static void freeze_array(struct r10conf *conf, int extra)
|
| 1045 | {
|
| 1046 | |
| 1047 | * go quiet.
|
| 1048 | * We increment barrier and nr_waiting, and then
|
| 1049 | * wait until nr_pending match nr_queued+extra
|
| 1050 | * This is called in the context of one normal IO request
|
| 1051 | * that has failed. Thus any sync request that might be pending
|
| 1052 | * will be blocked by nr_pending, and we need to wait for
|
| 1053 | * pending IO requests to complete or be queued for re-try.
|
| 1054 | * Thus the number queued (nr_queued) plus this request (extra)
|
| 1055 | * must match the number of pending IOs (nr_pending) before
|
| 1056 | * we continue.
|
| 1057 |
|
| 1058 | write_seqlock_irq(&conf->resync_lock);
|
| 1059 | conf->array_freeze_pending++;
|
| 1060 | WRITE_ONCE(conf->barrier, conf->barrier + 1);
|
| 1061 | conf->nr_waiting++;
|
| 1062 | wait_event_barrier_cmd(conf, atomic_read(&conf->nr_pending) ==
|
| 1063 | conf->nr_queued + extra, flush_pending_writes(conf));
|
| 1064 | conf->array_freeze_pending--;
|
| 1065 | write_sequnlock_irq(&conf->resync_lock);
|
| 1066 | }
|
| 1067 |
|
| 1068 | static void unfreeze_array(struct r10conf *conf)
|
| 1069 | {
|
| 1070 |
|
| 1071 | write_seqlock_irq(&conf->resync_lock);
|
| 1072 | WRITE_ONCE(conf->barrier, conf->barrier - 1);
|
| 1073 | conf->nr_waiting--;
|
| 1074 | wake_up(&conf->wait_barrier);
|
| 1075 | write_sequnlock_irq(&conf->resync_lock);
|
| 1076 | }
|
| 1077 |
|
| 1078 | static sector_t choose_data_offset(struct r10bio *r10_bio,
|
| 1079 | struct md_rdev *rdev)
|
| 1080 | {
|
| 1081 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) ||
|
| 1082 | test_bit(R10BIO_Previous, &r10_bio->state))
|
| 1083 | return rdev->data_offset;
|
| 1084 | else
|
| 1085 | return rdev->new_data_offset;
|
| 1086 | }
|
| 1087 |
|
| 1088 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule)
|
| 1089 | {
|
| 1090 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, cb);
|
| 1091 | struct mddev *mddev = plug->cb.data;
|
| 1092 | struct r10conf *conf = mddev->private;
|
| 1093 | struct bio *bio;
|
| 1094 |
|
| 1095 | if (from_schedule) {
|
| 1096 | spin_lock_irq(&conf->device_lock);
|
| 1097 | bio_list_merge(&conf->pending_bio_list, &plug->pending);
|
| 1098 | spin_unlock_irq(&conf->device_lock);
|
| 1099 | wake_up_barrier(conf);
|
| 1100 | md_wakeup_thread(mddev->thread);
|
| 1101 | kfree(plug);
|
| 1102 | return;
|
| 1103 | }
|
| 1104 |
|
| 1105 |
|
| 1106 | bio = bio_list_get(&plug->pending);
|
| 1107 | raid1_prepare_flush_writes(mddev);
|
| 1108 | wake_up_barrier(conf);
|
| 1109 |
|
| 1110 | while (bio) {
|
| 1111 | struct bio *next = bio->bi_next;
|
| 1112 |
|
| 1113 | raid1_submit_write(bio);
|
| 1114 | bio = next;
|
| 1115 | cond_resched();
|
| 1116 | }
|
| 1117 | kfree(plug);
|
| 1118 | }
|
| 1119 |
|
| 1120 | |
| 1121 | * 1. Register the new request and wait if the reconstruction thread has put
|
| 1122 | * up a bar for new requests. Continue immediately if no resync is active
|
| 1123 | * currently.
|
| 1124 | * 2. If IO spans the reshape position. Need to wait for reshape to pass.
|
| 1125 |
|
| 1126 | static void regular_request_wait(struct mddev *mddev, struct r10conf *conf,
|
| 1127 | struct bio *bio, sector_t sectors)
|
| 1128 | {
|
| 1129 | wait_barrier(conf);
|
| 1130 |
|
| 1131 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
|
| 1132 | bio->bi_iter.bi_sector < conf->reshape_progress &&
|
| 1133 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) {
|
| 1134 | allow_barrier(conf);
|
| 1135 | mddev_add_trace_msg(conf->mddev, "raid10 wait reshape");
|
| 1136 | wait_event(conf->wait_barrier,
|
| 1137 | conf->reshape_progress <= bio->bi_iter.bi_sector ||
|
| 1138 | conf->reshape_progress >= bio->bi_iter.bi_sector +
|
| 1139 | sectors);
|
| 1140 | wait_barrier(conf);
|
| 1141 | }
|
| 1142 | }
|
| 1143 |
|
| 1144 | static void raid10_read_request(struct mddev *mddev, struct bio *bio,
|
| 1145 | struct r10bio *r10_bio)
|
| 1146 | {
|
| 1147 | struct r10conf *conf = mddev->private;
|
| 1148 | struct bio *read_bio;
|
| 1149 | int max_sectors;
|
| 1150 | struct md_rdev *rdev;
|
| 1151 | char b[BDEVNAME_SIZE];
|
| 1152 | int slot = r10_bio->read_slot;
|
| 1153 | struct md_rdev *err_rdev = NULL;
|
| 1154 |
|
| 1155 | |
| 1156 | * An md cloned bio indicates we are in the error path.
|
| 1157 | * This is more reliable than checking slot, which might
|
| 1158 | * be -1 even in the error path if a failed bio was split.
|
| 1159 |
|
| 1160 | bool err_path = md_cloned_bio(mddev, bio);
|
| 1161 |
|
| 1162 | |
| 1163 | * If we are in the error path, we are blocking the raid10d
|
| 1164 | * thread so there is a tiny risk of deadlock. So ask for
|
| 1165 | * emergency memory if needed.
|
| 1166 |
|
| 1167 | gfp_t gfp = err_path ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO;
|
| 1168 |
|
| 1169 | if (slot >= 0 && r10_bio->devs[slot].rdev) {
|
| 1170 | |
| 1171 | * This is an error retry, but we cannot
|
| 1172 | * safely dereference the rdev in the r10_bio,
|
| 1173 | * we must use the one in conf.
|
| 1174 | * If it has already been disconnected (unlikely)
|
| 1175 | * we lose the device name in error messages.
|
| 1176 |
|
| 1177 | int disk;
|
| 1178 |
|
| 1179 | disk = r10_bio->devs[slot].devnum;
|
| 1180 | err_rdev = conf->mirrors[disk].rdev;
|
| 1181 | if (err_rdev)
|
| 1182 | snprintf(b, sizeof(b), "%pg", err_rdev->bdev);
|
| 1183 | else {
|
| 1184 | strcpy(b, "???");
|
| 1185 |
|
| 1186 | err_rdev = r10_bio->devs[slot].rdev;
|
| 1187 | }
|
| 1188 | }
|
| 1189 |
|
| 1190 | regular_request_wait(mddev, conf, bio, r10_bio->sectors);
|
| 1191 |
|
| 1192 | rdev = read_balance(conf, r10_bio, &max_sectors);
|
| 1193 | if (!rdev) {
|
| 1194 | if (err_rdev) {
|
| 1195 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n",
|
| 1196 | mdname(mddev), b,
|
| 1197 | (unsigned long long)r10_bio->sector);
|
| 1198 | }
|
| 1199 | raid_end_bio_io(r10_bio);
|
| 1200 | return;
|
| 1201 | }
|
| 1202 | if (err_rdev)
|
| 1203 | pr_err_ratelimited("md/raid10:%s: %pg: redirecting sector %llu to another mirror\n",
|
| 1204 | mdname(mddev),
|
| 1205 | rdev->bdev,
|
| 1206 | (unsigned long long)r10_bio->sector);
|
| 1207 | if (max_sectors < bio_sectors(bio)) {
|
| 1208 | allow_barrier(conf);
|
| 1209 | bio = bio_submit_split_bioset(bio, max_sectors,
|
| 1210 | &conf->bio_split);
|
| 1211 | wait_barrier(conf);
|
| 1212 | if (!bio) {
|
| 1213 | set_bit(R10BIO_Returned, &r10_bio->state);
|
| 1214 | goto err_handle;
|
| 1215 | }
|
| 1216 |
|
| 1217 | r10_bio->master_bio = bio;
|
| 1218 | r10_bio->sectors = max_sectors;
|
| 1219 | }
|
| 1220 | slot = r10_bio->read_slot;
|
| 1221 |
|
| 1222 | if (likely(!md_cloned_bio(mddev, bio))) {
|
| 1223 | md_account_bio(mddev, &bio);
|
| 1224 | r10_bio->master_bio = bio;
|
| 1225 | }
|
| 1226 | read_bio = bio_alloc_clone(rdev->bdev, bio, gfp, &mddev->bio_set);
|
| 1227 |
|
| 1228 | r10_bio->devs[slot].bio = read_bio;
|
| 1229 | r10_bio->devs[slot].rdev = rdev;
|
| 1230 |
|
| 1231 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr +
|
| 1232 | choose_data_offset(r10_bio, rdev);
|
| 1233 | read_bio->bi_end_io = raid10_end_read_request;
|
| 1234 | if (test_bit(FailFast, &rdev->flags) &&
|
| 1235 | test_bit(R10BIO_FailFast, &r10_bio->state))
|
| 1236 | read_bio->bi_opf |= MD_FAILFAST;
|
| 1237 | read_bio->bi_private = r10_bio;
|
| 1238 | mddev_trace_remap(mddev, read_bio, r10_bio->sector);
|
| 1239 | submit_bio_noacct(read_bio);
|
| 1240 | return;
|
| 1241 | err_handle:
|
| 1242 | atomic_dec(&rdev->nr_pending);
|
| 1243 | raid_end_bio_io(r10_bio);
|
| 1244 | }
|
| 1245 |
|
| 1246 | static void raid10_write_one_disk(struct mddev *mddev, struct r10bio *r10_bio,
|
| 1247 | struct bio *bio, bool replacement,
|
| 1248 | int n_copy)
|
| 1249 | {
|
| 1250 | unsigned long flags;
|
| 1251 | struct r10conf *conf = mddev->private;
|
| 1252 | struct md_rdev *rdev;
|
| 1253 | int devnum = r10_bio->devs[n_copy].devnum;
|
| 1254 | struct bio *mbio;
|
| 1255 |
|
| 1256 | rdev = replacement ? conf->mirrors[devnum].replacement :
|
| 1257 | conf->mirrors[devnum].rdev;
|
| 1258 |
|
| 1259 | mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, &mddev->bio_set);
|
| 1260 | if (replacement)
|
| 1261 | r10_bio->devs[n_copy].repl_bio = mbio;
|
| 1262 | else
|
| 1263 | r10_bio->devs[n_copy].bio = mbio;
|
| 1264 |
|
| 1265 | mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr +
|
| 1266 | choose_data_offset(r10_bio, rdev));
|
| 1267 | mbio->bi_end_io = raid10_end_write_request;
|
| 1268 | if (!replacement && test_bit(FailFast,
|
| 1269 | &conf->mirrors[devnum].rdev->flags)
|
| 1270 | && enough(conf, devnum))
|
| 1271 | mbio->bi_opf |= MD_FAILFAST;
|
| 1272 | mbio->bi_private = r10_bio;
|
| 1273 | mddev_trace_remap(mddev, mbio, r10_bio->sector);
|
| 1274 |
|
| 1275 | mbio->bi_bdev = (void *)rdev;
|
| 1276 |
|
| 1277 | atomic_inc(&r10_bio->remaining);
|
| 1278 |
|
| 1279 | if (!raid1_add_bio_to_plug(mddev, mbio, raid10_unplug, conf->copies)) {
|
| 1280 | spin_lock_irqsave(&conf->device_lock, flags);
|
| 1281 | bio_list_add(&conf->pending_bio_list, mbio);
|
| 1282 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 1283 | md_wakeup_thread(mddev->thread);
|
| 1284 | }
|
| 1285 | }
|
| 1286 |
|
| 1287 | static void wait_blocked_dev(struct mddev *mddev, struct r10bio *r10_bio)
|
| 1288 | {
|
| 1289 | struct r10conf *conf = mddev->private;
|
| 1290 | struct md_rdev *blocked_rdev;
|
| 1291 | int i;
|
| 1292 |
|
| 1293 | retry_wait:
|
| 1294 | blocked_rdev = NULL;
|
| 1295 | for (i = 0; i < conf->copies; i++) {
|
| 1296 | struct md_rdev *rdev, *rrdev;
|
| 1297 |
|
| 1298 | rdev = conf->mirrors[i].rdev;
|
| 1299 | if (rdev) {
|
| 1300 | sector_t dev_sector = r10_bio->devs[i].addr;
|
| 1301 |
|
| 1302 | |
| 1303 | * Discard request doesn't care the write result
|
| 1304 | * so it doesn't need to wait blocked disk here.
|
| 1305 |
|
| 1306 | if (test_bit(WriteErrorSeen, &rdev->flags) &&
|
| 1307 | r10_bio->sectors &&
|
| 1308 | rdev_has_badblock(rdev, dev_sector,
|
| 1309 | r10_bio->sectors) < 0)
|
| 1310 | |
| 1311 | * Mustn't write here until the bad
|
| 1312 | * block is acknowledged
|
| 1313 |
|
| 1314 | set_bit(BlockedBadBlocks, &rdev->flags);
|
| 1315 |
|
| 1316 | if (rdev_blocked(rdev)) {
|
| 1317 | blocked_rdev = rdev;
|
| 1318 | atomic_inc(&rdev->nr_pending);
|
| 1319 | break;
|
| 1320 | }
|
| 1321 | }
|
| 1322 |
|
| 1323 | rrdev = conf->mirrors[i].replacement;
|
| 1324 | if (rrdev && rdev_blocked(rrdev)) {
|
| 1325 | atomic_inc(&rrdev->nr_pending);
|
| 1326 | blocked_rdev = rrdev;
|
| 1327 | break;
|
| 1328 | }
|
| 1329 | }
|
| 1330 |
|
| 1331 | if (unlikely(blocked_rdev)) {
|
| 1332 |
|
| 1333 | allow_barrier(conf);
|
| 1334 | mddev_add_trace_msg(conf->mddev,
|
| 1335 | "raid10 %s wait rdev %d blocked",
|
| 1336 | __func__, blocked_rdev->raid_disk);
|
| 1337 | md_wait_for_blocked_rdev(blocked_rdev, mddev);
|
| 1338 | wait_barrier(conf);
|
| 1339 | goto retry_wait;
|
| 1340 | }
|
| 1341 | }
|
| 1342 |
|
| 1343 | static bool raid10_write_request(struct mddev *mddev, struct bio *bio,
|
| 1344 | struct r10bio *r10_bio)
|
| 1345 | {
|
| 1346 | struct r10conf *conf = mddev->private;
|
| 1347 | int i, k;
|
| 1348 | sector_t sectors;
|
| 1349 | int max_sectors;
|
| 1350 | bool atomic = bio->bi_opf & REQ_ATOMIC;
|
| 1351 |
|
| 1352 | if ((mddev_is_clustered(mddev) &&
|
| 1353 | mddev->cluster_ops->area_resyncing(mddev, WRITE,
|
| 1354 | bio->bi_iter.bi_sector,
|
| 1355 | bio_end_sector(bio)))) {
|
| 1356 | wait_event_idle(conf->wait_barrier,
|
| 1357 | !mddev->cluster_ops->area_resyncing(mddev, WRITE,
|
| 1358 | bio->bi_iter.bi_sector,
|
| 1359 | bio_end_sector(bio)));
|
| 1360 | }
|
| 1361 |
|
| 1362 | sectors = r10_bio->sectors;
|
| 1363 | regular_request_wait(mddev, conf, bio, sectors);
|
| 1364 |
|
| 1365 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
|
| 1366 | (mddev->reshape_backwards
|
| 1367 | ? (bio->bi_iter.bi_sector < conf->reshape_safe &&
|
| 1368 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress)
|
| 1369 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe &&
|
| 1370 | bio->bi_iter.bi_sector < conf->reshape_progress))) {
|
| 1371 |
|
| 1372 | mddev->reshape_position = conf->reshape_progress;
|
| 1373 | set_mask_bits(&mddev->sb_flags, 0,
|
| 1374 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
|
| 1375 | md_wakeup_thread(mddev->thread);
|
| 1376 | mddev_add_trace_msg(conf->mddev,
|
| 1377 | "raid10 wait reshape metadata");
|
| 1378 | wait_event(mddev->sb_wait,
|
| 1379 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
|
| 1380 |
|
| 1381 | conf->reshape_safe = mddev->reshape_position;
|
| 1382 | }
|
| 1383 |
|
| 1384 | |
| 1385 | * inc refcount on their rdev. Record them by setting
|
| 1386 | * bios[x] to bio
|
| 1387 | * If there are known/acknowledged bad blocks on any device
|
| 1388 | * on which we have seen a write error, we want to avoid
|
| 1389 | * writing to those blocks. This potentially requires several
|
| 1390 | * writes to write around the bad blocks. Each set of writes
|
| 1391 | * gets its own r10_bio with a set of bios attached.
|
| 1392 |
|
| 1393 |
|
| 1394 | r10_bio->read_slot = -1;
|
| 1395 | raid10_find_phys(conf, r10_bio);
|
| 1396 |
|
| 1397 | wait_blocked_dev(mddev, r10_bio);
|
| 1398 |
|
| 1399 | max_sectors = r10_bio->sectors;
|
| 1400 |
|
| 1401 | for (i = 0; i < conf->copies; i++) {
|
| 1402 | int d = r10_bio->devs[i].devnum;
|
| 1403 | struct md_rdev *rdev, *rrdev;
|
| 1404 |
|
| 1405 | rdev = conf->mirrors[d].rdev;
|
| 1406 | rrdev = conf->mirrors[d].replacement;
|
| 1407 | if (rdev && (test_bit(Faulty, &rdev->flags)))
|
| 1408 | rdev = NULL;
|
| 1409 | if (rrdev && (test_bit(Faulty, &rrdev->flags)))
|
| 1410 | rrdev = NULL;
|
| 1411 |
|
| 1412 | r10_bio->devs[i].bio = NULL;
|
| 1413 | r10_bio->devs[i].repl_bio = NULL;
|
| 1414 |
|
| 1415 | if (!rdev && !rrdev)
|
| 1416 | continue;
|
| 1417 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) {
|
| 1418 | sector_t first_bad;
|
| 1419 | sector_t dev_sector = r10_bio->devs[i].addr;
|
| 1420 | sector_t bad_sectors;
|
| 1421 | int is_bad;
|
| 1422 |
|
| 1423 | is_bad = is_badblock(rdev, dev_sector, max_sectors,
|
| 1424 | &first_bad, &bad_sectors);
|
| 1425 | if (is_bad && first_bad <= dev_sector) {
|
| 1426 |
|
| 1427 | bad_sectors -= (dev_sector - first_bad);
|
| 1428 | if (bad_sectors < max_sectors)
|
| 1429 | |
| 1430 | * to other devices yet
|
| 1431 |
|
| 1432 | max_sectors = bad_sectors;
|
| 1433 | continue;
|
| 1434 | }
|
| 1435 | if (is_bad) {
|
| 1436 | int good_sectors;
|
| 1437 |
|
| 1438 | good_sectors = first_bad - dev_sector;
|
| 1439 | if (good_sectors < max_sectors)
|
| 1440 | max_sectors = good_sectors;
|
| 1441 | }
|
| 1442 | }
|
| 1443 | if (rdev) {
|
| 1444 | r10_bio->devs[i].bio = bio;
|
| 1445 | atomic_inc(&rdev->nr_pending);
|
| 1446 | }
|
| 1447 | if (rrdev) {
|
| 1448 | r10_bio->devs[i].repl_bio = bio;
|
| 1449 | atomic_inc(&rrdev->nr_pending);
|
| 1450 | }
|
| 1451 | }
|
| 1452 |
|
| 1453 | if (max_sectors < r10_bio->sectors)
|
| 1454 | r10_bio->sectors = max_sectors;
|
| 1455 |
|
| 1456 | if (r10_bio->sectors < bio_sectors(bio)) {
|
| 1457 | if (atomic)
|
| 1458 | goto err_handle;
|
| 1459 |
|
| 1460 | bio = bio_submit_split_bioset(bio, r10_bio->sectors,
|
| 1461 | &conf->bio_split);
|
| 1462 | if (!bio) {
|
| 1463 | set_bit(R10BIO_Returned, &r10_bio->state);
|
| 1464 | goto err_handle;
|
| 1465 | }
|
| 1466 |
|
| 1467 | r10_bio->master_bio = bio;
|
| 1468 | }
|
| 1469 |
|
| 1470 | md_account_bio(mddev, &bio);
|
| 1471 | r10_bio->master_bio = bio;
|
| 1472 | atomic_set(&r10_bio->remaining, 1);
|
| 1473 |
|
| 1474 | for (i = 0; i < conf->copies; i++) {
|
| 1475 | if (r10_bio->devs[i].bio)
|
| 1476 | raid10_write_one_disk(mddev, r10_bio, bio, false, i);
|
| 1477 | if (r10_bio->devs[i].repl_bio)
|
| 1478 | raid10_write_one_disk(mddev, r10_bio, bio, true, i);
|
| 1479 | }
|
| 1480 | one_write_done(r10_bio);
|
| 1481 | return true;
|
| 1482 |
|
| 1483 | err_handle:
|
| 1484 | for (k = 0; k < i; k++) {
|
| 1485 | int d = r10_bio->devs[k].devnum;
|
| 1486 | struct md_rdev *rdev = conf->mirrors[d].rdev;
|
| 1487 | struct md_rdev *rrdev = conf->mirrors[d].replacement;
|
| 1488 |
|
| 1489 | if (r10_bio->devs[k].bio) {
|
| 1490 | rdev_dec_pending(rdev, mddev);
|
| 1491 | r10_bio->devs[k].bio = NULL;
|
| 1492 | }
|
| 1493 | if (r10_bio->devs[k].repl_bio) {
|
| 1494 | rdev_dec_pending(rrdev, mddev);
|
| 1495 | r10_bio->devs[k].repl_bio = NULL;
|
| 1496 | }
|
| 1497 | }
|
| 1498 |
|
| 1499 | raid_end_bio_io(r10_bio);
|
| 1500 | return false;
|
| 1501 | }
|
| 1502 |
|
| 1503 | static bool __make_request(struct mddev *mddev, struct bio *bio, int sectors)
|
| 1504 | {
|
| 1505 | bool ret;
|
| 1506 | struct r10conf *conf = mddev->private;
|
| 1507 | struct r10bio *r10_bio;
|
| 1508 |
|
| 1509 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
|
| 1510 |
|
| 1511 | r10_bio->master_bio = bio;
|
| 1512 | r10_bio->sectors = sectors;
|
| 1513 |
|
| 1514 | r10_bio->mddev = mddev;
|
| 1515 | r10_bio->sector = bio->bi_iter.bi_sector;
|
| 1516 | r10_bio->state = 0;
|
| 1517 | r10_bio->read_slot = -1;
|
| 1518 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) *
|
| 1519 | conf->geo.raid_disks);
|
| 1520 |
|
| 1521 | ret = true;
|
| 1522 | if (bio_data_dir(bio) == READ)
|
| 1523 | raid10_read_request(mddev, bio, r10_bio);
|
| 1524 | else
|
| 1525 | ret = raid10_write_request(mddev, bio, r10_bio);
|
| 1526 |
|
| 1527 | return ret;
|
| 1528 | }
|
| 1529 |
|
| 1530 | static void raid_end_discard_bio(struct r10bio *r10bio)
|
| 1531 | {
|
| 1532 | struct r10conf *conf = r10bio->mddev->private;
|
| 1533 | struct r10bio *first_r10bio;
|
| 1534 |
|
| 1535 | while (atomic_dec_and_test(&r10bio->remaining)) {
|
| 1536 |
|
| 1537 | allow_barrier(conf);
|
| 1538 |
|
| 1539 | if (!test_bit(R10BIO_Discard, &r10bio->state)) {
|
| 1540 | first_r10bio = (struct r10bio *)r10bio->master_bio;
|
| 1541 | free_r10bio(r10bio);
|
| 1542 | r10bio = first_r10bio;
|
| 1543 | } else {
|
| 1544 | struct bio *master_bio = r10bio->master_bio;
|
| 1545 |
|
| 1546 | md_write_end(r10bio->mddev);
|
| 1547 | free_r10bio(r10bio);
|
| 1548 | bio_endio(master_bio);
|
| 1549 | break;
|
| 1550 | }
|
| 1551 | }
|
| 1552 | }
|
| 1553 |
|
| 1554 | static void raid10_end_discard_request(struct bio *bio)
|
| 1555 | {
|
| 1556 | struct r10bio *r10_bio = bio->bi_private;
|
| 1557 | struct r10conf *conf = r10_bio->mddev->private;
|
| 1558 | struct md_rdev *rdev = NULL;
|
| 1559 | int dev;
|
| 1560 | int slot, repl;
|
| 1561 |
|
| 1562 | |
| 1563 | * We don't care the return value of discard bio
|
| 1564 |
|
| 1565 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
|
| 1566 | set_bit(R10BIO_Uptodate, &r10_bio->state);
|
| 1567 |
|
| 1568 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
|
| 1569 | rdev = repl ? conf->mirrors[dev].replacement :
|
| 1570 | conf->mirrors[dev].rdev;
|
| 1571 |
|
| 1572 | raid_end_discard_bio(r10_bio);
|
| 1573 | rdev_dec_pending(rdev, conf->mddev);
|
| 1574 | }
|
| 1575 |
|
| 1576 | |
| 1577 | * There are some limitations to handle discard bio
|
| 1578 | * 1st, the discard size is bigger than stripe_size*2.
|
| 1579 | * 2st, if the discard bio spans reshape progress, we use the old way to
|
| 1580 | * handle discard bio
|
| 1581 |
|
| 1582 | static int raid10_handle_discard(struct mddev *mddev, struct bio *bio)
|
| 1583 | {
|
| 1584 | struct r10conf *conf = mddev->private;
|
| 1585 | struct geom *geo = &conf->geo;
|
| 1586 | int far_copies = geo->far_copies;
|
| 1587 | bool first_copy = true;
|
| 1588 | struct r10bio *r10_bio, *first_r10bio;
|
| 1589 | struct bio *split;
|
| 1590 | int disk;
|
| 1591 | sector_t chunk;
|
| 1592 | unsigned int stripe_size;
|
| 1593 | unsigned int stripe_data_disks;
|
| 1594 | sector_t split_size;
|
| 1595 | sector_t bio_start, bio_end;
|
| 1596 | sector_t first_stripe_index, last_stripe_index;
|
| 1597 | sector_t start_disk_offset;
|
| 1598 | unsigned int start_disk_index;
|
| 1599 | sector_t end_disk_offset;
|
| 1600 | unsigned int end_disk_index;
|
| 1601 | unsigned int remainder;
|
| 1602 |
|
| 1603 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
|
| 1604 | return -EAGAIN;
|
| 1605 |
|
| 1606 | wait_barrier(conf);
|
| 1607 |
|
| 1608 | |
| 1609 | * Check reshape again to avoid reshape happens after checking
|
| 1610 | * MD_RECOVERY_RESHAPE and before wait_barrier
|
| 1611 |
|
| 1612 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
|
| 1613 | goto out;
|
| 1614 |
|
| 1615 | if (geo->near_copies)
|
| 1616 | stripe_data_disks = geo->raid_disks / geo->near_copies +
|
| 1617 | geo->raid_disks % geo->near_copies;
|
| 1618 | else
|
| 1619 | stripe_data_disks = geo->raid_disks;
|
| 1620 |
|
| 1621 | stripe_size = stripe_data_disks << geo->chunk_shift;
|
| 1622 |
|
| 1623 | bio_start = bio->bi_iter.bi_sector;
|
| 1624 | bio_end = bio_end_sector(bio);
|
| 1625 |
|
| 1626 | |
| 1627 | * Maybe one discard bio is smaller than strip size or across one
|
| 1628 | * stripe and discard region is larger than one stripe size. For far
|
| 1629 | * offset layout, if the discard region is not aligned with stripe
|
| 1630 | * size, there is hole when we submit discard bio to member disk.
|
| 1631 | * For simplicity, we only handle discard bio which discard region
|
| 1632 | * is bigger than stripe_size * 2
|
| 1633 |
|
| 1634 | if (bio_sectors(bio) < stripe_size*2)
|
| 1635 | goto out;
|
| 1636 |
|
| 1637 | |
| 1638 | * Keep bio aligned with strip size.
|
| 1639 |
|
| 1640 | div_u64_rem(bio_start, stripe_size, &remainder);
|
| 1641 | if (remainder) {
|
| 1642 | split_size = stripe_size - remainder;
|
| 1643 | split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split);
|
| 1644 | if (IS_ERR(split)) {
|
| 1645 | bio->bi_status = errno_to_blk_status(PTR_ERR(split));
|
| 1646 | bio_endio(bio);
|
| 1647 | md_write_end(mddev);
|
| 1648 | allow_barrier(conf);
|
| 1649 | return 0;
|
| 1650 | }
|
| 1651 |
|
| 1652 | bio_chain(split, bio);
|
| 1653 | trace_block_split(split, bio->bi_iter.bi_sector);
|
| 1654 | allow_barrier(conf);
|
| 1655 |
|
| 1656 | submit_bio_noacct(split);
|
| 1657 | wait_barrier(conf);
|
| 1658 | }
|
| 1659 | div_u64_rem(bio_end, stripe_size, &remainder);
|
| 1660 | if (remainder) {
|
| 1661 | split_size = bio_sectors(bio) - remainder;
|
| 1662 | split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split);
|
| 1663 | if (IS_ERR(split)) {
|
| 1664 | bio->bi_status = errno_to_blk_status(PTR_ERR(split));
|
| 1665 | bio_endio(bio);
|
| 1666 | md_write_end(mddev);
|
| 1667 | allow_barrier(conf);
|
| 1668 | return 0;
|
| 1669 | }
|
| 1670 |
|
| 1671 | bio_chain(split, bio);
|
| 1672 | trace_block_split(split, bio->bi_iter.bi_sector);
|
| 1673 | allow_barrier(conf);
|
| 1674 |
|
| 1675 | submit_bio_noacct(bio);
|
| 1676 | bio = split;
|
| 1677 | wait_barrier(conf);
|
| 1678 | }
|
| 1679 |
|
| 1680 | bio_start = bio->bi_iter.bi_sector;
|
| 1681 | bio_end = bio_end_sector(bio);
|
| 1682 |
|
| 1683 | |
| 1684 | * Raid10 uses chunk as the unit to store data. It's similar like raid0.
|
| 1685 | * One stripe contains the chunks from all member disk (one chunk from
|
| 1686 | * one disk at the same HBA address). For layout detail, see 'man md 4'
|
| 1687 |
|
| 1688 | chunk = bio_start >> geo->chunk_shift;
|
| 1689 | chunk *= geo->near_copies;
|
| 1690 | first_stripe_index = chunk;
|
| 1691 | start_disk_index = sector_div(first_stripe_index, geo->raid_disks);
|
| 1692 | if (geo->far_offset)
|
| 1693 | first_stripe_index *= geo->far_copies;
|
| 1694 | start_disk_offset = (bio_start & geo->chunk_mask) +
|
| 1695 | (first_stripe_index << geo->chunk_shift);
|
| 1696 |
|
| 1697 | chunk = bio_end >> geo->chunk_shift;
|
| 1698 | chunk *= geo->near_copies;
|
| 1699 | last_stripe_index = chunk;
|
| 1700 | end_disk_index = sector_div(last_stripe_index, geo->raid_disks);
|
| 1701 | if (geo->far_offset)
|
| 1702 | last_stripe_index *= geo->far_copies;
|
| 1703 | end_disk_offset = (bio_end & geo->chunk_mask) +
|
| 1704 | (last_stripe_index << geo->chunk_shift);
|
| 1705 |
|
| 1706 | retry_discard:
|
| 1707 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO);
|
| 1708 | r10_bio->mddev = mddev;
|
| 1709 | r10_bio->state = 0;
|
| 1710 | r10_bio->sectors = 0;
|
| 1711 | r10_bio->read_slot = -1;
|
| 1712 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * geo->raid_disks);
|
| 1713 | wait_blocked_dev(mddev, r10_bio);
|
| 1714 |
|
| 1715 | |
| 1716 | * For far layout it needs more than one r10bio to cover all regions.
|
| 1717 | * Inspired by raid10_sync_request, we can use the first r10bio->master_bio
|
| 1718 | * to record the discard bio. Other r10bio->master_bio record the first
|
| 1719 | * r10bio. The first r10bio only release after all other r10bios finish.
|
| 1720 | * The discard bio returns only first r10bio finishes
|
| 1721 |
|
| 1722 | if (first_copy) {
|
| 1723 | md_account_bio(mddev, &bio);
|
| 1724 | r10_bio->master_bio = bio;
|
| 1725 | set_bit(R10BIO_Discard, &r10_bio->state);
|
| 1726 | first_copy = false;
|
| 1727 | first_r10bio = r10_bio;
|
| 1728 | } else
|
| 1729 | r10_bio->master_bio = (struct bio *)first_r10bio;
|
| 1730 |
|
| 1731 | |
| 1732 | * first select target devices under rcu_lock and
|
| 1733 | * inc refcount on their rdev. Record them by setting
|
| 1734 | * bios[x] to bio
|
| 1735 |
|
| 1736 | for (disk = 0; disk < geo->raid_disks; disk++) {
|
| 1737 | struct md_rdev *rdev, *rrdev;
|
| 1738 |
|
| 1739 | rdev = conf->mirrors[disk].rdev;
|
| 1740 | rrdev = conf->mirrors[disk].replacement;
|
| 1741 | r10_bio->devs[disk].bio = NULL;
|
| 1742 | r10_bio->devs[disk].repl_bio = NULL;
|
| 1743 |
|
| 1744 | if (rdev && (test_bit(Faulty, &rdev->flags)))
|
| 1745 | rdev = NULL;
|
| 1746 | if (rrdev && (test_bit(Faulty, &rrdev->flags)))
|
| 1747 | rrdev = NULL;
|
| 1748 | if (!rdev && !rrdev)
|
| 1749 | continue;
|
| 1750 |
|
| 1751 | if (rdev) {
|
| 1752 | r10_bio->devs[disk].bio = bio;
|
| 1753 | atomic_inc(&rdev->nr_pending);
|
| 1754 | }
|
| 1755 | if (rrdev) {
|
| 1756 | r10_bio->devs[disk].repl_bio = bio;
|
| 1757 | atomic_inc(&rrdev->nr_pending);
|
| 1758 | }
|
| 1759 | }
|
| 1760 |
|
| 1761 | atomic_set(&r10_bio->remaining, 1);
|
| 1762 | for (disk = 0; disk < geo->raid_disks; disk++) {
|
| 1763 | sector_t dev_start, dev_end;
|
| 1764 | struct bio *mbio, *rbio = NULL;
|
| 1765 |
|
| 1766 | |
| 1767 | * Now start to calculate the start and end address for each disk.
|
| 1768 | * The space between dev_start and dev_end is the discard region.
|
| 1769 | *
|
| 1770 | * For dev_start, it needs to consider three conditions:
|
| 1771 | * 1st, the disk is before start_disk, you can imagine the disk in
|
| 1772 | * the next stripe. So the dev_start is the start address of next
|
| 1773 | * stripe.
|
| 1774 | * 2st, the disk is after start_disk, it means the disk is at the
|
| 1775 | * same stripe of first disk
|
| 1776 | * 3st, the first disk itself, we can use start_disk_offset directly
|
| 1777 |
|
| 1778 | if (disk < start_disk_index)
|
| 1779 | dev_start = (first_stripe_index + 1) * mddev->chunk_sectors;
|
| 1780 | else if (disk > start_disk_index)
|
| 1781 | dev_start = first_stripe_index * mddev->chunk_sectors;
|
| 1782 | else
|
| 1783 | dev_start = start_disk_offset;
|
| 1784 |
|
| 1785 | if (disk < end_disk_index)
|
| 1786 | dev_end = (last_stripe_index + 1) * mddev->chunk_sectors;
|
| 1787 | else if (disk > end_disk_index)
|
| 1788 | dev_end = last_stripe_index * mddev->chunk_sectors;
|
| 1789 | else
|
| 1790 | dev_end = end_disk_offset;
|
| 1791 |
|
| 1792 | |
| 1793 | * It only handles discard bio which size is >= stripe size, so
|
| 1794 | * dev_end > dev_start all the time.
|
| 1795 | * It doesn't need to use rcu lock to get rdev here. We already
|
| 1796 | * add rdev->nr_pending in the first loop.
|
| 1797 |
|
| 1798 | if (r10_bio->devs[disk].bio) {
|
| 1799 | struct md_rdev *rdev = conf->mirrors[disk].rdev;
|
| 1800 | mbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO,
|
| 1801 | &mddev->bio_set);
|
| 1802 | mbio->bi_end_io = raid10_end_discard_request;
|
| 1803 | mbio->bi_private = r10_bio;
|
| 1804 | r10_bio->devs[disk].bio = mbio;
|
| 1805 | r10_bio->devs[disk].devnum = disk;
|
| 1806 | atomic_inc(&r10_bio->remaining);
|
| 1807 | md_submit_discard_bio(mddev, rdev, mbio,
|
| 1808 | dev_start + choose_data_offset(r10_bio, rdev),
|
| 1809 | dev_end - dev_start);
|
| 1810 | bio_endio(mbio);
|
| 1811 | }
|
| 1812 | if (r10_bio->devs[disk].repl_bio) {
|
| 1813 | struct md_rdev *rrdev = conf->mirrors[disk].replacement;
|
| 1814 | rbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO,
|
| 1815 | &mddev->bio_set);
|
| 1816 | rbio->bi_end_io = raid10_end_discard_request;
|
| 1817 | rbio->bi_private = r10_bio;
|
| 1818 | r10_bio->devs[disk].repl_bio = rbio;
|
| 1819 | r10_bio->devs[disk].devnum = disk;
|
| 1820 | atomic_inc(&r10_bio->remaining);
|
| 1821 | md_submit_discard_bio(mddev, rrdev, rbio,
|
| 1822 | dev_start + choose_data_offset(r10_bio, rrdev),
|
| 1823 | dev_end - dev_start);
|
| 1824 | bio_endio(rbio);
|
| 1825 | }
|
| 1826 | }
|
| 1827 |
|
| 1828 | if (!geo->far_offset && --far_copies) {
|
| 1829 | first_stripe_index += geo->stride >> geo->chunk_shift;
|
| 1830 | start_disk_offset += geo->stride;
|
| 1831 | last_stripe_index += geo->stride >> geo->chunk_shift;
|
| 1832 | end_disk_offset += geo->stride;
|
| 1833 | atomic_inc(&first_r10bio->remaining);
|
| 1834 | raid_end_discard_bio(r10_bio);
|
| 1835 | wait_barrier(conf);
|
| 1836 | goto retry_discard;
|
| 1837 | }
|
| 1838 |
|
| 1839 | raid_end_discard_bio(r10_bio);
|
| 1840 |
|
| 1841 | return 0;
|
| 1842 | out:
|
| 1843 | allow_barrier(conf);
|
| 1844 | return -EAGAIN;
|
| 1845 | }
|
| 1846 |
|
| 1847 | static bool raid10_make_request(struct mddev *mddev, struct bio *bio)
|
| 1848 | {
|
| 1849 | struct r10conf *conf = mddev->private;
|
| 1850 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask);
|
| 1851 | const int rw = bio_data_dir(bio);
|
| 1852 | int chunk_sects = chunk_mask + 1;
|
| 1853 | int sectors = bio_sectors(bio);
|
| 1854 |
|
| 1855 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)
|
| 1856 | && md_flush_request(mddev, bio))
|
| 1857 | return true;
|
| 1858 |
|
| 1859 | md_write_start(mddev, bio);
|
| 1860 |
|
| 1861 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
|
| 1862 | if (!raid10_handle_discard(mddev, bio))
|
| 1863 | return true;
|
| 1864 |
|
| 1865 | |
| 1866 | * If this request crosses a chunk boundary, we need to split
|
| 1867 | * it.
|
| 1868 |
|
| 1869 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) +
|
| 1870 | sectors > chunk_sects
|
| 1871 | && (conf->geo.near_copies < conf->geo.raid_disks
|
| 1872 | || conf->prev.near_copies <
|
| 1873 | conf->prev.raid_disks)))
|
| 1874 | sectors = chunk_sects -
|
| 1875 | (bio->bi_iter.bi_sector &
|
| 1876 | (chunk_sects - 1));
|
| 1877 |
|
| 1878 | bio = mddev_bio_split_at_reshape_offset(mddev, bio, §ors,
|
| 1879 | &conf->bio_split);
|
| 1880 | if (!bio) {
|
| 1881 | if (rw == WRITE)
|
| 1882 | md_write_end(mddev);
|
| 1883 | return true;
|
| 1884 | }
|
| 1885 |
|
| 1886 | if (!__make_request(mddev, bio, sectors))
|
| 1887 | md_write_end(mddev);
|
| 1888 |
|
| 1889 |
|
| 1890 | wake_up_barrier(conf);
|
| 1891 | return true;
|
| 1892 | }
|
| 1893 |
|
| 1894 | static void raid10_status(struct seq_file *seq, struct mddev *mddev)
|
| 1895 | {
|
| 1896 | struct r10conf *conf = mddev->private;
|
| 1897 | int i;
|
| 1898 |
|
| 1899 | lockdep_assert_held(&mddev->lock);
|
| 1900 |
|
| 1901 | if (conf->geo.near_copies < conf->geo.raid_disks)
|
| 1902 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2);
|
| 1903 | if (conf->geo.near_copies > 1)
|
| 1904 | seq_printf(seq, " %d near-copies", conf->geo.near_copies);
|
| 1905 | if (conf->geo.far_copies > 1) {
|
| 1906 | if (conf->geo.far_offset)
|
| 1907 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies);
|
| 1908 | else
|
| 1909 | seq_printf(seq, " %d far-copies", conf->geo.far_copies);
|
| 1910 | if (conf->geo.far_set_size != conf->geo.raid_disks)
|
| 1911 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size);
|
| 1912 | }
|
| 1913 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks,
|
| 1914 | conf->geo.raid_disks - mddev->degraded);
|
| 1915 | for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 1916 | struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev);
|
| 1917 |
|
| 1918 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
|
| 1919 | }
|
| 1920 | seq_printf(seq, "]");
|
| 1921 | }
|
| 1922 |
|
| 1923 | |
| 1924 | * every block to appear on atleast one.
|
| 1925 | * Don't consider the device numbered 'ignore'
|
| 1926 | * as we might be about to remove it.
|
| 1927 |
|
| 1928 | static int _enough(struct r10conf *conf, int previous, int ignore)
|
| 1929 | {
|
| 1930 | int first = 0;
|
| 1931 | int has_enough = 0;
|
| 1932 | int disks, ncopies;
|
| 1933 | if (previous) {
|
| 1934 | disks = conf->prev.raid_disks;
|
| 1935 | ncopies = conf->prev.near_copies;
|
| 1936 | } else {
|
| 1937 | disks = conf->geo.raid_disks;
|
| 1938 | ncopies = conf->geo.near_copies;
|
| 1939 | }
|
| 1940 |
|
| 1941 | do {
|
| 1942 | int n = conf->copies;
|
| 1943 | int cnt = 0;
|
| 1944 | int this = first;
|
| 1945 | while (n--) {
|
| 1946 | struct md_rdev *rdev;
|
| 1947 | if (this != ignore &&
|
| 1948 | (rdev = conf->mirrors[this].rdev) &&
|
| 1949 | test_bit(In_sync, &rdev->flags))
|
| 1950 | cnt++;
|
| 1951 | this = (this+1) % disks;
|
| 1952 | }
|
| 1953 | if (cnt == 0)
|
| 1954 | goto out;
|
| 1955 | first = (first + ncopies) % disks;
|
| 1956 | } while (first != 0);
|
| 1957 | has_enough = 1;
|
| 1958 | out:
|
| 1959 | return has_enough;
|
| 1960 | }
|
| 1961 |
|
| 1962 | static int enough(struct r10conf *conf, int ignore)
|
| 1963 | {
|
| 1964 | |
| 1965 | * be stable.
|
| 1966 | * This is ensured if ->reconfig_mutex or ->device_lock
|
| 1967 | * is held.
|
| 1968 |
|
| 1969 | return _enough(conf, 0, ignore) &&
|
| 1970 | _enough(conf, 1, ignore);
|
| 1971 | }
|
| 1972 |
|
| 1973 | |
| 1974 | * raid10_error() - RAID10 error handler.
|
| 1975 | * @mddev: affected md device.
|
| 1976 | * @rdev: member device to fail.
|
| 1977 | *
|
| 1978 | * The routine acknowledges &rdev failure and determines new @mddev state.
|
| 1979 | * If it failed, then:
|
| 1980 | * - &MD_BROKEN flag is set in &mddev->flags.
|
| 1981 | * Otherwise, it must be degraded:
|
| 1982 | * - recovery is interrupted.
|
| 1983 | * - &mddev->degraded is bumped.
|
| 1984 | *
|
| 1985 | * @rdev is marked as &Faulty excluding case when array is failed and
|
| 1986 | * MD_FAILLAST_DEV is not set.
|
| 1987 |
|
| 1988 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev)
|
| 1989 | {
|
| 1990 | struct r10conf *conf = mddev->private;
|
| 1991 | unsigned long flags;
|
| 1992 |
|
| 1993 | spin_lock_irqsave(&conf->device_lock, flags);
|
| 1994 |
|
| 1995 | if (test_bit(In_sync, &rdev->flags) && !enough(conf, rdev->raid_disk)) {
|
| 1996 | set_bit(MD_BROKEN, &mddev->flags);
|
| 1997 |
|
| 1998 | if (!test_bit(MD_FAILLAST_DEV, &mddev->flags)) {
|
| 1999 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 2000 | return;
|
| 2001 | }
|
| 2002 | }
|
| 2003 | if (test_and_clear_bit(In_sync, &rdev->flags))
|
| 2004 | mddev->degraded++;
|
| 2005 |
|
| 2006 | set_bit(MD_RECOVERY_INTR, &mddev->recovery);
|
| 2007 | set_bit(Blocked, &rdev->flags);
|
| 2008 | set_bit(Faulty, &rdev->flags);
|
| 2009 | set_mask_bits(&mddev->sb_flags, 0,
|
| 2010 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
|
| 2011 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 2012 | pr_crit("md/raid10:%s: Disk failure on %pg, disabling device.\n"
|
| 2013 | "md/raid10:%s: Operation continuing on %d devices.\n",
|
| 2014 | mdname(mddev), rdev->bdev,
|
| 2015 | mdname(mddev), conf->geo.raid_disks - mddev->degraded);
|
| 2016 | }
|
| 2017 |
|
| 2018 | static void print_conf(struct r10conf *conf)
|
| 2019 | {
|
| 2020 | int i;
|
| 2021 | struct md_rdev *rdev;
|
| 2022 |
|
| 2023 | pr_debug("RAID10 conf printout:\n");
|
| 2024 | if (!conf) {
|
| 2025 | pr_debug("(!conf)\n");
|
| 2026 | return;
|
| 2027 | }
|
| 2028 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded,
|
| 2029 | conf->geo.raid_disks);
|
| 2030 |
|
| 2031 | lockdep_assert_held(&conf->mddev->reconfig_mutex);
|
| 2032 | for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 2033 | rdev = conf->mirrors[i].rdev;
|
| 2034 | if (rdev)
|
| 2035 | pr_debug(" disk %d, wo:%d, o:%d, dev:%pg\n",
|
| 2036 | i, !test_bit(In_sync, &rdev->flags),
|
| 2037 | !test_bit(Faulty, &rdev->flags),
|
| 2038 | rdev->bdev);
|
| 2039 | }
|
| 2040 | }
|
| 2041 |
|
| 2042 | static void close_sync(struct r10conf *conf)
|
| 2043 | {
|
| 2044 | wait_barrier(conf);
|
| 2045 | allow_barrier(conf);
|
| 2046 |
|
| 2047 | mempool_exit(&conf->r10buf_pool);
|
| 2048 | }
|
| 2049 |
|
| 2050 | static int raid10_spare_active(struct mddev *mddev)
|
| 2051 | {
|
| 2052 | int i;
|
| 2053 | struct r10conf *conf = mddev->private;
|
| 2054 | struct raid10_info *tmp;
|
| 2055 | int count = 0;
|
| 2056 | unsigned long flags;
|
| 2057 |
|
| 2058 | |
| 2059 | * Find all non-in_sync disks within the RAID10 configuration
|
| 2060 | * and mark them in_sync
|
| 2061 |
|
| 2062 | for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 2063 | tmp = conf->mirrors + i;
|
| 2064 | if (tmp->replacement
|
| 2065 | && tmp->replacement->recovery_offset == MaxSector
|
| 2066 | && !test_bit(Faulty, &tmp->replacement->flags)
|
| 2067 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) {
|
| 2068 |
|
| 2069 | if (!tmp->rdev
|
| 2070 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags))
|
| 2071 | count++;
|
| 2072 | if (tmp->rdev) {
|
| 2073 | |
| 2074 | * but we need to be sure it gets removed
|
| 2075 | * and never re-added.
|
| 2076 |
|
| 2077 | set_bit(Faulty, &tmp->rdev->flags);
|
| 2078 | sysfs_notify_dirent_safe(
|
| 2079 | tmp->rdev->sysfs_state);
|
| 2080 | }
|
| 2081 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state);
|
| 2082 | } else if (tmp->rdev
|
| 2083 | && tmp->rdev->recovery_offset == MaxSector
|
| 2084 | && !test_bit(Faulty, &tmp->rdev->flags)
|
| 2085 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) {
|
| 2086 | count++;
|
| 2087 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state);
|
| 2088 | }
|
| 2089 | }
|
| 2090 | spin_lock_irqsave(&conf->device_lock, flags);
|
| 2091 | mddev->degraded -= count;
|
| 2092 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 2093 |
|
| 2094 | print_conf(conf);
|
| 2095 | return count;
|
| 2096 | }
|
| 2097 |
|
| 2098 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev)
|
| 2099 | {
|
| 2100 | struct r10conf *conf = mddev->private;
|
| 2101 | int err = -EEXIST;
|
| 2102 | int mirror, repl_slot = -1;
|
| 2103 | int first = 0;
|
| 2104 | int last = conf->geo.raid_disks - 1;
|
| 2105 | struct raid10_info *p;
|
| 2106 |
|
| 2107 | if (mddev->resync_offset < MaxSector)
|
| 2108 | |
| 2109 | * very different from resync
|
| 2110 |
|
| 2111 | return -EBUSY;
|
| 2112 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1))
|
| 2113 | return -EINVAL;
|
| 2114 |
|
| 2115 | if (rdev->raid_disk >= 0)
|
| 2116 | first = last = rdev->raid_disk;
|
| 2117 |
|
| 2118 | if (rdev->saved_raid_disk >= first &&
|
| 2119 | rdev->saved_raid_disk < conf->geo.raid_disks &&
|
| 2120 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
|
| 2121 | mirror = rdev->saved_raid_disk;
|
| 2122 | else
|
| 2123 | mirror = first;
|
| 2124 | for ( ; mirror <= last ; mirror++) {
|
| 2125 | p = &conf->mirrors[mirror];
|
| 2126 | if (p->rdev) {
|
| 2127 | if (test_bit(WantReplacement, &p->rdev->flags) &&
|
| 2128 | p->replacement == NULL && repl_slot < 0)
|
| 2129 | repl_slot = mirror;
|
| 2130 | continue;
|
| 2131 | }
|
| 2132 |
|
| 2133 | err = mddev_stack_new_rdev(mddev, rdev);
|
| 2134 | if (err)
|
| 2135 | return err;
|
| 2136 | p->head_position = 0;
|
| 2137 | rdev->raid_disk = mirror;
|
| 2138 | err = 0;
|
| 2139 | if (rdev->saved_raid_disk != mirror)
|
| 2140 | conf->fullsync = 1;
|
| 2141 | WRITE_ONCE(p->rdev, rdev);
|
| 2142 | break;
|
| 2143 | }
|
| 2144 |
|
| 2145 | if (err && repl_slot >= 0) {
|
| 2146 | p = &conf->mirrors[repl_slot];
|
| 2147 | clear_bit(In_sync, &rdev->flags);
|
| 2148 | set_bit(Replacement, &rdev->flags);
|
| 2149 | rdev->raid_disk = repl_slot;
|
| 2150 | err = mddev_stack_new_rdev(mddev, rdev);
|
| 2151 | if (err)
|
| 2152 | return err;
|
| 2153 | conf->fullsync = 1;
|
| 2154 | WRITE_ONCE(p->replacement, rdev);
|
| 2155 | }
|
| 2156 |
|
| 2157 | print_conf(conf);
|
| 2158 | return err;
|
| 2159 | }
|
| 2160 |
|
| 2161 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
|
| 2162 | {
|
| 2163 | struct r10conf *conf = mddev->private;
|
| 2164 | int err = 0;
|
| 2165 | int number = rdev->raid_disk;
|
| 2166 | struct md_rdev **rdevp;
|
| 2167 | struct raid10_info *p;
|
| 2168 |
|
| 2169 | print_conf(conf);
|
| 2170 | if (unlikely(number >= mddev->raid_disks))
|
| 2171 | return 0;
|
| 2172 | p = conf->mirrors + number;
|
| 2173 | if (rdev == p->rdev)
|
| 2174 | rdevp = &p->rdev;
|
| 2175 | else if (rdev == p->replacement)
|
| 2176 | rdevp = &p->replacement;
|
| 2177 | else
|
| 2178 | return 0;
|
| 2179 |
|
| 2180 | if (test_bit(In_sync, &rdev->flags) ||
|
| 2181 | atomic_read(&rdev->nr_pending)) {
|
| 2182 | err = -EBUSY;
|
| 2183 | goto abort;
|
| 2184 | }
|
| 2185 | |
| 2186 | * is not possible.
|
| 2187 |
|
| 2188 | if (!test_bit(Faulty, &rdev->flags) &&
|
| 2189 | (!p->replacement || p->replacement == rdev) &&
|
| 2190 | number < conf->geo.raid_disks &&
|
| 2191 | enough(conf, -1)) {
|
| 2192 | err = -EBUSY;
|
| 2193 | goto abort;
|
| 2194 | }
|
| 2195 | WRITE_ONCE(*rdevp, NULL);
|
| 2196 | if (p->replacement) {
|
| 2197 |
|
| 2198 | WRITE_ONCE(p->rdev, p->replacement);
|
| 2199 | clear_bit(Replacement, &p->replacement->flags);
|
| 2200 | WRITE_ONCE(p->replacement, NULL);
|
| 2201 | }
|
| 2202 |
|
| 2203 | clear_bit(WantReplacement, &rdev->flags);
|
| 2204 | err = md_integrity_register(mddev);
|
| 2205 |
|
| 2206 | abort:
|
| 2207 |
|
| 2208 | print_conf(conf);
|
| 2209 | return err;
|
| 2210 | }
|
| 2211 |
|
| 2212 | static void __end_sync_read(struct r10bio *r10_bio, struct bio *bio, int d)
|
| 2213 | {
|
| 2214 | struct r10conf *conf = r10_bio->mddev->private;
|
| 2215 |
|
| 2216 | if (!bio->bi_status)
|
| 2217 | set_bit(R10BIO_Uptodate, &r10_bio->state);
|
| 2218 | else
|
| 2219 | |
| 2220 | * R10BIO_Uptodate and record any errors etc
|
| 2221 |
|
| 2222 | atomic_add(r10_bio->sectors,
|
| 2223 | &conf->mirrors[d].rdev->corrected_errors);
|
| 2224 |
|
| 2225 | |
| 2226 | * for resync, only after all reads
|
| 2227 |
|
| 2228 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev);
|
| 2229 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) ||
|
| 2230 | atomic_dec_and_test(&r10_bio->remaining)) {
|
| 2231 | |
| 2232 | * do the comparison in process context in raid10d
|
| 2233 |
|
| 2234 | reschedule_retry(r10_bio);
|
| 2235 | }
|
| 2236 | }
|
| 2237 |
|
| 2238 | static void end_sync_read(struct bio *bio)
|
| 2239 | {
|
| 2240 | struct r10bio *r10_bio = get_resync_r10bio(bio);
|
| 2241 | struct r10conf *conf = r10_bio->mddev->private;
|
| 2242 | int d = find_bio_disk(conf, r10_bio, bio, NULL, NULL);
|
| 2243 |
|
| 2244 | __end_sync_read(r10_bio, bio, d);
|
| 2245 | }
|
| 2246 |
|
| 2247 | static void end_reshape_read(struct bio *bio)
|
| 2248 | {
|
| 2249 |
|
| 2250 | struct r10bio *r10_bio = bio->bi_private;
|
| 2251 |
|
| 2252 | __end_sync_read(r10_bio, bio, r10_bio->read_slot);
|
| 2253 | }
|
| 2254 |
|
| 2255 | static void end_sync_request(struct r10bio *r10_bio)
|
| 2256 | {
|
| 2257 | struct mddev *mddev = r10_bio->mddev;
|
| 2258 |
|
| 2259 | while (atomic_dec_and_test(&r10_bio->remaining)) {
|
| 2260 | if (r10_bio->master_bio == NULL) {
|
| 2261 |
|
| 2262 | sector_t s = r10_bio->sectors;
|
| 2263 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
|
| 2264 | test_bit(R10BIO_WriteError, &r10_bio->state))
|
| 2265 | reschedule_retry(r10_bio);
|
| 2266 | else
|
| 2267 | put_buf(r10_bio);
|
| 2268 | md_done_sync(mddev, s);
|
| 2269 | break;
|
| 2270 | } else {
|
| 2271 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio;
|
| 2272 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
|
| 2273 | test_bit(R10BIO_WriteError, &r10_bio->state))
|
| 2274 | reschedule_retry(r10_bio);
|
| 2275 | else
|
| 2276 | put_buf(r10_bio);
|
| 2277 | r10_bio = r10_bio2;
|
| 2278 | }
|
| 2279 | }
|
| 2280 | }
|
| 2281 |
|
| 2282 | static void end_sync_write(struct bio *bio)
|
| 2283 | {
|
| 2284 | struct r10bio *r10_bio = get_resync_r10bio(bio);
|
| 2285 | struct mddev *mddev = r10_bio->mddev;
|
| 2286 | struct r10conf *conf = mddev->private;
|
| 2287 | int d;
|
| 2288 | int slot;
|
| 2289 | int repl;
|
| 2290 | struct md_rdev *rdev = NULL;
|
| 2291 |
|
| 2292 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
|
| 2293 | if (repl)
|
| 2294 | rdev = conf->mirrors[d].replacement;
|
| 2295 | else
|
| 2296 | rdev = conf->mirrors[d].rdev;
|
| 2297 |
|
| 2298 | if (bio->bi_status) {
|
| 2299 | if (repl)
|
| 2300 | md_error(mddev, rdev);
|
| 2301 | else {
|
| 2302 | set_bit(WriteErrorSeen, &rdev->flags);
|
| 2303 | if (!test_and_set_bit(WantReplacement, &rdev->flags))
|
| 2304 | set_bit(MD_RECOVERY_NEEDED,
|
| 2305 | &rdev->mddev->recovery);
|
| 2306 | set_bit(R10BIO_WriteError, &r10_bio->state);
|
| 2307 | }
|
| 2308 | } else if (rdev_has_badblock(rdev, r10_bio->devs[slot].addr,
|
| 2309 | r10_bio->sectors)) {
|
| 2310 | set_bit(R10BIO_MadeGood, &r10_bio->state);
|
| 2311 | }
|
| 2312 |
|
| 2313 | rdev_dec_pending(rdev, mddev);
|
| 2314 |
|
| 2315 | end_sync_request(r10_bio);
|
| 2316 | }
|
| 2317 |
|
| 2318 | |
| 2319 | * Note: sync and recover and handled very differently for raid10
|
| 2320 | * This code is for resync.
|
| 2321 | * For resync, we read through virtual addresses and read all blocks.
|
| 2322 | * If there is any error, we schedule a write. The lowest numbered
|
| 2323 | * drive is authoritative.
|
| 2324 | * However requests come for physical address, so we need to map.
|
| 2325 | * For every physical address there are raid_disks/copies virtual addresses,
|
| 2326 | * which is always are least one, but is not necessarly an integer.
|
| 2327 | * This means that a physical address can span multiple chunks, so we may
|
| 2328 | * have to submit multiple io requests for a single sync request.
|
| 2329 |
|
| 2330 | |
| 2331 | * We check if all blocks are in-sync and only write to blocks that
|
| 2332 | * aren't in sync
|
| 2333 |
|
| 2334 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio)
|
| 2335 | {
|
| 2336 | struct r10conf *conf = mddev->private;
|
| 2337 | int i, first;
|
| 2338 | struct bio *tbio, *fbio;
|
| 2339 | int vcnt;
|
| 2340 | struct page **tpages, **fpages;
|
| 2341 |
|
| 2342 | atomic_set(&r10_bio->remaining, 1);
|
| 2343 |
|
| 2344 |
|
| 2345 | for (i=0; i<conf->copies; i++)
|
| 2346 | if (!r10_bio->devs[i].bio->bi_status)
|
| 2347 | break;
|
| 2348 |
|
| 2349 | if (i == conf->copies)
|
| 2350 | goto done;
|
| 2351 |
|
| 2352 | first = i;
|
| 2353 | fbio = r10_bio->devs[i].bio;
|
| 2354 | fbio->bi_iter.bi_size = r10_bio->sectors << 9;
|
| 2355 | fbio->bi_iter.bi_idx = 0;
|
| 2356 | fpages = get_resync_pages(fbio)->pages;
|
| 2357 |
|
| 2358 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9);
|
| 2359 |
|
| 2360 | for (i=0 ; i < conf->copies ; i++) {
|
| 2361 | int j, d;
|
| 2362 | struct md_rdev *rdev;
|
| 2363 | struct resync_pages *rp;
|
| 2364 |
|
| 2365 | tbio = r10_bio->devs[i].bio;
|
| 2366 |
|
| 2367 | if (tbio->bi_end_io != end_sync_read)
|
| 2368 | continue;
|
| 2369 | if (i == first)
|
| 2370 | continue;
|
| 2371 |
|
| 2372 | tpages = get_resync_pages(tbio)->pages;
|
| 2373 | d = r10_bio->devs[i].devnum;
|
| 2374 | rdev = conf->mirrors[d].rdev;
|
| 2375 | if (!r10_bio->devs[i].bio->bi_status) {
|
| 2376 | |
| 2377 | * both 'first' and 'i', so we just compare them.
|
| 2378 | * All vec entries are PAGE_SIZE;
|
| 2379 |
|
| 2380 | int sectors = r10_bio->sectors;
|
| 2381 | for (j = 0; j < vcnt; j++) {
|
| 2382 | int len = PAGE_SIZE;
|
| 2383 | if (sectors < (len / 512))
|
| 2384 | len = sectors * 512;
|
| 2385 | if (memcmp(page_address(fpages[j]),
|
| 2386 | page_address(tpages[j]),
|
| 2387 | len))
|
| 2388 | break;
|
| 2389 | sectors -= len/512;
|
| 2390 | }
|
| 2391 | if (j == vcnt)
|
| 2392 | continue;
|
| 2393 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches);
|
| 2394 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
|
| 2395 |
|
| 2396 | continue;
|
| 2397 | } else if (test_bit(FailFast, &rdev->flags)) {
|
| 2398 |
|
| 2399 | md_error(rdev->mddev, rdev);
|
| 2400 | continue;
|
| 2401 | }
|
| 2402 | |
| 2403 | * inconsistency or to correct an unreadable block.
|
| 2404 | * First we need to fixup bv_offset, bv_len and
|
| 2405 | * bi_vecs, as the read request might have corrupted these
|
| 2406 |
|
| 2407 | rp = get_resync_pages(tbio);
|
| 2408 | bio_reset(tbio, conf->mirrors[d].rdev->bdev, REQ_OP_WRITE);
|
| 2409 |
|
| 2410 | md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size);
|
| 2411 |
|
| 2412 | rp->raid_bio = r10_bio;
|
| 2413 | tbio->bi_private = rp;
|
| 2414 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr;
|
| 2415 | tbio->bi_end_io = end_sync_write;
|
| 2416 |
|
| 2417 | bio_copy_data(tbio, fbio);
|
| 2418 |
|
| 2419 | atomic_inc(&conf->mirrors[d].rdev->nr_pending);
|
| 2420 | atomic_inc(&r10_bio->remaining);
|
| 2421 |
|
| 2422 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags))
|
| 2423 | tbio->bi_opf |= MD_FAILFAST;
|
| 2424 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset;
|
| 2425 | submit_bio_noacct(tbio);
|
| 2426 | }
|
| 2427 |
|
| 2428 | |
| 2429 | * that are active
|
| 2430 |
|
| 2431 | for (i = 0; i < conf->copies; i++) {
|
| 2432 | tbio = r10_bio->devs[i].repl_bio;
|
| 2433 | if (!tbio || !tbio->bi_end_io)
|
| 2434 | continue;
|
| 2435 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write
|
| 2436 | && r10_bio->devs[i].bio != fbio)
|
| 2437 | bio_copy_data(tbio, fbio);
|
| 2438 | atomic_inc(&r10_bio->remaining);
|
| 2439 | submit_bio_noacct(tbio);
|
| 2440 | }
|
| 2441 |
|
| 2442 | done:
|
| 2443 | if (atomic_dec_and_test(&r10_bio->remaining)) {
|
| 2444 | md_done_sync(mddev, r10_bio->sectors);
|
| 2445 | put_buf(r10_bio);
|
| 2446 | }
|
| 2447 | }
|
| 2448 |
|
| 2449 | |
| 2450 | * Now for the recovery code.
|
| 2451 | * Recovery happens across physical sectors.
|
| 2452 | * We recover all non-is_sync drives by finding the virtual address of
|
| 2453 | * each, and then choose a working drive that also has that virt address.
|
| 2454 | * There is a separate r10_bio for each non-in_sync drive.
|
| 2455 | * Only the first two slots are in use. The first for reading,
|
| 2456 | * The second for writing.
|
| 2457 | *
|
| 2458 |
|
| 2459 | static void fix_recovery_read_error(struct r10bio *r10_bio)
|
| 2460 | {
|
| 2461 | |
| 2462 | * We repeat the read in smaller page-sized sections.
|
| 2463 | * If a read succeeds, write it to the new device or record
|
| 2464 | * a bad block if we cannot.
|
| 2465 | * If a read fails, record a bad block on both old and
|
| 2466 | * new devices.
|
| 2467 |
|
| 2468 | struct mddev *mddev = r10_bio->mddev;
|
| 2469 | struct r10conf *conf = mddev->private;
|
| 2470 | struct bio *bio = r10_bio->devs[0].bio;
|
| 2471 | sector_t sect = 0;
|
| 2472 | int sectors = r10_bio->sectors;
|
| 2473 | int idx = 0;
|
| 2474 | int dr = r10_bio->devs[0].devnum;
|
| 2475 | int dw = r10_bio->devs[1].devnum;
|
| 2476 | struct page **pages = get_resync_pages(bio)->pages;
|
| 2477 |
|
| 2478 | while (sectors) {
|
| 2479 | int s = sectors;
|
| 2480 | struct md_rdev *rdev;
|
| 2481 | sector_t addr;
|
| 2482 | int ok;
|
| 2483 |
|
| 2484 | if (s > (PAGE_SIZE>>9))
|
| 2485 | s = PAGE_SIZE >> 9;
|
| 2486 |
|
| 2487 | rdev = conf->mirrors[dr].rdev;
|
| 2488 | addr = r10_bio->devs[0].addr + sect;
|
| 2489 | ok = sync_page_io(rdev,
|
| 2490 | addr,
|
| 2491 | s << 9,
|
| 2492 | pages[idx],
|
| 2493 | REQ_OP_READ, false);
|
| 2494 | if (ok) {
|
| 2495 | rdev = conf->mirrors[dw].rdev;
|
| 2496 | addr = r10_bio->devs[1].addr + sect;
|
| 2497 | ok = sync_page_io(rdev,
|
| 2498 | addr,
|
| 2499 | s << 9,
|
| 2500 | pages[idx],
|
| 2501 | REQ_OP_WRITE, false);
|
| 2502 | if (!ok) {
|
| 2503 | set_bit(WriteErrorSeen, &rdev->flags);
|
| 2504 | if (!test_and_set_bit(WantReplacement,
|
| 2505 | &rdev->flags))
|
| 2506 | set_bit(MD_RECOVERY_NEEDED,
|
| 2507 | &rdev->mddev->recovery);
|
| 2508 | }
|
| 2509 | }
|
| 2510 | if (!ok) {
|
| 2511 | |
| 2512 | * it really is bad so there is no loss in not
|
| 2513 | * recording it yet
|
| 2514 |
|
| 2515 | rdev_set_badblocks(rdev, addr, s, 0);
|
| 2516 |
|
| 2517 | if (rdev != conf->mirrors[dw].rdev) {
|
| 2518 |
|
| 2519 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev;
|
| 2520 | addr = r10_bio->devs[1].addr + sect;
|
| 2521 | ok = rdev_set_badblocks(rdev2, addr, s, 0);
|
| 2522 | if (!ok) {
|
| 2523 |
|
| 2524 | pr_notice("md/raid10:%s: recovery aborted due to read error\n",
|
| 2525 | mdname(mddev));
|
| 2526 |
|
| 2527 | set_bit(MD_RECOVERY_INTR,
|
| 2528 | &mddev->recovery);
|
| 2529 | break;
|
| 2530 | }
|
| 2531 | }
|
| 2532 | }
|
| 2533 |
|
| 2534 | sectors -= s;
|
| 2535 | sect += s;
|
| 2536 | idx++;
|
| 2537 | }
|
| 2538 | }
|
| 2539 |
|
| 2540 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio)
|
| 2541 | {
|
| 2542 | struct r10conf *conf = mddev->private;
|
| 2543 | int d;
|
| 2544 | struct bio *wbio = r10_bio->devs[1].bio;
|
| 2545 | struct bio *wbio2 = r10_bio->devs[1].repl_bio;
|
| 2546 |
|
| 2547 | |
| 2548 | * submit_bio_noacct as if the former is NULL,
|
| 2549 | * the latter is free to free wbio2.
|
| 2550 |
|
| 2551 | if (wbio2 && !wbio2->bi_end_io)
|
| 2552 | wbio2 = NULL;
|
| 2553 |
|
| 2554 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) {
|
| 2555 | fix_recovery_read_error(r10_bio);
|
| 2556 | if (wbio->bi_end_io)
|
| 2557 | end_sync_request(r10_bio);
|
| 2558 | if (wbio2)
|
| 2559 | end_sync_request(r10_bio);
|
| 2560 | return;
|
| 2561 | }
|
| 2562 |
|
| 2563 | |
| 2564 | * share the pages with the first bio
|
| 2565 | * and submit the write request
|
| 2566 |
|
| 2567 | d = r10_bio->devs[1].devnum;
|
| 2568 | if (wbio->bi_end_io) {
|
| 2569 | atomic_inc(&conf->mirrors[d].rdev->nr_pending);
|
| 2570 | submit_bio_noacct(wbio);
|
| 2571 | }
|
| 2572 | if (wbio2) {
|
| 2573 | atomic_inc(&conf->mirrors[d].replacement->nr_pending);
|
| 2574 | submit_bio_noacct(wbio2);
|
| 2575 | }
|
| 2576 | }
|
| 2577 |
|
| 2578 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector,
|
| 2579 | int sectors, struct page *page, enum req_op op)
|
| 2580 | {
|
| 2581 | if (rdev_has_badblock(rdev, sector, sectors) &&
|
| 2582 | (op == REQ_OP_READ || test_bit(WriteErrorSeen, &rdev->flags)))
|
| 2583 | return -1;
|
| 2584 | if (sync_page_io(rdev, sector, sectors << 9, page, op, false))
|
| 2585 |
|
| 2586 | return 1;
|
| 2587 | if (op == REQ_OP_WRITE) {
|
| 2588 | set_bit(WriteErrorSeen, &rdev->flags);
|
| 2589 | if (!test_and_set_bit(WantReplacement, &rdev->flags))
|
| 2590 | set_bit(MD_RECOVERY_NEEDED,
|
| 2591 | &rdev->mddev->recovery);
|
| 2592 | }
|
| 2593 |
|
| 2594 | rdev_set_badblocks(rdev, sector, sectors, 0);
|
| 2595 | return 0;
|
| 2596 | }
|
| 2597 |
|
| 2598 | |
| 2599 | * This is a kernel thread which:
|
| 2600 | *
|
| 2601 | * 1. Retries failed read operations on working mirrors.
|
| 2602 | * 2. Updates the raid superblock when problems encounter.
|
| 2603 | * 3. Performs writes following reads for array synchronising.
|
| 2604 |
|
| 2605 |
|
| 2606 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio)
|
| 2607 | {
|
| 2608 | int sect = 0;
|
| 2609 | int sectors = r10_bio->sectors, slot = r10_bio->read_slot;
|
| 2610 | struct md_rdev *rdev;
|
| 2611 | int d = r10_bio->devs[slot].devnum;
|
| 2612 |
|
| 2613 | |
| 2614 | * have been cleared recently.
|
| 2615 |
|
| 2616 | rdev = conf->mirrors[d].rdev;
|
| 2617 |
|
| 2618 | if (test_bit(Faulty, &rdev->flags))
|
| 2619 | |
| 2620 | more fix_read_error() attempts
|
| 2621 | return;
|
| 2622 |
|
| 2623 | if (exceed_read_errors(mddev, rdev)) {
|
| 2624 | r10_bio->devs[slot].bio = IO_BLOCKED;
|
| 2625 | return;
|
| 2626 | }
|
| 2627 |
|
| 2628 | while(sectors) {
|
| 2629 | int s = sectors;
|
| 2630 | int sl = slot;
|
| 2631 | int success = 0;
|
| 2632 | int start;
|
| 2633 |
|
| 2634 | if (s > (PAGE_SIZE>>9))
|
| 2635 | s = PAGE_SIZE >> 9;
|
| 2636 |
|
| 2637 | do {
|
| 2638 | d = r10_bio->devs[sl].devnum;
|
| 2639 | rdev = conf->mirrors[d].rdev;
|
| 2640 | if (rdev &&
|
| 2641 | test_bit(In_sync, &rdev->flags) &&
|
| 2642 | !test_bit(Faulty, &rdev->flags) &&
|
| 2643 | rdev_has_badblock(rdev,
|
| 2644 | r10_bio->devs[sl].addr + sect,
|
| 2645 | s) == 0) {
|
| 2646 | atomic_inc(&rdev->nr_pending);
|
| 2647 | success = sync_page_io(rdev,
|
| 2648 | r10_bio->devs[sl].addr +
|
| 2649 | sect,
|
| 2650 | s<<9,
|
| 2651 | conf->tmppage,
|
| 2652 | REQ_OP_READ, false);
|
| 2653 | rdev_dec_pending(rdev, mddev);
|
| 2654 | if (success)
|
| 2655 | break;
|
| 2656 | }
|
| 2657 | sl++;
|
| 2658 | if (sl == conf->copies)
|
| 2659 | sl = 0;
|
| 2660 | } while (sl != slot);
|
| 2661 |
|
| 2662 | if (!success) {
|
| 2663 | |
| 2664 | * as bad on the first device to discourage future
|
| 2665 | * reads.
|
| 2666 |
|
| 2667 | int dn = r10_bio->devs[slot].devnum;
|
| 2668 | rdev = conf->mirrors[dn].rdev;
|
| 2669 |
|
| 2670 | if (!rdev_set_badblocks(
|
| 2671 | rdev,
|
| 2672 | r10_bio->devs[slot].addr
|
| 2673 | + sect,
|
| 2674 | s, 0)) {
|
| 2675 | r10_bio->devs[slot].bio
|
| 2676 | = IO_BLOCKED;
|
| 2677 | }
|
| 2678 | break;
|
| 2679 | }
|
| 2680 |
|
| 2681 | start = sl;
|
| 2682 |
|
| 2683 | while (sl != slot) {
|
| 2684 | if (sl==0)
|
| 2685 | sl = conf->copies;
|
| 2686 | sl--;
|
| 2687 | d = r10_bio->devs[sl].devnum;
|
| 2688 | rdev = conf->mirrors[d].rdev;
|
| 2689 | if (!rdev ||
|
| 2690 | test_bit(Faulty, &rdev->flags) ||
|
| 2691 | !test_bit(In_sync, &rdev->flags))
|
| 2692 | continue;
|
| 2693 |
|
| 2694 | atomic_inc(&rdev->nr_pending);
|
| 2695 | if (r10_sync_page_io(rdev,
|
| 2696 | r10_bio->devs[sl].addr +
|
| 2697 | sect,
|
| 2698 | s, conf->tmppage, REQ_OP_WRITE)
|
| 2699 | == 0) {
|
| 2700 |
|
| 2701 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %pg)\n",
|
| 2702 | mdname(mddev), s,
|
| 2703 | (unsigned long long)(
|
| 2704 | sect +
|
| 2705 | choose_data_offset(r10_bio,
|
| 2706 | rdev)),
|
| 2707 | rdev->bdev);
|
| 2708 | pr_notice("md/raid10:%s: %pg: failing drive\n",
|
| 2709 | mdname(mddev),
|
| 2710 | rdev->bdev);
|
| 2711 | }
|
| 2712 | rdev_dec_pending(rdev, mddev);
|
| 2713 | }
|
| 2714 | sl = start;
|
| 2715 | while (sl != slot) {
|
| 2716 | if (sl==0)
|
| 2717 | sl = conf->copies;
|
| 2718 | sl--;
|
| 2719 | d = r10_bio->devs[sl].devnum;
|
| 2720 | rdev = conf->mirrors[d].rdev;
|
| 2721 | if (!rdev ||
|
| 2722 | test_bit(Faulty, &rdev->flags) ||
|
| 2723 | !test_bit(In_sync, &rdev->flags))
|
| 2724 | continue;
|
| 2725 |
|
| 2726 | atomic_inc(&rdev->nr_pending);
|
| 2727 | switch (r10_sync_page_io(rdev,
|
| 2728 | r10_bio->devs[sl].addr +
|
| 2729 | sect,
|
| 2730 | s, conf->tmppage, REQ_OP_READ)) {
|
| 2731 | case 0:
|
| 2732 |
|
| 2733 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %pg)\n",
|
| 2734 | mdname(mddev), s,
|
| 2735 | (unsigned long long)(
|
| 2736 | sect +
|
| 2737 | choose_data_offset(r10_bio, rdev)),
|
| 2738 | rdev->bdev);
|
| 2739 | pr_notice("md/raid10:%s: %pg: failing drive\n",
|
| 2740 | mdname(mddev),
|
| 2741 | rdev->bdev);
|
| 2742 | break;
|
| 2743 | case 1:
|
| 2744 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %pg)\n",
|
| 2745 | mdname(mddev), s,
|
| 2746 | (unsigned long long)(
|
| 2747 | sect +
|
| 2748 | choose_data_offset(r10_bio, rdev)),
|
| 2749 | rdev->bdev);
|
| 2750 | atomic_add(s, &rdev->corrected_errors);
|
| 2751 | }
|
| 2752 |
|
| 2753 | rdev_dec_pending(rdev, mddev);
|
| 2754 | }
|
| 2755 |
|
| 2756 | sectors -= s;
|
| 2757 | sect += s;
|
| 2758 | }
|
| 2759 | }
|
| 2760 |
|
| 2761 | static void narrow_write_error(struct r10bio *r10_bio, int i)
|
| 2762 | {
|
| 2763 | struct bio *bio = r10_bio->master_bio;
|
| 2764 | struct mddev *mddev = r10_bio->mddev;
|
| 2765 | struct r10conf *conf = mddev->private;
|
| 2766 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev;
|
| 2767 | |
| 2768 | * we just recently had a write error.
|
| 2769 | * We repeatedly clone the bio and trim down to one block,
|
| 2770 | * then try the write. Where the write fails we record
|
| 2771 | * a bad block.
|
| 2772 | * It is conceivable that the bio doesn't exactly align with
|
| 2773 | * blocks. We must handle this.
|
| 2774 | *
|
| 2775 | * We currently own a reference to the rdev.
|
| 2776 |
|
| 2777 |
|
| 2778 | int block_sectors, lbs = bdev_logical_block_size(rdev->bdev) >> 9;
|
| 2779 | sector_t sector;
|
| 2780 | int sectors;
|
| 2781 | int sect_to_write = r10_bio->sectors;
|
| 2782 |
|
| 2783 | if (rdev->badblocks.shift < 0)
|
| 2784 | block_sectors = lbs;
|
| 2785 | else
|
| 2786 | block_sectors = roundup(1 << rdev->badblocks.shift, lbs);
|
| 2787 |
|
| 2788 | sector = r10_bio->sector;
|
| 2789 | sectors = ((r10_bio->sector + block_sectors)
|
| 2790 | & ~(sector_t)(block_sectors - 1))
|
| 2791 | - sector;
|
| 2792 |
|
| 2793 | while (sect_to_write) {
|
| 2794 | struct bio *wbio;
|
| 2795 | sector_t wsector;
|
| 2796 | if (sectors > sect_to_write)
|
| 2797 | sectors = sect_to_write;
|
| 2798 |
|
| 2799 | wbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO,
|
| 2800 | &mddev->bio_set);
|
| 2801 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors);
|
| 2802 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector);
|
| 2803 | wbio->bi_iter.bi_sector = wsector +
|
| 2804 | choose_data_offset(r10_bio, rdev);
|
| 2805 | wbio->bi_opf = REQ_OP_WRITE;
|
| 2806 |
|
| 2807 | if (submit_bio_wait(wbio) &&
|
| 2808 | !rdev_set_badblocks(rdev, wsector, sectors, 0)) {
|
| 2809 | |
| 2810 | * Badblocks set failed, disk marked Faulty.
|
| 2811 | * No further operations needed.
|
| 2812 |
|
| 2813 | bio_put(wbio);
|
| 2814 | break;
|
| 2815 | }
|
| 2816 |
|
| 2817 | bio_put(wbio);
|
| 2818 | sect_to_write -= sectors;
|
| 2819 | sector += sectors;
|
| 2820 | sectors = block_sectors;
|
| 2821 | }
|
| 2822 | }
|
| 2823 |
|
| 2824 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio)
|
| 2825 | {
|
| 2826 | int slot = r10_bio->read_slot;
|
| 2827 | struct bio *bio;
|
| 2828 | struct r10conf *conf = mddev->private;
|
| 2829 | struct md_rdev *rdev = r10_bio->devs[slot].rdev;
|
| 2830 |
|
| 2831 | |
| 2832 | * the block and we can fix it.
|
| 2833 | * We freeze all other IO, and try reading the block from
|
| 2834 | * other devices. When we find one, we re-write
|
| 2835 | * and check it that fixes the read error.
|
| 2836 | * This is all done synchronously while the array is
|
| 2837 | * frozen.
|
| 2838 |
|
| 2839 | bio = r10_bio->devs[slot].bio;
|
| 2840 | bio_put(bio);
|
| 2841 | r10_bio->devs[slot].bio = NULL;
|
| 2842 |
|
| 2843 | if (mddev->ro)
|
| 2844 | r10_bio->devs[slot].bio = IO_BLOCKED;
|
| 2845 | else if (!test_bit(FailFast, &rdev->flags)) {
|
| 2846 | freeze_array(conf, 1);
|
| 2847 | fix_read_error(conf, mddev, r10_bio);
|
| 2848 | unfreeze_array(conf);
|
| 2849 | } else
|
| 2850 | md_error(mddev, rdev);
|
| 2851 |
|
| 2852 | rdev_dec_pending(rdev, mddev);
|
| 2853 | r10_bio->state = 0;
|
| 2854 | raid10_read_request(mddev, r10_bio->master_bio, r10_bio);
|
| 2855 | |
| 2856 | * allow_barrier after re-submit to ensure no sync io
|
| 2857 | * can be issued while regular io pending.
|
| 2858 |
|
| 2859 | allow_barrier(conf);
|
| 2860 | }
|
| 2861 |
|
| 2862 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio)
|
| 2863 | {
|
| 2864 | |
| 2865 | * succeeded in writing where we thought there was a
|
| 2866 | * bad block. So forget the bad block.
|
| 2867 | * Or possibly if failed and we need to record
|
| 2868 | * a bad block.
|
| 2869 |
|
| 2870 | int m;
|
| 2871 | struct md_rdev *rdev;
|
| 2872 |
|
| 2873 | if (test_bit(R10BIO_IsSync, &r10_bio->state) ||
|
| 2874 | test_bit(R10BIO_IsRecover, &r10_bio->state)) {
|
| 2875 | for (m = 0; m < conf->copies; m++) {
|
| 2876 | int dev = r10_bio->devs[m].devnum;
|
| 2877 | rdev = conf->mirrors[dev].rdev;
|
| 2878 | if (r10_bio->devs[m].bio == NULL ||
|
| 2879 | r10_bio->devs[m].bio->bi_end_io == NULL)
|
| 2880 | continue;
|
| 2881 | if (!r10_bio->devs[m].bio->bi_status)
|
| 2882 | rdev_clear_badblocks(
|
| 2883 | rdev,
|
| 2884 | r10_bio->devs[m].addr,
|
| 2885 | r10_bio->sectors, 0);
|
| 2886 | else
|
| 2887 | rdev_set_badblocks(rdev,
|
| 2888 | r10_bio->devs[m].addr,
|
| 2889 | r10_bio->sectors, 0);
|
| 2890 | rdev = conf->mirrors[dev].replacement;
|
| 2891 | if (r10_bio->devs[m].repl_bio == NULL ||
|
| 2892 | r10_bio->devs[m].repl_bio->bi_end_io == NULL)
|
| 2893 | continue;
|
| 2894 |
|
| 2895 | if (!r10_bio->devs[m].repl_bio->bi_status)
|
| 2896 | rdev_clear_badblocks(
|
| 2897 | rdev,
|
| 2898 | r10_bio->devs[m].addr,
|
| 2899 | r10_bio->sectors, 0);
|
| 2900 | else
|
| 2901 | rdev_set_badblocks(rdev,
|
| 2902 | r10_bio->devs[m].addr,
|
| 2903 | r10_bio->sectors, 0);
|
| 2904 | }
|
| 2905 | put_buf(r10_bio);
|
| 2906 | } else {
|
| 2907 | bool fail = false;
|
| 2908 | for (m = 0; m < conf->copies; m++) {
|
| 2909 | int dev = r10_bio->devs[m].devnum;
|
| 2910 | struct bio *bio = r10_bio->devs[m].bio;
|
| 2911 | rdev = conf->mirrors[dev].rdev;
|
| 2912 | if (bio == IO_MADE_GOOD) {
|
| 2913 | rdev_clear_badblocks(
|
| 2914 | rdev,
|
| 2915 | r10_bio->devs[m].addr,
|
| 2916 | r10_bio->sectors, 0);
|
| 2917 | rdev_dec_pending(rdev, conf->mddev);
|
| 2918 | } else if (bio != NULL && bio->bi_status) {
|
| 2919 | fail = true;
|
| 2920 | narrow_write_error(r10_bio, m);
|
| 2921 | rdev_dec_pending(rdev, conf->mddev);
|
| 2922 | }
|
| 2923 | bio = r10_bio->devs[m].repl_bio;
|
| 2924 | rdev = conf->mirrors[dev].replacement;
|
| 2925 | if (rdev && bio == IO_MADE_GOOD) {
|
| 2926 | rdev_clear_badblocks(
|
| 2927 | rdev,
|
| 2928 | r10_bio->devs[m].addr,
|
| 2929 | r10_bio->sectors, 0);
|
| 2930 | rdev_dec_pending(rdev, conf->mddev);
|
| 2931 | }
|
| 2932 | }
|
| 2933 | if (fail) {
|
| 2934 | spin_lock_irq(&conf->device_lock);
|
| 2935 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list);
|
| 2936 | conf->nr_queued++;
|
| 2937 | spin_unlock_irq(&conf->device_lock);
|
| 2938 | |
| 2939 | * In case freeze_array() is waiting for condition
|
| 2940 | * nr_pending == nr_queued + extra to be true.
|
| 2941 |
|
| 2942 | wake_up(&conf->wait_barrier);
|
| 2943 | md_wakeup_thread(conf->mddev->thread);
|
| 2944 | } else {
|
| 2945 | if (test_bit(R10BIO_WriteError,
|
| 2946 | &r10_bio->state))
|
| 2947 | close_write(r10_bio);
|
| 2948 | raid_end_bio_io(r10_bio);
|
| 2949 | }
|
| 2950 | }
|
| 2951 | }
|
| 2952 |
|
| 2953 | static void raid10d(struct md_thread *thread)
|
| 2954 | {
|
| 2955 | struct mddev *mddev = thread->mddev;
|
| 2956 | struct r10bio *r10_bio;
|
| 2957 | unsigned long flags;
|
| 2958 | struct r10conf *conf = mddev->private;
|
| 2959 | struct list_head *head = &conf->retry_list;
|
| 2960 | struct blk_plug plug;
|
| 2961 |
|
| 2962 | md_check_recovery(mddev);
|
| 2963 |
|
| 2964 | if (!list_empty_careful(&conf->bio_end_io_list) &&
|
| 2965 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
|
| 2966 | LIST_HEAD(tmp);
|
| 2967 | spin_lock_irqsave(&conf->device_lock, flags);
|
| 2968 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
|
| 2969 | while (!list_empty(&conf->bio_end_io_list)) {
|
| 2970 | list_move(conf->bio_end_io_list.prev, &tmp);
|
| 2971 | conf->nr_queued--;
|
| 2972 | }
|
| 2973 | }
|
| 2974 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 2975 | while (!list_empty(&tmp)) {
|
| 2976 | r10_bio = list_first_entry(&tmp, struct r10bio,
|
| 2977 | retry_list);
|
| 2978 | list_del(&r10_bio->retry_list);
|
| 2979 |
|
| 2980 | if (test_bit(R10BIO_WriteError,
|
| 2981 | &r10_bio->state))
|
| 2982 | close_write(r10_bio);
|
| 2983 | raid_end_bio_io(r10_bio);
|
| 2984 | }
|
| 2985 | }
|
| 2986 |
|
| 2987 | blk_start_plug(&plug);
|
| 2988 | for (;;) {
|
| 2989 |
|
| 2990 | flush_pending_writes(conf);
|
| 2991 |
|
| 2992 | spin_lock_irqsave(&conf->device_lock, flags);
|
| 2993 | if (list_empty(head)) {
|
| 2994 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 2995 | break;
|
| 2996 | }
|
| 2997 | r10_bio = list_entry(head->prev, struct r10bio, retry_list);
|
| 2998 | list_del(head->prev);
|
| 2999 | conf->nr_queued--;
|
| 3000 | spin_unlock_irqrestore(&conf->device_lock, flags);
|
| 3001 |
|
| 3002 | mddev = r10_bio->mddev;
|
| 3003 | conf = mddev->private;
|
| 3004 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) ||
|
| 3005 | test_bit(R10BIO_WriteError, &r10_bio->state))
|
| 3006 | handle_write_completed(conf, r10_bio);
|
| 3007 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state))
|
| 3008 | reshape_request_write(mddev, r10_bio);
|
| 3009 | else if (test_bit(R10BIO_IsSync, &r10_bio->state))
|
| 3010 | sync_request_write(mddev, r10_bio);
|
| 3011 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state))
|
| 3012 | recovery_request_write(mddev, r10_bio);
|
| 3013 | else if (test_bit(R10BIO_ReadError, &r10_bio->state))
|
| 3014 | handle_read_error(mddev, r10_bio);
|
| 3015 | else
|
| 3016 | WARN_ON_ONCE(1);
|
| 3017 |
|
| 3018 | cond_resched();
|
| 3019 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING))
|
| 3020 | md_check_recovery(mddev);
|
| 3021 | }
|
| 3022 | blk_finish_plug(&plug);
|
| 3023 | }
|
| 3024 |
|
| 3025 | static int init_resync(struct r10conf *conf)
|
| 3026 | {
|
| 3027 | int ret, buffs, i;
|
| 3028 |
|
| 3029 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
|
| 3030 | BUG_ON(mempool_initialized(&conf->r10buf_pool));
|
| 3031 | conf->have_replacement = 0;
|
| 3032 | for (i = 0; i < conf->geo.raid_disks; i++)
|
| 3033 | if (conf->mirrors[i].replacement)
|
| 3034 | conf->have_replacement = 1;
|
| 3035 | ret = mempool_init(&conf->r10buf_pool, buffs,
|
| 3036 | r10buf_pool_alloc, r10buf_pool_free, conf);
|
| 3037 | if (ret)
|
| 3038 | return ret;
|
| 3039 | conf->next_resync = 0;
|
| 3040 | return 0;
|
| 3041 | }
|
| 3042 |
|
| 3043 | static struct r10bio *raid10_alloc_init_r10buf(struct r10conf *conf)
|
| 3044 | {
|
| 3045 | struct r10bio *r10bio = mempool_alloc(&conf->r10buf_pool, GFP_NOIO);
|
| 3046 | struct rsync_pages *rp;
|
| 3047 | struct bio *bio;
|
| 3048 | int nalloc;
|
| 3049 | int i;
|
| 3050 |
|
| 3051 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) ||
|
| 3052 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery))
|
| 3053 | nalloc = conf->copies;
|
| 3054 | else
|
| 3055 | nalloc = 2;
|
| 3056 |
|
| 3057 | for (i = 0; i < nalloc; i++) {
|
| 3058 | bio = r10bio->devs[i].bio;
|
| 3059 | rp = bio->bi_private;
|
| 3060 | bio_reset(bio, NULL, 0);
|
| 3061 | bio->bi_private = rp;
|
| 3062 | bio = r10bio->devs[i].repl_bio;
|
| 3063 | if (bio) {
|
| 3064 | rp = bio->bi_private;
|
| 3065 | bio_reset(bio, NULL, 0);
|
| 3066 | bio->bi_private = rp;
|
| 3067 | }
|
| 3068 | }
|
| 3069 | return r10bio;
|
| 3070 | }
|
| 3071 |
|
| 3072 | |
| 3073 | * Set cluster_sync_high since we need other nodes to add the
|
| 3074 | * range [cluster_sync_low, cluster_sync_high] to suspend list.
|
| 3075 |
|
| 3076 | static void raid10_set_cluster_sync_high(struct r10conf *conf)
|
| 3077 | {
|
| 3078 | sector_t window_size;
|
| 3079 | int extra_chunk, chunks;
|
| 3080 |
|
| 3081 | |
| 3082 | * First, here we define "stripe" as a unit which across
|
| 3083 | * all member devices one time, so we get chunks by use
|
| 3084 | * raid_disks / near_copies. Otherwise, if near_copies is
|
| 3085 | * close to raid_disks, then resync window could increases
|
| 3086 | * linearly with the increase of raid_disks, which means
|
| 3087 | * we will suspend a really large IO window while it is not
|
| 3088 | * necessary. If raid_disks is not divisible by near_copies,
|
| 3089 | * an extra chunk is needed to ensure the whole "stripe" is
|
| 3090 | * covered.
|
| 3091 |
|
| 3092 |
|
| 3093 | chunks = conf->geo.raid_disks / conf->geo.near_copies;
|
| 3094 | if (conf->geo.raid_disks % conf->geo.near_copies == 0)
|
| 3095 | extra_chunk = 0;
|
| 3096 | else
|
| 3097 | extra_chunk = 1;
|
| 3098 | window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors;
|
| 3099 |
|
| 3100 | |
| 3101 | * At least use a 32M window to align with raid1's resync window
|
| 3102 |
|
| 3103 | window_size = (CLUSTER_RESYNC_WINDOW_SECTORS > window_size) ?
|
| 3104 | CLUSTER_RESYNC_WINDOW_SECTORS : window_size;
|
| 3105 |
|
| 3106 | conf->cluster_sync_high = conf->cluster_sync_low + window_size;
|
| 3107 | }
|
| 3108 |
|
| 3109 | |
| 3110 | * perform a "sync" on one "block"
|
| 3111 | *
|
| 3112 | * We need to make sure that no normal I/O request - particularly write
|
| 3113 | * requests - conflict with active sync requests.
|
| 3114 | *
|
| 3115 | * This is achieved by tracking pending requests and a 'barrier' concept
|
| 3116 | * that can be installed to exclude normal IO requests.
|
| 3117 | *
|
| 3118 | * Resync and recovery are handled very differently.
|
| 3119 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
|
| 3120 | *
|
| 3121 | * For resync, we iterate over virtual addresses, read all copies,
|
| 3122 | * and update if there are differences. If only one copy is live,
|
| 3123 | * skip it.
|
| 3124 | * For recovery, we iterate over physical addresses, read a good
|
| 3125 | * value for each non-in_sync drive, and over-write.
|
| 3126 | *
|
| 3127 | * So, for recovery we may have several outstanding complex requests for a
|
| 3128 | * given address, one for each out-of-sync device. We model this by allocating
|
| 3129 | * a number of r10_bio structures, one for each out-of-sync device.
|
| 3130 | * As we setup these structures, we collect all bio's together into a list
|
| 3131 | * which we then process collectively to add pages, and then process again
|
| 3132 | * to pass to submit_bio_noacct.
|
| 3133 | *
|
| 3134 | * The r10_bio structures are linked using a borrowed master_bio pointer.
|
| 3135 | * This link is counted in ->remaining. When the r10_bio that points to NULL
|
| 3136 | * has its remaining count decremented to 0, the whole complex operation
|
| 3137 | * is complete.
|
| 3138 | *
|
| 3139 |
|
| 3140 |
|
| 3141 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr,
|
| 3142 | sector_t max_sector, int *skipped)
|
| 3143 | {
|
| 3144 | struct r10conf *conf = mddev->private;
|
| 3145 | struct r10bio *r10_bio;
|
| 3146 | struct bio *biolist = NULL, *bio;
|
| 3147 | sector_t nr_sectors;
|
| 3148 | int i;
|
| 3149 | int max_sync;
|
| 3150 | sector_t sync_blocks;
|
| 3151 | sector_t chunk_mask = conf->geo.chunk_mask;
|
| 3152 | int page_idx = 0;
|
| 3153 |
|
| 3154 | |
| 3155 | * Allow skipping a full rebuild for incremental assembly
|
| 3156 | * of a clean array, like RAID1 does.
|
| 3157 |
|
| 3158 | if (mddev->bitmap == NULL &&
|
| 3159 | mddev->resync_offset == MaxSector &&
|
| 3160 | mddev->reshape_position == MaxSector &&
|
| 3161 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
|
| 3162 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
|
| 3163 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
|
| 3164 | conf->fullsync == 0) {
|
| 3165 | *skipped = 1;
|
| 3166 | return mddev->dev_sectors - sector_nr;
|
| 3167 | }
|
| 3168 |
|
| 3169 | if (!mempool_initialized(&conf->r10buf_pool))
|
| 3170 | if (init_resync(conf))
|
| 3171 | return 0;
|
| 3172 |
|
| 3173 | if (sector_nr >= max_sector) {
|
| 3174 | conf->cluster_sync_low = 0;
|
| 3175 | conf->cluster_sync_high = 0;
|
| 3176 |
|
| 3177 | |
| 3178 | * sync on the 'current' bitmap chucks (there can
|
| 3179 | * be several when recovering multiple devices).
|
| 3180 | * as we may have started syncing it but not finished.
|
| 3181 | * We can find the current address in
|
| 3182 | * mddev->curr_resync, but for recovery,
|
| 3183 | * we need to convert that to several
|
| 3184 | * virtual addresses.
|
| 3185 |
|
| 3186 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
|
| 3187 | end_reshape(conf);
|
| 3188 | close_sync(conf);
|
| 3189 | return 0;
|
| 3190 | }
|
| 3191 |
|
| 3192 | if (mddev->curr_resync < max_sector) {
|
| 3193 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
|
| 3194 | md_bitmap_end_sync(mddev, mddev->curr_resync,
|
| 3195 | &sync_blocks);
|
| 3196 | else for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 3197 | sector_t sect =
|
| 3198 | raid10_find_virt(conf, mddev->curr_resync, i);
|
| 3199 |
|
| 3200 | md_bitmap_end_sync(mddev, sect, &sync_blocks);
|
| 3201 | }
|
| 3202 | } else {
|
| 3203 |
|
| 3204 | if ((!mddev->bitmap || conf->fullsync)
|
| 3205 | && conf->have_replacement
|
| 3206 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
|
| 3207 | |
| 3208 | * are now fully recovered.
|
| 3209 |
|
| 3210 | for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 3211 | struct md_rdev *rdev =
|
| 3212 | conf->mirrors[i].replacement;
|
| 3213 |
|
| 3214 | if (rdev)
|
| 3215 | rdev->recovery_offset = MaxSector;
|
| 3216 | }
|
| 3217 | }
|
| 3218 | conf->fullsync = 0;
|
| 3219 | }
|
| 3220 | if (md_bitmap_enabled(mddev, false))
|
| 3221 | mddev->bitmap_ops->close_sync(mddev);
|
| 3222 | close_sync(conf);
|
| 3223 | *skipped = 1;
|
| 3224 | return 0;
|
| 3225 | }
|
| 3226 |
|
| 3227 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
|
| 3228 | return reshape_request(mddev, sector_nr, skipped);
|
| 3229 |
|
| 3230 | if (max_sector > mddev->resync_max)
|
| 3231 | max_sector = mddev->resync_max;
|
| 3232 |
|
| 3233 | |
| 3234 | * are meaningful
|
| 3235 |
|
| 3236 | if (conf->geo.near_copies < conf->geo.raid_disks &&
|
| 3237 | max_sector > (sector_nr | chunk_mask))
|
| 3238 | max_sector = (sector_nr | chunk_mask) + 1;
|
| 3239 |
|
| 3240 | |
| 3241 | * If there is non-resync activity waiting for a turn, then let it
|
| 3242 | * though before starting on this new sync request.
|
| 3243 |
|
| 3244 | if (conf->nr_waiting)
|
| 3245 | schedule_timeout_uninterruptible(1);
|
| 3246 |
|
| 3247 | |
| 3248 | * Both must result in an r10bio with a list of bios that
|
| 3249 | * have bi_end_io, bi_sector, bi_bdev set,
|
| 3250 | * and bi_private set to the r10bio.
|
| 3251 | * For recovery, we may actually create several r10bios
|
| 3252 | * with 2 bios in each, that correspond to the bios in the main one.
|
| 3253 | * In this case, the subordinate r10bios link back through a
|
| 3254 | * borrowed master_bio pointer, and the counter in the master
|
| 3255 | * includes a ref from each subordinate.
|
| 3256 |
|
| 3257 | |
| 3258 | * To end_sync_read if we want to read, and
|
| 3259 | * end_sync_write if we will want to write.
|
| 3260 |
|
| 3261 |
|
| 3262 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9);
|
| 3263 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
|
| 3264 |
|
| 3265 | int j;
|
| 3266 | r10_bio = NULL;
|
| 3267 |
|
| 3268 | for (i = 0 ; i < conf->geo.raid_disks; i++) {
|
| 3269 | bool still_degraded;
|
| 3270 | struct r10bio *rb2;
|
| 3271 | sector_t sect;
|
| 3272 | bool must_sync;
|
| 3273 | int any_working;
|
| 3274 | struct raid10_info *mirror = &conf->mirrors[i];
|
| 3275 | struct md_rdev *mrdev, *mreplace;
|
| 3276 |
|
| 3277 | mrdev = mirror->rdev;
|
| 3278 | mreplace = mirror->replacement;
|
| 3279 |
|
| 3280 | if (mrdev && (test_bit(Faulty, &mrdev->flags) ||
|
| 3281 | test_bit(In_sync, &mrdev->flags)))
|
| 3282 | mrdev = NULL;
|
| 3283 | if (mreplace && test_bit(Faulty, &mreplace->flags))
|
| 3284 | mreplace = NULL;
|
| 3285 |
|
| 3286 | if (!mrdev && !mreplace)
|
| 3287 | continue;
|
| 3288 |
|
| 3289 | still_degraded = false;
|
| 3290 |
|
| 3291 | rb2 = r10_bio;
|
| 3292 | sect = raid10_find_virt(conf, sector_nr, i);
|
| 3293 | if (sect >= mddev->resync_max_sectors)
|
| 3294 | |
| 3295 | * try to recover this sector.
|
| 3296 |
|
| 3297 | continue;
|
| 3298 | |
| 3299 | * we only need to recover the block if it is set in
|
| 3300 | * the bitmap
|
| 3301 |
|
| 3302 | must_sync = md_bitmap_start_sync(mddev, sect,
|
| 3303 | &sync_blocks, true);
|
| 3304 | if (sync_blocks < max_sync)
|
| 3305 | max_sync = sync_blocks;
|
| 3306 | if (!must_sync &&
|
| 3307 | mreplace == NULL &&
|
| 3308 | !conf->fullsync) {
|
| 3309 | |
| 3310 | * that there will never be anything to do here
|
| 3311 |
|
| 3312 | continue;
|
| 3313 | }
|
| 3314 | if (mrdev)
|
| 3315 | atomic_inc(&mrdev->nr_pending);
|
| 3316 | if (mreplace)
|
| 3317 | atomic_inc(&mreplace->nr_pending);
|
| 3318 |
|
| 3319 | r10_bio = raid10_alloc_init_r10buf(conf);
|
| 3320 | r10_bio->state = 0;
|
| 3321 | raise_barrier(conf, rb2 != NULL);
|
| 3322 | atomic_set(&r10_bio->remaining, 0);
|
| 3323 |
|
| 3324 | r10_bio->master_bio = (struct bio*)rb2;
|
| 3325 | if (rb2)
|
| 3326 | atomic_inc(&rb2->remaining);
|
| 3327 | r10_bio->mddev = mddev;
|
| 3328 | set_bit(R10BIO_IsRecover, &r10_bio->state);
|
| 3329 | r10_bio->sector = sect;
|
| 3330 |
|
| 3331 | raid10_find_phys(conf, r10_bio);
|
| 3332 |
|
| 3333 | |
| 3334 | * degraded
|
| 3335 |
|
| 3336 | for (j = 0; j < conf->geo.raid_disks; j++) {
|
| 3337 | struct md_rdev *rdev = conf->mirrors[j].rdev;
|
| 3338 |
|
| 3339 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) {
|
| 3340 | still_degraded = true;
|
| 3341 | break;
|
| 3342 | }
|
| 3343 | }
|
| 3344 |
|
| 3345 | md_bitmap_start_sync(mddev, sect, &sync_blocks,
|
| 3346 | still_degraded);
|
| 3347 | any_working = 0;
|
| 3348 | for (j=0; j<conf->copies;j++) {
|
| 3349 | int k;
|
| 3350 | int d = r10_bio->devs[j].devnum;
|
| 3351 | sector_t from_addr, to_addr;
|
| 3352 | struct md_rdev *rdev = conf->mirrors[d].rdev;
|
| 3353 | sector_t sector, first_bad;
|
| 3354 | sector_t bad_sectors;
|
| 3355 | if (!rdev ||
|
| 3356 | !test_bit(In_sync, &rdev->flags))
|
| 3357 | continue;
|
| 3358 |
|
| 3359 | sector = r10_bio->devs[j].addr;
|
| 3360 |
|
| 3361 | if (is_badblock(rdev, sector, max_sync,
|
| 3362 | &first_bad, &bad_sectors)) {
|
| 3363 | if (first_bad > sector)
|
| 3364 | max_sync = first_bad - sector;
|
| 3365 | else {
|
| 3366 | bad_sectors -= (sector
|
| 3367 | - first_bad);
|
| 3368 | if (max_sync > bad_sectors)
|
| 3369 | max_sync = bad_sectors;
|
| 3370 | continue;
|
| 3371 | }
|
| 3372 | }
|
| 3373 | any_working = 1;
|
| 3374 | bio = r10_bio->devs[0].bio;
|
| 3375 | bio->bi_next = biolist;
|
| 3376 | biolist = bio;
|
| 3377 | bio->bi_end_io = end_sync_read;
|
| 3378 | bio->bi_opf = REQ_OP_READ;
|
| 3379 | if (test_bit(FailFast, &rdev->flags))
|
| 3380 | bio->bi_opf |= MD_FAILFAST;
|
| 3381 | from_addr = r10_bio->devs[j].addr;
|
| 3382 | bio->bi_iter.bi_sector = from_addr +
|
| 3383 | rdev->data_offset;
|
| 3384 | bio_set_dev(bio, rdev->bdev);
|
| 3385 | atomic_inc(&rdev->nr_pending);
|
| 3386 |
|
| 3387 |
|
| 3388 | for (k=0; k<conf->copies; k++)
|
| 3389 | if (r10_bio->devs[k].devnum == i)
|
| 3390 | break;
|
| 3391 | BUG_ON(k == conf->copies);
|
| 3392 | to_addr = r10_bio->devs[k].addr;
|
| 3393 | r10_bio->devs[0].devnum = d;
|
| 3394 | r10_bio->devs[0].addr = from_addr;
|
| 3395 | r10_bio->devs[1].devnum = i;
|
| 3396 | r10_bio->devs[1].addr = to_addr;
|
| 3397 |
|
| 3398 | if (mrdev) {
|
| 3399 | bio = r10_bio->devs[1].bio;
|
| 3400 | bio->bi_next = biolist;
|
| 3401 | biolist = bio;
|
| 3402 | bio->bi_end_io = end_sync_write;
|
| 3403 | bio->bi_opf = REQ_OP_WRITE;
|
| 3404 | bio->bi_iter.bi_sector = to_addr
|
| 3405 | + mrdev->data_offset;
|
| 3406 | bio_set_dev(bio, mrdev->bdev);
|
| 3407 | atomic_inc(&r10_bio->remaining);
|
| 3408 | } else
|
| 3409 | r10_bio->devs[1].bio->bi_end_io = NULL;
|
| 3410 |
|
| 3411 |
|
| 3412 | bio = r10_bio->devs[1].repl_bio;
|
| 3413 | if (bio)
|
| 3414 | bio->bi_end_io = NULL;
|
| 3415 | |
| 3416 | * cannot be NULL as r10buf_pool_alloc will
|
| 3417 | * have allocated it.
|
| 3418 |
|
| 3419 | if (!mreplace)
|
| 3420 | break;
|
| 3421 | bio->bi_next = biolist;
|
| 3422 | biolist = bio;
|
| 3423 | bio->bi_end_io = end_sync_write;
|
| 3424 | bio->bi_opf = REQ_OP_WRITE;
|
| 3425 | bio->bi_iter.bi_sector = to_addr +
|
| 3426 | mreplace->data_offset;
|
| 3427 | bio_set_dev(bio, mreplace->bdev);
|
| 3428 | atomic_inc(&r10_bio->remaining);
|
| 3429 | break;
|
| 3430 | }
|
| 3431 | if (j == conf->copies) {
|
| 3432 | |
| 3433 | * record a bad block
|
| 3434 | if (any_working) {
|
| 3435 | |
| 3436 | * on other device(s)
|
| 3437 |
|
| 3438 | int k;
|
| 3439 | for (k = 0; k < conf->copies; k++)
|
| 3440 | if (r10_bio->devs[k].devnum == i)
|
| 3441 | break;
|
| 3442 | if (mrdev &&
|
| 3443 | !test_bit(In_sync, &mrdev->flags))
|
| 3444 | rdev_set_badblocks(
|
| 3445 | mrdev,
|
| 3446 | r10_bio->devs[k].addr,
|
| 3447 | max_sync, 0);
|
| 3448 | if (mreplace)
|
| 3449 | rdev_set_badblocks(
|
| 3450 | mreplace,
|
| 3451 | r10_bio->devs[k].addr,
|
| 3452 | max_sync, 0);
|
| 3453 | pr_warn("md/raid10:%s: cannot recovery sector %llu + %d.\n",
|
| 3454 | mdname(mddev), r10_bio->devs[k].addr, max_sync);
|
| 3455 | }
|
| 3456 | put_buf(r10_bio);
|
| 3457 | if (rb2)
|
| 3458 | atomic_dec(&rb2->remaining);
|
| 3459 | r10_bio = rb2;
|
| 3460 | if (mrdev)
|
| 3461 | rdev_dec_pending(mrdev, mddev);
|
| 3462 | if (mreplace)
|
| 3463 | rdev_dec_pending(mreplace, mddev);
|
| 3464 | break;
|
| 3465 | }
|
| 3466 | if (mrdev)
|
| 3467 | rdev_dec_pending(mrdev, mddev);
|
| 3468 | if (mreplace)
|
| 3469 | rdev_dec_pending(mreplace, mddev);
|
| 3470 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) {
|
| 3471 | |
| 3472 | * read from. 'j' is currently the first
|
| 3473 | * readable copy.
|
| 3474 |
|
| 3475 | int targets = 1;
|
| 3476 | for (; j < conf->copies; j++) {
|
| 3477 | int d = r10_bio->devs[j].devnum;
|
| 3478 | if (conf->mirrors[d].rdev &&
|
| 3479 | test_bit(In_sync,
|
| 3480 | &conf->mirrors[d].rdev->flags))
|
| 3481 | targets++;
|
| 3482 | }
|
| 3483 | if (targets == 1)
|
| 3484 | r10_bio->devs[0].bio->bi_opf
|
| 3485 | &= ~MD_FAILFAST;
|
| 3486 | }
|
| 3487 | }
|
| 3488 | if (biolist == NULL) {
|
| 3489 | while (r10_bio) {
|
| 3490 | struct r10bio *rb2 = r10_bio;
|
| 3491 | r10_bio = (struct r10bio*) rb2->master_bio;
|
| 3492 | rb2->master_bio = NULL;
|
| 3493 | put_buf(rb2);
|
| 3494 | }
|
| 3495 | *skipped = 1;
|
| 3496 | return max_sync;
|
| 3497 | }
|
| 3498 | } else {
|
| 3499 |
|
| 3500 | int count = 0;
|
| 3501 |
|
| 3502 | |
| 3503 | * Since curr_resync_completed could probably not update in
|
| 3504 | * time, and we will set cluster_sync_low based on it.
|
| 3505 | * Let's check against "sector_nr + 2 * RESYNC_SECTORS" for
|
| 3506 | * safety reason, which ensures curr_resync_completed is
|
| 3507 | * updated in bitmap_cond_end_sync.
|
| 3508 |
|
| 3509 | if (md_bitmap_enabled(mddev, false))
|
| 3510 | mddev->bitmap_ops->cond_end_sync(mddev, sector_nr,
|
| 3511 | mddev_is_clustered(mddev) &&
|
| 3512 | (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
|
| 3513 |
|
| 3514 | if (!md_bitmap_start_sync(mddev, sector_nr, &sync_blocks,
|
| 3515 | mddev->degraded) &&
|
| 3516 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED,
|
| 3517 | &mddev->recovery)) {
|
| 3518 |
|
| 3519 | *skipped = 1;
|
| 3520 | return sync_blocks;
|
| 3521 | }
|
| 3522 | if (sync_blocks < max_sync)
|
| 3523 | max_sync = sync_blocks;
|
| 3524 | r10_bio = raid10_alloc_init_r10buf(conf);
|
| 3525 | r10_bio->state = 0;
|
| 3526 |
|
| 3527 | r10_bio->mddev = mddev;
|
| 3528 | atomic_set(&r10_bio->remaining, 0);
|
| 3529 | raise_barrier(conf, 0);
|
| 3530 | conf->next_resync = sector_nr;
|
| 3531 |
|
| 3532 | r10_bio->master_bio = NULL;
|
| 3533 | r10_bio->sector = sector_nr;
|
| 3534 | set_bit(R10BIO_IsSync, &r10_bio->state);
|
| 3535 | raid10_find_phys(conf, r10_bio);
|
| 3536 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1;
|
| 3537 |
|
| 3538 | for (i = 0; i < conf->copies; i++) {
|
| 3539 | int d = r10_bio->devs[i].devnum;
|
| 3540 | sector_t first_bad, sector;
|
| 3541 | sector_t bad_sectors;
|
| 3542 | struct md_rdev *rdev;
|
| 3543 |
|
| 3544 | if (r10_bio->devs[i].repl_bio)
|
| 3545 | r10_bio->devs[i].repl_bio->bi_end_io = NULL;
|
| 3546 |
|
| 3547 | bio = r10_bio->devs[i].bio;
|
| 3548 | bio->bi_status = BLK_STS_IOERR;
|
| 3549 | rdev = conf->mirrors[d].rdev;
|
| 3550 | if (rdev == NULL || test_bit(Faulty, &rdev->flags))
|
| 3551 | continue;
|
| 3552 |
|
| 3553 | sector = r10_bio->devs[i].addr;
|
| 3554 | if (is_badblock(rdev, sector, max_sync,
|
| 3555 | &first_bad, &bad_sectors)) {
|
| 3556 | if (first_bad > sector)
|
| 3557 | max_sync = first_bad - sector;
|
| 3558 | else {
|
| 3559 | bad_sectors -= (sector - first_bad);
|
| 3560 | if (max_sync > bad_sectors)
|
| 3561 | max_sync = bad_sectors;
|
| 3562 | continue;
|
| 3563 | }
|
| 3564 | }
|
| 3565 | atomic_inc(&rdev->nr_pending);
|
| 3566 | atomic_inc(&r10_bio->remaining);
|
| 3567 | bio->bi_next = biolist;
|
| 3568 | biolist = bio;
|
| 3569 | bio->bi_end_io = end_sync_read;
|
| 3570 | bio->bi_opf = REQ_OP_READ;
|
| 3571 | if (test_bit(FailFast, &rdev->flags))
|
| 3572 | bio->bi_opf |= MD_FAILFAST;
|
| 3573 | bio->bi_iter.bi_sector = sector + rdev->data_offset;
|
| 3574 | bio_set_dev(bio, rdev->bdev);
|
| 3575 | count++;
|
| 3576 |
|
| 3577 | rdev = conf->mirrors[d].replacement;
|
| 3578 | if (rdev == NULL || test_bit(Faulty, &rdev->flags))
|
| 3579 | continue;
|
| 3580 |
|
| 3581 | atomic_inc(&rdev->nr_pending);
|
| 3582 |
|
| 3583 |
|
| 3584 | bio = r10_bio->devs[i].repl_bio;
|
| 3585 | bio->bi_status = BLK_STS_IOERR;
|
| 3586 |
|
| 3587 | sector = r10_bio->devs[i].addr;
|
| 3588 | bio->bi_next = biolist;
|
| 3589 | biolist = bio;
|
| 3590 | bio->bi_end_io = end_sync_write;
|
| 3591 | bio->bi_opf = REQ_OP_WRITE;
|
| 3592 | if (test_bit(FailFast, &rdev->flags))
|
| 3593 | bio->bi_opf |= MD_FAILFAST;
|
| 3594 | bio->bi_iter.bi_sector = sector + rdev->data_offset;
|
| 3595 | bio_set_dev(bio, rdev->bdev);
|
| 3596 | count++;
|
| 3597 | }
|
| 3598 |
|
| 3599 | if (count < 2) {
|
| 3600 | for (i=0; i<conf->copies; i++) {
|
| 3601 | int d = r10_bio->devs[i].devnum;
|
| 3602 | if (r10_bio->devs[i].bio->bi_end_io)
|
| 3603 | rdev_dec_pending(conf->mirrors[d].rdev,
|
| 3604 | mddev);
|
| 3605 | if (r10_bio->devs[i].repl_bio &&
|
| 3606 | r10_bio->devs[i].repl_bio->bi_end_io)
|
| 3607 | rdev_dec_pending(
|
| 3608 | conf->mirrors[d].replacement,
|
| 3609 | mddev);
|
| 3610 | }
|
| 3611 | put_buf(r10_bio);
|
| 3612 | *skipped = 1;
|
| 3613 | return max_sync;
|
| 3614 | }
|
| 3615 | }
|
| 3616 |
|
| 3617 | nr_sectors = 0;
|
| 3618 | if (sector_nr + max_sync < max_sector)
|
| 3619 | max_sector = sector_nr + max_sync;
|
| 3620 | do {
|
| 3621 | struct page *page;
|
| 3622 | int len = PAGE_SIZE;
|
| 3623 | if (sector_nr + (len>>9) > max_sector)
|
| 3624 | len = (max_sector - sector_nr) << 9;
|
| 3625 | if (len == 0)
|
| 3626 | break;
|
| 3627 | for (bio= biolist ; bio ; bio=bio->bi_next) {
|
| 3628 | struct resync_pages *rp = get_resync_pages(bio);
|
| 3629 | page = resync_fetch_page(rp, page_idx);
|
| 3630 | if (WARN_ON(!bio_add_page(bio, page, len, 0))) {
|
| 3631 | bio->bi_status = BLK_STS_RESOURCE;
|
| 3632 | bio_endio(bio);
|
| 3633 | *skipped = 1;
|
| 3634 | return max_sync;
|
| 3635 | }
|
| 3636 | }
|
| 3637 | nr_sectors += len>>9;
|
| 3638 | sector_nr += len>>9;
|
| 3639 | } while (++page_idx < RESYNC_PAGES);
|
| 3640 | r10_bio->sectors = nr_sectors;
|
| 3641 |
|
| 3642 | if (mddev_is_clustered(mddev) &&
|
| 3643 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
|
| 3644 |
|
| 3645 | if (conf->cluster_sync_high < sector_nr + nr_sectors) {
|
| 3646 | conf->cluster_sync_low = mddev->curr_resync_completed;
|
| 3647 | raid10_set_cluster_sync_high(conf);
|
| 3648 |
|
| 3649 | mddev->cluster_ops->resync_info_update(mddev,
|
| 3650 | conf->cluster_sync_low,
|
| 3651 | conf->cluster_sync_high);
|
| 3652 | }
|
| 3653 | } else if (mddev_is_clustered(mddev)) {
|
| 3654 |
|
| 3655 | sector_t sect_va1, sect_va2;
|
| 3656 | bool broadcast_msg = false;
|
| 3657 |
|
| 3658 | for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 3659 | |
| 3660 | * sector_nr is a device address for recovery, so we
|
| 3661 | * need translate it to array address before compare
|
| 3662 | * with cluster_sync_high.
|
| 3663 |
|
| 3664 | sect_va1 = raid10_find_virt(conf, sector_nr, i);
|
| 3665 |
|
| 3666 | if (conf->cluster_sync_high < sect_va1 + nr_sectors) {
|
| 3667 | broadcast_msg = true;
|
| 3668 | |
| 3669 | * curr_resync_completed is similar as
|
| 3670 | * sector_nr, so make the translation too.
|
| 3671 |
|
| 3672 | sect_va2 = raid10_find_virt(conf,
|
| 3673 | mddev->curr_resync_completed, i);
|
| 3674 |
|
| 3675 | if (conf->cluster_sync_low == 0 ||
|
| 3676 | conf->cluster_sync_low > sect_va2)
|
| 3677 | conf->cluster_sync_low = sect_va2;
|
| 3678 | }
|
| 3679 | }
|
| 3680 | if (broadcast_msg) {
|
| 3681 | raid10_set_cluster_sync_high(conf);
|
| 3682 | mddev->cluster_ops->resync_info_update(mddev,
|
| 3683 | conf->cluster_sync_low,
|
| 3684 | conf->cluster_sync_high);
|
| 3685 | }
|
| 3686 | }
|
| 3687 |
|
| 3688 | while (biolist) {
|
| 3689 | bio = biolist;
|
| 3690 | biolist = biolist->bi_next;
|
| 3691 |
|
| 3692 | bio->bi_next = NULL;
|
| 3693 | r10_bio = get_resync_r10bio(bio);
|
| 3694 | r10_bio->sectors = nr_sectors;
|
| 3695 |
|
| 3696 | if (bio->bi_end_io == end_sync_read) {
|
| 3697 | bio->bi_status = 0;
|
| 3698 | submit_bio_noacct(bio);
|
| 3699 | }
|
| 3700 | }
|
| 3701 |
|
| 3702 | return nr_sectors;
|
| 3703 | }
|
| 3704 |
|
| 3705 | static sector_t
|
| 3706 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks)
|
| 3707 | {
|
| 3708 | sector_t size;
|
| 3709 | struct r10conf *conf = mddev->private;
|
| 3710 |
|
| 3711 | if (!raid_disks)
|
| 3712 | raid_disks = min(conf->geo.raid_disks,
|
| 3713 | conf->prev.raid_disks);
|
| 3714 | if (!sectors)
|
| 3715 | sectors = conf->dev_sectors;
|
| 3716 |
|
| 3717 | size = sectors >> conf->geo.chunk_shift;
|
| 3718 | sector_div(size, conf->geo.far_copies);
|
| 3719 | size = size * raid_disks;
|
| 3720 | sector_div(size, conf->geo.near_copies);
|
| 3721 |
|
| 3722 | return size << conf->geo.chunk_shift;
|
| 3723 | }
|
| 3724 |
|
| 3725 | static void calc_sectors(struct r10conf *conf, sector_t size)
|
| 3726 | {
|
| 3727 | |
| 3728 | * actually be used, and set conf->dev_sectors and
|
| 3729 | * conf->stride
|
| 3730 |
|
| 3731 |
|
| 3732 | size = size >> conf->geo.chunk_shift;
|
| 3733 | sector_div(size, conf->geo.far_copies);
|
| 3734 | size = size * conf->geo.raid_disks;
|
| 3735 | sector_div(size, conf->geo.near_copies);
|
| 3736 |
|
| 3737 |
|
| 3738 | size = size * conf->copies;
|
| 3739 |
|
| 3740 | |
| 3741 | * get the stride size.
|
| 3742 |
|
| 3743 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks);
|
| 3744 |
|
| 3745 | conf->dev_sectors = size << conf->geo.chunk_shift;
|
| 3746 |
|
| 3747 | if (conf->geo.far_offset)
|
| 3748 | conf->geo.stride = 1 << conf->geo.chunk_shift;
|
| 3749 | else {
|
| 3750 | sector_div(size, conf->geo.far_copies);
|
| 3751 | conf->geo.stride = size << conf->geo.chunk_shift;
|
| 3752 | }
|
| 3753 | }
|
| 3754 |
|
| 3755 | enum geo_type {geo_new, geo_old, geo_start};
|
| 3756 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new)
|
| 3757 | {
|
| 3758 | int nc, fc, fo;
|
| 3759 | int layout, chunk, disks;
|
| 3760 | switch (new) {
|
| 3761 | case geo_old:
|
| 3762 | layout = mddev->layout;
|
| 3763 | chunk = mddev->chunk_sectors;
|
| 3764 | disks = mddev->raid_disks - mddev->delta_disks;
|
| 3765 | break;
|
| 3766 | case geo_new:
|
| 3767 | layout = mddev->new_layout;
|
| 3768 | chunk = mddev->new_chunk_sectors;
|
| 3769 | disks = mddev->raid_disks;
|
| 3770 | break;
|
| 3771 | default:
|
| 3772 | case geo_start: |
| 3773 | * updated yet.
|
| 3774 | layout = mddev->new_layout;
|
| 3775 | chunk = mddev->new_chunk_sectors;
|
| 3776 | disks = mddev->raid_disks + mddev->delta_disks;
|
| 3777 | break;
|
| 3778 | }
|
| 3779 | if (layout >> 19)
|
| 3780 | return -1;
|
| 3781 | if (chunk < (PAGE_SIZE >> 9) ||
|
| 3782 | !is_power_of_2(chunk))
|
| 3783 | return -2;
|
| 3784 | nc = layout & 255;
|
| 3785 | fc = (layout >> 8) & 255;
|
| 3786 | fo = layout & (1<<16);
|
| 3787 | if (!nc || !fc)
|
| 3788 | return -1;
|
| 3789 | geo->raid_disks = disks;
|
| 3790 | geo->near_copies = nc;
|
| 3791 | geo->far_copies = fc;
|
| 3792 | geo->far_offset = fo;
|
| 3793 | switch (layout >> 17) {
|
| 3794 | case 0:
|
| 3795 | geo->far_set_size = disks;
|
| 3796 | break;
|
| 3797 | case 1: |
| 3798 | * actually using this, but leave code here just in case.
|
| 3799 | geo->far_set_size = disks/fc;
|
| 3800 | WARN(geo->far_set_size < fc,
|
| 3801 | "This RAID10 layout does not provide data safety - please backup and create new array\n");
|
| 3802 | break;
|
| 3803 | case 2:
|
| 3804 | geo->far_set_size = fc * nc;
|
| 3805 | break;
|
| 3806 | default:
|
| 3807 | return -1;
|
| 3808 | }
|
| 3809 | geo->chunk_mask = chunk - 1;
|
| 3810 | geo->chunk_shift = ffz(~chunk);
|
| 3811 | return nc*fc;
|
| 3812 | }
|
| 3813 |
|
| 3814 | static void raid10_free_conf(struct r10conf *conf)
|
| 3815 | {
|
| 3816 | if (!conf)
|
| 3817 | return;
|
| 3818 |
|
| 3819 | mempool_destroy(conf->r10bio_pool);
|
| 3820 | kfree(conf->mirrors);
|
| 3821 | kfree(conf->mirrors_old);
|
| 3822 | kfree(conf->mirrors_new);
|
| 3823 | safe_put_page(conf->tmppage);
|
| 3824 | bioset_exit(&conf->bio_split);
|
| 3825 | kfree(conf);
|
| 3826 | }
|
| 3827 |
|
| 3828 | static struct r10conf *setup_conf(struct mddev *mddev)
|
| 3829 | {
|
| 3830 | struct r10conf *conf = NULL;
|
| 3831 | int err = -EINVAL;
|
| 3832 | struct geom geo;
|
| 3833 | int copies;
|
| 3834 |
|
| 3835 | copies = setup_geo(&geo, mddev, geo_new);
|
| 3836 |
|
| 3837 | if (copies == -2) {
|
| 3838 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n",
|
| 3839 | mdname(mddev), PAGE_SIZE);
|
| 3840 | goto out;
|
| 3841 | }
|
| 3842 |
|
| 3843 | if (copies < 2 || copies > mddev->raid_disks) {
|
| 3844 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n",
|
| 3845 | mdname(mddev), mddev->new_layout);
|
| 3846 | goto out;
|
| 3847 | }
|
| 3848 |
|
| 3849 | err = -ENOMEM;
|
| 3850 | conf = kzalloc_obj(struct r10conf);
|
| 3851 | if (!conf)
|
| 3852 | goto out;
|
| 3853 |
|
| 3854 |
|
| 3855 | conf->mirrors = kzalloc_objs(struct raid10_info,
|
| 3856 | mddev->raid_disks + max(0, -mddev->delta_disks));
|
| 3857 | if (!conf->mirrors)
|
| 3858 | goto out;
|
| 3859 |
|
| 3860 | conf->tmppage = alloc_page(GFP_KERNEL);
|
| 3861 | if (!conf->tmppage)
|
| 3862 | goto out;
|
| 3863 |
|
| 3864 | conf->geo = geo;
|
| 3865 | conf->copies = copies;
|
| 3866 | conf->r10bio_pool = create_r10bio_pool(conf->geo.raid_disks);
|
| 3867 | if (!conf->r10bio_pool)
|
| 3868 | goto out;
|
| 3869 |
|
| 3870 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0);
|
| 3871 | if (err)
|
| 3872 | goto out;
|
| 3873 |
|
| 3874 | calc_sectors(conf, mddev->dev_sectors);
|
| 3875 | if (mddev->reshape_position == MaxSector) {
|
| 3876 | conf->prev = conf->geo;
|
| 3877 | conf->reshape_progress = MaxSector;
|
| 3878 | } else {
|
| 3879 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) {
|
| 3880 | err = -EINVAL;
|
| 3881 | goto out;
|
| 3882 | }
|
| 3883 | conf->reshape_progress = mddev->reshape_position;
|
| 3884 | if (conf->prev.far_offset)
|
| 3885 | conf->prev.stride = 1 << conf->prev.chunk_shift;
|
| 3886 | else
|
| 3887 |
|
| 3888 | conf->prev.stride = conf->dev_sectors;
|
| 3889 | }
|
| 3890 | conf->reshape_safe = conf->reshape_progress;
|
| 3891 | spin_lock_init(&conf->device_lock);
|
| 3892 | INIT_LIST_HEAD(&conf->retry_list);
|
| 3893 | INIT_LIST_HEAD(&conf->bio_end_io_list);
|
| 3894 |
|
| 3895 | seqlock_init(&conf->resync_lock);
|
| 3896 | init_waitqueue_head(&conf->wait_barrier);
|
| 3897 | atomic_set(&conf->nr_pending, 0);
|
| 3898 |
|
| 3899 | err = -ENOMEM;
|
| 3900 | rcu_assign_pointer(conf->thread,
|
| 3901 | md_register_thread(raid10d, mddev, "raid10"));
|
| 3902 | if (!conf->thread)
|
| 3903 | goto out;
|
| 3904 |
|
| 3905 | conf->mddev = mddev;
|
| 3906 | return conf;
|
| 3907 |
|
| 3908 | out:
|
| 3909 | raid10_free_conf(conf);
|
| 3910 | return ERR_PTR(err);
|
| 3911 | }
|
| 3912 |
|
| 3913 | static unsigned int raid10_nr_stripes(struct r10conf *conf)
|
| 3914 | {
|
| 3915 | unsigned int raid_disks = conf->geo.raid_disks;
|
| 3916 |
|
| 3917 | if (conf->geo.raid_disks % conf->geo.near_copies)
|
| 3918 | return raid_disks;
|
| 3919 | return raid_disks / conf->geo.near_copies;
|
| 3920 | }
|
| 3921 |
|
| 3922 | static int raid10_set_queue_limits(struct mddev *mddev)
|
| 3923 | {
|
| 3924 | struct r10conf *conf = mddev->private;
|
| 3925 | struct queue_limits lim;
|
| 3926 | int err;
|
| 3927 |
|
| 3928 | md_init_stacking_limits(&lim);
|
| 3929 | lim.max_write_zeroes_sectors = 0;
|
| 3930 | lim.max_hw_wzeroes_unmap_sectors = 0;
|
| 3931 | lim.logical_block_size = mddev->logical_block_size;
|
| 3932 | lim.io_min = mddev->chunk_sectors << 9;
|
| 3933 | lim.chunk_sectors = mddev->chunk_sectors;
|
| 3934 | lim.io_opt = lim.io_min * raid10_nr_stripes(conf);
|
| 3935 | lim.features |= BLK_FEAT_ATOMIC_WRITES;
|
| 3936 | lim.features |= BLK_FEAT_PCI_P2PDMA;
|
| 3937 | err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
|
| 3938 | if (err)
|
| 3939 | return err;
|
| 3940 | return queue_limits_set(mddev->gendisk->queue, &lim);
|
| 3941 | }
|
| 3942 |
|
| 3943 | static int raid10_run(struct mddev *mddev)
|
| 3944 | {
|
| 3945 | struct r10conf *conf;
|
| 3946 | int i, disk_idx;
|
| 3947 | struct raid10_info *disk;
|
| 3948 | struct md_rdev *rdev;
|
| 3949 | sector_t size;
|
| 3950 | sector_t min_offset_diff = 0;
|
| 3951 | int first = 1;
|
| 3952 | int ret = -EIO;
|
| 3953 |
|
| 3954 | if (mddev->private == NULL) {
|
| 3955 | conf = setup_conf(mddev);
|
| 3956 | if (IS_ERR(conf))
|
| 3957 | return PTR_ERR(conf);
|
| 3958 | mddev->private = conf;
|
| 3959 | }
|
| 3960 | conf = mddev->private;
|
| 3961 | if (!conf)
|
| 3962 | goto out;
|
| 3963 |
|
| 3964 | rcu_assign_pointer(mddev->thread, conf->thread);
|
| 3965 | rcu_assign_pointer(conf->thread, NULL);
|
| 3966 |
|
| 3967 | if (mddev_is_clustered(conf->mddev)) {
|
| 3968 | int fc, fo;
|
| 3969 |
|
| 3970 | fc = (mddev->layout >> 8) & 255;
|
| 3971 | fo = mddev->layout & (1<<16);
|
| 3972 | if (fc > 1 || fo > 0) {
|
| 3973 | pr_err("only near layout is supported by clustered"
|
| 3974 | " raid10\n");
|
| 3975 | goto out_free_conf;
|
| 3976 | }
|
| 3977 | }
|
| 3978 |
|
| 3979 | rdev_for_each(rdev, mddev) {
|
| 3980 | long long diff;
|
| 3981 |
|
| 3982 | disk_idx = rdev->raid_disk;
|
| 3983 | if (disk_idx < 0)
|
| 3984 | continue;
|
| 3985 | if (disk_idx >= conf->geo.raid_disks &&
|
| 3986 | disk_idx >= conf->prev.raid_disks)
|
| 3987 | continue;
|
| 3988 | disk = conf->mirrors + disk_idx;
|
| 3989 |
|
| 3990 | if (test_bit(Replacement, &rdev->flags)) {
|
| 3991 | if (disk->replacement)
|
| 3992 | goto out_free_conf;
|
| 3993 | disk->replacement = rdev;
|
| 3994 | } else {
|
| 3995 | if (disk->rdev)
|
| 3996 | goto out_free_conf;
|
| 3997 | disk->rdev = rdev;
|
| 3998 | }
|
| 3999 | diff = (rdev->new_data_offset - rdev->data_offset);
|
| 4000 | if (!mddev->reshape_backwards)
|
| 4001 | diff = -diff;
|
| 4002 | if (diff < 0)
|
| 4003 | diff = 0;
|
| 4004 | if (first || diff < min_offset_diff)
|
| 4005 | min_offset_diff = diff;
|
| 4006 |
|
| 4007 | disk->head_position = 0;
|
| 4008 | first = 0;
|
| 4009 | }
|
| 4010 |
|
| 4011 | if (!mddev_is_dm(conf->mddev)) {
|
| 4012 | int err = raid10_set_queue_limits(mddev);
|
| 4013 |
|
| 4014 | if (err) {
|
| 4015 | ret = err;
|
| 4016 | goto out_free_conf;
|
| 4017 | }
|
| 4018 | }
|
| 4019 |
|
| 4020 |
|
| 4021 | if (!enough(conf, -1)) {
|
| 4022 | pr_err("md/raid10:%s: not enough operational mirrors.\n",
|
| 4023 | mdname(mddev));
|
| 4024 | goto out_free_conf;
|
| 4025 | }
|
| 4026 |
|
| 4027 | if (conf->reshape_progress != MaxSector) {
|
| 4028 |
|
| 4029 | if (conf->geo.far_copies != 1 &&
|
| 4030 | conf->geo.far_offset == 0)
|
| 4031 | goto out_free_conf;
|
| 4032 | if (conf->prev.far_copies != 1 &&
|
| 4033 | conf->prev.far_offset == 0)
|
| 4034 | goto out_free_conf;
|
| 4035 | }
|
| 4036 |
|
| 4037 | mddev->degraded = 0;
|
| 4038 | for (i = 0;
|
| 4039 | i < conf->geo.raid_disks
|
| 4040 | || i < conf->prev.raid_disks;
|
| 4041 | i++) {
|
| 4042 |
|
| 4043 | disk = conf->mirrors + i;
|
| 4044 |
|
| 4045 | if (!disk->rdev && disk->replacement) {
|
| 4046 |
|
| 4047 | disk->rdev = disk->replacement;
|
| 4048 | disk->replacement = NULL;
|
| 4049 | clear_bit(Replacement, &disk->rdev->flags);
|
| 4050 | }
|
| 4051 |
|
| 4052 | if (!disk->rdev ||
|
| 4053 | !test_bit(In_sync, &disk->rdev->flags)) {
|
| 4054 | disk->head_position = 0;
|
| 4055 | mddev->degraded++;
|
| 4056 | if (disk->rdev &&
|
| 4057 | disk->rdev->saved_raid_disk < 0)
|
| 4058 | conf->fullsync = 1;
|
| 4059 | }
|
| 4060 |
|
| 4061 | if (disk->replacement &&
|
| 4062 | !test_bit(In_sync, &disk->replacement->flags) &&
|
| 4063 | disk->replacement->saved_raid_disk < 0) {
|
| 4064 | conf->fullsync = 1;
|
| 4065 | }
|
| 4066 | }
|
| 4067 |
|
| 4068 | if (mddev->resync_offset != MaxSector)
|
| 4069 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n",
|
| 4070 | mdname(mddev));
|
| 4071 | pr_info("md/raid10:%s: active with %d out of %d devices\n",
|
| 4072 | mdname(mddev), conf->geo.raid_disks - mddev->degraded,
|
| 4073 | conf->geo.raid_disks);
|
| 4074 | |
| 4075 | * Ok, everything is just fine now
|
| 4076 |
|
| 4077 | mddev->dev_sectors = conf->dev_sectors;
|
| 4078 | size = raid10_size(mddev, 0, 0);
|
| 4079 | md_set_array_sectors(mddev, size);
|
| 4080 | mddev->resync_max_sectors = size;
|
| 4081 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
|
| 4082 |
|
| 4083 | if (md_integrity_register(mddev))
|
| 4084 | goto out_free_conf;
|
| 4085 |
|
| 4086 | if (conf->reshape_progress != MaxSector) {
|
| 4087 | unsigned long before_length, after_length;
|
| 4088 |
|
| 4089 | before_length = ((1 << conf->prev.chunk_shift) *
|
| 4090 | conf->prev.far_copies);
|
| 4091 | after_length = ((1 << conf->geo.chunk_shift) *
|
| 4092 | conf->geo.far_copies);
|
| 4093 |
|
| 4094 | if (max(before_length, after_length) > min_offset_diff) {
|
| 4095 |
|
| 4096 | pr_warn("md/raid10: offset difference not enough to continue reshape\n");
|
| 4097 | goto out_free_conf;
|
| 4098 | }
|
| 4099 | conf->offset_diff = min_offset_diff;
|
| 4100 |
|
| 4101 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
|
| 4102 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
|
| 4103 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
|
| 4104 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
|
| 4105 | }
|
| 4106 |
|
| 4107 | return 0;
|
| 4108 |
|
| 4109 | out_free_conf:
|
| 4110 | md_unregister_thread(mddev, &mddev->thread);
|
| 4111 | raid10_free_conf(conf);
|
| 4112 | mddev->private = NULL;
|
| 4113 | out:
|
| 4114 | return ret;
|
| 4115 | }
|
| 4116 |
|
| 4117 | static void raid10_free(struct mddev *mddev, void *priv)
|
| 4118 | {
|
| 4119 | raid10_free_conf(priv);
|
| 4120 | }
|
| 4121 |
|
| 4122 | static void raid10_quiesce(struct mddev *mddev, int quiesce)
|
| 4123 | {
|
| 4124 | struct r10conf *conf = mddev->private;
|
| 4125 |
|
| 4126 | if (quiesce)
|
| 4127 | raise_barrier(conf, 0);
|
| 4128 | else
|
| 4129 | lower_barrier(conf);
|
| 4130 | }
|
| 4131 |
|
| 4132 | static int raid10_resize(struct mddev *mddev, sector_t sectors)
|
| 4133 | {
|
| 4134 | |
| 4135 | * For 'near' and 'offset' arrays we can set the
|
| 4136 | * number of sectors used to be an appropriate multiple
|
| 4137 | * of the chunk size.
|
| 4138 | * For 'offset', this is far_copies*chunksize.
|
| 4139 | * For 'near' the multiplier is the LCM of
|
| 4140 | * near_copies and raid_disks.
|
| 4141 | * So if far_copies > 1 && !far_offset, fail.
|
| 4142 | * Else find LCM(raid_disks, near_copy)*far_copies and
|
| 4143 | * multiply by chunk_size. Then round to this number.
|
| 4144 | * This is mostly done by raid10_size()
|
| 4145 |
|
| 4146 | struct r10conf *conf = mddev->private;
|
| 4147 | sector_t oldsize, size;
|
| 4148 |
|
| 4149 | if (mddev->reshape_position != MaxSector)
|
| 4150 | return -EBUSY;
|
| 4151 |
|
| 4152 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset)
|
| 4153 | return -EINVAL;
|
| 4154 |
|
| 4155 | oldsize = raid10_size(mddev, 0, 0);
|
| 4156 | size = raid10_size(mddev, sectors, 0);
|
| 4157 | if (mddev->external_size &&
|
| 4158 | mddev->array_sectors > size)
|
| 4159 | return -EINVAL;
|
| 4160 |
|
| 4161 | if (md_bitmap_enabled(mddev, false)) {
|
| 4162 | int ret = mddev->bitmap_ops->resize(mddev, size, 0);
|
| 4163 |
|
| 4164 | if (ret)
|
| 4165 | return ret;
|
| 4166 | }
|
| 4167 |
|
| 4168 | md_set_array_sectors(mddev, size);
|
| 4169 | if (sectors > mddev->dev_sectors &&
|
| 4170 | mddev->resync_offset > oldsize) {
|
| 4171 | mddev->resync_offset = oldsize;
|
| 4172 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
|
| 4173 | }
|
| 4174 | calc_sectors(conf, sectors);
|
| 4175 | mddev->dev_sectors = conf->dev_sectors;
|
| 4176 | mddev->resync_max_sectors = size;
|
| 4177 | return 0;
|
| 4178 | }
|
| 4179 |
|
| 4180 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs)
|
| 4181 | {
|
| 4182 | struct md_rdev *rdev;
|
| 4183 | struct r10conf *conf;
|
| 4184 |
|
| 4185 | if (mddev->degraded > 0) {
|
| 4186 | pr_warn("md/raid10:%s: Error: degraded raid0!\n",
|
| 4187 | mdname(mddev));
|
| 4188 | return ERR_PTR(-EINVAL);
|
| 4189 | }
|
| 4190 | sector_div(size, devs);
|
| 4191 |
|
| 4192 |
|
| 4193 | mddev->new_level = 10;
|
| 4194 |
|
| 4195 | mddev->new_layout = (1<<8) + 2;
|
| 4196 | mddev->new_chunk_sectors = mddev->chunk_sectors;
|
| 4197 | mddev->delta_disks = mddev->raid_disks;
|
| 4198 | mddev->raid_disks *= 2;
|
| 4199 |
|
| 4200 | mddev->resync_offset = MaxSector;
|
| 4201 | mddev->dev_sectors = size;
|
| 4202 |
|
| 4203 | conf = setup_conf(mddev);
|
| 4204 | if (!IS_ERR(conf)) {
|
| 4205 | rdev_for_each(rdev, mddev)
|
| 4206 | if (rdev->raid_disk >= 0) {
|
| 4207 | rdev->new_raid_disk = rdev->raid_disk * 2;
|
| 4208 | rdev->sectors = size;
|
| 4209 | }
|
| 4210 | }
|
| 4211 |
|
| 4212 | return conf;
|
| 4213 | }
|
| 4214 |
|
| 4215 | static void *raid10_takeover(struct mddev *mddev)
|
| 4216 | {
|
| 4217 | struct r0conf *raid0_conf;
|
| 4218 |
|
| 4219 | |
| 4220 | * raid0 - providing it has only two drives
|
| 4221 |
|
| 4222 | if (mddev->level == 0) {
|
| 4223 |
|
| 4224 | raid0_conf = mddev->private;
|
| 4225 | if (raid0_conf->nr_strip_zones > 1) {
|
| 4226 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n",
|
| 4227 | mdname(mddev));
|
| 4228 | return ERR_PTR(-EINVAL);
|
| 4229 | }
|
| 4230 | return raid10_takeover_raid0(mddev,
|
| 4231 | raid0_conf->strip_zone->zone_end,
|
| 4232 | raid0_conf->strip_zone->nb_dev);
|
| 4233 | }
|
| 4234 | return ERR_PTR(-EINVAL);
|
| 4235 | }
|
| 4236 |
|
| 4237 | static int raid10_check_reshape(struct mddev *mddev)
|
| 4238 | {
|
| 4239 | |
| 4240 | * - layout (to ->new_layout)
|
| 4241 | * - chunk size (to ->new_chunk_sectors)
|
| 4242 | * - raid_disks (by delta_disks)
|
| 4243 | * or when trying to restart a reshape that was ongoing.
|
| 4244 | *
|
| 4245 | * We need to validate the request and possibly allocate
|
| 4246 | * space if that might be an issue later.
|
| 4247 | *
|
| 4248 | * Currently we reject any reshape of a 'far' mode array,
|
| 4249 | * allow chunk size to change if new is generally acceptable,
|
| 4250 | * allow raid_disks to increase, and allow
|
| 4251 | * a switch between 'near' mode and 'offset' mode.
|
| 4252 |
|
| 4253 | struct r10conf *conf = mddev->private;
|
| 4254 | struct geom geo;
|
| 4255 |
|
| 4256 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset)
|
| 4257 | return -EINVAL;
|
| 4258 | if (mddev->bitmap_id == ID_LLBITMAP &&
|
| 4259 | mddev->new_chunk_sectors &&
|
| 4260 | mddev->new_chunk_sectors < mddev->chunk_sectors)
|
| 4261 | return -EOPNOTSUPP;
|
| 4262 |
|
| 4263 | if (setup_geo(&geo, mddev, geo_start) != conf->copies)
|
| 4264 |
|
| 4265 | return -EINVAL;
|
| 4266 | if (geo.far_copies > 1 && !geo.far_offset)
|
| 4267 |
|
| 4268 | return -EINVAL;
|
| 4269 |
|
| 4270 | if (mddev->array_sectors & geo.chunk_mask)
|
| 4271 |
|
| 4272 | return -EINVAL;
|
| 4273 |
|
| 4274 | if (!enough(conf, -1))
|
| 4275 | return -EINVAL;
|
| 4276 |
|
| 4277 | kfree(conf->mirrors_new);
|
| 4278 | conf->mirrors_new = NULL;
|
| 4279 | if (mddev->delta_disks > 0) {
|
| 4280 |
|
| 4281 | conf->mirrors_new =
|
| 4282 | kzalloc_objs(struct raid10_info,
|
| 4283 | mddev->raid_disks + mddev->delta_disks);
|
| 4284 | if (!conf->mirrors_new)
|
| 4285 | return -ENOMEM;
|
| 4286 | }
|
| 4287 | return 0;
|
| 4288 | }
|
| 4289 |
|
| 4290 | |
| 4291 | * Need to check if array has failed when deciding whether to:
|
| 4292 | * - start an array
|
| 4293 | * - remove non-faulty devices
|
| 4294 | * - add a spare
|
| 4295 | * - allow a reshape
|
| 4296 | * This determination is simple when no reshape is happening.
|
| 4297 | * However if there is a reshape, we need to carefully check
|
| 4298 | * both the before and after sections.
|
| 4299 | * This is because some failed devices may only affect one
|
| 4300 | * of the two sections, and some non-in_sync devices may
|
| 4301 | * be insync in the section most affected by failed devices.
|
| 4302 |
|
| 4303 | static int calc_degraded(struct r10conf *conf)
|
| 4304 | {
|
| 4305 | int degraded, degraded2;
|
| 4306 | int i;
|
| 4307 |
|
| 4308 | degraded = 0;
|
| 4309 |
|
| 4310 | for (i = 0; i < conf->prev.raid_disks; i++) {
|
| 4311 | struct md_rdev *rdev = conf->mirrors[i].rdev;
|
| 4312 |
|
| 4313 | if (!rdev || test_bit(Faulty, &rdev->flags))
|
| 4314 | degraded++;
|
| 4315 | else if (!test_bit(In_sync, &rdev->flags))
|
| 4316 | |
| 4317 | * an array, this might not contribute to
|
| 4318 | * 'degraded'. It does now.
|
| 4319 |
|
| 4320 | degraded++;
|
| 4321 | }
|
| 4322 | if (conf->geo.raid_disks == conf->prev.raid_disks)
|
| 4323 | return degraded;
|
| 4324 | degraded2 = 0;
|
| 4325 | for (i = 0; i < conf->geo.raid_disks; i++) {
|
| 4326 | struct md_rdev *rdev = conf->mirrors[i].rdev;
|
| 4327 |
|
| 4328 | if (!rdev || test_bit(Faulty, &rdev->flags))
|
| 4329 | degraded2++;
|
| 4330 | else if (!test_bit(In_sync, &rdev->flags)) {
|
| 4331 | |
| 4332 | * this section has already been recovered, so
|
| 4333 | * it doesn't contribute to degraded.
|
| 4334 | * else it does.
|
| 4335 |
|
| 4336 | if (conf->geo.raid_disks <= conf->prev.raid_disks)
|
| 4337 | degraded2++;
|
| 4338 | }
|
| 4339 | }
|
| 4340 | if (degraded2 > degraded)
|
| 4341 | return degraded2;
|
| 4342 | return degraded;
|
| 4343 | }
|
| 4344 |
|
| 4345 | static int raid10_start_reshape(struct mddev *mddev)
|
| 4346 | {
|
| 4347 | |
| 4348 | * the various 'new' fields and sets MD_RECOVER_RESHAPE
|
| 4349 | * This also checks if there are enough spares and adds them
|
| 4350 | * to the array.
|
| 4351 | * We currently require enough spares to make the final
|
| 4352 | * array non-degraded. We also require that the difference
|
| 4353 | * between old and new data_offset - on each device - is
|
| 4354 | * enough that we never risk over-writing.
|
| 4355 |
|
| 4356 |
|
| 4357 | unsigned long before_length, after_length;
|
| 4358 | sector_t min_offset_diff = 0;
|
| 4359 | int first = 1;
|
| 4360 | struct geom new;
|
| 4361 | struct r10conf *conf = mddev->private;
|
| 4362 | struct md_rdev *rdev;
|
| 4363 | int spares = 0;
|
| 4364 | int ret;
|
| 4365 | mempool_t *new_pool = NULL;
|
| 4366 |
|
| 4367 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
|
| 4368 | return -EBUSY;
|
| 4369 | if (md_bitmap_enabled(mddev, false) &&
|
| 4370 | mddev->bitmap_ops->reshape_can_start) {
|
| 4371 | ret = mddev->bitmap_ops->reshape_can_start(mddev);
|
| 4372 | if (ret)
|
| 4373 | return ret;
|
| 4374 | }
|
| 4375 |
|
| 4376 | if (setup_geo(&new, mddev, geo_start) != conf->copies)
|
| 4377 | return -EINVAL;
|
| 4378 |
|
| 4379 | before_length = ((1 << conf->prev.chunk_shift) *
|
| 4380 | conf->prev.far_copies);
|
| 4381 | after_length = ((1 << conf->geo.chunk_shift) *
|
| 4382 | conf->geo.far_copies);
|
| 4383 |
|
| 4384 | rdev_for_each(rdev, mddev) {
|
| 4385 | if (!test_bit(In_sync, &rdev->flags)
|
| 4386 | && !test_bit(Faulty, &rdev->flags))
|
| 4387 | spares++;
|
| 4388 | if (rdev->raid_disk >= 0) {
|
| 4389 | long long diff = (rdev->new_data_offset
|
| 4390 | - rdev->data_offset);
|
| 4391 | if (!mddev->reshape_backwards)
|
| 4392 | diff = -diff;
|
| 4393 | if (diff < 0)
|
| 4394 | diff = 0;
|
| 4395 | if (first || diff < min_offset_diff)
|
| 4396 | min_offset_diff = diff;
|
| 4397 | first = 0;
|
| 4398 | }
|
| 4399 | }
|
| 4400 |
|
| 4401 | if (max(before_length, after_length) > min_offset_diff)
|
| 4402 | return -EINVAL;
|
| 4403 |
|
| 4404 | if (spares < mddev->delta_disks)
|
| 4405 | return -EINVAL;
|
| 4406 |
|
| 4407 | conf->offset_diff = min_offset_diff;
|
| 4408 | if (mddev->delta_disks > 0) {
|
| 4409 | new_pool = create_r10bio_pool(new.raid_disks);
|
| 4410 | if (!new_pool)
|
| 4411 | return -ENOMEM;
|
| 4412 | }
|
| 4413 | spin_lock_irq(&conf->device_lock);
|
| 4414 | if (conf->mirrors_new) {
|
| 4415 | memcpy(conf->mirrors_new, conf->mirrors,
|
| 4416 | sizeof(struct raid10_info)*conf->prev.raid_disks);
|
| 4417 | smp_mb();
|
| 4418 | kfree(conf->mirrors_old);
|
| 4419 | conf->mirrors_old = conf->mirrors;
|
| 4420 | conf->mirrors = conf->mirrors_new;
|
| 4421 | conf->mirrors_new = NULL;
|
| 4422 | }
|
| 4423 | setup_geo(&conf->geo, mddev, geo_start);
|
| 4424 | smp_mb();
|
| 4425 | if (mddev->reshape_backwards) {
|
| 4426 | sector_t size = raid10_size(mddev, 0, 0);
|
| 4427 | if (size < mddev->array_sectors) {
|
| 4428 | spin_unlock_irq(&conf->device_lock);
|
| 4429 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n",
|
| 4430 | mdname(mddev));
|
| 4431 | return -EINVAL;
|
| 4432 | }
|
| 4433 | mddev->resync_max_sectors = size;
|
| 4434 | conf->reshape_progress = size;
|
| 4435 | } else
|
| 4436 | conf->reshape_progress = 0;
|
| 4437 | conf->reshape_safe = conf->reshape_progress;
|
| 4438 | spin_unlock_irq(&conf->device_lock);
|
| 4439 |
|
| 4440 | if (mddev->delta_disks && mddev->bitmap) {
|
| 4441 | struct mdp_superblock_1 *sb = NULL;
|
| 4442 | sector_t oldsize, newsize;
|
| 4443 |
|
| 4444 | oldsize = raid10_size(mddev, 0, 0);
|
| 4445 | newsize = raid10_size(mddev, 0, conf->geo.raid_disks);
|
| 4446 |
|
| 4447 | if (!mddev_is_clustered(mddev) &&
|
| 4448 | md_bitmap_enabled(mddev, false)) {
|
| 4449 | ret = mddev->bitmap_ops->resize(mddev, newsize, 0);
|
| 4450 | if (ret)
|
| 4451 | goto abort;
|
| 4452 | else
|
| 4453 | goto out;
|
| 4454 | }
|
| 4455 |
|
| 4456 | rdev_for_each(rdev, mddev) {
|
| 4457 | if (rdev->raid_disk > -1 &&
|
| 4458 | !test_bit(Faulty, &rdev->flags))
|
| 4459 | sb = page_address(rdev->sb_page);
|
| 4460 | }
|
| 4461 |
|
| 4462 | |
| 4463 | * some node is already performing reshape, and no need to
|
| 4464 | * call bitmap_ops->resize again since it should be called when
|
| 4465 | * receiving BITMAP_RESIZE msg
|
| 4466 |
|
| 4467 | if ((sb && (le32_to_cpu(sb->feature_map) &
|
| 4468 | MD_FEATURE_RESHAPE_ACTIVE)) || (oldsize == newsize))
|
| 4469 | goto out;
|
| 4470 |
|
| 4471 |
|
| 4472 | ret = mddev->bitmap_ops->resize(mddev, newsize, 0);
|
| 4473 | if (ret)
|
| 4474 | goto abort;
|
| 4475 |
|
| 4476 | ret = mddev->cluster_ops->resize_bitmaps(mddev, newsize, oldsize);
|
| 4477 | if (ret) {
|
| 4478 | mddev->bitmap_ops->resize(mddev, oldsize, 0);
|
| 4479 | goto abort;
|
| 4480 | }
|
| 4481 | }
|
| 4482 | out:
|
| 4483 | if (mddev->delta_disks > 0) {
|
| 4484 | rdev_for_each(rdev, mddev)
|
| 4485 | if (rdev->raid_disk < 0 &&
|
| 4486 | !test_bit(Faulty, &rdev->flags)) {
|
| 4487 | if (raid10_add_disk(mddev, rdev) == 0) {
|
| 4488 | if (rdev->raid_disk >=
|
| 4489 | conf->prev.raid_disks)
|
| 4490 | set_bit(In_sync, &rdev->flags);
|
| 4491 | else
|
| 4492 | rdev->recovery_offset = 0;
|
| 4493 |
|
| 4494 |
|
| 4495 | sysfs_link_rdev(mddev, rdev);
|
| 4496 | }
|
| 4497 | } else if (rdev->raid_disk >= conf->prev.raid_disks
|
| 4498 | && !test_bit(Faulty, &rdev->flags)) {
|
| 4499 |
|
| 4500 | set_bit(In_sync, &rdev->flags);
|
| 4501 | }
|
| 4502 | }
|
| 4503 | |
| 4504 | * ->degraded is measured against the larger of the
|
| 4505 | * pre and post numbers.
|
| 4506 |
|
| 4507 | spin_lock_irq(&conf->device_lock);
|
| 4508 | mddev->degraded = calc_degraded(conf);
|
| 4509 | spin_unlock_irq(&conf->device_lock);
|
| 4510 | mddev->raid_disks = conf->geo.raid_disks;
|
| 4511 | mddev->reshape_position = conf->reshape_progress;
|
| 4512 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
|
| 4513 | if (new_pool) {
|
| 4514 | mempool_destroy(conf->r10bio_pool);
|
| 4515 | conf->r10bio_pool = new_pool;
|
| 4516 | }
|
| 4517 |
|
| 4518 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
|
| 4519 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
|
| 4520 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
|
| 4521 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
|
| 4522 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
|
| 4523 | conf->reshape_checkpoint = jiffies;
|
| 4524 | md_new_event();
|
| 4525 | return 0;
|
| 4526 |
|
| 4527 | abort:
|
| 4528 | mddev->recovery = 0;
|
| 4529 | spin_lock_irq(&conf->device_lock);
|
| 4530 | conf->geo = conf->prev;
|
| 4531 | mddev->raid_disks = conf->geo.raid_disks;
|
| 4532 | rdev_for_each(rdev, mddev)
|
| 4533 | rdev->new_data_offset = rdev->data_offset;
|
| 4534 | smp_wmb();
|
| 4535 | conf->reshape_progress = MaxSector;
|
| 4536 | conf->reshape_safe = MaxSector;
|
| 4537 | mddev->reshape_position = MaxSector;
|
| 4538 | spin_unlock_irq(&conf->device_lock);
|
| 4539 | mempool_destroy(new_pool);
|
| 4540 | return ret;
|
| 4541 | }
|
| 4542 |
|
| 4543 | |
| 4544 | * any block from the chunk that includes the array-address 's'
|
| 4545 | * and report the next address.
|
| 4546 | * i.e. the address returned will be chunk-aligned and after
|
| 4547 | * any data that is in the chunk containing 's'.
|
| 4548 |
|
| 4549 | static sector_t last_dev_address(sector_t s, struct geom *geo)
|
| 4550 | {
|
| 4551 | s = (s | geo->chunk_mask) + 1;
|
| 4552 | s >>= geo->chunk_shift;
|
| 4553 | s *= geo->near_copies;
|
| 4554 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks);
|
| 4555 | s *= geo->far_copies;
|
| 4556 | s <<= geo->chunk_shift;
|
| 4557 | return s;
|
| 4558 | }
|
| 4559 |
|
| 4560 | |
| 4561 | * any block from the chunk that includes the array-address 's'.
|
| 4562 | * This too will be the start of a chunk
|
| 4563 |
|
| 4564 | static sector_t first_dev_address(sector_t s, struct geom *geo)
|
| 4565 | {
|
| 4566 | s >>= geo->chunk_shift;
|
| 4567 | s *= geo->near_copies;
|
| 4568 | sector_div(s, geo->raid_disks);
|
| 4569 | s *= geo->far_copies;
|
| 4570 | s <<= geo->chunk_shift;
|
| 4571 | return s;
|
| 4572 | }
|
| 4573 |
|
| 4574 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
|
| 4575 | int *skipped)
|
| 4576 | {
|
| 4577 | |
| 4578 | * at a time, possibly less if that exceeds RESYNC_PAGES,
|
| 4579 | * or we hit a bad block or something.
|
| 4580 | * This might mean we pause for normal IO in the middle of
|
| 4581 | * a chunk, but that is not a problem as mddev->reshape_position
|
| 4582 | * can record any location.
|
| 4583 | *
|
| 4584 | * If we will want to write to a location that isn't
|
| 4585 | * yet recorded as 'safe' (i.e. in metadata on disk) then
|
| 4586 | * we need to flush all reshape requests and update the metadata.
|
| 4587 | *
|
| 4588 | * When reshaping forwards (e.g. to more devices), we interpret
|
| 4589 | * 'safe' as the earliest block which might not have been copied
|
| 4590 | * down yet. We divide this by previous stripe size and multiply
|
| 4591 | * by previous stripe length to get lowest device offset that we
|
| 4592 | * cannot write to yet.
|
| 4593 | * We interpret 'sector_nr' as an address that we want to write to.
|
| 4594 | * From this we use last_device_address() to find where we might
|
| 4595 | * write to, and first_device_address on the 'safe' position.
|
| 4596 | * If this 'next' write position is after the 'safe' position,
|
| 4597 | * we must update the metadata to increase the 'safe' position.
|
| 4598 | *
|
| 4599 | * When reshaping backwards, we round in the opposite direction
|
| 4600 | * and perform the reverse test: next write position must not be
|
| 4601 | * less than current safe position.
|
| 4602 | *
|
| 4603 | * In all this the minimum difference in data offsets
|
| 4604 | * (conf->offset_diff - always positive) allows a bit of slack,
|
| 4605 | * so next can be after 'safe', but not by more than offset_diff
|
| 4606 | *
|
| 4607 | * We need to prepare all the bios here before we start any IO
|
| 4608 | * to ensure the size we choose is acceptable to all devices.
|
| 4609 | * The means one for each copy for write-out and an extra one for
|
| 4610 | * read-in.
|
| 4611 | * We store the read-in bio in ->master_bio and the others in
|
| 4612 | * ->devs[x].bio and ->devs[x].repl_bio.
|
| 4613 |
|
| 4614 | struct r10conf *conf = mddev->private;
|
| 4615 | struct r10bio *r10_bio;
|
| 4616 | sector_t next, safe, last;
|
| 4617 | int max_sectors;
|
| 4618 | int nr_sectors;
|
| 4619 | int s;
|
| 4620 | struct md_rdev *rdev;
|
| 4621 | int need_flush = 0;
|
| 4622 | struct bio *blist;
|
| 4623 | struct bio *bio, *read_bio;
|
| 4624 | int sectors_done = 0;
|
| 4625 | struct page **pages;
|
| 4626 |
|
| 4627 | if (sector_nr == 0) {
|
| 4628 |
|
| 4629 | if (mddev->reshape_backwards &&
|
| 4630 | conf->reshape_progress < raid10_size(mddev, 0, 0)) {
|
| 4631 | sector_nr = (raid10_size(mddev, 0, 0)
|
| 4632 | - conf->reshape_progress);
|
| 4633 | } else if (!mddev->reshape_backwards &&
|
| 4634 | conf->reshape_progress > 0)
|
| 4635 | sector_nr = conf->reshape_progress;
|
| 4636 | if (sector_nr) {
|
| 4637 | mddev->curr_resync_completed = sector_nr;
|
| 4638 | sysfs_notify_dirent_safe(mddev->sysfs_completed);
|
| 4639 | *skipped = 1;
|
| 4640 | return sector_nr;
|
| 4641 | }
|
| 4642 | }
|
| 4643 |
|
| 4644 | |
| 4645 | * as that doesn't work well for ->reshape_backwards.
|
| 4646 | * So just use ->reshape_progress.
|
| 4647 |
|
| 4648 | if (mddev->reshape_backwards) {
|
| 4649 | |
| 4650 | * write to for this chunk in the new layout
|
| 4651 |
|
| 4652 | next = first_dev_address(conf->reshape_progress - 1,
|
| 4653 | &conf->geo);
|
| 4654 |
|
| 4655 | |
| 4656 | * in the old layout after a restart
|
| 4657 |
|
| 4658 | safe = last_dev_address(conf->reshape_safe - 1,
|
| 4659 | &conf->prev);
|
| 4660 |
|
| 4661 | if (next + conf->offset_diff < safe)
|
| 4662 | need_flush = 1;
|
| 4663 |
|
| 4664 | last = conf->reshape_progress - 1;
|
| 4665 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask
|
| 4666 | & conf->prev.chunk_mask);
|
| 4667 | if (sector_nr + RESYNC_SECTORS < last)
|
| 4668 | sector_nr = last + 1 - RESYNC_SECTORS;
|
| 4669 | } else {
|
| 4670 | |
| 4671 | * might write to for this chunk in the new layout
|
| 4672 |
|
| 4673 | next = last_dev_address(conf->reshape_progress, &conf->geo);
|
| 4674 |
|
| 4675 | |
| 4676 | * read from in the old layout after a restart
|
| 4677 |
|
| 4678 | safe = first_dev_address(conf->reshape_safe, &conf->prev);
|
| 4679 |
|
| 4680 | |
| 4681 | * as that would possibly corrupt data
|
| 4682 |
|
| 4683 | if (next > safe + conf->offset_diff)
|
| 4684 | need_flush = 1;
|
| 4685 |
|
| 4686 | sector_nr = conf->reshape_progress;
|
| 4687 | last = sector_nr | (conf->geo.chunk_mask
|
| 4688 | & conf->prev.chunk_mask);
|
| 4689 |
|
| 4690 | if (sector_nr + RESYNC_SECTORS <= last)
|
| 4691 | last = sector_nr + RESYNC_SECTORS - 1;
|
| 4692 | }
|
| 4693 |
|
| 4694 | if (need_flush ||
|
| 4695 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
|
| 4696 |
|
| 4697 | wait_barrier(conf);
|
| 4698 | if (md_bitmap_enabled(mddev, false) &&
|
| 4699 | mddev->bitmap_ops->reshape_mark &&
|
| 4700 | conf->reshape_safe != conf->reshape_progress) {
|
| 4701 | mddev->bitmap_ops->reshape_mark(mddev, conf->reshape_safe,
|
| 4702 | conf->reshape_progress);
|
| 4703 | mddev->bitmap_ops->unplug(mddev, true);
|
| 4704 | }
|
| 4705 | mddev->reshape_position = conf->reshape_progress;
|
| 4706 | if (mddev->reshape_backwards)
|
| 4707 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0)
|
| 4708 | - conf->reshape_progress;
|
| 4709 | else
|
| 4710 | mddev->curr_resync_completed = conf->reshape_progress;
|
| 4711 | conf->reshape_checkpoint = jiffies;
|
| 4712 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
|
| 4713 | md_wakeup_thread(mddev->thread);
|
| 4714 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 ||
|
| 4715 | test_bit(MD_RECOVERY_INTR, &mddev->recovery));
|
| 4716 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
|
| 4717 | allow_barrier(conf);
|
| 4718 | return sectors_done;
|
| 4719 | }
|
| 4720 | conf->reshape_safe = mddev->reshape_position;
|
| 4721 | allow_barrier(conf);
|
| 4722 | }
|
| 4723 |
|
| 4724 | raise_barrier(conf, 0);
|
| 4725 | read_more:
|
| 4726 |
|
| 4727 | r10_bio = raid10_alloc_init_r10buf(conf);
|
| 4728 | r10_bio->state = 0;
|
| 4729 | raise_barrier(conf, 1);
|
| 4730 | atomic_set(&r10_bio->remaining, 0);
|
| 4731 | r10_bio->mddev = mddev;
|
| 4732 | r10_bio->sector = sector_nr;
|
| 4733 | set_bit(R10BIO_IsReshape, &r10_bio->state);
|
| 4734 | r10_bio->sectors = last - sector_nr + 1;
|
| 4735 | rdev = read_balance(conf, r10_bio, &max_sectors);
|
| 4736 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state));
|
| 4737 |
|
| 4738 | if (!rdev) {
|
| 4739 | |
| 4740 | * on all the target devices.
|
| 4741 |
|
| 4742 | |
| 4743 | mempool_free(r10_bio, &conf->r10buf_pool);
|
| 4744 | set_bit(MD_RECOVERY_INTR, &mddev->recovery);
|
| 4745 | return sectors_done;
|
| 4746 | }
|
| 4747 |
|
| 4748 | read_bio = bio_alloc_bioset(rdev->bdev, RESYNC_PAGES, REQ_OP_READ,
|
| 4749 | GFP_KERNEL, &mddev->bio_set);
|
| 4750 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr
|
| 4751 | + rdev->data_offset);
|
| 4752 | read_bio->bi_private = r10_bio;
|
| 4753 | read_bio->bi_end_io = end_reshape_read;
|
| 4754 | r10_bio->master_bio = read_bio;
|
| 4755 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum;
|
| 4756 |
|
| 4757 | |
| 4758 | * Broadcast RESYNC message to other nodes, so all nodes would not
|
| 4759 | * write to the region to avoid conflict.
|
| 4760 |
|
| 4761 | if (mddev_is_clustered(mddev) && conf->cluster_sync_high <= sector_nr) {
|
| 4762 | struct mdp_superblock_1 *sb = NULL;
|
| 4763 | int sb_reshape_pos = 0;
|
| 4764 |
|
| 4765 | conf->cluster_sync_low = sector_nr;
|
| 4766 | conf->cluster_sync_high = sector_nr + CLUSTER_RESYNC_WINDOW_SECTORS;
|
| 4767 | sb = page_address(rdev->sb_page);
|
| 4768 | if (sb) {
|
| 4769 | sb_reshape_pos = le64_to_cpu(sb->reshape_position);
|
| 4770 | |
| 4771 | * Set cluster_sync_low again if next address for array
|
| 4772 | * reshape is less than cluster_sync_low. Since we can't
|
| 4773 | * update cluster_sync_low until it has finished reshape.
|
| 4774 |
|
| 4775 | if (sb_reshape_pos < conf->cluster_sync_low)
|
| 4776 | conf->cluster_sync_low = sb_reshape_pos;
|
| 4777 | }
|
| 4778 |
|
| 4779 | mddev->cluster_ops->resync_info_update(mddev, conf->cluster_sync_low,
|
| 4780 | conf->cluster_sync_high);
|
| 4781 | }
|
| 4782 |
|
| 4783 |
|
| 4784 | __raid10_find_phys(&conf->geo, r10_bio);
|
| 4785 |
|
| 4786 | blist = read_bio;
|
| 4787 | read_bio->bi_next = NULL;
|
| 4788 |
|
| 4789 | for (s = 0; s < conf->copies*2; s++) {
|
| 4790 | struct bio *b;
|
| 4791 | int d = r10_bio->devs[s/2].devnum;
|
| 4792 | struct md_rdev *rdev2;
|
| 4793 | if (s&1) {
|
| 4794 | rdev2 = conf->mirrors[d].replacement;
|
| 4795 | b = r10_bio->devs[s/2].repl_bio;
|
| 4796 | } else {
|
| 4797 | rdev2 = conf->mirrors[d].rdev;
|
| 4798 | b = r10_bio->devs[s/2].bio;
|
| 4799 | }
|
| 4800 | if (!rdev2 || test_bit(Faulty, &rdev2->flags))
|
| 4801 | continue;
|
| 4802 |
|
| 4803 | bio_set_dev(b, rdev2->bdev);
|
| 4804 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr +
|
| 4805 | rdev2->new_data_offset;
|
| 4806 | b->bi_end_io = end_reshape_write;
|
| 4807 | b->bi_opf = REQ_OP_WRITE;
|
| 4808 | b->bi_next = blist;
|
| 4809 | blist = b;
|
| 4810 | }
|
| 4811 |
|
| 4812 |
|
| 4813 |
|
| 4814 | nr_sectors = 0;
|
| 4815 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages;
|
| 4816 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) {
|
| 4817 | struct page *page = pages[s / (PAGE_SIZE >> 9)];
|
| 4818 | int len = (max_sectors - s) << 9;
|
| 4819 | if (len > PAGE_SIZE)
|
| 4820 | len = PAGE_SIZE;
|
| 4821 | for (bio = blist; bio ; bio = bio->bi_next) {
|
| 4822 | if (WARN_ON(!bio_add_page(bio, page, len, 0))) {
|
| 4823 | bio->bi_status = BLK_STS_RESOURCE;
|
| 4824 | bio_endio(bio);
|
| 4825 | return sectors_done;
|
| 4826 | }
|
| 4827 | }
|
| 4828 | sector_nr += len >> 9;
|
| 4829 | nr_sectors += len >> 9;
|
| 4830 | }
|
| 4831 | r10_bio->sectors = nr_sectors;
|
| 4832 |
|
| 4833 |
|
| 4834 | atomic_inc(&r10_bio->remaining);
|
| 4835 | read_bio->bi_next = NULL;
|
| 4836 | submit_bio_noacct(read_bio);
|
| 4837 | sectors_done += nr_sectors;
|
| 4838 | if (sector_nr <= last)
|
| 4839 | goto read_more;
|
| 4840 |
|
| 4841 | lower_barrier(conf);
|
| 4842 |
|
| 4843 | |
| 4844 | * update reshape_progress
|
| 4845 |
|
| 4846 | if (mddev->reshape_backwards)
|
| 4847 | conf->reshape_progress -= sectors_done;
|
| 4848 | else
|
| 4849 | conf->reshape_progress += sectors_done;
|
| 4850 |
|
| 4851 | return sectors_done;
|
| 4852 | }
|
| 4853 |
|
| 4854 | static void end_reshape_request(struct r10bio *r10_bio);
|
| 4855 | static int handle_reshape_read_error(struct mddev *mddev,
|
| 4856 | struct r10bio *r10_bio);
|
| 4857 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio)
|
| 4858 | {
|
| 4859 | |
| 4860 | * to write out.
|
| 4861 | * If we got a read error then we do sync 1-page reads from
|
| 4862 | * elsewhere until we find the data - or give up.
|
| 4863 |
|
| 4864 | struct r10conf *conf = mddev->private;
|
| 4865 | int s;
|
| 4866 |
|
| 4867 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state))
|
| 4868 | if (handle_reshape_read_error(mddev, r10_bio) < 0) {
|
| 4869 |
|
| 4870 | md_done_sync(mddev, r10_bio->sectors);
|
| 4871 | md_sync_error(mddev);
|
| 4872 | return;
|
| 4873 | }
|
| 4874 |
|
| 4875 | |
| 4876 | * writes.
|
| 4877 |
|
| 4878 | atomic_set(&r10_bio->remaining, 1);
|
| 4879 | for (s = 0; s < conf->copies*2; s++) {
|
| 4880 | struct bio *b;
|
| 4881 | int d = r10_bio->devs[s/2].devnum;
|
| 4882 | struct md_rdev *rdev;
|
| 4883 | if (s&1) {
|
| 4884 | rdev = conf->mirrors[d].replacement;
|
| 4885 | b = r10_bio->devs[s/2].repl_bio;
|
| 4886 | } else {
|
| 4887 | rdev = conf->mirrors[d].rdev;
|
| 4888 | b = r10_bio->devs[s/2].bio;
|
| 4889 | }
|
| 4890 | if (!rdev || test_bit(Faulty, &rdev->flags))
|
| 4891 | continue;
|
| 4892 |
|
| 4893 | atomic_inc(&rdev->nr_pending);
|
| 4894 | atomic_inc(&r10_bio->remaining);
|
| 4895 | b->bi_next = NULL;
|
| 4896 | submit_bio_noacct(b);
|
| 4897 | }
|
| 4898 | end_reshape_request(r10_bio);
|
| 4899 | }
|
| 4900 |
|
| 4901 | static void end_reshape(struct r10conf *conf)
|
| 4902 | {
|
| 4903 | struct mddev *mddev = conf->mddev;
|
| 4904 |
|
| 4905 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery))
|
| 4906 | return;
|
| 4907 |
|
| 4908 | if (md_bitmap_enabled(mddev, false) &&
|
| 4909 | mddev->bitmap_ops->reshape_mark &&
|
| 4910 | conf->reshape_safe != conf->reshape_progress) {
|
| 4911 | mddev->bitmap_ops->reshape_mark(mddev, conf->reshape_safe,
|
| 4912 | conf->reshape_progress);
|
| 4913 | mddev->bitmap_ops->unplug(mddev, true);
|
| 4914 | }
|
| 4915 |
|
| 4916 | spin_lock_irq(&conf->device_lock);
|
| 4917 | conf->prev = conf->geo;
|
| 4918 | md_finish_reshape(conf->mddev);
|
| 4919 | smp_wmb();
|
| 4920 | conf->reshape_progress = MaxSector;
|
| 4921 | conf->reshape_safe = MaxSector;
|
| 4922 | spin_unlock_irq(&conf->device_lock);
|
| 4923 |
|
| 4924 | mddev_update_io_opt(conf->mddev, raid10_nr_stripes(conf));
|
| 4925 | conf->fullsync = 0;
|
| 4926 | }
|
| 4927 |
|
| 4928 | static void raid10_update_reshape_pos(struct mddev *mddev)
|
| 4929 | {
|
| 4930 | struct r10conf *conf = mddev->private;
|
| 4931 | sector_t lo, hi;
|
| 4932 |
|
| 4933 | mddev->cluster_ops->resync_info_get(mddev, &lo, &hi);
|
| 4934 | if (((mddev->reshape_position <= hi) && (mddev->reshape_position >= lo))
|
| 4935 | || mddev->reshape_position == MaxSector)
|
| 4936 | conf->reshape_progress = mddev->reshape_position;
|
| 4937 | else
|
| 4938 | WARN_ON_ONCE(1);
|
| 4939 | }
|
| 4940 |
|
| 4941 | static int handle_reshape_read_error(struct mddev *mddev,
|
| 4942 | struct r10bio *r10_bio)
|
| 4943 | {
|
| 4944 |
|
| 4945 | int sectors = r10_bio->sectors;
|
| 4946 | struct r10conf *conf = mddev->private;
|
| 4947 | struct r10bio *r10b;
|
| 4948 | int slot = 0;
|
| 4949 | int idx = 0;
|
| 4950 | struct page **pages;
|
| 4951 |
|
| 4952 | r10b = kmalloc_flex(*r10b, devs, conf->copies, GFP_NOIO);
|
| 4953 | if (!r10b) {
|
| 4954 | set_bit(MD_RECOVERY_INTR, &mddev->recovery);
|
| 4955 | return -ENOMEM;
|
| 4956 | }
|
| 4957 |
|
| 4958 |
|
| 4959 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages;
|
| 4960 |
|
| 4961 | r10b->sector = r10_bio->sector;
|
| 4962 | __raid10_find_phys(&conf->prev, r10b);
|
| 4963 |
|
| 4964 | while (sectors) {
|
| 4965 | int s = sectors;
|
| 4966 | int success = 0;
|
| 4967 | int first_slot = slot;
|
| 4968 |
|
| 4969 | if (s > (PAGE_SIZE >> 9))
|
| 4970 | s = PAGE_SIZE >> 9;
|
| 4971 |
|
| 4972 | while (!success) {
|
| 4973 | int d = r10b->devs[slot].devnum;
|
| 4974 | struct md_rdev *rdev = conf->mirrors[d].rdev;
|
| 4975 | sector_t addr;
|
| 4976 | if (rdev == NULL ||
|
| 4977 | test_bit(Faulty, &rdev->flags) ||
|
| 4978 | !test_bit(In_sync, &rdev->flags))
|
| 4979 | goto failed;
|
| 4980 |
|
| 4981 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE;
|
| 4982 | atomic_inc(&rdev->nr_pending);
|
| 4983 | success = sync_page_io(rdev,
|
| 4984 | addr,
|
| 4985 | s << 9,
|
| 4986 | pages[idx],
|
| 4987 | REQ_OP_READ, false);
|
| 4988 | rdev_dec_pending(rdev, mddev);
|
| 4989 | if (success)
|
| 4990 | break;
|
| 4991 | failed:
|
| 4992 | slot++;
|
| 4993 | if (slot >= conf->copies)
|
| 4994 | slot = 0;
|
| 4995 | if (slot == first_slot)
|
| 4996 | break;
|
| 4997 | }
|
| 4998 | if (!success) {
|
| 4999 |
|
| 5000 | set_bit(MD_RECOVERY_INTR,
|
| 5001 | &mddev->recovery);
|
| 5002 | kfree(r10b);
|
| 5003 | return -EIO;
|
| 5004 | }
|
| 5005 | sectors -= s;
|
| 5006 | idx++;
|
| 5007 | }
|
| 5008 | kfree(r10b);
|
| 5009 | return 0;
|
| 5010 | }
|
| 5011 |
|
| 5012 | static void end_reshape_write(struct bio *bio)
|
| 5013 | {
|
| 5014 | struct r10bio *r10_bio = get_resync_r10bio(bio);
|
| 5015 | struct mddev *mddev = r10_bio->mddev;
|
| 5016 | struct r10conf *conf = mddev->private;
|
| 5017 | int d;
|
| 5018 | int slot;
|
| 5019 | int repl;
|
| 5020 | struct md_rdev *rdev = NULL;
|
| 5021 |
|
| 5022 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
|
| 5023 | rdev = repl ? conf->mirrors[d].replacement :
|
| 5024 | conf->mirrors[d].rdev;
|
| 5025 |
|
| 5026 | if (bio->bi_status) {
|
| 5027 |
|
| 5028 | md_error(mddev, rdev);
|
| 5029 | }
|
| 5030 |
|
| 5031 | rdev_dec_pending(rdev, mddev);
|
| 5032 | end_reshape_request(r10_bio);
|
| 5033 | }
|
| 5034 |
|
| 5035 | static void end_reshape_request(struct r10bio *r10_bio)
|
| 5036 | {
|
| 5037 | if (!atomic_dec_and_test(&r10_bio->remaining))
|
| 5038 | return;
|
| 5039 | md_done_sync(r10_bio->mddev, r10_bio->sectors);
|
| 5040 | bio_put(r10_bio->master_bio);
|
| 5041 | put_buf(r10_bio);
|
| 5042 | }
|
| 5043 |
|
| 5044 | static void raid10_finish_reshape(struct mddev *mddev)
|
| 5045 | {
|
| 5046 | struct r10conf *conf = mddev->private;
|
| 5047 | bool llbitmap = mddev->bitmap_id == ID_LLBITMAP &&
|
| 5048 | md_bitmap_enabled(mddev, false);
|
| 5049 |
|
| 5050 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
|
| 5051 | return;
|
| 5052 |
|
| 5053 | if (llbitmap && mddev->bitmap_ops->reshape_finish)
|
| 5054 | mddev->bitmap_ops->reshape_finish(mddev);
|
| 5055 |
|
| 5056 | if (mddev->delta_disks > 0) {
|
| 5057 | if (mddev->resync_offset > mddev->resync_max_sectors) {
|
| 5058 | mddev->resync_offset = mddev->resync_max_sectors;
|
| 5059 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
|
| 5060 | }
|
| 5061 | mddev->resync_max_sectors = mddev->array_sectors;
|
| 5062 | } else {
|
| 5063 | int d;
|
| 5064 | for (d = conf->geo.raid_disks ;
|
| 5065 | d < conf->geo.raid_disks - mddev->delta_disks;
|
| 5066 | d++) {
|
| 5067 | struct md_rdev *rdev = conf->mirrors[d].rdev;
|
| 5068 | if (rdev)
|
| 5069 | clear_bit(In_sync, &rdev->flags);
|
| 5070 | rdev = conf->mirrors[d].replacement;
|
| 5071 | if (rdev)
|
| 5072 | clear_bit(In_sync, &rdev->flags);
|
| 5073 | }
|
| 5074 | }
|
| 5075 | mddev->layout = mddev->new_layout;
|
| 5076 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift;
|
| 5077 | mddev->reshape_position = MaxSector;
|
| 5078 | mddev->delta_disks = 0;
|
| 5079 | mddev->reshape_backwards = 0;
|
| 5080 | }
|
| 5081 |
|
| 5082 | static sector_t raid10_bitmap_sync_size(struct mddev *mddev, bool previous)
|
| 5083 | {
|
| 5084 | struct r10conf *conf = mddev->private;
|
| 5085 |
|
| 5086 | if (previous)
|
| 5087 | return raid10_size(mddev, 0, 0);
|
| 5088 | return raid10_size(mddev, 0, conf->geo.raid_disks);
|
| 5089 | }
|
| 5090 |
|
| 5091 | static struct md_personality raid10_personality =
|
| 5092 | {
|
| 5093 | .head = {
|
| 5094 | .type = MD_PERSONALITY,
|
| 5095 | .id = ID_RAID10,
|
| 5096 | .name = "raid10",
|
| 5097 | .owner = THIS_MODULE,
|
| 5098 | },
|
| 5099 |
|
| 5100 | .make_request = raid10_make_request,
|
| 5101 | .run = raid10_run,
|
| 5102 | .free = raid10_free,
|
| 5103 | .status = raid10_status,
|
| 5104 | .error_handler = raid10_error,
|
| 5105 | .hot_add_disk = raid10_add_disk,
|
| 5106 | .hot_remove_disk= raid10_remove_disk,
|
| 5107 | .spare_active = raid10_spare_active,
|
| 5108 | .sync_request = raid10_sync_request,
|
| 5109 | .quiesce = raid10_quiesce,
|
| 5110 | .size = raid10_size,
|
| 5111 | .resize = raid10_resize,
|
| 5112 | .takeover = raid10_takeover,
|
| 5113 | .check_reshape = raid10_check_reshape,
|
| 5114 | .start_reshape = raid10_start_reshape,
|
| 5115 | .finish_reshape = raid10_finish_reshape,
|
| 5116 | .update_reshape_pos = raid10_update_reshape_pos,
|
| 5117 | .bitmap_sync_size = raid10_bitmap_sync_size,
|
| 5118 | .bitmap_array_sectors = raid10_bitmap_sync_size,
|
| 5119 | };
|
| 5120 |
|
| 5121 | static int __init raid10_init(void)
|
| 5122 | {
|
| 5123 | return register_md_submodule(&raid10_personality.head);
|
| 5124 | }
|
| 5125 |
|
| 5126 | static void __exit raid10_exit(void)
|
| 5127 | {
|
| 5128 | unregister_md_submodule(&raid10_personality.head);
|
| 5129 | }
|
| 5130 |
|
| 5131 | module_init(raid10_init);
|
| 5132 | module_exit(raid10_exit);
|
| 5133 | MODULE_LICENSE("GPL");
|
| 5134 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD");
|
| 5135 | MODULE_ALIAS("md-personality-9");
|
| 5136 | MODULE_ALIAS("md-raid10");
|
| 5137 | MODULE_ALIAS("md-level-10");
|