Commit | Line | Data |
---|---|---|
af1a8899 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
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 | * | |
25985edc | 9 | * Base on code in raid1.c. See raid1.c for further copyright information. |
1da177e4 LT |
10 | */ |
11 | ||
5a0e3ad6 | 12 | #include <linux/slab.h> |
25570727 | 13 | #include <linux/delay.h> |
bff61975 | 14 | #include <linux/blkdev.h> |
056075c7 | 15 | #include <linux/module.h> |
bff61975 | 16 | #include <linux/seq_file.h> |
8bda470e | 17 | #include <linux/ratelimit.h> |
3ea7daa5 | 18 | #include <linux/kthread.h> |
afd75628 | 19 | #include <linux/raid/md_p.h> |
109e3765 | 20 | #include <trace/events/block.h> |
43b2e5d8 | 21 | #include "md.h" |
ef740c37 | 22 | #include "raid10.h" |
dab8b292 | 23 | #include "raid0.h" |
935fe098 | 24 | #include "md-bitmap.h" |
1da177e4 LT |
25 | |
26 | /* | |
27 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
28 | * The layout of data is defined by | |
29 | * chunk_size | |
30 | * raid_disks | |
31 | * near_copies (stored in low byte of layout) | |
32 | * far_copies (stored in second byte of layout) | |
c93983bf | 33 | * far_offset (stored in bit 16 of layout ) |
475901af | 34 | * use_far_sets (stored in bit 17 of layout ) |
8bce6d35 | 35 | * use_far_sets_bugfixed (stored in bit 18 of layout ) |
1da177e4 | 36 | * |
475901af JB |
37 | * The data to be stored is divided into chunks using chunksize. Each device |
38 | * is divided into far_copies sections. In each section, chunks are laid out | |
39 | * in a style similar to raid0, but near_copies copies of each chunk is stored | |
40 | * (each on a different drive). The starting device for each section is offset | |
41 | * near_copies from the starting device of the previous section. Thus there | |
42 | * are (near_copies * far_copies) of each chunk, and each is on a different | |
43 | * drive. near_copies and far_copies must be at least one, and their product | |
44 | * is at most raid_disks. | |
c93983bf N |
45 | * |
46 | * If far_offset is true, then the far_copies are handled a bit differently. | |
475901af JB |
47 | * The copies are still in different stripes, but instead of being very far |
48 | * apart on disk, there are adjacent stripes. | |
49 | * | |
50 | * The far and offset algorithms are handled slightly differently if | |
51 | * 'use_far_sets' is true. In this case, the array's devices are grouped into | |
52 | * sets that are (near_copies * far_copies) in size. The far copied stripes | |
53 | * are still shifted by 'near_copies' devices, but this shifting stays confined | |
54 | * to the set rather than the entire array. This is done to improve the number | |
55 | * of device combinations that can fail without causing the array to fail. | |
56 | * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk | |
57 | * on a device): | |
58 | * A B C D A B C D E | |
59 | * ... ... | |
60 | * D A B C E A B C D | |
61 | * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s): | |
62 | * [A B] [C D] [A B] [C D E] | |
63 | * |...| |...| |...| | ... | | |
64 | * [B A] [D C] [B A] [E C D] | |
1da177e4 LT |
65 | */ |
66 | ||
e879a879 N |
67 | static void allow_barrier(struct r10conf *conf); |
68 | static void lower_barrier(struct r10conf *conf); | |
635f6416 | 69 | static int _enough(struct r10conf *conf, int previous, int ignore); |
1919cbb2 | 70 | static int enough(struct r10conf *conf, int ignore); |
3ea7daa5 N |
71 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, |
72 | int *skipped); | |
73 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio); | |
4246a0b6 | 74 | static void end_reshape_write(struct bio *bio); |
3ea7daa5 | 75 | static void end_reshape(struct r10conf *conf); |
0a27ec96 | 76 | |
578b54ad N |
77 | #define raid10_log(md, fmt, args...) \ |
78 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid10 " fmt, ##args); } while (0) | |
79 | ||
fb0eb5df ML |
80 | #include "raid1-10.c" |
81 | ||
f0250618 ML |
82 | /* |
83 | * for resync bio, r10bio pointer can be retrieved from the per-bio | |
84 | * 'struct resync_pages'. | |
85 | */ | |
86 | static inline struct r10bio *get_resync_r10bio(struct bio *bio) | |
87 | { | |
88 | return get_resync_pages(bio)->raid_bio; | |
89 | } | |
90 | ||
dd0fc66f | 91 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 92 | { |
e879a879 | 93 | struct r10conf *conf = data; |
c2968285 | 94 | int size = offsetof(struct r10bio, devs[conf->geo.raid_disks]); |
1da177e4 | 95 | |
69335ef3 N |
96 | /* allocate a r10bio with room for raid_disks entries in the |
97 | * bios array */ | |
7eaceacc | 98 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
99 | } |
100 | ||
8db87912 | 101 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
0310fa21 N |
102 | /* amount of memory to reserve for resync requests */ |
103 | #define RESYNC_WINDOW (1024*1024) | |
104 | /* maximum number of concurrent requests, memory permitting */ | |
105 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
4b242e97 | 106 | #define CLUSTER_RESYNC_WINDOW (32 * RESYNC_WINDOW) |
8db87912 | 107 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) |
1da177e4 LT |
108 | |
109 | /* | |
110 | * When performing a resync, we need to read and compare, so | |
111 | * we need as many pages are there are copies. | |
112 | * When performing a recovery, we need 2 bios, one for read, | |
113 | * one for write (we recover only one drive per r10buf) | |
114 | * | |
115 | */ | |
dd0fc66f | 116 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 117 | { |
e879a879 | 118 | struct r10conf *conf = data; |
9f2c9d12 | 119 | struct r10bio *r10_bio; |
1da177e4 | 120 | struct bio *bio; |
f0250618 ML |
121 | int j; |
122 | int nalloc, nalloc_rp; | |
123 | struct resync_pages *rps; | |
1da177e4 LT |
124 | |
125 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 126 | if (!r10_bio) |
1da177e4 | 127 | return NULL; |
1da177e4 | 128 | |
3ea7daa5 N |
129 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || |
130 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
1da177e4 LT |
131 | nalloc = conf->copies; /* resync */ |
132 | else | |
133 | nalloc = 2; /* recovery */ | |
134 | ||
f0250618 ML |
135 | /* allocate once for all bios */ |
136 | if (!conf->have_replacement) | |
137 | nalloc_rp = nalloc; | |
138 | else | |
139 | nalloc_rp = nalloc * 2; | |
6da2ec56 | 140 | rps = kmalloc_array(nalloc_rp, sizeof(struct resync_pages), gfp_flags); |
f0250618 ML |
141 | if (!rps) |
142 | goto out_free_r10bio; | |
143 | ||
1da177e4 LT |
144 | /* |
145 | * Allocate bios. | |
146 | */ | |
147 | for (j = nalloc ; j-- ; ) { | |
6746557f | 148 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
149 | if (!bio) |
150 | goto out_free_bio; | |
151 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
152 | if (!conf->have_replacement) |
153 | continue; | |
154 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
155 | if (!bio) | |
156 | goto out_free_bio; | |
157 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
158 | } |
159 | /* | |
160 | * Allocate RESYNC_PAGES data pages and attach them | |
161 | * where needed. | |
162 | */ | |
f0250618 | 163 | for (j = 0; j < nalloc; j++) { |
69335ef3 | 164 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
f0250618 ML |
165 | struct resync_pages *rp, *rp_repl; |
166 | ||
167 | rp = &rps[j]; | |
168 | if (rbio) | |
169 | rp_repl = &rps[nalloc + j]; | |
170 | ||
1da177e4 | 171 | bio = r10_bio->devs[j].bio; |
f0250618 ML |
172 | |
173 | if (!j || test_bit(MD_RECOVERY_SYNC, | |
174 | &conf->mddev->recovery)) { | |
175 | if (resync_alloc_pages(rp, gfp_flags)) | |
1da177e4 | 176 | goto out_free_pages; |
f0250618 ML |
177 | } else { |
178 | memcpy(rp, &rps[0], sizeof(*rp)); | |
179 | resync_get_all_pages(rp); | |
180 | } | |
1da177e4 | 181 | |
f0250618 ML |
182 | rp->raid_bio = r10_bio; |
183 | bio->bi_private = rp; | |
184 | if (rbio) { | |
185 | memcpy(rp_repl, rp, sizeof(*rp)); | |
186 | rbio->bi_private = rp_repl; | |
1da177e4 LT |
187 | } |
188 | } | |
189 | ||
190 | return r10_bio; | |
191 | ||
192 | out_free_pages: | |
f0250618 | 193 | while (--j >= 0) |
45422b70 | 194 | resync_free_pages(&rps[j]); |
f0250618 | 195 | |
5fdd2cf8 | 196 | j = 0; |
1da177e4 | 197 | out_free_bio: |
5fdd2cf8 | 198 | for ( ; j < nalloc; j++) { |
199 | if (r10_bio->devs[j].bio) | |
200 | bio_put(r10_bio->devs[j].bio); | |
69335ef3 N |
201 | if (r10_bio->devs[j].repl_bio) |
202 | bio_put(r10_bio->devs[j].repl_bio); | |
203 | } | |
f0250618 ML |
204 | kfree(rps); |
205 | out_free_r10bio: | |
c7afa803 | 206 | rbio_pool_free(r10_bio, conf); |
1da177e4 LT |
207 | return NULL; |
208 | } | |
209 | ||
210 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
211 | { | |
e879a879 | 212 | struct r10conf *conf = data; |
9f2c9d12 | 213 | struct r10bio *r10bio = __r10_bio; |
1da177e4 | 214 | int j; |
f0250618 | 215 | struct resync_pages *rp = NULL; |
1da177e4 | 216 | |
f0250618 | 217 | for (j = conf->copies; j--; ) { |
1da177e4 | 218 | struct bio *bio = r10bio->devs[j].bio; |
f0250618 | 219 | |
eb81b328 GJ |
220 | if (bio) { |
221 | rp = get_resync_pages(bio); | |
222 | resync_free_pages(rp); | |
223 | bio_put(bio); | |
224 | } | |
f0250618 | 225 | |
69335ef3 N |
226 | bio = r10bio->devs[j].repl_bio; |
227 | if (bio) | |
228 | bio_put(bio); | |
1da177e4 | 229 | } |
f0250618 ML |
230 | |
231 | /* resync pages array stored in the 1st bio's .bi_private */ | |
232 | kfree(rp); | |
233 | ||
c7afa803 | 234 | rbio_pool_free(r10bio, conf); |
1da177e4 LT |
235 | } |
236 | ||
e879a879 | 237 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
238 | { |
239 | int i; | |
240 | ||
c2968285 | 241 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 242 | struct bio **bio = & r10_bio->devs[i].bio; |
749c55e9 | 243 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
244 | bio_put(*bio); |
245 | *bio = NULL; | |
69335ef3 N |
246 | bio = &r10_bio->devs[i].repl_bio; |
247 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
248 | bio_put(*bio); | |
249 | *bio = NULL; | |
1da177e4 LT |
250 | } |
251 | } | |
252 | ||
9f2c9d12 | 253 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 254 | { |
e879a879 | 255 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 256 | |
1da177e4 | 257 | put_all_bios(conf, r10_bio); |
afeee514 | 258 | mempool_free(r10_bio, &conf->r10bio_pool); |
1da177e4 LT |
259 | } |
260 | ||
9f2c9d12 | 261 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 262 | { |
e879a879 | 263 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 264 | |
afeee514 | 265 | mempool_free(r10_bio, &conf->r10buf_pool); |
1da177e4 | 266 | |
0a27ec96 | 267 | lower_barrier(conf); |
1da177e4 LT |
268 | } |
269 | ||
9f2c9d12 | 270 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
271 | { |
272 | unsigned long flags; | |
fd01b88c | 273 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 274 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
275 | |
276 | spin_lock_irqsave(&conf->device_lock, flags); | |
277 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 278 | conf->nr_queued ++; |
1da177e4 LT |
279 | spin_unlock_irqrestore(&conf->device_lock, flags); |
280 | ||
388667be AJ |
281 | /* wake up frozen array... */ |
282 | wake_up(&conf->wait_barrier); | |
283 | ||
1da177e4 LT |
284 | md_wakeup_thread(mddev->thread); |
285 | } | |
286 | ||
287 | /* | |
288 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
289 | * operation and are ready to return a success/failure code to the buffer | |
290 | * cache layer. | |
291 | */ | |
9f2c9d12 | 292 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
293 | { |
294 | struct bio *bio = r10_bio->master_bio; | |
e879a879 | 295 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 296 | |
856e08e2 | 297 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) |
4e4cbee9 | 298 | bio->bi_status = BLK_STS_IOERR; |
fd16f2e8 | 299 | |
528bc2cf GJ |
300 | if (blk_queue_io_stat(bio->bi_bdev->bd_disk->queue)) |
301 | bio_end_io_acct(bio, r10_bio->start_time); | |
fd16f2e8 N |
302 | bio_endio(bio); |
303 | /* | |
304 | * Wake up any possible resync thread that waits for the device | |
305 | * to go idle. | |
306 | */ | |
307 | allow_barrier(conf); | |
308 | ||
1da177e4 LT |
309 | free_r10bio(r10_bio); |
310 | } | |
311 | ||
312 | /* | |
313 | * Update disk head position estimator based on IRQ completion info. | |
314 | */ | |
9f2c9d12 | 315 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 316 | { |
e879a879 | 317 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
318 | |
319 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
320 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
321 | } | |
322 | ||
778ca018 NK |
323 | /* |
324 | * Find the disk number which triggered given bio | |
325 | */ | |
e879a879 | 326 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 327 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
328 | { |
329 | int slot; | |
69335ef3 | 330 | int repl = 0; |
778ca018 | 331 | |
c2968285 | 332 | for (slot = 0; slot < conf->geo.raid_disks; slot++) { |
778ca018 NK |
333 | if (r10_bio->devs[slot].bio == bio) |
334 | break; | |
69335ef3 N |
335 | if (r10_bio->devs[slot].repl_bio == bio) { |
336 | repl = 1; | |
337 | break; | |
338 | } | |
339 | } | |
778ca018 | 340 | |
778ca018 NK |
341 | update_head_pos(slot, r10_bio); |
342 | ||
749c55e9 N |
343 | if (slotp) |
344 | *slotp = slot; | |
69335ef3 N |
345 | if (replp) |
346 | *replp = repl; | |
778ca018 NK |
347 | return r10_bio->devs[slot].devnum; |
348 | } | |
349 | ||
4246a0b6 | 350 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 351 | { |
4e4cbee9 | 352 | int uptodate = !bio->bi_status; |
9f2c9d12 | 353 | struct r10bio *r10_bio = bio->bi_private; |
a0e764c5 | 354 | int slot; |
abbf098e | 355 | struct md_rdev *rdev; |
e879a879 | 356 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 357 | |
1da177e4 | 358 | slot = r10_bio->read_slot; |
abbf098e | 359 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
360 | /* |
361 | * this branch is our 'one mirror IO has finished' event handler: | |
362 | */ | |
4443ae10 N |
363 | update_head_pos(slot, r10_bio); |
364 | ||
365 | if (uptodate) { | |
1da177e4 LT |
366 | /* |
367 | * Set R10BIO_Uptodate in our master bio, so that | |
368 | * we will return a good error code to the higher | |
369 | * levels even if IO on some other mirrored buffer fails. | |
370 | * | |
371 | * The 'master' represents the composite IO operation to | |
372 | * user-side. So if something waits for IO, then it will | |
373 | * wait for the 'master' bio. | |
374 | */ | |
375 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
376 | } else { |
377 | /* If all other devices that store this block have | |
378 | * failed, we want to return the error upwards rather | |
379 | * than fail the last device. Here we redefine | |
380 | * "uptodate" to mean "Don't want to retry" | |
381 | */ | |
635f6416 N |
382 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
383 | rdev->raid_disk)) | |
fae8cc5e | 384 | uptodate = 1; |
fae8cc5e N |
385 | } |
386 | if (uptodate) { | |
1da177e4 | 387 | raid_end_bio_io(r10_bio); |
abbf098e | 388 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 389 | } else { |
1da177e4 | 390 | /* |
7c4e06ff | 391 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
392 | */ |
393 | char b[BDEVNAME_SIZE]; | |
08464e09 | 394 | pr_err_ratelimited("md/raid10:%s: %s: rescheduling sector %llu\n", |
8bda470e | 395 | mdname(conf->mddev), |
abbf098e | 396 | bdevname(rdev->bdev, b), |
8bda470e | 397 | (unsigned long long)r10_bio->sector); |
856e08e2 | 398 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
399 | reschedule_retry(r10_bio); |
400 | } | |
1da177e4 LT |
401 | } |
402 | ||
9f2c9d12 | 403 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
404 | { |
405 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
406 | md_bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, |
407 | r10_bio->sectors, | |
408 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
409 | 0); | |
bd870a16 N |
410 | md_write_end(r10_bio->mddev); |
411 | } | |
412 | ||
9f2c9d12 | 413 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
414 | { |
415 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
416 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
417 | reschedule_retry(r10_bio); | |
418 | else { | |
419 | close_write(r10_bio); | |
420 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
421 | reschedule_retry(r10_bio); | |
422 | else | |
423 | raid_end_bio_io(r10_bio); | |
424 | } | |
425 | } | |
426 | } | |
427 | ||
4246a0b6 | 428 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 429 | { |
9f2c9d12 | 430 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 431 | int dev; |
749c55e9 | 432 | int dec_rdev = 1; |
e879a879 | 433 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 434 | int slot, repl; |
4ca40c2c | 435 | struct md_rdev *rdev = NULL; |
1919cbb2 | 436 | struct bio *to_put = NULL; |
579ed34f SL |
437 | bool discard_error; |
438 | ||
4e4cbee9 | 439 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 440 | |
475b0321 | 441 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 442 | |
475b0321 N |
443 | if (repl) |
444 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
445 | if (!rdev) { |
446 | smp_rmb(); | |
447 | repl = 0; | |
475b0321 | 448 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 449 | } |
1da177e4 LT |
450 | /* |
451 | * this branch is our 'one mirror IO has finished' event handler: | |
452 | */ | |
4e4cbee9 | 453 | if (bio->bi_status && !discard_error) { |
475b0321 N |
454 | if (repl) |
455 | /* Never record new bad blocks to replacement, | |
456 | * just fail it. | |
457 | */ | |
458 | md_error(rdev->mddev, rdev); | |
459 | else { | |
460 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
461 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
462 | set_bit(MD_RECOVERY_NEEDED, | |
463 | &rdev->mddev->recovery); | |
1919cbb2 | 464 | |
475b0321 | 465 | dec_rdev = 0; |
1919cbb2 N |
466 | if (test_bit(FailFast, &rdev->flags) && |
467 | (bio->bi_opf & MD_FAILFAST)) { | |
468 | md_error(rdev->mddev, rdev); | |
7cee6d4e YY |
469 | } |
470 | ||
471 | /* | |
472 | * When the device is faulty, it is not necessary to | |
473 | * handle write error. | |
7cee6d4e YY |
474 | */ |
475 | if (!test_bit(Faulty, &rdev->flags)) | |
1919cbb2 | 476 | set_bit(R10BIO_WriteError, &r10_bio->state); |
7cee6d4e | 477 | else { |
5ba03936 WS |
478 | /* Fail the request */ |
479 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
7cee6d4e YY |
480 | r10_bio->devs[slot].bio = NULL; |
481 | to_put = bio; | |
482 | dec_rdev = 1; | |
483 | } | |
475b0321 | 484 | } |
749c55e9 | 485 | } else { |
1da177e4 LT |
486 | /* |
487 | * Set R10BIO_Uptodate in our master bio, so that | |
488 | * we will return a good error code for to the higher | |
489 | * levels even if IO on some other mirrored buffer fails. | |
490 | * | |
491 | * The 'master' represents the composite IO operation to | |
492 | * user-side. So if something waits for IO, then it will | |
493 | * wait for the 'master' bio. | |
494 | */ | |
749c55e9 N |
495 | sector_t first_bad; |
496 | int bad_sectors; | |
497 | ||
3056e3ae AL |
498 | /* |
499 | * Do not set R10BIO_Uptodate if the current device is | |
500 | * rebuilding or Faulty. This is because we cannot use | |
501 | * such device for properly reading the data back (we could | |
502 | * potentially use it, if the current write would have felt | |
503 | * before rdev->recovery_offset, but for simplicity we don't | |
504 | * check this here. | |
505 | */ | |
506 | if (test_bit(In_sync, &rdev->flags) && | |
507 | !test_bit(Faulty, &rdev->flags)) | |
508 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 509 | |
749c55e9 | 510 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 511 | if (is_badblock(rdev, |
749c55e9 N |
512 | r10_bio->devs[slot].addr, |
513 | r10_bio->sectors, | |
579ed34f | 514 | &first_bad, &bad_sectors) && !discard_error) { |
749c55e9 | 515 | bio_put(bio); |
475b0321 N |
516 | if (repl) |
517 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
518 | else | |
519 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
520 | dec_rdev = 0; |
521 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
522 | } | |
523 | } | |
524 | ||
1da177e4 LT |
525 | /* |
526 | * | |
527 | * Let's see if all mirrored write operations have finished | |
528 | * already. | |
529 | */ | |
19d5f834 | 530 | one_write_done(r10_bio); |
749c55e9 | 531 | if (dec_rdev) |
884162df | 532 | rdev_dec_pending(rdev, conf->mddev); |
1919cbb2 N |
533 | if (to_put) |
534 | bio_put(to_put); | |
1da177e4 LT |
535 | } |
536 | ||
1da177e4 LT |
537 | /* |
538 | * RAID10 layout manager | |
25985edc | 539 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
540 | * parameters: near_copies and far_copies. |
541 | * near_copies * far_copies must be <= raid_disks. | |
542 | * Normally one of these will be 1. | |
543 | * If both are 1, we get raid0. | |
544 | * If near_copies == raid_disks, we get raid1. | |
545 | * | |
25985edc | 546 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
547 | * first chunk, followed by near_copies copies of the next chunk and |
548 | * so on. | |
549 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
550 | * as described above, we start again with a device offset of near_copies. | |
551 | * So we effectively have another copy of the whole array further down all | |
552 | * the drives, but with blocks on different drives. | |
553 | * With this layout, and block is never stored twice on the one device. | |
554 | * | |
555 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 556 | * on each device that it is on. |
1da177e4 LT |
557 | * |
558 | * raid10_find_virt does the reverse mapping, from a device and a | |
559 | * sector offset to a virtual address | |
560 | */ | |
561 | ||
f8c9e74f | 562 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
563 | { |
564 | int n,f; | |
565 | sector_t sector; | |
566 | sector_t chunk; | |
567 | sector_t stripe; | |
568 | int dev; | |
1da177e4 | 569 | int slot = 0; |
9a3152ab JB |
570 | int last_far_set_start, last_far_set_size; |
571 | ||
572 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
573 | last_far_set_start *= geo->far_set_size; | |
574 | ||
575 | last_far_set_size = geo->far_set_size; | |
576 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
577 | |
578 | /* now calculate first sector/dev */ | |
5cf00fcd N |
579 | chunk = r10bio->sector >> geo->chunk_shift; |
580 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 581 | |
5cf00fcd | 582 | chunk *= geo->near_copies; |
1da177e4 | 583 | stripe = chunk; |
5cf00fcd N |
584 | dev = sector_div(stripe, geo->raid_disks); |
585 | if (geo->far_offset) | |
586 | stripe *= geo->far_copies; | |
1da177e4 | 587 | |
5cf00fcd | 588 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
589 | |
590 | /* and calculate all the others */ | |
5cf00fcd | 591 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 592 | int d = dev; |
475901af | 593 | int set; |
1da177e4 | 594 | sector_t s = sector; |
1da177e4 | 595 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 596 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
597 | slot++; |
598 | ||
5cf00fcd | 599 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 600 | set = d / geo->far_set_size; |
5cf00fcd | 601 | d += geo->near_copies; |
475901af | 602 | |
9a3152ab JB |
603 | if ((geo->raid_disks % geo->far_set_size) && |
604 | (d > last_far_set_start)) { | |
605 | d -= last_far_set_start; | |
606 | d %= last_far_set_size; | |
607 | d += last_far_set_start; | |
608 | } else { | |
609 | d %= geo->far_set_size; | |
610 | d += geo->far_set_size * set; | |
611 | } | |
5cf00fcd | 612 | s += geo->stride; |
1da177e4 LT |
613 | r10bio->devs[slot].devnum = d; |
614 | r10bio->devs[slot].addr = s; | |
615 | slot++; | |
616 | } | |
617 | dev++; | |
5cf00fcd | 618 | if (dev >= geo->raid_disks) { |
1da177e4 | 619 | dev = 0; |
5cf00fcd | 620 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
621 | } |
622 | } | |
f8c9e74f N |
623 | } |
624 | ||
625 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
626 | { | |
627 | struct geom *geo = &conf->geo; | |
628 | ||
629 | if (conf->reshape_progress != MaxSector && | |
630 | ((r10bio->sector >= conf->reshape_progress) != | |
631 | conf->mddev->reshape_backwards)) { | |
632 | set_bit(R10BIO_Previous, &r10bio->state); | |
633 | geo = &conf->prev; | |
634 | } else | |
635 | clear_bit(R10BIO_Previous, &r10bio->state); | |
636 | ||
637 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
638 | } |
639 | ||
e879a879 | 640 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
641 | { |
642 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
643 | /* Never use conf->prev as this is only called during resync |
644 | * or recovery, so reshape isn't happening | |
645 | */ | |
5cf00fcd | 646 | struct geom *geo = &conf->geo; |
475901af JB |
647 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
648 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
649 | int last_far_set_start; |
650 | ||
651 | if (geo->raid_disks % geo->far_set_size) { | |
652 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
653 | last_far_set_start *= geo->far_set_size; | |
654 | ||
655 | if (dev >= last_far_set_start) { | |
656 | far_set_size = geo->far_set_size; | |
657 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
658 | far_set_start = last_far_set_start; | |
659 | } | |
660 | } | |
1da177e4 | 661 | |
5cf00fcd N |
662 | offset = sector & geo->chunk_mask; |
663 | if (geo->far_offset) { | |
c93983bf | 664 | int fc; |
5cf00fcd N |
665 | chunk = sector >> geo->chunk_shift; |
666 | fc = sector_div(chunk, geo->far_copies); | |
667 | dev -= fc * geo->near_copies; | |
475901af JB |
668 | if (dev < far_set_start) |
669 | dev += far_set_size; | |
c93983bf | 670 | } else { |
5cf00fcd N |
671 | while (sector >= geo->stride) { |
672 | sector -= geo->stride; | |
475901af JB |
673 | if (dev < (geo->near_copies + far_set_start)) |
674 | dev += far_set_size - geo->near_copies; | |
c93983bf | 675 | else |
5cf00fcd | 676 | dev -= geo->near_copies; |
c93983bf | 677 | } |
5cf00fcd | 678 | chunk = sector >> geo->chunk_shift; |
c93983bf | 679 | } |
5cf00fcd N |
680 | vchunk = chunk * geo->raid_disks + dev; |
681 | sector_div(vchunk, geo->near_copies); | |
682 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
683 | } |
684 | ||
1da177e4 LT |
685 | /* |
686 | * This routine returns the disk from which the requested read should | |
687 | * be done. There is a per-array 'next expected sequential IO' sector | |
688 | * number - if this matches on the next IO then we use the last disk. | |
689 | * There is also a per-disk 'last know head position' sector that is | |
690 | * maintained from IRQ contexts, both the normal and the resync IO | |
691 | * completion handlers update this position correctly. If there is no | |
692 | * perfect sequential match then we pick the disk whose head is closest. | |
693 | * | |
694 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
695 | * because position is mirror, not device based. | |
696 | * | |
697 | * The rdev for the device selected will have nr_pending incremented. | |
698 | */ | |
699 | ||
700 | /* | |
701 | * FIXME: possibly should rethink readbalancing and do it differently | |
702 | * depending on near_copies / far_copies geometry. | |
703 | */ | |
96c3fd1f N |
704 | static struct md_rdev *read_balance(struct r10conf *conf, |
705 | struct r10bio *r10_bio, | |
706 | int *max_sectors) | |
1da177e4 | 707 | { |
af3a2cd6 | 708 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 709 | int disk, slot; |
856e08e2 N |
710 | int sectors = r10_bio->sectors; |
711 | int best_good_sectors; | |
56d99121 | 712 | sector_t new_distance, best_dist; |
e9eeba28 | 713 | struct md_rdev *best_dist_rdev, *best_pending_rdev, *rdev = NULL; |
56d99121 | 714 | int do_balance; |
e9eeba28 GJ |
715 | int best_dist_slot, best_pending_slot; |
716 | bool has_nonrot_disk = false; | |
717 | unsigned int min_pending; | |
5cf00fcd | 718 | struct geom *geo = &conf->geo; |
1da177e4 LT |
719 | |
720 | raid10_find_phys(conf, r10_bio); | |
721 | rcu_read_lock(); | |
e9eeba28 GJ |
722 | best_dist_slot = -1; |
723 | min_pending = UINT_MAX; | |
724 | best_dist_rdev = NULL; | |
725 | best_pending_rdev = NULL; | |
56d99121 | 726 | best_dist = MaxSector; |
856e08e2 | 727 | best_good_sectors = 0; |
56d99121 | 728 | do_balance = 1; |
8d3ca83d | 729 | clear_bit(R10BIO_FailFast, &r10_bio->state); |
1da177e4 LT |
730 | /* |
731 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
732 | * device if no resync is going on (recovery is ok), or below |
733 | * the resync window. We take the first readable disk when | |
734 | * above the resync window. | |
1da177e4 | 735 | */ |
d4098c72 GJ |
736 | if ((conf->mddev->recovery_cp < MaxSector |
737 | && (this_sector + sectors >= conf->next_resync)) || | |
738 | (mddev_is_clustered(conf->mddev) && | |
739 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, | |
740 | this_sector + sectors))) | |
56d99121 | 741 | do_balance = 0; |
1da177e4 | 742 | |
56d99121 | 743 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
744 | sector_t first_bad; |
745 | int bad_sectors; | |
746 | sector_t dev_sector; | |
e9eeba28 GJ |
747 | unsigned int pending; |
748 | bool nonrot; | |
856e08e2 | 749 | |
56d99121 N |
750 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
751 | continue; | |
1da177e4 | 752 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
753 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
754 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
755 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
756 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 | 757 | if (rdev == NULL || |
8ae12666 | 758 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
759 | continue; |
760 | if (!test_bit(In_sync, &rdev->flags) && | |
761 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
762 | continue; |
763 | ||
856e08e2 N |
764 | dev_sector = r10_bio->devs[slot].addr; |
765 | if (is_badblock(rdev, dev_sector, sectors, | |
766 | &first_bad, &bad_sectors)) { | |
767 | if (best_dist < MaxSector) | |
768 | /* Already have a better slot */ | |
769 | continue; | |
770 | if (first_bad <= dev_sector) { | |
771 | /* Cannot read here. If this is the | |
772 | * 'primary' device, then we must not read | |
773 | * beyond 'bad_sectors' from another device. | |
774 | */ | |
775 | bad_sectors -= (dev_sector - first_bad); | |
776 | if (!do_balance && sectors > bad_sectors) | |
777 | sectors = bad_sectors; | |
778 | if (best_good_sectors > sectors) | |
779 | best_good_sectors = sectors; | |
780 | } else { | |
781 | sector_t good_sectors = | |
782 | first_bad - dev_sector; | |
783 | if (good_sectors > best_good_sectors) { | |
784 | best_good_sectors = good_sectors; | |
e9eeba28 GJ |
785 | best_dist_slot = slot; |
786 | best_dist_rdev = rdev; | |
856e08e2 N |
787 | } |
788 | if (!do_balance) | |
789 | /* Must read from here */ | |
790 | break; | |
791 | } | |
792 | continue; | |
793 | } else | |
794 | best_good_sectors = sectors; | |
795 | ||
56d99121 N |
796 | if (!do_balance) |
797 | break; | |
1da177e4 | 798 | |
e9eeba28 GJ |
799 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
800 | has_nonrot_disk |= nonrot; | |
801 | pending = atomic_read(&rdev->nr_pending); | |
802 | if (min_pending > pending && nonrot) { | |
803 | min_pending = pending; | |
804 | best_pending_slot = slot; | |
805 | best_pending_rdev = rdev; | |
806 | } | |
807 | ||
808 | if (best_dist_slot >= 0) | |
8d3ca83d N |
809 | /* At least 2 disks to choose from so failfast is OK */ |
810 | set_bit(R10BIO_FailFast, &r10_bio->state); | |
22dfdf52 N |
811 | /* This optimisation is debatable, and completely destroys |
812 | * sequential read speed for 'far copies' arrays. So only | |
813 | * keep it for 'near' arrays, and review those later. | |
814 | */ | |
e9eeba28 | 815 | if (geo->near_copies > 1 && !pending) |
8d3ca83d | 816 | new_distance = 0; |
8ed3a195 KS |
817 | |
818 | /* for far > 1 always use the lowest address */ | |
8d3ca83d | 819 | else if (geo->far_copies > 1) |
56d99121 | 820 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 821 | else |
56d99121 N |
822 | new_distance = abs(r10_bio->devs[slot].addr - |
823 | conf->mirrors[disk].head_position); | |
e9eeba28 | 824 | |
56d99121 N |
825 | if (new_distance < best_dist) { |
826 | best_dist = new_distance; | |
e9eeba28 GJ |
827 | best_dist_slot = slot; |
828 | best_dist_rdev = rdev; | |
1da177e4 LT |
829 | } |
830 | } | |
abbf098e | 831 | if (slot >= conf->copies) { |
e9eeba28 GJ |
832 | if (has_nonrot_disk) { |
833 | slot = best_pending_slot; | |
834 | rdev = best_pending_rdev; | |
835 | } else { | |
836 | slot = best_dist_slot; | |
837 | rdev = best_dist_rdev; | |
838 | } | |
abbf098e | 839 | } |
1da177e4 | 840 | |
56d99121 | 841 | if (slot >= 0) { |
56d99121 | 842 | atomic_inc(&rdev->nr_pending); |
56d99121 N |
843 | r10_bio->read_slot = slot; |
844 | } else | |
96c3fd1f | 845 | rdev = NULL; |
1da177e4 | 846 | rcu_read_unlock(); |
856e08e2 | 847 | *max_sectors = best_good_sectors; |
1da177e4 | 848 | |
96c3fd1f | 849 | return rdev; |
1da177e4 LT |
850 | } |
851 | ||
e879a879 | 852 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
853 | { |
854 | /* Any writes that have been queued but are awaiting | |
855 | * bitmap updates get flushed here. | |
a35e63ef | 856 | */ |
a35e63ef N |
857 | spin_lock_irq(&conf->device_lock); |
858 | ||
859 | if (conf->pending_bio_list.head) { | |
18022a1b | 860 | struct blk_plug plug; |
a35e63ef | 861 | struct bio *bio; |
18022a1b | 862 | |
a35e63ef | 863 | bio = bio_list_get(&conf->pending_bio_list); |
34db0cd6 | 864 | conf->pending_count = 0; |
a35e63ef | 865 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
866 | |
867 | /* | |
868 | * As this is called in a wait_event() loop (see freeze_array), | |
869 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
870 | * cause a warning when we prepare to wait again. As it is | |
871 | * rare that this path is taken, it is perfectly safe to force | |
872 | * us to go around the wait_event() loop again, so the warning | |
873 | * is a false-positive. Silence the warning by resetting | |
874 | * thread state | |
875 | */ | |
876 | __set_current_state(TASK_RUNNING); | |
877 | ||
18022a1b | 878 | blk_start_plug(&plug); |
a35e63ef N |
879 | /* flush any pending bitmap writes to disk |
880 | * before proceeding w/ I/O */ | |
e64e4018 | 881 | md_bitmap_unplug(conf->mddev->bitmap); |
34db0cd6 | 882 | wake_up(&conf->wait_barrier); |
a35e63ef N |
883 | |
884 | while (bio) { /* submit pending writes */ | |
885 | struct bio *next = bio->bi_next; | |
309dca30 | 886 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 887 | bio->bi_next = NULL; |
74d46992 | 888 | bio_set_dev(bio, rdev->bdev); |
a9ae93c8 | 889 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 890 | bio_io_error(bio); |
a9ae93c8 | 891 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
309dca30 | 892 | !blk_queue_discard(bio->bi_bdev->bd_disk->queue))) |
532a2a3f | 893 | /* Just ignore it */ |
4246a0b6 | 894 | bio_endio(bio); |
532a2a3f | 895 | else |
ed00aabd | 896 | submit_bio_noacct(bio); |
a35e63ef N |
897 | bio = next; |
898 | } | |
18022a1b | 899 | blk_finish_plug(&plug); |
a35e63ef N |
900 | } else |
901 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 902 | } |
7eaceacc | 903 | |
0a27ec96 N |
904 | /* Barriers.... |
905 | * Sometimes we need to suspend IO while we do something else, | |
906 | * either some resync/recovery, or reconfigure the array. | |
907 | * To do this we raise a 'barrier'. | |
908 | * The 'barrier' is a counter that can be raised multiple times | |
909 | * to count how many activities are happening which preclude | |
910 | * normal IO. | |
911 | * We can only raise the barrier if there is no pending IO. | |
912 | * i.e. if nr_pending == 0. | |
913 | * We choose only to raise the barrier if no-one is waiting for the | |
914 | * barrier to go down. This means that as soon as an IO request | |
915 | * is ready, no other operations which require a barrier will start | |
916 | * until the IO request has had a chance. | |
917 | * | |
918 | * So: regular IO calls 'wait_barrier'. When that returns there | |
919 | * is no backgroup IO happening, It must arrange to call | |
920 | * allow_barrier when it has finished its IO. | |
921 | * backgroup IO calls must call raise_barrier. Once that returns | |
922 | * there is no normal IO happeing. It must arrange to call | |
923 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 924 | */ |
1da177e4 | 925 | |
e879a879 | 926 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 927 | { |
6cce3b23 | 928 | BUG_ON(force && !conf->barrier); |
1da177e4 | 929 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 930 | |
6cce3b23 N |
931 | /* Wait until no block IO is waiting (unless 'force') */ |
932 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
eed8c02e | 933 | conf->resync_lock); |
0a27ec96 N |
934 | |
935 | /* block any new IO from starting */ | |
936 | conf->barrier++; | |
937 | ||
c3b328ac | 938 | /* Now wait for all pending IO to complete */ |
0a27ec96 | 939 | wait_event_lock_irq(conf->wait_barrier, |
0e5313e2 | 940 | !atomic_read(&conf->nr_pending) && conf->barrier < RESYNC_DEPTH, |
eed8c02e | 941 | conf->resync_lock); |
0a27ec96 N |
942 | |
943 | spin_unlock_irq(&conf->resync_lock); | |
944 | } | |
945 | ||
e879a879 | 946 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
947 | { |
948 | unsigned long flags; | |
949 | spin_lock_irqsave(&conf->resync_lock, flags); | |
950 | conf->barrier--; | |
951 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
952 | wake_up(&conf->wait_barrier); | |
953 | } | |
954 | ||
e879a879 | 955 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
956 | { |
957 | spin_lock_irq(&conf->resync_lock); | |
958 | if (conf->barrier) { | |
fe630de0 | 959 | struct bio_list *bio_list = current->bio_list; |
0a27ec96 | 960 | conf->nr_waiting++; |
d6b42dcb N |
961 | /* Wait for the barrier to drop. |
962 | * However if there are already pending | |
963 | * requests (preventing the barrier from | |
964 | * rising completely), and the | |
965 | * pre-process bio queue isn't empty, | |
966 | * then don't wait, as we need to empty | |
967 | * that queue to get the nr_pending | |
968 | * count down. | |
969 | */ | |
578b54ad | 970 | raid10_log(conf->mddev, "wait barrier"); |
d6b42dcb N |
971 | wait_event_lock_irq(conf->wait_barrier, |
972 | !conf->barrier || | |
0e5313e2 | 973 | (atomic_read(&conf->nr_pending) && |
fe630de0 VM |
974 | bio_list && |
975 | (!bio_list_empty(&bio_list[0]) || | |
976 | !bio_list_empty(&bio_list[1]))) || | |
977 | /* move on if recovery thread is | |
978 | * blocked by us | |
979 | */ | |
980 | (conf->mddev->thread->tsk == current && | |
981 | test_bit(MD_RECOVERY_RUNNING, | |
982 | &conf->mddev->recovery) && | |
983 | conf->nr_queued > 0), | |
eed8c02e | 984 | conf->resync_lock); |
0a27ec96 | 985 | conf->nr_waiting--; |
0e5313e2 TM |
986 | if (!conf->nr_waiting) |
987 | wake_up(&conf->wait_barrier); | |
1da177e4 | 988 | } |
0e5313e2 | 989 | atomic_inc(&conf->nr_pending); |
1da177e4 LT |
990 | spin_unlock_irq(&conf->resync_lock); |
991 | } | |
992 | ||
e879a879 | 993 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 | 994 | { |
0e5313e2 TM |
995 | if ((atomic_dec_and_test(&conf->nr_pending)) || |
996 | (conf->array_freeze_pending)) | |
997 | wake_up(&conf->wait_barrier); | |
0a27ec96 N |
998 | } |
999 | ||
e2d59925 | 1000 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
1001 | { |
1002 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 1003 | * go quiet. |
4443ae10 | 1004 | * We increment barrier and nr_waiting, and then |
e2d59925 | 1005 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
1006 | * This is called in the context of one normal IO request |
1007 | * that has failed. Thus any sync request that might be pending | |
1008 | * will be blocked by nr_pending, and we need to wait for | |
1009 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 1010 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
1011 | * must match the number of pending IOs (nr_pending) before |
1012 | * we continue. | |
4443ae10 N |
1013 | */ |
1014 | spin_lock_irq(&conf->resync_lock); | |
0e5313e2 | 1015 | conf->array_freeze_pending++; |
4443ae10 N |
1016 | conf->barrier++; |
1017 | conf->nr_waiting++; | |
eed8c02e | 1018 | wait_event_lock_irq_cmd(conf->wait_barrier, |
0e5313e2 | 1019 | atomic_read(&conf->nr_pending) == conf->nr_queued+extra, |
eed8c02e LC |
1020 | conf->resync_lock, |
1021 | flush_pending_writes(conf)); | |
c3b328ac | 1022 | |
0e5313e2 | 1023 | conf->array_freeze_pending--; |
4443ae10 N |
1024 | spin_unlock_irq(&conf->resync_lock); |
1025 | } | |
1026 | ||
e879a879 | 1027 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1028 | { |
1029 | /* reverse the effect of the freeze */ | |
1030 | spin_lock_irq(&conf->resync_lock); | |
1031 | conf->barrier--; | |
1032 | conf->nr_waiting--; | |
1033 | wake_up(&conf->wait_barrier); | |
1034 | spin_unlock_irq(&conf->resync_lock); | |
1035 | } | |
1036 | ||
f8c9e74f N |
1037 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1038 | struct md_rdev *rdev) | |
1039 | { | |
1040 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1041 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1042 | return rdev->data_offset; | |
1043 | else | |
1044 | return rdev->new_data_offset; | |
1045 | } | |
1046 | ||
57c67df4 N |
1047 | struct raid10_plug_cb { |
1048 | struct blk_plug_cb cb; | |
1049 | struct bio_list pending; | |
1050 | int pending_cnt; | |
1051 | }; | |
1052 | ||
1053 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1054 | { | |
1055 | struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb, | |
1056 | cb); | |
1057 | struct mddev *mddev = plug->cb.data; | |
1058 | struct r10conf *conf = mddev->private; | |
1059 | struct bio *bio; | |
1060 | ||
874807a8 | 1061 | if (from_schedule || current->bio_list) { |
57c67df4 N |
1062 | spin_lock_irq(&conf->device_lock); |
1063 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1064 | conf->pending_count += plug->pending_cnt; | |
1065 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1066 | wake_up(&conf->wait_barrier); |
57c67df4 N |
1067 | md_wakeup_thread(mddev->thread); |
1068 | kfree(plug); | |
1069 | return; | |
1070 | } | |
1071 | ||
1072 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1073 | bio = bio_list_get(&plug->pending); | |
e64e4018 | 1074 | md_bitmap_unplug(mddev->bitmap); |
57c67df4 N |
1075 | wake_up(&conf->wait_barrier); |
1076 | ||
1077 | while (bio) { /* submit pending writes */ | |
1078 | struct bio *next = bio->bi_next; | |
309dca30 | 1079 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
57c67df4 | 1080 | bio->bi_next = NULL; |
74d46992 | 1081 | bio_set_dev(bio, rdev->bdev); |
a9ae93c8 | 1082 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 1083 | bio_io_error(bio); |
a9ae93c8 | 1084 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
309dca30 | 1085 | !blk_queue_discard(bio->bi_bdev->bd_disk->queue))) |
32f9f570 | 1086 | /* Just ignore it */ |
4246a0b6 | 1087 | bio_endio(bio); |
32f9f570 | 1088 | else |
ed00aabd | 1089 | submit_bio_noacct(bio); |
57c67df4 N |
1090 | bio = next; |
1091 | } | |
1092 | kfree(plug); | |
1093 | } | |
1094 | ||
caea3c47 GJ |
1095 | /* |
1096 | * 1. Register the new request and wait if the reconstruction thread has put | |
1097 | * up a bar for new requests. Continue immediately if no resync is active | |
1098 | * currently. | |
1099 | * 2. If IO spans the reshape position. Need to wait for reshape to pass. | |
1100 | */ | |
1101 | static void regular_request_wait(struct mddev *mddev, struct r10conf *conf, | |
1102 | struct bio *bio, sector_t sectors) | |
1103 | { | |
1104 | wait_barrier(conf); | |
1105 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
1106 | bio->bi_iter.bi_sector < conf->reshape_progress && | |
1107 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
1108 | raid10_log(conf->mddev, "wait reshape"); | |
1109 | allow_barrier(conf); | |
1110 | wait_event(conf->wait_barrier, | |
1111 | conf->reshape_progress <= bio->bi_iter.bi_sector || | |
1112 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1113 | sectors); | |
1114 | wait_barrier(conf); | |
1115 | } | |
1116 | } | |
1117 | ||
bb5f1ed7 RL |
1118 | static void raid10_read_request(struct mddev *mddev, struct bio *bio, |
1119 | struct r10bio *r10_bio) | |
1da177e4 | 1120 | { |
e879a879 | 1121 | struct r10conf *conf = mddev->private; |
1da177e4 | 1122 | struct bio *read_bio; |
bb5f1ed7 RL |
1123 | const int op = bio_op(bio); |
1124 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
bb5f1ed7 | 1125 | int max_sectors; |
bb5f1ed7 | 1126 | struct md_rdev *rdev; |
545250f2 N |
1127 | char b[BDEVNAME_SIZE]; |
1128 | int slot = r10_bio->read_slot; | |
1129 | struct md_rdev *err_rdev = NULL; | |
1130 | gfp_t gfp = GFP_NOIO; | |
bb5f1ed7 | 1131 | |
93decc56 | 1132 | if (slot >= 0 && r10_bio->devs[slot].rdev) { |
545250f2 N |
1133 | /* |
1134 | * This is an error retry, but we cannot | |
1135 | * safely dereference the rdev in the r10_bio, | |
1136 | * we must use the one in conf. | |
1137 | * If it has already been disconnected (unlikely) | |
1138 | * we lose the device name in error messages. | |
1139 | */ | |
1140 | int disk; | |
1141 | /* | |
1142 | * As we are blocking raid10, it is a little safer to | |
1143 | * use __GFP_HIGH. | |
1144 | */ | |
1145 | gfp = GFP_NOIO | __GFP_HIGH; | |
1146 | ||
1147 | rcu_read_lock(); | |
1148 | disk = r10_bio->devs[slot].devnum; | |
1149 | err_rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
1150 | if (err_rdev) | |
1151 | bdevname(err_rdev->bdev, b); | |
1152 | else { | |
1153 | strcpy(b, "???"); | |
1154 | /* This never gets dereferenced */ | |
1155 | err_rdev = r10_bio->devs[slot].rdev; | |
1156 | } | |
1157 | rcu_read_unlock(); | |
1158 | } | |
bb5f1ed7 | 1159 | |
caea3c47 | 1160 | regular_request_wait(mddev, conf, bio, r10_bio->sectors); |
bb5f1ed7 RL |
1161 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1162 | if (!rdev) { | |
545250f2 N |
1163 | if (err_rdev) { |
1164 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1165 | mdname(mddev), b, | |
1166 | (unsigned long long)r10_bio->sector); | |
1167 | } | |
bb5f1ed7 RL |
1168 | raid_end_bio_io(r10_bio); |
1169 | return; | |
1170 | } | |
545250f2 N |
1171 | if (err_rdev) |
1172 | pr_err_ratelimited("md/raid10:%s: %s: redirecting sector %llu to another mirror\n", | |
1173 | mdname(mddev), | |
1174 | bdevname(rdev->bdev, b), | |
1175 | (unsigned long long)r10_bio->sector); | |
fc9977dd N |
1176 | if (max_sectors < bio_sectors(bio)) { |
1177 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1178 | gfp, &conf->bio_split); |
fc9977dd | 1179 | bio_chain(split, bio); |
e820d55c | 1180 | allow_barrier(conf); |
ed00aabd | 1181 | submit_bio_noacct(bio); |
e820d55c | 1182 | wait_barrier(conf); |
fc9977dd N |
1183 | bio = split; |
1184 | r10_bio->master_bio = bio; | |
1185 | r10_bio->sectors = max_sectors; | |
1186 | } | |
bb5f1ed7 RL |
1187 | slot = r10_bio->read_slot; |
1188 | ||
528bc2cf GJ |
1189 | if (blk_queue_io_stat(bio->bi_bdev->bd_disk->queue)) |
1190 | r10_bio->start_time = bio_start_io_acct(bio); | |
afeee514 | 1191 | read_bio = bio_clone_fast(bio, gfp, &mddev->bio_set); |
bb5f1ed7 RL |
1192 | |
1193 | r10_bio->devs[slot].bio = read_bio; | |
1194 | r10_bio->devs[slot].rdev = rdev; | |
1195 | ||
1196 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + | |
1197 | choose_data_offset(r10_bio, rdev); | |
74d46992 | 1198 | bio_set_dev(read_bio, rdev->bdev); |
bb5f1ed7 RL |
1199 | read_bio->bi_end_io = raid10_end_read_request; |
1200 | bio_set_op_attrs(read_bio, op, do_sync); | |
1201 | if (test_bit(FailFast, &rdev->flags) && | |
1202 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
1203 | read_bio->bi_opf |= MD_FAILFAST; | |
1204 | read_bio->bi_private = r10_bio; | |
1205 | ||
1206 | if (mddev->gendisk) | |
1c02fca6 | 1207 | trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), |
bb5f1ed7 | 1208 | r10_bio->sector); |
ed00aabd | 1209 | submit_bio_noacct(read_bio); |
bb5f1ed7 RL |
1210 | return; |
1211 | } | |
1212 | ||
27f26a0f GJ |
1213 | static void raid10_write_one_disk(struct mddev *mddev, struct r10bio *r10_bio, |
1214 | struct bio *bio, bool replacement, | |
fc9977dd | 1215 | int n_copy) |
bb5f1ed7 | 1216 | { |
796a5cf0 | 1217 | const int op = bio_op(bio); |
1eff9d32 JA |
1218 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1219 | const unsigned long do_fua = (bio->bi_opf & REQ_FUA); | |
6cce3b23 | 1220 | unsigned long flags; |
57c67df4 N |
1221 | struct blk_plug_cb *cb; |
1222 | struct raid10_plug_cb *plug = NULL; | |
27f26a0f GJ |
1223 | struct r10conf *conf = mddev->private; |
1224 | struct md_rdev *rdev; | |
1225 | int devnum = r10_bio->devs[n_copy].devnum; | |
1226 | struct bio *mbio; | |
1227 | ||
1228 | if (replacement) { | |
1229 | rdev = conf->mirrors[devnum].replacement; | |
1230 | if (rdev == NULL) { | |
1231 | /* Replacement just got moved to main 'rdev' */ | |
1232 | smp_mb(); | |
1233 | rdev = conf->mirrors[devnum].rdev; | |
1234 | } | |
1235 | } else | |
1236 | rdev = conf->mirrors[devnum].rdev; | |
1237 | ||
afeee514 | 1238 | mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
27f26a0f GJ |
1239 | if (replacement) |
1240 | r10_bio->devs[n_copy].repl_bio = mbio; | |
1241 | else | |
1242 | r10_bio->devs[n_copy].bio = mbio; | |
1243 | ||
1244 | mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + | |
1245 | choose_data_offset(r10_bio, rdev)); | |
74d46992 | 1246 | bio_set_dev(mbio, rdev->bdev); |
27f26a0f GJ |
1247 | mbio->bi_end_io = raid10_end_write_request; |
1248 | bio_set_op_attrs(mbio, op, do_sync | do_fua); | |
1249 | if (!replacement && test_bit(FailFast, | |
1250 | &conf->mirrors[devnum].rdev->flags) | |
1251 | && enough(conf, devnum)) | |
1252 | mbio->bi_opf |= MD_FAILFAST; | |
1253 | mbio->bi_private = r10_bio; | |
1254 | ||
1255 | if (conf->mddev->gendisk) | |
1c02fca6 | 1256 | trace_block_bio_remap(mbio, disk_devt(conf->mddev->gendisk), |
27f26a0f GJ |
1257 | r10_bio->sector); |
1258 | /* flush_pending_writes() needs access to the rdev so...*/ | |
309dca30 | 1259 | mbio->bi_bdev = (void *)rdev; |
27f26a0f GJ |
1260 | |
1261 | atomic_inc(&r10_bio->remaining); | |
1262 | ||
1263 | cb = blk_check_plugged(raid10_unplug, mddev, sizeof(*plug)); | |
1264 | if (cb) | |
1265 | plug = container_of(cb, struct raid10_plug_cb, cb); | |
1266 | else | |
1267 | plug = NULL; | |
27f26a0f GJ |
1268 | if (plug) { |
1269 | bio_list_add(&plug->pending, mbio); | |
1270 | plug->pending_cnt++; | |
1271 | } else { | |
23b245c0 | 1272 | spin_lock_irqsave(&conf->device_lock, flags); |
27f26a0f GJ |
1273 | bio_list_add(&conf->pending_bio_list, mbio); |
1274 | conf->pending_count++; | |
23b245c0 | 1275 | spin_unlock_irqrestore(&conf->device_lock, flags); |
27f26a0f | 1276 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1277 | } |
27f26a0f GJ |
1278 | } |
1279 | ||
f2e7e269 XN |
1280 | static void wait_blocked_dev(struct mddev *mddev, struct r10bio *r10_bio) |
1281 | { | |
1282 | int i; | |
1283 | struct r10conf *conf = mddev->private; | |
1284 | struct md_rdev *blocked_rdev; | |
1285 | ||
1286 | retry_wait: | |
1287 | blocked_rdev = NULL; | |
1288 | rcu_read_lock(); | |
1289 | for (i = 0; i < conf->copies; i++) { | |
1290 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1291 | struct md_rdev *rrdev = rcu_dereference( | |
1292 | conf->mirrors[i].replacement); | |
1293 | if (rdev == rrdev) | |
1294 | rrdev = NULL; | |
1295 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { | |
1296 | atomic_inc(&rdev->nr_pending); | |
1297 | blocked_rdev = rdev; | |
1298 | break; | |
1299 | } | |
1300 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { | |
1301 | atomic_inc(&rrdev->nr_pending); | |
1302 | blocked_rdev = rrdev; | |
1303 | break; | |
1304 | } | |
1305 | ||
1306 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { | |
1307 | sector_t first_bad; | |
1308 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1309 | int bad_sectors; | |
1310 | int is_bad; | |
1311 | ||
1312 | /* | |
1313 | * Discard request doesn't care the write result | |
1314 | * so it doesn't need to wait blocked disk here. | |
1315 | */ | |
1316 | if (!r10_bio->sectors) | |
1317 | continue; | |
1318 | ||
1319 | is_bad = is_badblock(rdev, dev_sector, r10_bio->sectors, | |
1320 | &first_bad, &bad_sectors); | |
1321 | if (is_bad < 0) { | |
1322 | /* | |
1323 | * Mustn't write here until the bad block | |
1324 | * is acknowledged | |
1325 | */ | |
1326 | atomic_inc(&rdev->nr_pending); | |
1327 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1328 | blocked_rdev = rdev; | |
1329 | break; | |
1330 | } | |
1331 | } | |
1332 | } | |
1333 | rcu_read_unlock(); | |
1334 | ||
1335 | if (unlikely(blocked_rdev)) { | |
1336 | /* Have to wait for this device to get unblocked, then retry */ | |
1337 | allow_barrier(conf); | |
1338 | raid10_log(conf->mddev, "%s wait rdev %d blocked", | |
1339 | __func__, blocked_rdev->raid_disk); | |
1340 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1341 | wait_barrier(conf); | |
1342 | goto retry_wait; | |
1343 | } | |
1344 | } | |
1345 | ||
27f26a0f GJ |
1346 | static void raid10_write_request(struct mddev *mddev, struct bio *bio, |
1347 | struct r10bio *r10_bio) | |
1348 | { | |
1349 | struct r10conf *conf = mddev->private; | |
1350 | int i; | |
bb5f1ed7 | 1351 | sector_t sectors; |
d4432c23 | 1352 | int max_sectors; |
1da177e4 | 1353 | |
cb8a7a7e GJ |
1354 | if ((mddev_is_clustered(mddev) && |
1355 | md_cluster_ops->area_resyncing(mddev, WRITE, | |
1356 | bio->bi_iter.bi_sector, | |
1357 | bio_end_sector(bio)))) { | |
1358 | DEFINE_WAIT(w); | |
1359 | for (;;) { | |
1360 | prepare_to_wait(&conf->wait_barrier, | |
1361 | &w, TASK_IDLE); | |
1362 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, | |
1363 | bio->bi_iter.bi_sector, bio_end_sector(bio))) | |
1364 | break; | |
1365 | schedule(); | |
1366 | } | |
1367 | finish_wait(&conf->wait_barrier, &w); | |
1368 | } | |
1369 | ||
fc9977dd | 1370 | sectors = r10_bio->sectors; |
caea3c47 | 1371 | regular_request_wait(mddev, conf, bio, sectors); |
3ea7daa5 | 1372 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
3ea7daa5 | 1373 | (mddev->reshape_backwards |
4f024f37 KO |
1374 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1375 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1376 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1377 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1378 | /* Need to update reshape_position in metadata */ |
1379 | mddev->reshape_position = conf->reshape_progress; | |
2953079c SL |
1380 | set_mask_bits(&mddev->sb_flags, 0, |
1381 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
3ea7daa5 | 1382 | md_wakeup_thread(mddev->thread); |
578b54ad | 1383 | raid10_log(conf->mddev, "wait reshape metadata"); |
3ea7daa5 | 1384 | wait_event(mddev->sb_wait, |
2953079c | 1385 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
3ea7daa5 N |
1386 | |
1387 | conf->reshape_safe = mddev->reshape_position; | |
1388 | } | |
1389 | ||
34db0cd6 N |
1390 | if (conf->pending_count >= max_queued_requests) { |
1391 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1392 | raid10_log(mddev, "wait queued"); |
34db0cd6 N |
1393 | wait_event(conf->wait_barrier, |
1394 | conf->pending_count < max_queued_requests); | |
1395 | } | |
6bfe0b49 | 1396 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1397 | * inc refcount on their rdev. Record them by setting |
1398 | * bios[x] to bio | |
d4432c23 N |
1399 | * If there are known/acknowledged bad blocks on any device |
1400 | * on which we have seen a write error, we want to avoid | |
1401 | * writing to those blocks. This potentially requires several | |
1402 | * writes to write around the bad blocks. Each set of writes | |
fd16f2e8 | 1403 | * gets its own r10_bio with a set of bios attached. |
1da177e4 | 1404 | */ |
c3b328ac | 1405 | |
69335ef3 | 1406 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1407 | raid10_find_phys(conf, r10_bio); |
f2e7e269 XN |
1408 | |
1409 | wait_blocked_dev(mddev, r10_bio); | |
1410 | ||
1da177e4 | 1411 | rcu_read_lock(); |
d4432c23 N |
1412 | max_sectors = r10_bio->sectors; |
1413 | ||
1da177e4 LT |
1414 | for (i = 0; i < conf->copies; i++) { |
1415 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1416 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1417 | struct md_rdev *rrdev = rcu_dereference( |
1418 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1419 | if (rdev == rrdev) |
1420 | rrdev = NULL; | |
8ae12666 | 1421 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1422 | rdev = NULL; |
8ae12666 | 1423 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1424 | rrdev = NULL; |
1425 | ||
d4432c23 | 1426 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1427 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1428 | |
1429 | if (!rdev && !rrdev) { | |
6cce3b23 | 1430 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1431 | continue; |
1432 | } | |
e7c0c3fa | 1433 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1434 | sector_t first_bad; |
1435 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1436 | int bad_sectors; | |
1437 | int is_bad; | |
1438 | ||
bb5f1ed7 | 1439 | is_bad = is_badblock(rdev, dev_sector, max_sectors, |
d4432c23 | 1440 | &first_bad, &bad_sectors); |
d4432c23 N |
1441 | if (is_bad && first_bad <= dev_sector) { |
1442 | /* Cannot write here at all */ | |
1443 | bad_sectors -= (dev_sector - first_bad); | |
1444 | if (bad_sectors < max_sectors) | |
1445 | /* Mustn't write more than bad_sectors | |
1446 | * to other devices yet | |
1447 | */ | |
1448 | max_sectors = bad_sectors; | |
1449 | /* We don't set R10BIO_Degraded as that | |
1450 | * only applies if the disk is missing, | |
1451 | * so it might be re-added, and we want to | |
1452 | * know to recover this chunk. | |
1453 | * In this case the device is here, and the | |
1454 | * fact that this chunk is not in-sync is | |
1455 | * recorded in the bad block log. | |
1456 | */ | |
1457 | continue; | |
1458 | } | |
1459 | if (is_bad) { | |
1460 | int good_sectors = first_bad - dev_sector; | |
1461 | if (good_sectors < max_sectors) | |
1462 | max_sectors = good_sectors; | |
1463 | } | |
6cce3b23 | 1464 | } |
e7c0c3fa N |
1465 | if (rdev) { |
1466 | r10_bio->devs[i].bio = bio; | |
1467 | atomic_inc(&rdev->nr_pending); | |
1468 | } | |
475b0321 N |
1469 | if (rrdev) { |
1470 | r10_bio->devs[i].repl_bio = bio; | |
1471 | atomic_inc(&rrdev->nr_pending); | |
1472 | } | |
1da177e4 LT |
1473 | } |
1474 | rcu_read_unlock(); | |
1475 | ||
6b6c8110 | 1476 | if (max_sectors < r10_bio->sectors) |
d4432c23 | 1477 | r10_bio->sectors = max_sectors; |
fc9977dd N |
1478 | |
1479 | if (r10_bio->sectors < bio_sectors(bio)) { | |
1480 | struct bio *split = bio_split(bio, r10_bio->sectors, | |
afeee514 | 1481 | GFP_NOIO, &conf->bio_split); |
fc9977dd | 1482 | bio_chain(split, bio); |
e820d55c | 1483 | allow_barrier(conf); |
ed00aabd | 1484 | submit_bio_noacct(bio); |
e820d55c | 1485 | wait_barrier(conf); |
fc9977dd N |
1486 | bio = split; |
1487 | r10_bio->master_bio = bio; | |
d4432c23 | 1488 | } |
d4432c23 | 1489 | |
528bc2cf GJ |
1490 | if (blk_queue_io_stat(bio->bi_bdev->bd_disk->queue)) |
1491 | r10_bio->start_time = bio_start_io_acct(bio); | |
4e78064f | 1492 | atomic_set(&r10_bio->remaining, 1); |
e64e4018 | 1493 | md_bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1494 | |
1da177e4 | 1495 | for (i = 0; i < conf->copies; i++) { |
27f26a0f | 1496 | if (r10_bio->devs[i].bio) |
fc9977dd | 1497 | raid10_write_one_disk(mddev, r10_bio, bio, false, i); |
27f26a0f | 1498 | if (r10_bio->devs[i].repl_bio) |
fc9977dd | 1499 | raid10_write_one_disk(mddev, r10_bio, bio, true, i); |
d4432c23 | 1500 | } |
079fa166 | 1501 | one_write_done(r10_bio); |
20d0189b KO |
1502 | } |
1503 | ||
fc9977dd | 1504 | static void __make_request(struct mddev *mddev, struct bio *bio, int sectors) |
bb5f1ed7 RL |
1505 | { |
1506 | struct r10conf *conf = mddev->private; | |
1507 | struct r10bio *r10_bio; | |
1508 | ||
afeee514 | 1509 | r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); |
bb5f1ed7 RL |
1510 | |
1511 | r10_bio->master_bio = bio; | |
fc9977dd | 1512 | r10_bio->sectors = sectors; |
bb5f1ed7 RL |
1513 | |
1514 | r10_bio->mddev = mddev; | |
1515 | r10_bio->sector = bio->bi_iter.bi_sector; | |
1516 | r10_bio->state = 0; | |
93decc56 | 1517 | r10_bio->read_slot = -1; |
c2968285 XN |
1518 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * |
1519 | conf->geo.raid_disks); | |
bb5f1ed7 RL |
1520 | |
1521 | if (bio_data_dir(bio) == READ) | |
1522 | raid10_read_request(mddev, bio, r10_bio); | |
1523 | else | |
1524 | raid10_write_request(mddev, bio, r10_bio); | |
1525 | } | |
1526 | ||
254c271d XN |
1527 | static void raid_end_discard_bio(struct r10bio *r10bio) |
1528 | { | |
1529 | struct r10conf *conf = r10bio->mddev->private; | |
1530 | struct r10bio *first_r10bio; | |
1531 | ||
1532 | while (atomic_dec_and_test(&r10bio->remaining)) { | |
1533 | ||
1534 | allow_barrier(conf); | |
1535 | ||
1536 | if (!test_bit(R10BIO_Discard, &r10bio->state)) { | |
1537 | first_r10bio = (struct r10bio *)r10bio->master_bio; | |
1538 | free_r10bio(r10bio); | |
1539 | r10bio = first_r10bio; | |
1540 | } else { | |
1541 | md_write_end(r10bio->mddev); | |
1542 | bio_endio(r10bio->master_bio); | |
1543 | free_r10bio(r10bio); | |
1544 | break; | |
1545 | } | |
1546 | } | |
1547 | } | |
1548 | ||
d30588b2 XN |
1549 | static void raid10_end_discard_request(struct bio *bio) |
1550 | { | |
1551 | struct r10bio *r10_bio = bio->bi_private; | |
1552 | struct r10conf *conf = r10_bio->mddev->private; | |
1553 | struct md_rdev *rdev = NULL; | |
1554 | int dev; | |
1555 | int slot, repl; | |
1556 | ||
1557 | /* | |
1558 | * We don't care the return value of discard bio | |
1559 | */ | |
1560 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
1561 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1562 | ||
1563 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
1564 | if (repl) | |
1565 | rdev = conf->mirrors[dev].replacement; | |
1566 | if (!rdev) { | |
1567 | /* | |
1568 | * raid10_remove_disk uses smp_mb to make sure rdev is set to | |
1569 | * replacement before setting replacement to NULL. It can read | |
1570 | * rdev first without barrier protect even replacment is NULL | |
1571 | */ | |
1572 | smp_rmb(); | |
1573 | rdev = conf->mirrors[dev].rdev; | |
1574 | } | |
1575 | ||
254c271d | 1576 | raid_end_discard_bio(r10_bio); |
d30588b2 XN |
1577 | rdev_dec_pending(rdev, conf->mddev); |
1578 | } | |
1579 | ||
1580 | /* | |
1581 | * There are some limitations to handle discard bio | |
1582 | * 1st, the discard size is bigger than stripe_size*2. | |
1583 | * 2st, if the discard bio spans reshape progress, we use the old way to | |
1584 | * handle discard bio | |
1585 | */ | |
1586 | static int raid10_handle_discard(struct mddev *mddev, struct bio *bio) | |
1587 | { | |
1588 | struct r10conf *conf = mddev->private; | |
1589 | struct geom *geo = &conf->geo; | |
254c271d XN |
1590 | int far_copies = geo->far_copies; |
1591 | bool first_copy = true; | |
1592 | struct r10bio *r10_bio, *first_r10bio; | |
d30588b2 XN |
1593 | struct bio *split; |
1594 | int disk; | |
1595 | sector_t chunk; | |
1596 | unsigned int stripe_size; | |
1597 | unsigned int stripe_data_disks; | |
1598 | sector_t split_size; | |
1599 | sector_t bio_start, bio_end; | |
1600 | sector_t first_stripe_index, last_stripe_index; | |
1601 | sector_t start_disk_offset; | |
1602 | unsigned int start_disk_index; | |
1603 | sector_t end_disk_offset; | |
1604 | unsigned int end_disk_index; | |
1605 | unsigned int remainder; | |
1606 | ||
1607 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1608 | return -EAGAIN; | |
1609 | ||
1610 | wait_barrier(conf); | |
1611 | ||
1612 | /* | |
1613 | * Check reshape again to avoid reshape happens after checking | |
1614 | * MD_RECOVERY_RESHAPE and before wait_barrier | |
1615 | */ | |
1616 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1617 | goto out; | |
1618 | ||
1619 | if (geo->near_copies) | |
1620 | stripe_data_disks = geo->raid_disks / geo->near_copies + | |
1621 | geo->raid_disks % geo->near_copies; | |
1622 | else | |
1623 | stripe_data_disks = geo->raid_disks; | |
1624 | ||
1625 | stripe_size = stripe_data_disks << geo->chunk_shift; | |
1626 | ||
1627 | bio_start = bio->bi_iter.bi_sector; | |
1628 | bio_end = bio_end_sector(bio); | |
1629 | ||
1630 | /* | |
1631 | * Maybe one discard bio is smaller than strip size or across one | |
1632 | * stripe and discard region is larger than one stripe size. For far | |
1633 | * offset layout, if the discard region is not aligned with stripe | |
1634 | * size, there is hole when we submit discard bio to member disk. | |
1635 | * For simplicity, we only handle discard bio which discard region | |
1636 | * is bigger than stripe_size * 2 | |
1637 | */ | |
1638 | if (bio_sectors(bio) < stripe_size*2) | |
1639 | goto out; | |
1640 | ||
1641 | /* | |
1642 | * Keep bio aligned with strip size. | |
1643 | */ | |
1644 | div_u64_rem(bio_start, stripe_size, &remainder); | |
1645 | if (remainder) { | |
1646 | split_size = stripe_size - remainder; | |
1647 | split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); | |
1648 | bio_chain(split, bio); | |
1649 | allow_barrier(conf); | |
1650 | /* Resend the fist split part */ | |
1651 | submit_bio_noacct(split); | |
1652 | wait_barrier(conf); | |
1653 | } | |
1654 | div_u64_rem(bio_end, stripe_size, &remainder); | |
1655 | if (remainder) { | |
1656 | split_size = bio_sectors(bio) - remainder; | |
1657 | split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); | |
1658 | bio_chain(split, bio); | |
1659 | allow_barrier(conf); | |
1660 | /* Resend the second split part */ | |
1661 | submit_bio_noacct(bio); | |
1662 | bio = split; | |
1663 | wait_barrier(conf); | |
1664 | } | |
1665 | ||
d30588b2 XN |
1666 | bio_start = bio->bi_iter.bi_sector; |
1667 | bio_end = bio_end_sector(bio); | |
1668 | ||
1669 | /* | |
1670 | * Raid10 uses chunk as the unit to store data. It's similar like raid0. | |
1671 | * One stripe contains the chunks from all member disk (one chunk from | |
1672 | * one disk at the same HBA address). For layout detail, see 'man md 4' | |
1673 | */ | |
1674 | chunk = bio_start >> geo->chunk_shift; | |
1675 | chunk *= geo->near_copies; | |
1676 | first_stripe_index = chunk; | |
1677 | start_disk_index = sector_div(first_stripe_index, geo->raid_disks); | |
1678 | if (geo->far_offset) | |
1679 | first_stripe_index *= geo->far_copies; | |
1680 | start_disk_offset = (bio_start & geo->chunk_mask) + | |
1681 | (first_stripe_index << geo->chunk_shift); | |
1682 | ||
1683 | chunk = bio_end >> geo->chunk_shift; | |
1684 | chunk *= geo->near_copies; | |
1685 | last_stripe_index = chunk; | |
1686 | end_disk_index = sector_div(last_stripe_index, geo->raid_disks); | |
1687 | if (geo->far_offset) | |
1688 | last_stripe_index *= geo->far_copies; | |
1689 | end_disk_offset = (bio_end & geo->chunk_mask) + | |
1690 | (last_stripe_index << geo->chunk_shift); | |
1691 | ||
254c271d XN |
1692 | retry_discard: |
1693 | r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); | |
1694 | r10_bio->mddev = mddev; | |
1695 | r10_bio->state = 0; | |
1696 | r10_bio->sectors = 0; | |
1697 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * geo->raid_disks); | |
1698 | wait_blocked_dev(mddev, r10_bio); | |
1699 | ||
1700 | /* | |
1701 | * For far layout it needs more than one r10bio to cover all regions. | |
1702 | * Inspired by raid10_sync_request, we can use the first r10bio->master_bio | |
1703 | * to record the discard bio. Other r10bio->master_bio record the first | |
1704 | * r10bio. The first r10bio only release after all other r10bios finish. | |
1705 | * The discard bio returns only first r10bio finishes | |
1706 | */ | |
1707 | if (first_copy) { | |
1708 | r10_bio->master_bio = bio; | |
1709 | set_bit(R10BIO_Discard, &r10_bio->state); | |
1710 | first_copy = false; | |
1711 | first_r10bio = r10_bio; | |
1712 | } else | |
1713 | r10_bio->master_bio = (struct bio *)first_r10bio; | |
1714 | ||
46d4703b XN |
1715 | /* |
1716 | * first select target devices under rcu_lock and | |
1717 | * inc refcount on their rdev. Record them by setting | |
1718 | * bios[x] to bio | |
1719 | */ | |
d30588b2 XN |
1720 | rcu_read_lock(); |
1721 | for (disk = 0; disk < geo->raid_disks; disk++) { | |
1722 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
1723 | struct md_rdev *rrdev = rcu_dereference( | |
1724 | conf->mirrors[disk].replacement); | |
1725 | ||
1726 | r10_bio->devs[disk].bio = NULL; | |
1727 | r10_bio->devs[disk].repl_bio = NULL; | |
1728 | ||
1729 | if (rdev && (test_bit(Faulty, &rdev->flags))) | |
1730 | rdev = NULL; | |
1731 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) | |
1732 | rrdev = NULL; | |
1733 | if (!rdev && !rrdev) | |
1734 | continue; | |
1735 | ||
1736 | if (rdev) { | |
1737 | r10_bio->devs[disk].bio = bio; | |
1738 | atomic_inc(&rdev->nr_pending); | |
1739 | } | |
1740 | if (rrdev) { | |
1741 | r10_bio->devs[disk].repl_bio = bio; | |
1742 | atomic_inc(&rrdev->nr_pending); | |
1743 | } | |
1744 | } | |
1745 | rcu_read_unlock(); | |
1746 | ||
1747 | atomic_set(&r10_bio->remaining, 1); | |
1748 | for (disk = 0; disk < geo->raid_disks; disk++) { | |
1749 | sector_t dev_start, dev_end; | |
1750 | struct bio *mbio, *rbio = NULL; | |
d30588b2 XN |
1751 | |
1752 | /* | |
1753 | * Now start to calculate the start and end address for each disk. | |
1754 | * The space between dev_start and dev_end is the discard region. | |
1755 | * | |
1756 | * For dev_start, it needs to consider three conditions: | |
1757 | * 1st, the disk is before start_disk, you can imagine the disk in | |
1758 | * the next stripe. So the dev_start is the start address of next | |
1759 | * stripe. | |
1760 | * 2st, the disk is after start_disk, it means the disk is at the | |
1761 | * same stripe of first disk | |
1762 | * 3st, the first disk itself, we can use start_disk_offset directly | |
1763 | */ | |
1764 | if (disk < start_disk_index) | |
1765 | dev_start = (first_stripe_index + 1) * mddev->chunk_sectors; | |
1766 | else if (disk > start_disk_index) | |
1767 | dev_start = first_stripe_index * mddev->chunk_sectors; | |
1768 | else | |
1769 | dev_start = start_disk_offset; | |
1770 | ||
1771 | if (disk < end_disk_index) | |
1772 | dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; | |
1773 | else if (disk > end_disk_index) | |
1774 | dev_end = last_stripe_index * mddev->chunk_sectors; | |
1775 | else | |
1776 | dev_end = end_disk_offset; | |
1777 | ||
1778 | /* | |
1779 | * It only handles discard bio which size is >= stripe size, so | |
46d4703b XN |
1780 | * dev_end > dev_start all the time. |
1781 | * It doesn't need to use rcu lock to get rdev here. We already | |
1782 | * add rdev->nr_pending in the first loop. | |
d30588b2 XN |
1783 | */ |
1784 | if (r10_bio->devs[disk].bio) { | |
46d4703b | 1785 | struct md_rdev *rdev = conf->mirrors[disk].rdev; |
d30588b2 XN |
1786 | mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
1787 | mbio->bi_end_io = raid10_end_discard_request; | |
1788 | mbio->bi_private = r10_bio; | |
1789 | r10_bio->devs[disk].bio = mbio; | |
1790 | r10_bio->devs[disk].devnum = disk; | |
1791 | atomic_inc(&r10_bio->remaining); | |
1792 | md_submit_discard_bio(mddev, rdev, mbio, | |
1793 | dev_start + choose_data_offset(r10_bio, rdev), | |
1794 | dev_end - dev_start); | |
1795 | bio_endio(mbio); | |
1796 | } | |
1797 | if (r10_bio->devs[disk].repl_bio) { | |
46d4703b | 1798 | struct md_rdev *rrdev = conf->mirrors[disk].replacement; |
d30588b2 XN |
1799 | rbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
1800 | rbio->bi_end_io = raid10_end_discard_request; | |
1801 | rbio->bi_private = r10_bio; | |
1802 | r10_bio->devs[disk].repl_bio = rbio; | |
1803 | r10_bio->devs[disk].devnum = disk; | |
1804 | atomic_inc(&r10_bio->remaining); | |
1805 | md_submit_discard_bio(mddev, rrdev, rbio, | |
1806 | dev_start + choose_data_offset(r10_bio, rrdev), | |
1807 | dev_end - dev_start); | |
1808 | bio_endio(rbio); | |
1809 | } | |
1810 | } | |
1811 | ||
254c271d XN |
1812 | if (!geo->far_offset && --far_copies) { |
1813 | first_stripe_index += geo->stride >> geo->chunk_shift; | |
1814 | start_disk_offset += geo->stride; | |
1815 | last_stripe_index += geo->stride >> geo->chunk_shift; | |
1816 | end_disk_offset += geo->stride; | |
1817 | atomic_inc(&first_r10bio->remaining); | |
1818 | raid_end_discard_bio(r10_bio); | |
1819 | wait_barrier(conf); | |
1820 | goto retry_discard; | |
d30588b2 XN |
1821 | } |
1822 | ||
254c271d XN |
1823 | raid_end_discard_bio(r10_bio); |
1824 | ||
d30588b2 XN |
1825 | return 0; |
1826 | out: | |
1827 | allow_barrier(conf); | |
1828 | return -EAGAIN; | |
1829 | } | |
1830 | ||
cc27b0c7 | 1831 | static bool raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1832 | { |
1833 | struct r10conf *conf = mddev->private; | |
1834 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1835 | int chunk_sects = chunk_mask + 1; | |
fc9977dd | 1836 | int sectors = bio_sectors(bio); |
20d0189b | 1837 | |
775d7831 DJ |
1838 | if (unlikely(bio->bi_opf & REQ_PREFLUSH) |
1839 | && md_flush_request(mddev, bio)) | |
cc27b0c7 | 1840 | return true; |
20d0189b | 1841 | |
cc27b0c7 N |
1842 | if (!md_write_start(mddev, bio)) |
1843 | return false; | |
1844 | ||
d30588b2 XN |
1845 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
1846 | if (!raid10_handle_discard(mddev, bio)) | |
1847 | return true; | |
1848 | ||
fc9977dd N |
1849 | /* |
1850 | * If this request crosses a chunk boundary, we need to split | |
1851 | * it. | |
1852 | */ | |
1853 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1854 | sectors > chunk_sects | |
1855 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1856 | || conf->prev.near_copies < | |
1857 | conf->prev.raid_disks))) | |
1858 | sectors = chunk_sects - | |
1859 | (bio->bi_iter.bi_sector & | |
1860 | (chunk_sects - 1)); | |
1861 | __make_request(mddev, bio, sectors); | |
079fa166 N |
1862 | |
1863 | /* In case raid10d snuck in to freeze_array */ | |
1864 | wake_up(&conf->wait_barrier); | |
cc27b0c7 | 1865 | return true; |
1da177e4 LT |
1866 | } |
1867 | ||
849674e4 | 1868 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1869 | { |
e879a879 | 1870 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1871 | int i; |
1872 | ||
5cf00fcd | 1873 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1874 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1875 | if (conf->geo.near_copies > 1) |
1876 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1877 | if (conf->geo.far_copies > 1) { | |
1878 | if (conf->geo.far_offset) | |
1879 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1880 | else |
5cf00fcd | 1881 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1882 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1883 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1884 | } |
5cf00fcd N |
1885 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1886 | conf->geo.raid_disks - mddev->degraded); | |
d44b0a92 N |
1887 | rcu_read_lock(); |
1888 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
1889 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1890 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
1891 | } | |
1892 | rcu_read_unlock(); | |
1da177e4 LT |
1893 | seq_printf(seq, "]"); |
1894 | } | |
1895 | ||
700c7213 N |
1896 | /* check if there are enough drives for |
1897 | * every block to appear on atleast one. | |
1898 | * Don't consider the device numbered 'ignore' | |
1899 | * as we might be about to remove it. | |
1900 | */ | |
635f6416 | 1901 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1902 | { |
1903 | int first = 0; | |
725d6e57 | 1904 | int has_enough = 0; |
635f6416 N |
1905 | int disks, ncopies; |
1906 | if (previous) { | |
1907 | disks = conf->prev.raid_disks; | |
1908 | ncopies = conf->prev.near_copies; | |
1909 | } else { | |
1910 | disks = conf->geo.raid_disks; | |
1911 | ncopies = conf->geo.near_copies; | |
1912 | } | |
700c7213 | 1913 | |
725d6e57 | 1914 | rcu_read_lock(); |
700c7213 N |
1915 | do { |
1916 | int n = conf->copies; | |
1917 | int cnt = 0; | |
80b48124 | 1918 | int this = first; |
700c7213 | 1919 | while (n--) { |
725d6e57 N |
1920 | struct md_rdev *rdev; |
1921 | if (this != ignore && | |
1922 | (rdev = rcu_dereference(conf->mirrors[this].rdev)) && | |
1923 | test_bit(In_sync, &rdev->flags)) | |
700c7213 | 1924 | cnt++; |
635f6416 | 1925 | this = (this+1) % disks; |
700c7213 N |
1926 | } |
1927 | if (cnt == 0) | |
725d6e57 | 1928 | goto out; |
635f6416 | 1929 | first = (first + ncopies) % disks; |
700c7213 | 1930 | } while (first != 0); |
725d6e57 N |
1931 | has_enough = 1; |
1932 | out: | |
1933 | rcu_read_unlock(); | |
1934 | return has_enough; | |
700c7213 N |
1935 | } |
1936 | ||
f8c9e74f N |
1937 | static int enough(struct r10conf *conf, int ignore) |
1938 | { | |
635f6416 N |
1939 | /* when calling 'enough', both 'prev' and 'geo' must |
1940 | * be stable. | |
1941 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1942 | * is held. | |
1943 | */ | |
1944 | return _enough(conf, 0, ignore) && | |
1945 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1946 | } |
1947 | ||
849674e4 | 1948 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1949 | { |
1950 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1951 | struct r10conf *conf = mddev->private; |
635f6416 | 1952 | unsigned long flags; |
1da177e4 LT |
1953 | |
1954 | /* | |
1955 | * If it is not operational, then we have already marked it as dead | |
9a567843 GJ |
1956 | * else if it is the last working disks with "fail_last_dev == false", |
1957 | * ignore the error, let the next level up know. | |
1da177e4 LT |
1958 | * else mark the drive as failed |
1959 | */ | |
635f6416 | 1960 | spin_lock_irqsave(&conf->device_lock, flags); |
9a567843 | 1961 | if (test_bit(In_sync, &rdev->flags) && !mddev->fail_last_dev |
635f6416 | 1962 | && !enough(conf, rdev->raid_disk)) { |
1da177e4 LT |
1963 | /* |
1964 | * Don't fail the drive, just return an IO error. | |
1da177e4 | 1965 | */ |
635f6416 | 1966 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1967 | return; |
635f6416 | 1968 | } |
2446dba0 | 1969 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1970 | mddev->degraded++; |
2446dba0 N |
1971 | /* |
1972 | * If recovery is running, make sure it aborts. | |
1973 | */ | |
1974 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
de393cde | 1975 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1976 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
1977 | set_mask_bits(&mddev->sb_flags, 0, |
1978 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
635f6416 | 1979 | spin_unlock_irqrestore(&conf->device_lock, flags); |
08464e09 N |
1980 | pr_crit("md/raid10:%s: Disk failure on %s, disabling device.\n" |
1981 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
1982 | mdname(mddev), bdevname(rdev->bdev, b), | |
1983 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); | |
1da177e4 LT |
1984 | } |
1985 | ||
e879a879 | 1986 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1987 | { |
1988 | int i; | |
4056ca51 | 1989 | struct md_rdev *rdev; |
1da177e4 | 1990 | |
08464e09 | 1991 | pr_debug("RAID10 conf printout:\n"); |
1da177e4 | 1992 | if (!conf) { |
08464e09 | 1993 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1994 | return; |
1995 | } | |
08464e09 N |
1996 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1997 | conf->geo.raid_disks); | |
1da177e4 | 1998 | |
4056ca51 N |
1999 | /* This is only called with ->reconfix_mutex held, so |
2000 | * rcu protection of rdev is not needed */ | |
5cf00fcd | 2001 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 2002 | char b[BDEVNAME_SIZE]; |
4056ca51 N |
2003 | rdev = conf->mirrors[i].rdev; |
2004 | if (rdev) | |
08464e09 N |
2005 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
2006 | i, !test_bit(In_sync, &rdev->flags), | |
2007 | !test_bit(Faulty, &rdev->flags), | |
2008 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
2009 | } |
2010 | } | |
2011 | ||
e879a879 | 2012 | static void close_sync(struct r10conf *conf) |
1da177e4 | 2013 | { |
0a27ec96 N |
2014 | wait_barrier(conf); |
2015 | allow_barrier(conf); | |
1da177e4 | 2016 | |
afeee514 | 2017 | mempool_exit(&conf->r10buf_pool); |
1da177e4 LT |
2018 | } |
2019 | ||
fd01b88c | 2020 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
2021 | { |
2022 | int i; | |
e879a879 | 2023 | struct r10conf *conf = mddev->private; |
dc280d98 | 2024 | struct raid10_info *tmp; |
6b965620 N |
2025 | int count = 0; |
2026 | unsigned long flags; | |
1da177e4 LT |
2027 | |
2028 | /* | |
2029 | * Find all non-in_sync disks within the RAID10 configuration | |
2030 | * and mark them in_sync | |
2031 | */ | |
5cf00fcd | 2032 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 2033 | tmp = conf->mirrors + i; |
4ca40c2c N |
2034 | if (tmp->replacement |
2035 | && tmp->replacement->recovery_offset == MaxSector | |
2036 | && !test_bit(Faulty, &tmp->replacement->flags) | |
2037 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
2038 | /* Replacement has just become active */ | |
2039 | if (!tmp->rdev | |
2040 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
2041 | count++; | |
2042 | if (tmp->rdev) { | |
2043 | /* Replaced device not technically faulty, | |
2044 | * but we need to be sure it gets removed | |
2045 | * and never re-added. | |
2046 | */ | |
2047 | set_bit(Faulty, &tmp->rdev->flags); | |
2048 | sysfs_notify_dirent_safe( | |
2049 | tmp->rdev->sysfs_state); | |
2050 | } | |
2051 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
2052 | } else if (tmp->rdev | |
61e4947c | 2053 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
2054 | && !test_bit(Faulty, &tmp->rdev->flags) |
2055 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 2056 | count++; |
2863b9eb | 2057 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
2058 | } |
2059 | } | |
6b965620 N |
2060 | spin_lock_irqsave(&conf->device_lock, flags); |
2061 | mddev->degraded -= count; | |
2062 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
2063 | |
2064 | print_conf(conf); | |
6b965620 | 2065 | return count; |
1da177e4 LT |
2066 | } |
2067 | ||
fd01b88c | 2068 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 2069 | { |
e879a879 | 2070 | struct r10conf *conf = mddev->private; |
199050ea | 2071 | int err = -EEXIST; |
1da177e4 | 2072 | int mirror; |
6c2fce2e | 2073 | int first = 0; |
5cf00fcd | 2074 | int last = conf->geo.raid_disks - 1; |
1da177e4 LT |
2075 | |
2076 | if (mddev->recovery_cp < MaxSector) | |
2077 | /* only hot-add to in-sync arrays, as recovery is | |
2078 | * very different from resync | |
2079 | */ | |
199050ea | 2080 | return -EBUSY; |
635f6416 | 2081 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 2082 | return -EINVAL; |
1da177e4 | 2083 | |
1501efad DW |
2084 | if (md_integrity_add_rdev(rdev, mddev)) |
2085 | return -ENXIO; | |
2086 | ||
a53a6c85 | 2087 | if (rdev->raid_disk >= 0) |
6c2fce2e | 2088 | first = last = rdev->raid_disk; |
1da177e4 | 2089 | |
2c4193df | 2090 | if (rdev->saved_raid_disk >= first && |
9e753ba9 | 2091 | rdev->saved_raid_disk < conf->geo.raid_disks && |
6cce3b23 N |
2092 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
2093 | mirror = rdev->saved_raid_disk; | |
2094 | else | |
6c2fce2e | 2095 | mirror = first; |
2bb77736 | 2096 | for ( ; mirror <= last ; mirror++) { |
dc280d98 | 2097 | struct raid10_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
2098 | if (p->recovery_disabled == mddev->recovery_disabled) |
2099 | continue; | |
b7044d41 N |
2100 | if (p->rdev) { |
2101 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
2102 | p->replacement != NULL) | |
2103 | continue; | |
2104 | clear_bit(In_sync, &rdev->flags); | |
2105 | set_bit(Replacement, &rdev->flags); | |
2106 | rdev->raid_disk = mirror; | |
2107 | err = 0; | |
9092c02d JB |
2108 | if (mddev->gendisk) |
2109 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
2110 | rdev->data_offset << 9); | |
b7044d41 N |
2111 | conf->fullsync = 1; |
2112 | rcu_assign_pointer(p->replacement, rdev); | |
2113 | break; | |
2114 | } | |
1da177e4 | 2115 | |
9092c02d JB |
2116 | if (mddev->gendisk) |
2117 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
2118 | rdev->data_offset << 9); | |
1da177e4 | 2119 | |
2bb77736 | 2120 | p->head_position = 0; |
d890fa2b | 2121 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
2122 | rdev->raid_disk = mirror; |
2123 | err = 0; | |
2124 | if (rdev->saved_raid_disk != mirror) | |
2125 | conf->fullsync = 1; | |
2126 | rcu_assign_pointer(p->rdev, rdev); | |
2127 | break; | |
2128 | } | |
ed30be07 | 2129 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
8b904b5b | 2130 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue); |
532a2a3f | 2131 | |
1da177e4 | 2132 | print_conf(conf); |
199050ea | 2133 | return err; |
1da177e4 LT |
2134 | } |
2135 | ||
b8321b68 | 2136 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 2137 | { |
e879a879 | 2138 | struct r10conf *conf = mddev->private; |
1da177e4 | 2139 | int err = 0; |
b8321b68 | 2140 | int number = rdev->raid_disk; |
c8ab903e | 2141 | struct md_rdev **rdevp; |
dc280d98 | 2142 | struct raid10_info *p = conf->mirrors + number; |
1da177e4 LT |
2143 | |
2144 | print_conf(conf); | |
c8ab903e N |
2145 | if (rdev == p->rdev) |
2146 | rdevp = &p->rdev; | |
2147 | else if (rdev == p->replacement) | |
2148 | rdevp = &p->replacement; | |
2149 | else | |
2150 | return 0; | |
2151 | ||
2152 | if (test_bit(In_sync, &rdev->flags) || | |
2153 | atomic_read(&rdev->nr_pending)) { | |
2154 | err = -EBUSY; | |
2155 | goto abort; | |
2156 | } | |
d787be40 | 2157 | /* Only remove non-faulty devices if recovery |
c8ab903e N |
2158 | * is not possible. |
2159 | */ | |
2160 | if (!test_bit(Faulty, &rdev->flags) && | |
2161 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 2162 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 2163 | number < conf->geo.raid_disks && |
c8ab903e N |
2164 | enough(conf, -1)) { |
2165 | err = -EBUSY; | |
2166 | goto abort; | |
1da177e4 | 2167 | } |
c8ab903e | 2168 | *rdevp = NULL; |
d787be40 N |
2169 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
2170 | synchronize_rcu(); | |
2171 | if (atomic_read(&rdev->nr_pending)) { | |
2172 | /* lost the race, try later */ | |
2173 | err = -EBUSY; | |
2174 | *rdevp = rdev; | |
2175 | goto abort; | |
2176 | } | |
2177 | } | |
2178 | if (p->replacement) { | |
4ca40c2c N |
2179 | /* We must have just cleared 'rdev' */ |
2180 | p->rdev = p->replacement; | |
2181 | clear_bit(Replacement, &p->replacement->flags); | |
2182 | smp_mb(); /* Make sure other CPUs may see both as identical | |
2183 | * but will never see neither -- if they are careful. | |
2184 | */ | |
2185 | p->replacement = NULL; | |
e5bc9c3c | 2186 | } |
4ca40c2c | 2187 | |
e5bc9c3c | 2188 | clear_bit(WantReplacement, &rdev->flags); |
c8ab903e N |
2189 | err = md_integrity_register(mddev); |
2190 | ||
1da177e4 LT |
2191 | abort: |
2192 | ||
2193 | print_conf(conf); | |
2194 | return err; | |
2195 | } | |
2196 | ||
81fa1520 | 2197 | static void __end_sync_read(struct r10bio *r10_bio, struct bio *bio, int d) |
1da177e4 | 2198 | { |
e879a879 | 2199 | struct r10conf *conf = r10_bio->mddev->private; |
0eb3ff12 | 2200 | |
4e4cbee9 | 2201 | if (!bio->bi_status) |
0eb3ff12 | 2202 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
2203 | else |
2204 | /* The write handler will notice the lack of | |
2205 | * R10BIO_Uptodate and record any errors etc | |
2206 | */ | |
4dbcdc75 N |
2207 | atomic_add(r10_bio->sectors, |
2208 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
2209 | |
2210 | /* for reconstruct, we always reschedule after a read. | |
2211 | * for resync, only after all reads | |
2212 | */ | |
73d5c38a | 2213 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
2214 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
2215 | atomic_dec_and_test(&r10_bio->remaining)) { | |
2216 | /* we have read all the blocks, | |
2217 | * do the comparison in process context in raid10d | |
2218 | */ | |
2219 | reschedule_retry(r10_bio); | |
2220 | } | |
1da177e4 LT |
2221 | } |
2222 | ||
81fa1520 ML |
2223 | static void end_sync_read(struct bio *bio) |
2224 | { | |
f0250618 | 2225 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
81fa1520 ML |
2226 | struct r10conf *conf = r10_bio->mddev->private; |
2227 | int d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
2228 | ||
2229 | __end_sync_read(r10_bio, bio, d); | |
2230 | } | |
2231 | ||
2232 | static void end_reshape_read(struct bio *bio) | |
2233 | { | |
f0250618 | 2234 | /* reshape read bio isn't allocated from r10buf_pool */ |
81fa1520 ML |
2235 | struct r10bio *r10_bio = bio->bi_private; |
2236 | ||
2237 | __end_sync_read(r10_bio, bio, r10_bio->read_slot); | |
2238 | } | |
2239 | ||
9f2c9d12 | 2240 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 2241 | { |
fd01b88c | 2242 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 2243 | |
1da177e4 LT |
2244 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
2245 | if (r10_bio->master_bio == NULL) { | |
2246 | /* the primary of several recovery bios */ | |
73d5c38a | 2247 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
2248 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2249 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
2250 | reschedule_retry(r10_bio); |
2251 | else | |
2252 | put_buf(r10_bio); | |
73d5c38a | 2253 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
2254 | break; |
2255 | } else { | |
9f2c9d12 | 2256 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
2257 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2258 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
2259 | reschedule_retry(r10_bio); |
2260 | else | |
2261 | put_buf(r10_bio); | |
1da177e4 LT |
2262 | r10_bio = r10_bio2; |
2263 | } | |
2264 | } | |
1da177e4 LT |
2265 | } |
2266 | ||
4246a0b6 | 2267 | static void end_sync_write(struct bio *bio) |
5e570289 | 2268 | { |
f0250618 | 2269 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
fd01b88c | 2270 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2271 | struct r10conf *conf = mddev->private; |
5e570289 N |
2272 | int d; |
2273 | sector_t first_bad; | |
2274 | int bad_sectors; | |
2275 | int slot; | |
9ad1aefc | 2276 | int repl; |
4ca40c2c | 2277 | struct md_rdev *rdev = NULL; |
5e570289 | 2278 | |
9ad1aefc N |
2279 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
2280 | if (repl) | |
2281 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 2282 | else |
9ad1aefc | 2283 | rdev = conf->mirrors[d].rdev; |
5e570289 | 2284 | |
4e4cbee9 | 2285 | if (bio->bi_status) { |
9ad1aefc N |
2286 | if (repl) |
2287 | md_error(mddev, rdev); | |
2288 | else { | |
2289 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
2290 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2291 | set_bit(MD_RECOVERY_NEEDED, | |
2292 | &rdev->mddev->recovery); | |
9ad1aefc N |
2293 | set_bit(R10BIO_WriteError, &r10_bio->state); |
2294 | } | |
2295 | } else if (is_badblock(rdev, | |
5e570289 N |
2296 | r10_bio->devs[slot].addr, |
2297 | r10_bio->sectors, | |
2298 | &first_bad, &bad_sectors)) | |
2299 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
2300 | ||
9ad1aefc | 2301 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
2302 | |
2303 | end_sync_request(r10_bio); | |
2304 | } | |
2305 | ||
1da177e4 LT |
2306 | /* |
2307 | * Note: sync and recover and handled very differently for raid10 | |
2308 | * This code is for resync. | |
2309 | * For resync, we read through virtual addresses and read all blocks. | |
2310 | * If there is any error, we schedule a write. The lowest numbered | |
2311 | * drive is authoritative. | |
2312 | * However requests come for physical address, so we need to map. | |
2313 | * For every physical address there are raid_disks/copies virtual addresses, | |
2314 | * which is always are least one, but is not necessarly an integer. | |
2315 | * This means that a physical address can span multiple chunks, so we may | |
2316 | * have to submit multiple io requests for a single sync request. | |
2317 | */ | |
2318 | /* | |
2319 | * We check if all blocks are in-sync and only write to blocks that | |
2320 | * aren't in sync | |
2321 | */ | |
9f2c9d12 | 2322 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2323 | { |
e879a879 | 2324 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
2325 | int i, first; |
2326 | struct bio *tbio, *fbio; | |
f4380a91 | 2327 | int vcnt; |
cdb76be3 | 2328 | struct page **tpages, **fpages; |
1da177e4 LT |
2329 | |
2330 | atomic_set(&r10_bio->remaining, 1); | |
2331 | ||
2332 | /* find the first device with a block */ | |
2333 | for (i=0; i<conf->copies; i++) | |
4e4cbee9 | 2334 | if (!r10_bio->devs[i].bio->bi_status) |
1da177e4 LT |
2335 | break; |
2336 | ||
2337 | if (i == conf->copies) | |
2338 | goto done; | |
2339 | ||
2340 | first = i; | |
2341 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
2342 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
2343 | fbio->bi_iter.bi_idx = 0; | |
cdb76be3 | 2344 | fpages = get_resync_pages(fbio)->pages; |
1da177e4 | 2345 | |
f4380a91 | 2346 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 2347 | /* now find blocks with errors */ |
0eb3ff12 N |
2348 | for (i=0 ; i < conf->copies ; i++) { |
2349 | int j, d; | |
8d3ca83d | 2350 | struct md_rdev *rdev; |
f0250618 | 2351 | struct resync_pages *rp; |
1da177e4 | 2352 | |
1da177e4 | 2353 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
2354 | |
2355 | if (tbio->bi_end_io != end_sync_read) | |
2356 | continue; | |
2357 | if (i == first) | |
1da177e4 | 2358 | continue; |
cdb76be3 ML |
2359 | |
2360 | tpages = get_resync_pages(tbio)->pages; | |
8d3ca83d N |
2361 | d = r10_bio->devs[i].devnum; |
2362 | rdev = conf->mirrors[d].rdev; | |
4e4cbee9 | 2363 | if (!r10_bio->devs[i].bio->bi_status) { |
0eb3ff12 N |
2364 | /* We know that the bi_io_vec layout is the same for |
2365 | * both 'first' and 'i', so we just compare them. | |
2366 | * All vec entries are PAGE_SIZE; | |
2367 | */ | |
7bb23c49 N |
2368 | int sectors = r10_bio->sectors; |
2369 | for (j = 0; j < vcnt; j++) { | |
2370 | int len = PAGE_SIZE; | |
2371 | if (sectors < (len / 512)) | |
2372 | len = sectors * 512; | |
cdb76be3 ML |
2373 | if (memcmp(page_address(fpages[j]), |
2374 | page_address(tpages[j]), | |
7bb23c49 | 2375 | len)) |
0eb3ff12 | 2376 | break; |
7bb23c49 N |
2377 | sectors -= len/512; |
2378 | } | |
0eb3ff12 N |
2379 | if (j == vcnt) |
2380 | continue; | |
7f7583d4 | 2381 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
2382 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
2383 | /* Don't fix anything. */ | |
2384 | continue; | |
8d3ca83d N |
2385 | } else if (test_bit(FailFast, &rdev->flags)) { |
2386 | /* Just give up on this device */ | |
2387 | md_error(rdev->mddev, rdev); | |
2388 | continue; | |
0eb3ff12 | 2389 | } |
f84ee364 N |
2390 | /* Ok, we need to write this bio, either to correct an |
2391 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
2392 | * First we need to fixup bv_offset, bv_len and |
2393 | * bi_vecs, as the read request might have corrupted these | |
2394 | */ | |
f0250618 | 2395 | rp = get_resync_pages(tbio); |
8be185f2 KO |
2396 | bio_reset(tbio); |
2397 | ||
fb0eb5df ML |
2398 | md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size); |
2399 | ||
f0250618 ML |
2400 | rp->raid_bio = r10_bio; |
2401 | tbio->bi_private = rp; | |
4f024f37 | 2402 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 | 2403 | tbio->bi_end_io = end_sync_write; |
796a5cf0 | 2404 | bio_set_op_attrs(tbio, REQ_OP_WRITE, 0); |
1da177e4 | 2405 | |
c31df25f KO |
2406 | bio_copy_data(tbio, fbio); |
2407 | ||
1da177e4 LT |
2408 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2409 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2410 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2411 | |
1919cbb2 N |
2412 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
2413 | tbio->bi_opf |= MD_FAILFAST; | |
4f024f37 | 2414 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
74d46992 | 2415 | bio_set_dev(tbio, conf->mirrors[d].rdev->bdev); |
ed00aabd | 2416 | submit_bio_noacct(tbio); |
1da177e4 LT |
2417 | } |
2418 | ||
9ad1aefc N |
2419 | /* Now write out to any replacement devices |
2420 | * that are active | |
2421 | */ | |
2422 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2423 | int d; |
9ad1aefc N |
2424 | |
2425 | tbio = r10_bio->devs[i].repl_bio; | |
2426 | if (!tbio || !tbio->bi_end_io) | |
2427 | continue; | |
2428 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2429 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2430 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2431 | d = r10_bio->devs[i].devnum; |
2432 | atomic_inc(&r10_bio->remaining); | |
2433 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2434 | bio_sectors(tbio)); |
ed00aabd | 2435 | submit_bio_noacct(tbio); |
9ad1aefc N |
2436 | } |
2437 | ||
1da177e4 LT |
2438 | done: |
2439 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2440 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2441 | put_buf(r10_bio); | |
2442 | } | |
2443 | } | |
2444 | ||
2445 | /* | |
2446 | * Now for the recovery code. | |
2447 | * Recovery happens across physical sectors. | |
2448 | * We recover all non-is_sync drives by finding the virtual address of | |
2449 | * each, and then choose a working drive that also has that virt address. | |
2450 | * There is a separate r10_bio for each non-in_sync drive. | |
2451 | * Only the first two slots are in use. The first for reading, | |
2452 | * The second for writing. | |
2453 | * | |
2454 | */ | |
9f2c9d12 | 2455 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2456 | { |
2457 | /* We got a read error during recovery. | |
2458 | * We repeat the read in smaller page-sized sections. | |
2459 | * If a read succeeds, write it to the new device or record | |
2460 | * a bad block if we cannot. | |
2461 | * If a read fails, record a bad block on both old and | |
2462 | * new devices. | |
2463 | */ | |
fd01b88c | 2464 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2465 | struct r10conf *conf = mddev->private; |
5e570289 N |
2466 | struct bio *bio = r10_bio->devs[0].bio; |
2467 | sector_t sect = 0; | |
2468 | int sectors = r10_bio->sectors; | |
2469 | int idx = 0; | |
2470 | int dr = r10_bio->devs[0].devnum; | |
2471 | int dw = r10_bio->devs[1].devnum; | |
cdb76be3 | 2472 | struct page **pages = get_resync_pages(bio)->pages; |
5e570289 N |
2473 | |
2474 | while (sectors) { | |
2475 | int s = sectors; | |
3cb03002 | 2476 | struct md_rdev *rdev; |
5e570289 N |
2477 | sector_t addr; |
2478 | int ok; | |
2479 | ||
2480 | if (s > (PAGE_SIZE>>9)) | |
2481 | s = PAGE_SIZE >> 9; | |
2482 | ||
2483 | rdev = conf->mirrors[dr].rdev; | |
2484 | addr = r10_bio->devs[0].addr + sect, | |
2485 | ok = sync_page_io(rdev, | |
2486 | addr, | |
2487 | s << 9, | |
cdb76be3 | 2488 | pages[idx], |
796a5cf0 | 2489 | REQ_OP_READ, 0, false); |
5e570289 N |
2490 | if (ok) { |
2491 | rdev = conf->mirrors[dw].rdev; | |
2492 | addr = r10_bio->devs[1].addr + sect; | |
2493 | ok = sync_page_io(rdev, | |
2494 | addr, | |
2495 | s << 9, | |
cdb76be3 | 2496 | pages[idx], |
796a5cf0 | 2497 | REQ_OP_WRITE, 0, false); |
b7044d41 | 2498 | if (!ok) { |
5e570289 | 2499 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2500 | if (!test_and_set_bit(WantReplacement, |
2501 | &rdev->flags)) | |
2502 | set_bit(MD_RECOVERY_NEEDED, | |
2503 | &rdev->mddev->recovery); | |
2504 | } | |
5e570289 N |
2505 | } |
2506 | if (!ok) { | |
2507 | /* We don't worry if we cannot set a bad block - | |
2508 | * it really is bad so there is no loss in not | |
2509 | * recording it yet | |
2510 | */ | |
2511 | rdev_set_badblocks(rdev, addr, s, 0); | |
2512 | ||
2513 | if (rdev != conf->mirrors[dw].rdev) { | |
2514 | /* need bad block on destination too */ | |
3cb03002 | 2515 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2516 | addr = r10_bio->devs[1].addr + sect; |
2517 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2518 | if (!ok) { | |
2519 | /* just abort the recovery */ | |
08464e09 N |
2520 | pr_notice("md/raid10:%s: recovery aborted due to read error\n", |
2521 | mdname(mddev)); | |
5e570289 N |
2522 | |
2523 | conf->mirrors[dw].recovery_disabled | |
2524 | = mddev->recovery_disabled; | |
2525 | set_bit(MD_RECOVERY_INTR, | |
2526 | &mddev->recovery); | |
2527 | break; | |
2528 | } | |
2529 | } | |
2530 | } | |
2531 | ||
2532 | sectors -= s; | |
2533 | sect += s; | |
2534 | idx++; | |
2535 | } | |
2536 | } | |
1da177e4 | 2537 | |
9f2c9d12 | 2538 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2539 | { |
e879a879 | 2540 | struct r10conf *conf = mddev->private; |
c65060ad | 2541 | int d; |
24afd80d | 2542 | struct bio *wbio, *wbio2; |
1da177e4 | 2543 | |
5e570289 N |
2544 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2545 | fix_recovery_read_error(r10_bio); | |
2546 | end_sync_request(r10_bio); | |
2547 | return; | |
2548 | } | |
2549 | ||
c65060ad NK |
2550 | /* |
2551 | * share the pages with the first bio | |
1da177e4 LT |
2552 | * and submit the write request |
2553 | */ | |
1da177e4 | 2554 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2555 | wbio = r10_bio->devs[1].bio; |
2556 | wbio2 = r10_bio->devs[1].repl_bio; | |
0eb25bb0 | 2557 | /* Need to test wbio2->bi_end_io before we call |
ed00aabd | 2558 | * submit_bio_noacct as if the former is NULL, |
0eb25bb0 N |
2559 | * the latter is free to free wbio2. |
2560 | */ | |
2561 | if (wbio2 && !wbio2->bi_end_io) | |
2562 | wbio2 = NULL; | |
24afd80d N |
2563 | if (wbio->bi_end_io) { |
2564 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2565 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
ed00aabd | 2566 | submit_bio_noacct(wbio); |
24afd80d | 2567 | } |
0eb25bb0 | 2568 | if (wbio2) { |
24afd80d N |
2569 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2570 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2571 | bio_sectors(wbio2)); |
ed00aabd | 2572 | submit_bio_noacct(wbio2); |
24afd80d | 2573 | } |
1da177e4 LT |
2574 | } |
2575 | ||
1e50915f RB |
2576 | /* |
2577 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2578 | * We halve the read error count for every hour that has elapsed | |
2579 | * since the last recorded read error. | |
2580 | * | |
2581 | */ | |
fd01b88c | 2582 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f | 2583 | { |
0e3ef49e | 2584 | long cur_time_mon; |
1e50915f RB |
2585 | unsigned long hours_since_last; |
2586 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2587 | ||
0e3ef49e | 2588 | cur_time_mon = ktime_get_seconds(); |
1e50915f | 2589 | |
0e3ef49e | 2590 | if (rdev->last_read_error == 0) { |
1e50915f RB |
2591 | /* first time we've seen a read error */ |
2592 | rdev->last_read_error = cur_time_mon; | |
2593 | return; | |
2594 | } | |
2595 | ||
0e3ef49e AB |
2596 | hours_since_last = (long)(cur_time_mon - |
2597 | rdev->last_read_error) / 3600; | |
1e50915f RB |
2598 | |
2599 | rdev->last_read_error = cur_time_mon; | |
2600 | ||
2601 | /* | |
2602 | * if hours_since_last is > the number of bits in read_errors | |
2603 | * just set read errors to 0. We do this to avoid | |
2604 | * overflowing the shift of read_errors by hours_since_last. | |
2605 | */ | |
2606 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2607 | atomic_set(&rdev->read_errors, 0); | |
2608 | else | |
2609 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2610 | } | |
2611 | ||
3cb03002 | 2612 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2613 | int sectors, struct page *page, int rw) |
2614 | { | |
2615 | sector_t first_bad; | |
2616 | int bad_sectors; | |
2617 | ||
2618 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2619 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2620 | return -1; | |
796a5cf0 | 2621 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
58c54fcc N |
2622 | /* success */ |
2623 | return 1; | |
b7044d41 | 2624 | if (rw == WRITE) { |
58c54fcc | 2625 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2626 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2627 | set_bit(MD_RECOVERY_NEEDED, | |
2628 | &rdev->mddev->recovery); | |
2629 | } | |
58c54fcc N |
2630 | /* need to record an error - either for the block or the device */ |
2631 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2632 | md_error(rdev->mddev, rdev); | |
2633 | return 0; | |
2634 | } | |
2635 | ||
1da177e4 LT |
2636 | /* |
2637 | * This is a kernel thread which: | |
2638 | * | |
2639 | * 1. Retries failed read operations on working mirrors. | |
2640 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2641 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2642 | */ |
2643 | ||
e879a879 | 2644 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2645 | { |
2646 | int sect = 0; /* Offset from r10_bio->sector */ | |
2647 | int sectors = r10_bio->sectors; | |
13db16d7 | 2648 | struct md_rdev *rdev; |
1e50915f | 2649 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2650 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2651 | |
7c4e06ff N |
2652 | /* still own a reference to this rdev, so it cannot |
2653 | * have been cleared recently. | |
2654 | */ | |
2655 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2656 | |
7c4e06ff N |
2657 | if (test_bit(Faulty, &rdev->flags)) |
2658 | /* drive has already been failed, just ignore any | |
2659 | more fix_read_error() attempts */ | |
2660 | return; | |
1e50915f | 2661 | |
7c4e06ff N |
2662 | check_decay_read_errors(mddev, rdev); |
2663 | atomic_inc(&rdev->read_errors); | |
2664 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2665 | char b[BDEVNAME_SIZE]; | |
2666 | bdevname(rdev->bdev, b); | |
1e50915f | 2667 | |
08464e09 N |
2668 | pr_notice("md/raid10:%s: %s: Raid device exceeded read_error threshold [cur %d:max %d]\n", |
2669 | mdname(mddev), b, | |
2670 | atomic_read(&rdev->read_errors), max_read_errors); | |
2671 | pr_notice("md/raid10:%s: %s: Failing raid device\n", | |
2672 | mdname(mddev), b); | |
d683c8e0 | 2673 | md_error(mddev, rdev); |
fae8cc5e | 2674 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2675 | return; |
1e50915f | 2676 | } |
1e50915f | 2677 | |
6814d536 N |
2678 | while(sectors) { |
2679 | int s = sectors; | |
2680 | int sl = r10_bio->read_slot; | |
2681 | int success = 0; | |
2682 | int start; | |
2683 | ||
2684 | if (s > (PAGE_SIZE>>9)) | |
2685 | s = PAGE_SIZE >> 9; | |
2686 | ||
2687 | rcu_read_lock(); | |
2688 | do { | |
8dbed5ce N |
2689 | sector_t first_bad; |
2690 | int bad_sectors; | |
2691 | ||
0544a21d | 2692 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2693 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2694 | if (rdev && | |
8dbed5ce | 2695 | test_bit(In_sync, &rdev->flags) && |
f5b67ae8 | 2696 | !test_bit(Faulty, &rdev->flags) && |
8dbed5ce N |
2697 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, |
2698 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2699 | atomic_inc(&rdev->nr_pending); |
2700 | rcu_read_unlock(); | |
2b193363 | 2701 | success = sync_page_io(rdev, |
6814d536 | 2702 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2703 | sect, |
6814d536 | 2704 | s<<9, |
796a5cf0 MC |
2705 | conf->tmppage, |
2706 | REQ_OP_READ, 0, false); | |
6814d536 N |
2707 | rdev_dec_pending(rdev, mddev); |
2708 | rcu_read_lock(); | |
2709 | if (success) | |
2710 | break; | |
2711 | } | |
2712 | sl++; | |
2713 | if (sl == conf->copies) | |
2714 | sl = 0; | |
2715 | } while (!success && sl != r10_bio->read_slot); | |
2716 | rcu_read_unlock(); | |
2717 | ||
2718 | if (!success) { | |
58c54fcc N |
2719 | /* Cannot read from anywhere, just mark the block |
2720 | * as bad on the first device to discourage future | |
2721 | * reads. | |
2722 | */ | |
6814d536 | 2723 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2724 | rdev = conf->mirrors[dn].rdev; |
2725 | ||
2726 | if (!rdev_set_badblocks( | |
2727 | rdev, | |
2728 | r10_bio->devs[r10_bio->read_slot].addr | |
2729 | + sect, | |
fae8cc5e | 2730 | s, 0)) { |
58c54fcc | 2731 | md_error(mddev, rdev); |
fae8cc5e N |
2732 | r10_bio->devs[r10_bio->read_slot].bio |
2733 | = IO_BLOCKED; | |
2734 | } | |
6814d536 N |
2735 | break; |
2736 | } | |
2737 | ||
2738 | start = sl; | |
2739 | /* write it back and re-read */ | |
2740 | rcu_read_lock(); | |
2741 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2742 | char b[BDEVNAME_SIZE]; |
0544a21d | 2743 | |
6814d536 N |
2744 | if (sl==0) |
2745 | sl = conf->copies; | |
2746 | sl--; | |
2747 | d = r10_bio->devs[sl].devnum; | |
2748 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2749 | if (!rdev || |
f5b67ae8 | 2750 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2751 | !test_bit(In_sync, &rdev->flags)) |
2752 | continue; | |
2753 | ||
2754 | atomic_inc(&rdev->nr_pending); | |
2755 | rcu_read_unlock(); | |
58c54fcc N |
2756 | if (r10_sync_page_io(rdev, |
2757 | r10_bio->devs[sl].addr + | |
2758 | sect, | |
055d3747 | 2759 | s, conf->tmppage, WRITE) |
1294b9c9 N |
2760 | == 0) { |
2761 | /* Well, this device is dead */ | |
08464e09 N |
2762 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %s)\n", |
2763 | mdname(mddev), s, | |
2764 | (unsigned long long)( | |
2765 | sect + | |
2766 | choose_data_offset(r10_bio, | |
2767 | rdev)), | |
2768 | bdevname(rdev->bdev, b)); | |
2769 | pr_notice("md/raid10:%s: %s: failing drive\n", | |
2770 | mdname(mddev), | |
2771 | bdevname(rdev->bdev, b)); | |
6814d536 | 2772 | } |
1294b9c9 N |
2773 | rdev_dec_pending(rdev, mddev); |
2774 | rcu_read_lock(); | |
6814d536 N |
2775 | } |
2776 | sl = start; | |
2777 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2778 | char b[BDEVNAME_SIZE]; |
0544a21d | 2779 | |
6814d536 N |
2780 | if (sl==0) |
2781 | sl = conf->copies; | |
2782 | sl--; | |
2783 | d = r10_bio->devs[sl].devnum; | |
2784 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2785 | if (!rdev || |
f5b67ae8 | 2786 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2787 | !test_bit(In_sync, &rdev->flags)) |
2788 | continue; | |
6814d536 | 2789 | |
1294b9c9 N |
2790 | atomic_inc(&rdev->nr_pending); |
2791 | rcu_read_unlock(); | |
58c54fcc N |
2792 | switch (r10_sync_page_io(rdev, |
2793 | r10_bio->devs[sl].addr + | |
2794 | sect, | |
055d3747 | 2795 | s, conf->tmppage, |
58c54fcc N |
2796 | READ)) { |
2797 | case 0: | |
1294b9c9 | 2798 | /* Well, this device is dead */ |
08464e09 | 2799 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2800 | mdname(mddev), s, |
2801 | (unsigned long long)( | |
f8c9e74f N |
2802 | sect + |
2803 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 | 2804 | bdevname(rdev->bdev, b)); |
08464e09 | 2805 | pr_notice("md/raid10:%s: %s: failing drive\n", |
1294b9c9 N |
2806 | mdname(mddev), |
2807 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2808 | break; |
2809 | case 1: | |
08464e09 | 2810 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2811 | mdname(mddev), s, |
2812 | (unsigned long long)( | |
f8c9e74f N |
2813 | sect + |
2814 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2815 | bdevname(rdev->bdev, b)); |
2816 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2817 | } |
1294b9c9 N |
2818 | |
2819 | rdev_dec_pending(rdev, mddev); | |
2820 | rcu_read_lock(); | |
6814d536 N |
2821 | } |
2822 | rcu_read_unlock(); | |
2823 | ||
2824 | sectors -= s; | |
2825 | sect += s; | |
2826 | } | |
2827 | } | |
2828 | ||
9f2c9d12 | 2829 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2830 | { |
2831 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2832 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2833 | struct r10conf *conf = mddev->private; |
3cb03002 | 2834 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2835 | /* bio has the data to be written to slot 'i' where |
2836 | * we just recently had a write error. | |
2837 | * We repeatedly clone the bio and trim down to one block, | |
2838 | * then try the write. Where the write fails we record | |
2839 | * a bad block. | |
2840 | * It is conceivable that the bio doesn't exactly align with | |
2841 | * blocks. We must handle this. | |
2842 | * | |
2843 | * We currently own a reference to the rdev. | |
2844 | */ | |
2845 | ||
2846 | int block_sectors; | |
2847 | sector_t sector; | |
2848 | int sectors; | |
2849 | int sect_to_write = r10_bio->sectors; | |
2850 | int ok = 1; | |
2851 | ||
2852 | if (rdev->badblocks.shift < 0) | |
2853 | return 0; | |
2854 | ||
f04ebb0b N |
2855 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2856 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2857 | sector = r10_bio->sector; |
2858 | sectors = ((r10_bio->sector + block_sectors) | |
2859 | & ~(sector_t)(block_sectors - 1)) | |
2860 | - sector; | |
2861 | ||
2862 | while (sect_to_write) { | |
2863 | struct bio *wbio; | |
27028626 | 2864 | sector_t wsector; |
bd870a16 N |
2865 | if (sectors > sect_to_write) |
2866 | sectors = sect_to_write; | |
2867 | /* Write at 'sector' for 'sectors' */ | |
afeee514 | 2868 | wbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
4f024f37 | 2869 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
27028626 TM |
2870 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); |
2871 | wbio->bi_iter.bi_sector = wsector + | |
2872 | choose_data_offset(r10_bio, rdev); | |
74d46992 | 2873 | bio_set_dev(wbio, rdev->bdev); |
796a5cf0 | 2874 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
2875 | |
2876 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 | 2877 | /* Failure! */ |
27028626 | 2878 | ok = rdev_set_badblocks(rdev, wsector, |
bd870a16 N |
2879 | sectors, 0) |
2880 | && ok; | |
2881 | ||
2882 | bio_put(wbio); | |
2883 | sect_to_write -= sectors; | |
2884 | sector += sectors; | |
2885 | sectors = block_sectors; | |
2886 | } | |
2887 | return ok; | |
2888 | } | |
2889 | ||
9f2c9d12 | 2890 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2891 | { |
2892 | int slot = r10_bio->read_slot; | |
560f8e55 | 2893 | struct bio *bio; |
e879a879 | 2894 | struct r10conf *conf = mddev->private; |
abbf098e | 2895 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2896 | |
2897 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2898 | * the block and we can fix it. | |
2899 | * We freeze all other IO, and try reading the block from | |
2900 | * other devices. When we find one, we re-write | |
2901 | * and check it that fixes the read error. | |
2902 | * This is all done synchronously while the array is | |
2903 | * frozen. | |
2904 | */ | |
fae8cc5e | 2905 | bio = r10_bio->devs[slot].bio; |
fae8cc5e N |
2906 | bio_put(bio); |
2907 | r10_bio->devs[slot].bio = NULL; | |
2908 | ||
8d3ca83d N |
2909 | if (mddev->ro) |
2910 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2911 | else if (!test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2912 | freeze_array(conf, 1); |
560f8e55 N |
2913 | fix_read_error(conf, mddev, r10_bio); |
2914 | unfreeze_array(conf); | |
fae8cc5e | 2915 | } else |
8d3ca83d | 2916 | md_error(mddev, rdev); |
fae8cc5e | 2917 | |
abbf098e | 2918 | rdev_dec_pending(rdev, mddev); |
545250f2 N |
2919 | allow_barrier(conf); |
2920 | r10_bio->state = 0; | |
2921 | raid10_read_request(mddev, r10_bio->master_bio, r10_bio); | |
560f8e55 N |
2922 | } |
2923 | ||
e879a879 | 2924 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2925 | { |
2926 | /* Some sort of write request has finished and it | |
2927 | * succeeded in writing where we thought there was a | |
2928 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2929 | * Or possibly if failed and we need to record |
2930 | * a bad block. | |
749c55e9 N |
2931 | */ |
2932 | int m; | |
3cb03002 | 2933 | struct md_rdev *rdev; |
749c55e9 N |
2934 | |
2935 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2936 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2937 | for (m = 0; m < conf->copies; m++) { |
2938 | int dev = r10_bio->devs[m].devnum; | |
2939 | rdev = conf->mirrors[dev].rdev; | |
01a69cab YY |
2940 | if (r10_bio->devs[m].bio == NULL || |
2941 | r10_bio->devs[m].bio->bi_end_io == NULL) | |
1a0b7cd8 | 2942 | continue; |
4e4cbee9 | 2943 | if (!r10_bio->devs[m].bio->bi_status) { |
749c55e9 N |
2944 | rdev_clear_badblocks( |
2945 | rdev, | |
2946 | r10_bio->devs[m].addr, | |
c6563a8c | 2947 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2948 | } else { |
2949 | if (!rdev_set_badblocks( | |
2950 | rdev, | |
2951 | r10_bio->devs[m].addr, | |
2952 | r10_bio->sectors, 0)) | |
2953 | md_error(conf->mddev, rdev); | |
749c55e9 | 2954 | } |
9ad1aefc | 2955 | rdev = conf->mirrors[dev].replacement; |
01a69cab YY |
2956 | if (r10_bio->devs[m].repl_bio == NULL || |
2957 | r10_bio->devs[m].repl_bio->bi_end_io == NULL) | |
9ad1aefc | 2958 | continue; |
4246a0b6 | 2959 | |
4e4cbee9 | 2960 | if (!r10_bio->devs[m].repl_bio->bi_status) { |
9ad1aefc N |
2961 | rdev_clear_badblocks( |
2962 | rdev, | |
2963 | r10_bio->devs[m].addr, | |
c6563a8c | 2964 | r10_bio->sectors, 0); |
9ad1aefc N |
2965 | } else { |
2966 | if (!rdev_set_badblocks( | |
2967 | rdev, | |
2968 | r10_bio->devs[m].addr, | |
2969 | r10_bio->sectors, 0)) | |
2970 | md_error(conf->mddev, rdev); | |
2971 | } | |
1a0b7cd8 | 2972 | } |
749c55e9 N |
2973 | put_buf(r10_bio); |
2974 | } else { | |
95af587e | 2975 | bool fail = false; |
bd870a16 N |
2976 | for (m = 0; m < conf->copies; m++) { |
2977 | int dev = r10_bio->devs[m].devnum; | |
2978 | struct bio *bio = r10_bio->devs[m].bio; | |
2979 | rdev = conf->mirrors[dev].rdev; | |
2980 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2981 | rdev_clear_badblocks( |
2982 | rdev, | |
2983 | r10_bio->devs[m].addr, | |
c6563a8c | 2984 | r10_bio->sectors, 0); |
749c55e9 | 2985 | rdev_dec_pending(rdev, conf->mddev); |
4e4cbee9 | 2986 | } else if (bio != NULL && bio->bi_status) { |
95af587e | 2987 | fail = true; |
bd870a16 N |
2988 | if (!narrow_write_error(r10_bio, m)) { |
2989 | md_error(conf->mddev, rdev); | |
2990 | set_bit(R10BIO_Degraded, | |
2991 | &r10_bio->state); | |
2992 | } | |
2993 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2994 | } |
475b0321 N |
2995 | bio = r10_bio->devs[m].repl_bio; |
2996 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2997 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2998 | rdev_clear_badblocks( |
2999 | rdev, | |
3000 | r10_bio->devs[m].addr, | |
c6563a8c | 3001 | r10_bio->sectors, 0); |
475b0321 N |
3002 | rdev_dec_pending(rdev, conf->mddev); |
3003 | } | |
bd870a16 | 3004 | } |
95af587e N |
3005 | if (fail) { |
3006 | spin_lock_irq(&conf->device_lock); | |
3007 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 3008 | conf->nr_queued++; |
95af587e | 3009 | spin_unlock_irq(&conf->device_lock); |
cf25ae78 GJ |
3010 | /* |
3011 | * In case freeze_array() is waiting for condition | |
3012 | * nr_pending == nr_queued + extra to be true. | |
3013 | */ | |
3014 | wake_up(&conf->wait_barrier); | |
95af587e | 3015 | md_wakeup_thread(conf->mddev->thread); |
c340702c N |
3016 | } else { |
3017 | if (test_bit(R10BIO_WriteError, | |
3018 | &r10_bio->state)) | |
3019 | close_write(r10_bio); | |
95af587e | 3020 | raid_end_bio_io(r10_bio); |
c340702c | 3021 | } |
749c55e9 N |
3022 | } |
3023 | } | |
3024 | ||
4ed8731d | 3025 | static void raid10d(struct md_thread *thread) |
1da177e4 | 3026 | { |
4ed8731d | 3027 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 3028 | struct r10bio *r10_bio; |
1da177e4 | 3029 | unsigned long flags; |
e879a879 | 3030 | struct r10conf *conf = mddev->private; |
1da177e4 | 3031 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 3032 | struct blk_plug plug; |
1da177e4 LT |
3033 | |
3034 | md_check_recovery(mddev); | |
1da177e4 | 3035 | |
95af587e | 3036 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 3037 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
95af587e N |
3038 | LIST_HEAD(tmp); |
3039 | spin_lock_irqsave(&conf->device_lock, flags); | |
2953079c | 3040 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
23ddba80 SL |
3041 | while (!list_empty(&conf->bio_end_io_list)) { |
3042 | list_move(conf->bio_end_io_list.prev, &tmp); | |
3043 | conf->nr_queued--; | |
3044 | } | |
95af587e N |
3045 | } |
3046 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
3047 | while (!list_empty(&tmp)) { | |
a452744b MP |
3048 | r10_bio = list_first_entry(&tmp, struct r10bio, |
3049 | retry_list); | |
95af587e | 3050 | list_del(&r10_bio->retry_list); |
c340702c N |
3051 | if (mddev->degraded) |
3052 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
3053 | ||
3054 | if (test_bit(R10BIO_WriteError, | |
3055 | &r10_bio->state)) | |
3056 | close_write(r10_bio); | |
95af587e N |
3057 | raid_end_bio_io(r10_bio); |
3058 | } | |
3059 | } | |
3060 | ||
e1dfa0a2 | 3061 | blk_start_plug(&plug); |
1da177e4 | 3062 | for (;;) { |
6cce3b23 | 3063 | |
0021b7bc | 3064 | flush_pending_writes(conf); |
6cce3b23 | 3065 | |
a35e63ef N |
3066 | spin_lock_irqsave(&conf->device_lock, flags); |
3067 | if (list_empty(head)) { | |
3068 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 3069 | break; |
a35e63ef | 3070 | } |
9f2c9d12 | 3071 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 3072 | list_del(head->prev); |
4443ae10 | 3073 | conf->nr_queued--; |
1da177e4 LT |
3074 | spin_unlock_irqrestore(&conf->device_lock, flags); |
3075 | ||
3076 | mddev = r10_bio->mddev; | |
070ec55d | 3077 | conf = mddev->private; |
bd870a16 N |
3078 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
3079 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 3080 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
3081 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
3082 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 3083 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 3084 | sync_request_write(mddev, r10_bio); |
7eaceacc | 3085 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 3086 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 3087 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 3088 | handle_read_error(mddev, r10_bio); |
fc9977dd N |
3089 | else |
3090 | WARN_ON_ONCE(1); | |
560f8e55 | 3091 | |
1d9d5241 | 3092 | cond_resched(); |
2953079c | 3093 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 3094 | md_check_recovery(mddev); |
1da177e4 | 3095 | } |
e1dfa0a2 | 3096 | blk_finish_plug(&plug); |
1da177e4 LT |
3097 | } |
3098 | ||
e879a879 | 3099 | static int init_resync(struct r10conf *conf) |
1da177e4 | 3100 | { |
afeee514 | 3101 | int ret, buffs, i; |
1da177e4 LT |
3102 | |
3103 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 | 3104 | BUG_ON(mempool_initialized(&conf->r10buf_pool)); |
69335ef3 | 3105 | conf->have_replacement = 0; |
5cf00fcd | 3106 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
3107 | if (conf->mirrors[i].replacement) |
3108 | conf->have_replacement = 1; | |
afeee514 KO |
3109 | ret = mempool_init(&conf->r10buf_pool, buffs, |
3110 | r10buf_pool_alloc, r10buf_pool_free, conf); | |
3111 | if (ret) | |
3112 | return ret; | |
1da177e4 LT |
3113 | conf->next_resync = 0; |
3114 | return 0; | |
3115 | } | |
3116 | ||
208410b5 SL |
3117 | static struct r10bio *raid10_alloc_init_r10buf(struct r10conf *conf) |
3118 | { | |
afeee514 | 3119 | struct r10bio *r10bio = mempool_alloc(&conf->r10buf_pool, GFP_NOIO); |
208410b5 SL |
3120 | struct rsync_pages *rp; |
3121 | struct bio *bio; | |
3122 | int nalloc; | |
3123 | int i; | |
3124 | ||
3125 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || | |
3126 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
3127 | nalloc = conf->copies; /* resync */ | |
3128 | else | |
3129 | nalloc = 2; /* recovery */ | |
3130 | ||
3131 | for (i = 0; i < nalloc; i++) { | |
3132 | bio = r10bio->devs[i].bio; | |
3133 | rp = bio->bi_private; | |
3134 | bio_reset(bio); | |
3135 | bio->bi_private = rp; | |
3136 | bio = r10bio->devs[i].repl_bio; | |
3137 | if (bio) { | |
3138 | rp = bio->bi_private; | |
3139 | bio_reset(bio); | |
3140 | bio->bi_private = rp; | |
3141 | } | |
3142 | } | |
3143 | return r10bio; | |
3144 | } | |
3145 | ||
8db87912 GJ |
3146 | /* |
3147 | * Set cluster_sync_high since we need other nodes to add the | |
3148 | * range [cluster_sync_low, cluster_sync_high] to suspend list. | |
3149 | */ | |
3150 | static void raid10_set_cluster_sync_high(struct r10conf *conf) | |
3151 | { | |
3152 | sector_t window_size; | |
3153 | int extra_chunk, chunks; | |
3154 | ||
3155 | /* | |
3156 | * First, here we define "stripe" as a unit which across | |
3157 | * all member devices one time, so we get chunks by use | |
3158 | * raid_disks / near_copies. Otherwise, if near_copies is | |
3159 | * close to raid_disks, then resync window could increases | |
3160 | * linearly with the increase of raid_disks, which means | |
3161 | * we will suspend a really large IO window while it is not | |
3162 | * necessary. If raid_disks is not divisible by near_copies, | |
3163 | * an extra chunk is needed to ensure the whole "stripe" is | |
3164 | * covered. | |
3165 | */ | |
3166 | ||
3167 | chunks = conf->geo.raid_disks / conf->geo.near_copies; | |
3168 | if (conf->geo.raid_disks % conf->geo.near_copies == 0) | |
3169 | extra_chunk = 0; | |
3170 | else | |
3171 | extra_chunk = 1; | |
3172 | window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors; | |
3173 | ||
3174 | /* | |
3175 | * At least use a 32M window to align with raid1's resync window | |
3176 | */ | |
3177 | window_size = (CLUSTER_RESYNC_WINDOW_SECTORS > window_size) ? | |
3178 | CLUSTER_RESYNC_WINDOW_SECTORS : window_size; | |
3179 | ||
3180 | conf->cluster_sync_high = conf->cluster_sync_low + window_size; | |
3181 | } | |
3182 | ||
1da177e4 LT |
3183 | /* |
3184 | * perform a "sync" on one "block" | |
3185 | * | |
3186 | * We need to make sure that no normal I/O request - particularly write | |
3187 | * requests - conflict with active sync requests. | |
3188 | * | |
3189 | * This is achieved by tracking pending requests and a 'barrier' concept | |
3190 | * that can be installed to exclude normal IO requests. | |
3191 | * | |
3192 | * Resync and recovery are handled very differently. | |
3193 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
3194 | * | |
3195 | * For resync, we iterate over virtual addresses, read all copies, | |
3196 | * and update if there are differences. If only one copy is live, | |
3197 | * skip it. | |
3198 | * For recovery, we iterate over physical addresses, read a good | |
3199 | * value for each non-in_sync drive, and over-write. | |
3200 | * | |
3201 | * So, for recovery we may have several outstanding complex requests for a | |
3202 | * given address, one for each out-of-sync device. We model this by allocating | |
3203 | * a number of r10_bio structures, one for each out-of-sync device. | |
3204 | * As we setup these structures, we collect all bio's together into a list | |
3205 | * which we then process collectively to add pages, and then process again | |
ed00aabd | 3206 | * to pass to submit_bio_noacct. |
1da177e4 LT |
3207 | * |
3208 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
3209 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
3210 | * has its remaining count decremented to 0, the whole complex operation | |
3211 | * is complete. | |
3212 | * | |
3213 | */ | |
3214 | ||
849674e4 | 3215 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 3216 | int *skipped) |
1da177e4 | 3217 | { |
e879a879 | 3218 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 3219 | struct r10bio *r10_bio; |
1da177e4 LT |
3220 | struct bio *biolist = NULL, *bio; |
3221 | sector_t max_sector, nr_sectors; | |
1da177e4 | 3222 | int i; |
6cce3b23 | 3223 | int max_sync; |
57dab0bd | 3224 | sector_t sync_blocks; |
1da177e4 LT |
3225 | sector_t sectors_skipped = 0; |
3226 | int chunks_skipped = 0; | |
5cf00fcd | 3227 | sector_t chunk_mask = conf->geo.chunk_mask; |
022e510f | 3228 | int page_idx = 0; |
1da177e4 | 3229 | |
afeee514 | 3230 | if (!mempool_initialized(&conf->r10buf_pool)) |
1da177e4 | 3231 | if (init_resync(conf)) |
57afd89f | 3232 | return 0; |
1da177e4 | 3233 | |
7e83ccbe MW |
3234 | /* |
3235 | * Allow skipping a full rebuild for incremental assembly | |
3236 | * of a clean array, like RAID1 does. | |
3237 | */ | |
3238 | if (mddev->bitmap == NULL && | |
3239 | mddev->recovery_cp == MaxSector && | |
13765120 N |
3240 | mddev->reshape_position == MaxSector && |
3241 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 3242 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 3243 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
3244 | conf->fullsync == 0) { |
3245 | *skipped = 1; | |
13765120 | 3246 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
3247 | } |
3248 | ||
1da177e4 | 3249 | skipped: |
58c0fed4 | 3250 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
3251 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
3252 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
3253 | max_sector = mddev->resync_max_sectors; |
3254 | if (sector_nr >= max_sector) { | |
8db87912 GJ |
3255 | conf->cluster_sync_low = 0; |
3256 | conf->cluster_sync_high = 0; | |
3257 | ||
6cce3b23 N |
3258 | /* If we aborted, we need to abort the |
3259 | * sync on the 'current' bitmap chucks (there can | |
3260 | * be several when recovering multiple devices). | |
3261 | * as we may have started syncing it but not finished. | |
3262 | * We can find the current address in | |
3263 | * mddev->curr_resync, but for recovery, | |
3264 | * we need to convert that to several | |
3265 | * virtual addresses. | |
3266 | */ | |
3ea7daa5 N |
3267 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3268 | end_reshape(conf); | |
b3968552 | 3269 | close_sync(conf); |
3ea7daa5 N |
3270 | return 0; |
3271 | } | |
3272 | ||
6cce3b23 N |
3273 | if (mddev->curr_resync < max_sector) { /* aborted */ |
3274 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
e64e4018 AS |
3275 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
3276 | &sync_blocks, 1); | |
5cf00fcd | 3277 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
3278 | sector_t sect = |
3279 | raid10_find_virt(conf, mddev->curr_resync, i); | |
e64e4018 AS |
3280 | md_bitmap_end_sync(mddev->bitmap, sect, |
3281 | &sync_blocks, 1); | |
6cce3b23 | 3282 | } |
9ad1aefc N |
3283 | } else { |
3284 | /* completed sync */ | |
3285 | if ((!mddev->bitmap || conf->fullsync) | |
3286 | && conf->have_replacement | |
3287 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3288 | /* Completed a full sync so the replacements | |
3289 | * are now fully recovered. | |
3290 | */ | |
f90145f3 N |
3291 | rcu_read_lock(); |
3292 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3293 | struct md_rdev *rdev = | |
3294 | rcu_dereference(conf->mirrors[i].replacement); | |
3295 | if (rdev) | |
3296 | rdev->recovery_offset = MaxSector; | |
3297 | } | |
3298 | rcu_read_unlock(); | |
9ad1aefc | 3299 | } |
6cce3b23 | 3300 | conf->fullsync = 0; |
9ad1aefc | 3301 | } |
e64e4018 | 3302 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 3303 | close_sync(conf); |
57afd89f | 3304 | *skipped = 1; |
1da177e4 LT |
3305 | return sectors_skipped; |
3306 | } | |
3ea7daa5 N |
3307 | |
3308 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
3309 | return reshape_request(mddev, sector_nr, skipped); | |
3310 | ||
5cf00fcd | 3311 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
3312 | /* if there has been nothing to do on any drive, |
3313 | * then there is nothing to do at all.. | |
3314 | */ | |
57afd89f N |
3315 | *skipped = 1; |
3316 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
3317 | } |
3318 | ||
c6207277 N |
3319 | if (max_sector > mddev->resync_max) |
3320 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
3321 | ||
1da177e4 LT |
3322 | /* make sure whole request will fit in a chunk - if chunks |
3323 | * are meaningful | |
3324 | */ | |
5cf00fcd N |
3325 | if (conf->geo.near_copies < conf->geo.raid_disks && |
3326 | max_sector > (sector_nr | chunk_mask)) | |
3327 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 | 3328 | |
7ac50447 TM |
3329 | /* |
3330 | * If there is non-resync activity waiting for a turn, then let it | |
3331 | * though before starting on this new sync request. | |
3332 | */ | |
3333 | if (conf->nr_waiting) | |
3334 | schedule_timeout_uninterruptible(1); | |
3335 | ||
1da177e4 LT |
3336 | /* Again, very different code for resync and recovery. |
3337 | * Both must result in an r10bio with a list of bios that | |
309dca30 | 3338 | * have bi_end_io, bi_sector, bi_bdev set, |
1da177e4 LT |
3339 | * and bi_private set to the r10bio. |
3340 | * For recovery, we may actually create several r10bios | |
3341 | * with 2 bios in each, that correspond to the bios in the main one. | |
3342 | * In this case, the subordinate r10bios link back through a | |
3343 | * borrowed master_bio pointer, and the counter in the master | |
3344 | * includes a ref from each subordinate. | |
3345 | */ | |
3346 | /* First, we decide what to do and set ->bi_end_io | |
3347 | * To end_sync_read if we want to read, and | |
3348 | * end_sync_write if we will want to write. | |
3349 | */ | |
3350 | ||
6cce3b23 | 3351 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
3352 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
3353 | /* recovery... the complicated one */ | |
e875ecea | 3354 | int j; |
1da177e4 LT |
3355 | r10_bio = NULL; |
3356 | ||
5cf00fcd | 3357 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 3358 | int still_degraded; |
9f2c9d12 | 3359 | struct r10bio *rb2; |
ab9d47e9 N |
3360 | sector_t sect; |
3361 | int must_sync; | |
e875ecea | 3362 | int any_working; |
ee37d731 AW |
3363 | int need_recover = 0; |
3364 | int need_replace = 0; | |
dc280d98 | 3365 | struct raid10_info *mirror = &conf->mirrors[i]; |
f90145f3 | 3366 | struct md_rdev *mrdev, *mreplace; |
24afd80d | 3367 | |
f90145f3 N |
3368 | rcu_read_lock(); |
3369 | mrdev = rcu_dereference(mirror->rdev); | |
3370 | mreplace = rcu_dereference(mirror->replacement); | |
3371 | ||
ee37d731 AW |
3372 | if (mrdev != NULL && |
3373 | !test_bit(Faulty, &mrdev->flags) && | |
3374 | !test_bit(In_sync, &mrdev->flags)) | |
3375 | need_recover = 1; | |
3376 | if (mreplace != NULL && | |
3377 | !test_bit(Faulty, &mreplace->flags)) | |
3378 | need_replace = 1; | |
3379 | ||
3380 | if (!need_recover && !need_replace) { | |
f90145f3 | 3381 | rcu_read_unlock(); |
ab9d47e9 | 3382 | continue; |
f90145f3 | 3383 | } |
1da177e4 | 3384 | |
ab9d47e9 N |
3385 | still_degraded = 0; |
3386 | /* want to reconstruct this device */ | |
3387 | rb2 = r10_bio; | |
3388 | sect = raid10_find_virt(conf, sector_nr, i); | |
fc448a18 N |
3389 | if (sect >= mddev->resync_max_sectors) { |
3390 | /* last stripe is not complete - don't | |
3391 | * try to recover this sector. | |
3392 | */ | |
f90145f3 | 3393 | rcu_read_unlock(); |
fc448a18 N |
3394 | continue; |
3395 | } | |
f5b67ae8 N |
3396 | if (mreplace && test_bit(Faulty, &mreplace->flags)) |
3397 | mreplace = NULL; | |
24afd80d N |
3398 | /* Unless we are doing a full sync, or a replacement |
3399 | * we only need to recover the block if it is set in | |
3400 | * the bitmap | |
ab9d47e9 | 3401 | */ |
e64e4018 AS |
3402 | must_sync = md_bitmap_start_sync(mddev->bitmap, sect, |
3403 | &sync_blocks, 1); | |
ab9d47e9 N |
3404 | if (sync_blocks < max_sync) |
3405 | max_sync = sync_blocks; | |
3406 | if (!must_sync && | |
f90145f3 | 3407 | mreplace == NULL && |
ab9d47e9 N |
3408 | !conf->fullsync) { |
3409 | /* yep, skip the sync_blocks here, but don't assume | |
3410 | * that there will never be anything to do here | |
3411 | */ | |
3412 | chunks_skipped = -1; | |
f90145f3 | 3413 | rcu_read_unlock(); |
ab9d47e9 N |
3414 | continue; |
3415 | } | |
f90145f3 N |
3416 | atomic_inc(&mrdev->nr_pending); |
3417 | if (mreplace) | |
3418 | atomic_inc(&mreplace->nr_pending); | |
3419 | rcu_read_unlock(); | |
6cce3b23 | 3420 | |
208410b5 | 3421 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3422 | r10_bio->state = 0; |
ab9d47e9 N |
3423 | raise_barrier(conf, rb2 != NULL); |
3424 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 3425 | |
ab9d47e9 N |
3426 | r10_bio->master_bio = (struct bio*)rb2; |
3427 | if (rb2) | |
3428 | atomic_inc(&rb2->remaining); | |
3429 | r10_bio->mddev = mddev; | |
3430 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
3431 | r10_bio->sector = sect; | |
1da177e4 | 3432 | |
ab9d47e9 N |
3433 | raid10_find_phys(conf, r10_bio); |
3434 | ||
3435 | /* Need to check if the array will still be | |
3436 | * degraded | |
3437 | */ | |
f90145f3 N |
3438 | rcu_read_lock(); |
3439 | for (j = 0; j < conf->geo.raid_disks; j++) { | |
3440 | struct md_rdev *rdev = rcu_dereference( | |
3441 | conf->mirrors[j].rdev); | |
3442 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
ab9d47e9 | 3443 | still_degraded = 1; |
87fc767b | 3444 | break; |
1da177e4 | 3445 | } |
f90145f3 | 3446 | } |
ab9d47e9 | 3447 | |
e64e4018 AS |
3448 | must_sync = md_bitmap_start_sync(mddev->bitmap, sect, |
3449 | &sync_blocks, still_degraded); | |
ab9d47e9 | 3450 | |
e875ecea | 3451 | any_working = 0; |
ab9d47e9 | 3452 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3453 | int k; |
ab9d47e9 | 3454 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3455 | sector_t from_addr, to_addr; |
f90145f3 N |
3456 | struct md_rdev *rdev = |
3457 | rcu_dereference(conf->mirrors[d].rdev); | |
40c356ce N |
3458 | sector_t sector, first_bad; |
3459 | int bad_sectors; | |
f90145f3 N |
3460 | if (!rdev || |
3461 | !test_bit(In_sync, &rdev->flags)) | |
ab9d47e9 N |
3462 | continue; |
3463 | /* This is where we read from */ | |
e875ecea | 3464 | any_working = 1; |
40c356ce N |
3465 | sector = r10_bio->devs[j].addr; |
3466 | ||
3467 | if (is_badblock(rdev, sector, max_sync, | |
3468 | &first_bad, &bad_sectors)) { | |
3469 | if (first_bad > sector) | |
3470 | max_sync = first_bad - sector; | |
3471 | else { | |
3472 | bad_sectors -= (sector | |
3473 | - first_bad); | |
3474 | if (max_sync > bad_sectors) | |
3475 | max_sync = bad_sectors; | |
3476 | continue; | |
3477 | } | |
3478 | } | |
ab9d47e9 N |
3479 | bio = r10_bio->devs[0].bio; |
3480 | bio->bi_next = biolist; | |
3481 | biolist = bio; | |
ab9d47e9 | 3482 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3483 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3484 | if (test_bit(FailFast, &rdev->flags)) |
3485 | bio->bi_opf |= MD_FAILFAST; | |
5e570289 | 3486 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3487 | bio->bi_iter.bi_sector = from_addr + |
3488 | rdev->data_offset; | |
74d46992 | 3489 | bio_set_dev(bio, rdev->bdev); |
24afd80d N |
3490 | atomic_inc(&rdev->nr_pending); |
3491 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3492 | |
3493 | for (k=0; k<conf->copies; k++) | |
3494 | if (r10_bio->devs[k].devnum == i) | |
3495 | break; | |
3496 | BUG_ON(k == conf->copies); | |
5e570289 | 3497 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3498 | r10_bio->devs[0].devnum = d; |
5e570289 | 3499 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3500 | r10_bio->devs[1].devnum = i; |
5e570289 | 3501 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3502 | |
ee37d731 | 3503 | if (need_recover) { |
24afd80d N |
3504 | bio = r10_bio->devs[1].bio; |
3505 | bio->bi_next = biolist; | |
3506 | biolist = bio; | |
24afd80d | 3507 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3508 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3509 | bio->bi_iter.bi_sector = to_addr |
f90145f3 | 3510 | + mrdev->data_offset; |
74d46992 | 3511 | bio_set_dev(bio, mrdev->bdev); |
24afd80d N |
3512 | atomic_inc(&r10_bio->remaining); |
3513 | } else | |
3514 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3515 | ||
3516 | /* and maybe write to replacement */ | |
3517 | bio = r10_bio->devs[1].repl_bio; | |
3518 | if (bio) | |
3519 | bio->bi_end_io = NULL; | |
ee37d731 | 3520 | /* Note: if need_replace, then bio |
24afd80d N |
3521 | * cannot be NULL as r10buf_pool_alloc will |
3522 | * have allocated it. | |
24afd80d | 3523 | */ |
ee37d731 | 3524 | if (!need_replace) |
24afd80d N |
3525 | break; |
3526 | bio->bi_next = biolist; | |
3527 | biolist = bio; | |
24afd80d | 3528 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3529 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3530 | bio->bi_iter.bi_sector = to_addr + |
f90145f3 | 3531 | mreplace->data_offset; |
74d46992 | 3532 | bio_set_dev(bio, mreplace->bdev); |
24afd80d | 3533 | atomic_inc(&r10_bio->remaining); |
ab9d47e9 N |
3534 | break; |
3535 | } | |
f90145f3 | 3536 | rcu_read_unlock(); |
ab9d47e9 | 3537 | if (j == conf->copies) { |
e875ecea N |
3538 | /* Cannot recover, so abort the recovery or |
3539 | * record a bad block */ | |
e875ecea N |
3540 | if (any_working) { |
3541 | /* problem is that there are bad blocks | |
3542 | * on other device(s) | |
3543 | */ | |
3544 | int k; | |
3545 | for (k = 0; k < conf->copies; k++) | |
3546 | if (r10_bio->devs[k].devnum == i) | |
3547 | break; | |
24afd80d | 3548 | if (!test_bit(In_sync, |
f90145f3 | 3549 | &mrdev->flags) |
24afd80d | 3550 | && !rdev_set_badblocks( |
f90145f3 | 3551 | mrdev, |
24afd80d N |
3552 | r10_bio->devs[k].addr, |
3553 | max_sync, 0)) | |
3554 | any_working = 0; | |
f90145f3 | 3555 | if (mreplace && |
24afd80d | 3556 | !rdev_set_badblocks( |
f90145f3 | 3557 | mreplace, |
e875ecea N |
3558 | r10_bio->devs[k].addr, |
3559 | max_sync, 0)) | |
3560 | any_working = 0; | |
3561 | } | |
3562 | if (!any_working) { | |
3563 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3564 | &mddev->recovery)) | |
08464e09 | 3565 | pr_warn("md/raid10:%s: insufficient working devices for recovery.\n", |
e875ecea | 3566 | mdname(mddev)); |
24afd80d | 3567 | mirror->recovery_disabled |
e875ecea N |
3568 | = mddev->recovery_disabled; |
3569 | } | |
e8b84915 N |
3570 | put_buf(r10_bio); |
3571 | if (rb2) | |
3572 | atomic_dec(&rb2->remaining); | |
3573 | r10_bio = rb2; | |
f90145f3 N |
3574 | rdev_dec_pending(mrdev, mddev); |
3575 | if (mreplace) | |
3576 | rdev_dec_pending(mreplace, mddev); | |
ab9d47e9 | 3577 | break; |
1da177e4 | 3578 | } |
f90145f3 N |
3579 | rdev_dec_pending(mrdev, mddev); |
3580 | if (mreplace) | |
3581 | rdev_dec_pending(mreplace, mddev); | |
8d3ca83d N |
3582 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { |
3583 | /* Only want this if there is elsewhere to | |
3584 | * read from. 'j' is currently the first | |
3585 | * readable copy. | |
3586 | */ | |
3587 | int targets = 1; | |
3588 | for (; j < conf->copies; j++) { | |
3589 | int d = r10_bio->devs[j].devnum; | |
3590 | if (conf->mirrors[d].rdev && | |
3591 | test_bit(In_sync, | |
3592 | &conf->mirrors[d].rdev->flags)) | |
3593 | targets++; | |
3594 | } | |
3595 | if (targets == 1) | |
3596 | r10_bio->devs[0].bio->bi_opf | |
3597 | &= ~MD_FAILFAST; | |
3598 | } | |
ab9d47e9 | 3599 | } |
1da177e4 LT |
3600 | if (biolist == NULL) { |
3601 | while (r10_bio) { | |
9f2c9d12 N |
3602 | struct r10bio *rb2 = r10_bio; |
3603 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3604 | rb2->master_bio = NULL; |
3605 | put_buf(rb2); | |
3606 | } | |
3607 | goto giveup; | |
3608 | } | |
3609 | } else { | |
3610 | /* resync. Schedule a read for every block at this virt offset */ | |
3611 | int count = 0; | |
6cce3b23 | 3612 | |
8db87912 GJ |
3613 | /* |
3614 | * Since curr_resync_completed could probably not update in | |
3615 | * time, and we will set cluster_sync_low based on it. | |
3616 | * Let's check against "sector_nr + 2 * RESYNC_SECTORS" for | |
3617 | * safety reason, which ensures curr_resync_completed is | |
3618 | * updated in bitmap_cond_end_sync. | |
3619 | */ | |
e64e4018 AS |
3620 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
3621 | mddev_is_clustered(mddev) && | |
3622 | (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
78200d45 | 3623 | |
e64e4018 AS |
3624 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
3625 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3626 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3627 | &mddev->recovery)) { | |
6cce3b23 N |
3628 | /* We can skip this block */ |
3629 | *skipped = 1; | |
3630 | return sync_blocks + sectors_skipped; | |
3631 | } | |
3632 | if (sync_blocks < max_sync) | |
3633 | max_sync = sync_blocks; | |
208410b5 | 3634 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3635 | r10_bio->state = 0; |
1da177e4 | 3636 | |
1da177e4 LT |
3637 | r10_bio->mddev = mddev; |
3638 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3639 | raise_barrier(conf, 0); |
3640 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3641 | |
3642 | r10_bio->master_bio = NULL; | |
3643 | r10_bio->sector = sector_nr; | |
3644 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3645 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3646 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3647 | |
5cf00fcd | 3648 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3649 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3650 | sector_t first_bad, sector; |
3651 | int bad_sectors; | |
f90145f3 | 3652 | struct md_rdev *rdev; |
40c356ce | 3653 | |
9ad1aefc N |
3654 | if (r10_bio->devs[i].repl_bio) |
3655 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3656 | ||
1da177e4 | 3657 | bio = r10_bio->devs[i].bio; |
4e4cbee9 | 3658 | bio->bi_status = BLK_STS_IOERR; |
f90145f3 N |
3659 | rcu_read_lock(); |
3660 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
3661 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3662 | rcu_read_unlock(); | |
1da177e4 | 3663 | continue; |
f90145f3 | 3664 | } |
40c356ce | 3665 | sector = r10_bio->devs[i].addr; |
f90145f3 | 3666 | if (is_badblock(rdev, sector, max_sync, |
40c356ce N |
3667 | &first_bad, &bad_sectors)) { |
3668 | if (first_bad > sector) | |
3669 | max_sync = first_bad - sector; | |
3670 | else { | |
3671 | bad_sectors -= (sector - first_bad); | |
3672 | if (max_sync > bad_sectors) | |
91502f09 | 3673 | max_sync = bad_sectors; |
f90145f3 | 3674 | rcu_read_unlock(); |
40c356ce N |
3675 | continue; |
3676 | } | |
3677 | } | |
f90145f3 | 3678 | atomic_inc(&rdev->nr_pending); |
1da177e4 LT |
3679 | atomic_inc(&r10_bio->remaining); |
3680 | bio->bi_next = biolist; | |
3681 | biolist = bio; | |
1da177e4 | 3682 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3683 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
1cdd1257 | 3684 | if (test_bit(FailFast, &rdev->flags)) |
8d3ca83d | 3685 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3686 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3687 | bio_set_dev(bio, rdev->bdev); |
1da177e4 | 3688 | count++; |
9ad1aefc | 3689 | |
f90145f3 N |
3690 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3691 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3692 | rcu_read_unlock(); | |
9ad1aefc | 3693 | continue; |
f90145f3 N |
3694 | } |
3695 | atomic_inc(&rdev->nr_pending); | |
9ad1aefc N |
3696 | |
3697 | /* Need to set up for writing to the replacement */ | |
3698 | bio = r10_bio->devs[i].repl_bio; | |
4e4cbee9 | 3699 | bio->bi_status = BLK_STS_IOERR; |
9ad1aefc N |
3700 | |
3701 | sector = r10_bio->devs[i].addr; | |
9ad1aefc N |
3702 | bio->bi_next = biolist; |
3703 | biolist = bio; | |
9ad1aefc | 3704 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3705 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1cdd1257 | 3706 | if (test_bit(FailFast, &rdev->flags)) |
1919cbb2 | 3707 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3708 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3709 | bio_set_dev(bio, rdev->bdev); |
9ad1aefc | 3710 | count++; |
1cdd1257 | 3711 | rcu_read_unlock(); |
1da177e4 LT |
3712 | } |
3713 | ||
3714 | if (count < 2) { | |
3715 | for (i=0; i<conf->copies; i++) { | |
3716 | int d = r10_bio->devs[i].devnum; | |
3717 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3718 | rdev_dec_pending(conf->mirrors[d].rdev, |
3719 | mddev); | |
9ad1aefc N |
3720 | if (r10_bio->devs[i].repl_bio && |
3721 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3722 | rdev_dec_pending( | |
3723 | conf->mirrors[d].replacement, | |
3724 | mddev); | |
1da177e4 LT |
3725 | } |
3726 | put_buf(r10_bio); | |
3727 | biolist = NULL; | |
3728 | goto giveup; | |
3729 | } | |
3730 | } | |
3731 | ||
1da177e4 | 3732 | nr_sectors = 0; |
6cce3b23 N |
3733 | if (sector_nr + max_sync < max_sector) |
3734 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3735 | do { |
3736 | struct page *page; | |
3737 | int len = PAGE_SIZE; | |
1da177e4 LT |
3738 | if (sector_nr + (len>>9) > max_sector) |
3739 | len = (max_sector - sector_nr) << 9; | |
3740 | if (len == 0) | |
3741 | break; | |
3742 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
f0250618 | 3743 | struct resync_pages *rp = get_resync_pages(bio); |
022e510f | 3744 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
3745 | /* |
3746 | * won't fail because the vec table is big enough | |
3747 | * to hold all these pages | |
3748 | */ | |
3749 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
3750 | } |
3751 | nr_sectors += len>>9; | |
3752 | sector_nr += len>>9; | |
022e510f | 3753 | } while (++page_idx < RESYNC_PAGES); |
1da177e4 LT |
3754 | r10_bio->sectors = nr_sectors; |
3755 | ||
8db87912 GJ |
3756 | if (mddev_is_clustered(mddev) && |
3757 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3758 | /* It is resync not recovery */ | |
3759 | if (conf->cluster_sync_high < sector_nr + nr_sectors) { | |
3760 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
3761 | raid10_set_cluster_sync_high(conf); | |
3762 | /* Send resync message */ | |
3763 | md_cluster_ops->resync_info_update(mddev, | |
3764 | conf->cluster_sync_low, | |
3765 | conf->cluster_sync_high); | |
3766 | } | |
3767 | } else if (mddev_is_clustered(mddev)) { | |
3768 | /* This is recovery not resync */ | |
3769 | sector_t sect_va1, sect_va2; | |
3770 | bool broadcast_msg = false; | |
3771 | ||
3772 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3773 | /* | |
3774 | * sector_nr is a device address for recovery, so we | |
3775 | * need translate it to array address before compare | |
3776 | * with cluster_sync_high. | |
3777 | */ | |
3778 | sect_va1 = raid10_find_virt(conf, sector_nr, i); | |
3779 | ||
3780 | if (conf->cluster_sync_high < sect_va1 + nr_sectors) { | |
3781 | broadcast_msg = true; | |
3782 | /* | |
3783 | * curr_resync_completed is similar as | |
3784 | * sector_nr, so make the translation too. | |
3785 | */ | |
3786 | sect_va2 = raid10_find_virt(conf, | |
3787 | mddev->curr_resync_completed, i); | |
3788 | ||
3789 | if (conf->cluster_sync_low == 0 || | |
3790 | conf->cluster_sync_low > sect_va2) | |
3791 | conf->cluster_sync_low = sect_va2; | |
3792 | } | |
3793 | } | |
3794 | if (broadcast_msg) { | |
3795 | raid10_set_cluster_sync_high(conf); | |
3796 | md_cluster_ops->resync_info_update(mddev, | |
3797 | conf->cluster_sync_low, | |
3798 | conf->cluster_sync_high); | |
3799 | } | |
3800 | } | |
3801 | ||
1da177e4 LT |
3802 | while (biolist) { |
3803 | bio = biolist; | |
3804 | biolist = biolist->bi_next; | |
3805 | ||
3806 | bio->bi_next = NULL; | |
f0250618 | 3807 | r10_bio = get_resync_r10bio(bio); |
1da177e4 LT |
3808 | r10_bio->sectors = nr_sectors; |
3809 | ||
3810 | if (bio->bi_end_io == end_sync_read) { | |
74d46992 | 3811 | md_sync_acct_bio(bio, nr_sectors); |
4e4cbee9 | 3812 | bio->bi_status = 0; |
ed00aabd | 3813 | submit_bio_noacct(bio); |
1da177e4 LT |
3814 | } |
3815 | } | |
3816 | ||
57afd89f N |
3817 | if (sectors_skipped) |
3818 | /* pretend they weren't skipped, it makes | |
3819 | * no important difference in this case | |
3820 | */ | |
3821 | md_done_sync(mddev, sectors_skipped, 1); | |
3822 | ||
1da177e4 LT |
3823 | return sectors_skipped + nr_sectors; |
3824 | giveup: | |
3825 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3826 | * drives must be failed or in resync, all drives |
3827 | * have a bad block, so try the next chunk... | |
1da177e4 | 3828 | */ |
09b4068a N |
3829 | if (sector_nr + max_sync < max_sector) |
3830 | max_sector = sector_nr + max_sync; | |
3831 | ||
3832 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3833 | chunks_skipped ++; |
3834 | sector_nr = max_sector; | |
1da177e4 | 3835 | goto skipped; |
1da177e4 LT |
3836 | } |
3837 | ||
80c3a6ce | 3838 | static sector_t |
fd01b88c | 3839 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3840 | { |
3841 | sector_t size; | |
e879a879 | 3842 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3843 | |
3844 | if (!raid_disks) | |
3ea7daa5 N |
3845 | raid_disks = min(conf->geo.raid_disks, |
3846 | conf->prev.raid_disks); | |
80c3a6ce | 3847 | if (!sectors) |
dab8b292 | 3848 | sectors = conf->dev_sectors; |
80c3a6ce | 3849 | |
5cf00fcd N |
3850 | size = sectors >> conf->geo.chunk_shift; |
3851 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3852 | size = size * raid_disks; |
5cf00fcd | 3853 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3854 | |
5cf00fcd | 3855 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3856 | } |
3857 | ||
6508fdbf N |
3858 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3859 | { | |
3860 | /* Calculate the number of sectors-per-device that will | |
3861 | * actually be used, and set conf->dev_sectors and | |
3862 | * conf->stride | |
3863 | */ | |
3864 | ||
5cf00fcd N |
3865 | size = size >> conf->geo.chunk_shift; |
3866 | sector_div(size, conf->geo.far_copies); | |
3867 | size = size * conf->geo.raid_disks; | |
3868 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3869 | /* 'size' is now the number of chunks in the array */ |
3870 | /* calculate "used chunks per device" */ | |
3871 | size = size * conf->copies; | |
3872 | ||
3873 | /* We need to round up when dividing by raid_disks to | |
3874 | * get the stride size. | |
3875 | */ | |
5cf00fcd | 3876 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3877 | |
5cf00fcd | 3878 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3879 | |
5cf00fcd N |
3880 | if (conf->geo.far_offset) |
3881 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3882 | else { |
5cf00fcd N |
3883 | sector_div(size, conf->geo.far_copies); |
3884 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3885 | } |
3886 | } | |
dab8b292 | 3887 | |
deb200d0 N |
3888 | enum geo_type {geo_new, geo_old, geo_start}; |
3889 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3890 | { | |
3891 | int nc, fc, fo; | |
3892 | int layout, chunk, disks; | |
3893 | switch (new) { | |
3894 | case geo_old: | |
3895 | layout = mddev->layout; | |
3896 | chunk = mddev->chunk_sectors; | |
3897 | disks = mddev->raid_disks - mddev->delta_disks; | |
3898 | break; | |
3899 | case geo_new: | |
3900 | layout = mddev->new_layout; | |
3901 | chunk = mddev->new_chunk_sectors; | |
3902 | disks = mddev->raid_disks; | |
3903 | break; | |
3904 | default: /* avoid 'may be unused' warnings */ | |
3905 | case geo_start: /* new when starting reshape - raid_disks not | |
3906 | * updated yet. */ | |
3907 | layout = mddev->new_layout; | |
3908 | chunk = mddev->new_chunk_sectors; | |
3909 | disks = mddev->raid_disks + mddev->delta_disks; | |
3910 | break; | |
3911 | } | |
8bce6d35 | 3912 | if (layout >> 19) |
deb200d0 N |
3913 | return -1; |
3914 | if (chunk < (PAGE_SIZE >> 9) || | |
3915 | !is_power_of_2(chunk)) | |
3916 | return -2; | |
3917 | nc = layout & 255; | |
3918 | fc = (layout >> 8) & 255; | |
3919 | fo = layout & (1<<16); | |
3920 | geo->raid_disks = disks; | |
3921 | geo->near_copies = nc; | |
3922 | geo->far_copies = fc; | |
3923 | geo->far_offset = fo; | |
8bce6d35 N |
3924 | switch (layout >> 17) { |
3925 | case 0: /* original layout. simple but not always optimal */ | |
3926 | geo->far_set_size = disks; | |
3927 | break; | |
3928 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3929 | * actually using this, but leave code here just in case.*/ | |
3930 | geo->far_set_size = disks/fc; | |
3931 | WARN(geo->far_set_size < fc, | |
3932 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3933 | break; | |
3934 | case 2: /* "improved" layout fixed to match documentation */ | |
3935 | geo->far_set_size = fc * nc; | |
3936 | break; | |
3937 | default: /* Not a valid layout */ | |
3938 | return -1; | |
3939 | } | |
deb200d0 N |
3940 | geo->chunk_mask = chunk - 1; |
3941 | geo->chunk_shift = ffz(~chunk); | |
3942 | return nc*fc; | |
3943 | } | |
3944 | ||
e879a879 | 3945 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3946 | { |
e879a879 | 3947 | struct r10conf *conf = NULL; |
dab8b292 | 3948 | int err = -EINVAL; |
deb200d0 N |
3949 | struct geom geo; |
3950 | int copies; | |
3951 | ||
3952 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3953 | |
deb200d0 | 3954 | if (copies == -2) { |
08464e09 N |
3955 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n", |
3956 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3957 | goto out; |
1da177e4 | 3958 | } |
2604b703 | 3959 | |
deb200d0 | 3960 | if (copies < 2 || copies > mddev->raid_disks) { |
08464e09 N |
3961 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
3962 | mdname(mddev), mddev->new_layout); | |
1da177e4 LT |
3963 | goto out; |
3964 | } | |
dab8b292 TM |
3965 | |
3966 | err = -ENOMEM; | |
e879a879 | 3967 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3968 | if (!conf) |
1da177e4 | 3969 | goto out; |
dab8b292 | 3970 | |
3ea7daa5 | 3971 | /* FIXME calc properly */ |
6396bb22 KC |
3972 | conf->mirrors = kcalloc(mddev->raid_disks + max(0, -mddev->delta_disks), |
3973 | sizeof(struct raid10_info), | |
dab8b292 TM |
3974 | GFP_KERNEL); |
3975 | if (!conf->mirrors) | |
3976 | goto out; | |
4443ae10 N |
3977 | |
3978 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3979 | if (!conf->tmppage) | |
dab8b292 TM |
3980 | goto out; |
3981 | ||
deb200d0 N |
3982 | conf->geo = geo; |
3983 | conf->copies = copies; | |
3f677f9c | 3984 | err = mempool_init(&conf->r10bio_pool, NR_RAID_BIOS, r10bio_pool_alloc, |
c7afa803 | 3985 | rbio_pool_free, conf); |
afeee514 | 3986 | if (err) |
dab8b292 TM |
3987 | goto out; |
3988 | ||
afeee514 KO |
3989 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
3990 | if (err) | |
fc9977dd N |
3991 | goto out; |
3992 | ||
6508fdbf | 3993 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3994 | if (mddev->reshape_position == MaxSector) { |
3995 | conf->prev = conf->geo; | |
3996 | conf->reshape_progress = MaxSector; | |
3997 | } else { | |
3998 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3999 | err = -EINVAL; | |
4000 | goto out; | |
4001 | } | |
4002 | conf->reshape_progress = mddev->reshape_position; | |
4003 | if (conf->prev.far_offset) | |
4004 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
4005 | else | |
4006 | /* far_copies must be 1 */ | |
4007 | conf->prev.stride = conf->dev_sectors; | |
4008 | } | |
299b0685 | 4009 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 4010 | spin_lock_init(&conf->device_lock); |
dab8b292 | 4011 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 4012 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 TM |
4013 | |
4014 | spin_lock_init(&conf->resync_lock); | |
4015 | init_waitqueue_head(&conf->wait_barrier); | |
0e5313e2 | 4016 | atomic_set(&conf->nr_pending, 0); |
dab8b292 | 4017 | |
afeee514 | 4018 | err = -ENOMEM; |
0232605d | 4019 | conf->thread = md_register_thread(raid10d, mddev, "raid10"); |
dab8b292 TM |
4020 | if (!conf->thread) |
4021 | goto out; | |
4022 | ||
dab8b292 TM |
4023 | conf->mddev = mddev; |
4024 | return conf; | |
4025 | ||
4026 | out: | |
dab8b292 | 4027 | if (conf) { |
afeee514 | 4028 | mempool_exit(&conf->r10bio_pool); |
dab8b292 TM |
4029 | kfree(conf->mirrors); |
4030 | safe_put_page(conf->tmppage); | |
afeee514 | 4031 | bioset_exit(&conf->bio_split); |
dab8b292 TM |
4032 | kfree(conf); |
4033 | } | |
4034 | return ERR_PTR(err); | |
4035 | } | |
4036 | ||
16ef5101 CH |
4037 | static void raid10_set_io_opt(struct r10conf *conf) |
4038 | { | |
4039 | int raid_disks = conf->geo.raid_disks; | |
4040 | ||
4041 | if (!(conf->geo.raid_disks % conf->geo.near_copies)) | |
4042 | raid_disks /= conf->geo.near_copies; | |
4043 | blk_queue_io_opt(conf->mddev->queue, (conf->mddev->chunk_sectors << 9) * | |
4044 | raid_disks); | |
4045 | } | |
4046 | ||
849674e4 | 4047 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 4048 | { |
e879a879 | 4049 | struct r10conf *conf; |
16ef5101 | 4050 | int i, disk_idx; |
dc280d98 | 4051 | struct raid10_info *disk; |
3cb03002 | 4052 | struct md_rdev *rdev; |
dab8b292 | 4053 | sector_t size; |
3ea7daa5 N |
4054 | sector_t min_offset_diff = 0; |
4055 | int first = 1; | |
532a2a3f | 4056 | bool discard_supported = false; |
dab8b292 | 4057 | |
a415c0f1 N |
4058 | if (mddev_init_writes_pending(mddev) < 0) |
4059 | return -ENOMEM; | |
4060 | ||
dab8b292 TM |
4061 | if (mddev->private == NULL) { |
4062 | conf = setup_conf(mddev); | |
4063 | if (IS_ERR(conf)) | |
4064 | return PTR_ERR(conf); | |
4065 | mddev->private = conf; | |
4066 | } | |
4067 | conf = mddev->private; | |
4068 | if (!conf) | |
4069 | goto out; | |
4070 | ||
8db87912 GJ |
4071 | if (mddev_is_clustered(conf->mddev)) { |
4072 | int fc, fo; | |
4073 | ||
4074 | fc = (mddev->layout >> 8) & 255; | |
4075 | fo = mddev->layout & (1<<16); | |
4076 | if (fc > 1 || fo > 0) { | |
4077 | pr_err("only near layout is supported by clustered" | |
4078 | " raid10\n"); | |
43a52123 | 4079 | goto out_free_conf; |
8db87912 GJ |
4080 | } |
4081 | } | |
4082 | ||
dab8b292 TM |
4083 | mddev->thread = conf->thread; |
4084 | conf->thread = NULL; | |
4085 | ||
cc4d1efd | 4086 | if (mddev->queue) { |
532a2a3f | 4087 | blk_queue_max_discard_sectors(mddev->queue, |
d30588b2 | 4088 | UINT_MAX); |
5026d7a9 | 4089 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 4090 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
16ef5101 CH |
4091 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); |
4092 | raid10_set_io_opt(conf); | |
cc4d1efd | 4093 | } |
8f6c2e4b | 4094 | |
dafb20fa | 4095 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 4096 | long long diff; |
34b343cf | 4097 | |
1da177e4 | 4098 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
4099 | if (disk_idx < 0) |
4100 | continue; | |
4101 | if (disk_idx >= conf->geo.raid_disks && | |
4102 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
4103 | continue; |
4104 | disk = conf->mirrors + disk_idx; | |
4105 | ||
56a2559b N |
4106 | if (test_bit(Replacement, &rdev->flags)) { |
4107 | if (disk->replacement) | |
4108 | goto out_free_conf; | |
4109 | disk->replacement = rdev; | |
4110 | } else { | |
4111 | if (disk->rdev) | |
4112 | goto out_free_conf; | |
4113 | disk->rdev = rdev; | |
4114 | } | |
3ea7daa5 N |
4115 | diff = (rdev->new_data_offset - rdev->data_offset); |
4116 | if (!mddev->reshape_backwards) | |
4117 | diff = -diff; | |
4118 | if (diff < 0) | |
4119 | diff = 0; | |
4120 | if (first || diff < min_offset_diff) | |
4121 | min_offset_diff = diff; | |
56a2559b | 4122 | |
cc4d1efd JB |
4123 | if (mddev->gendisk) |
4124 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
4125 | rdev->data_offset << 9); | |
1da177e4 LT |
4126 | |
4127 | disk->head_position = 0; | |
532a2a3f SL |
4128 | |
4129 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) | |
4130 | discard_supported = true; | |
6f287ca6 | 4131 | first = 0; |
1da177e4 | 4132 | } |
3ea7daa5 | 4133 | |
ed30be07 JB |
4134 | if (mddev->queue) { |
4135 | if (discard_supported) | |
8b904b5b | 4136 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
ed30be07 JB |
4137 | mddev->queue); |
4138 | else | |
8b904b5b | 4139 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
ed30be07 JB |
4140 | mddev->queue); |
4141 | } | |
6d508242 | 4142 | /* need to check that every block has at least one working mirror */ |
700c7213 | 4143 | if (!enough(conf, -1)) { |
08464e09 | 4144 | pr_err("md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 4145 | mdname(mddev)); |
1da177e4 LT |
4146 | goto out_free_conf; |
4147 | } | |
4148 | ||
3ea7daa5 N |
4149 | if (conf->reshape_progress != MaxSector) { |
4150 | /* must ensure that shape change is supported */ | |
4151 | if (conf->geo.far_copies != 1 && | |
4152 | conf->geo.far_offset == 0) | |
4153 | goto out_free_conf; | |
4154 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 4155 | conf->prev.far_offset == 0) |
3ea7daa5 N |
4156 | goto out_free_conf; |
4157 | } | |
4158 | ||
1da177e4 | 4159 | mddev->degraded = 0; |
f8c9e74f N |
4160 | for (i = 0; |
4161 | i < conf->geo.raid_disks | |
4162 | || i < conf->prev.raid_disks; | |
4163 | i++) { | |
1da177e4 LT |
4164 | |
4165 | disk = conf->mirrors + i; | |
4166 | ||
56a2559b N |
4167 | if (!disk->rdev && disk->replacement) { |
4168 | /* The replacement is all we have - use it */ | |
4169 | disk->rdev = disk->replacement; | |
4170 | disk->replacement = NULL; | |
4171 | clear_bit(Replacement, &disk->rdev->flags); | |
4172 | } | |
4173 | ||
5fd6c1dc | 4174 | if (!disk->rdev || |
2e333e89 | 4175 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
4176 | disk->head_position = 0; |
4177 | mddev->degraded++; | |
0b59bb64 N |
4178 | if (disk->rdev && |
4179 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 4180 | conf->fullsync = 1; |
1da177e4 | 4181 | } |
bda31539 BC |
4182 | |
4183 | if (disk->replacement && | |
4184 | !test_bit(In_sync, &disk->replacement->flags) && | |
4185 | disk->replacement->saved_raid_disk < 0) { | |
4186 | conf->fullsync = 1; | |
4187 | } | |
4188 | ||
d890fa2b | 4189 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
4190 | } |
4191 | ||
8c6ac868 | 4192 | if (mddev->recovery_cp != MaxSector) |
08464e09 N |
4193 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", |
4194 | mdname(mddev)); | |
4195 | pr_info("md/raid10:%s: active with %d out of %d devices\n", | |
5cf00fcd N |
4196 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
4197 | conf->geo.raid_disks); | |
1da177e4 LT |
4198 | /* |
4199 | * Ok, everything is just fine now | |
4200 | */ | |
dab8b292 TM |
4201 | mddev->dev_sectors = conf->dev_sectors; |
4202 | size = raid10_size(mddev, 0, 0); | |
4203 | md_set_array_sectors(mddev, size); | |
4204 | mddev->resync_max_sectors = size; | |
46533ff7 | 4205 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
1da177e4 | 4206 | |
a91a2785 MP |
4207 | if (md_integrity_register(mddev)) |
4208 | goto out_free_conf; | |
4209 | ||
3ea7daa5 N |
4210 | if (conf->reshape_progress != MaxSector) { |
4211 | unsigned long before_length, after_length; | |
4212 | ||
4213 | before_length = ((1 << conf->prev.chunk_shift) * | |
4214 | conf->prev.far_copies); | |
4215 | after_length = ((1 << conf->geo.chunk_shift) * | |
4216 | conf->geo.far_copies); | |
4217 | ||
4218 | if (max(before_length, after_length) > min_offset_diff) { | |
4219 | /* This cannot work */ | |
08464e09 | 4220 | pr_warn("md/raid10: offset difference not enough to continue reshape\n"); |
3ea7daa5 N |
4221 | goto out_free_conf; |
4222 | } | |
4223 | conf->offset_diff = min_offset_diff; | |
4224 | ||
3ea7daa5 N |
4225 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
4226 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4227 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
4228 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4229 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4230 | "reshape"); | |
e406f12d AP |
4231 | if (!mddev->sync_thread) |
4232 | goto out_free_conf; | |
3ea7daa5 N |
4233 | } |
4234 | ||
1da177e4 LT |
4235 | return 0; |
4236 | ||
4237 | out_free_conf: | |
01f96c0a | 4238 | md_unregister_thread(&mddev->thread); |
afeee514 | 4239 | mempool_exit(&conf->r10bio_pool); |
1345b1d8 | 4240 | safe_put_page(conf->tmppage); |
990a8baf | 4241 | kfree(conf->mirrors); |
1da177e4 LT |
4242 | kfree(conf); |
4243 | mddev->private = NULL; | |
4244 | out: | |
4245 | return -EIO; | |
4246 | } | |
4247 | ||
afa0f557 | 4248 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 4249 | { |
afa0f557 | 4250 | struct r10conf *conf = priv; |
1da177e4 | 4251 | |
afeee514 | 4252 | mempool_exit(&conf->r10bio_pool); |
0fea7ed8 | 4253 | safe_put_page(conf->tmppage); |
990a8baf | 4254 | kfree(conf->mirrors); |
c4796e21 N |
4255 | kfree(conf->mirrors_old); |
4256 | kfree(conf->mirrors_new); | |
afeee514 | 4257 | bioset_exit(&conf->bio_split); |
1da177e4 | 4258 | kfree(conf); |
1da177e4 LT |
4259 | } |
4260 | ||
b03e0ccb | 4261 | static void raid10_quiesce(struct mddev *mddev, int quiesce) |
6cce3b23 | 4262 | { |
e879a879 | 4263 | struct r10conf *conf = mddev->private; |
6cce3b23 | 4264 | |
b03e0ccb | 4265 | if (quiesce) |
6cce3b23 | 4266 | raise_barrier(conf, 0); |
b03e0ccb | 4267 | else |
6cce3b23 | 4268 | lower_barrier(conf); |
6cce3b23 | 4269 | } |
1da177e4 | 4270 | |
006a09a0 N |
4271 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
4272 | { | |
4273 | /* Resize of 'far' arrays is not supported. | |
4274 | * For 'near' and 'offset' arrays we can set the | |
4275 | * number of sectors used to be an appropriate multiple | |
4276 | * of the chunk size. | |
4277 | * For 'offset', this is far_copies*chunksize. | |
4278 | * For 'near' the multiplier is the LCM of | |
4279 | * near_copies and raid_disks. | |
4280 | * So if far_copies > 1 && !far_offset, fail. | |
4281 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
4282 | * multiply by chunk_size. Then round to this number. | |
4283 | * This is mostly done by raid10_size() | |
4284 | */ | |
4285 | struct r10conf *conf = mddev->private; | |
4286 | sector_t oldsize, size; | |
4287 | ||
f8c9e74f N |
4288 | if (mddev->reshape_position != MaxSector) |
4289 | return -EBUSY; | |
4290 | ||
5cf00fcd | 4291 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
4292 | return -EINVAL; |
4293 | ||
4294 | oldsize = raid10_size(mddev, 0, 0); | |
4295 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
4296 | if (mddev->external_size && |
4297 | mddev->array_sectors > size) | |
006a09a0 | 4298 | return -EINVAL; |
a4a6125a | 4299 | if (mddev->bitmap) { |
e64e4018 | 4300 | int ret = md_bitmap_resize(mddev->bitmap, size, 0, 0); |
a4a6125a N |
4301 | if (ret) |
4302 | return ret; | |
4303 | } | |
4304 | md_set_array_sectors(mddev, size); | |
006a09a0 N |
4305 | if (sectors > mddev->dev_sectors && |
4306 | mddev->recovery_cp > oldsize) { | |
4307 | mddev->recovery_cp = oldsize; | |
4308 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4309 | } | |
6508fdbf N |
4310 | calc_sectors(conf, sectors); |
4311 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
4312 | mddev->resync_max_sectors = size; |
4313 | return 0; | |
4314 | } | |
4315 | ||
53a6ab4d | 4316 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 4317 | { |
3cb03002 | 4318 | struct md_rdev *rdev; |
e879a879 | 4319 | struct r10conf *conf; |
dab8b292 TM |
4320 | |
4321 | if (mddev->degraded > 0) { | |
08464e09 N |
4322 | pr_warn("md/raid10:%s: Error: degraded raid0!\n", |
4323 | mdname(mddev)); | |
dab8b292 TM |
4324 | return ERR_PTR(-EINVAL); |
4325 | } | |
53a6ab4d | 4326 | sector_div(size, devs); |
dab8b292 | 4327 | |
dab8b292 TM |
4328 | /* Set new parameters */ |
4329 | mddev->new_level = 10; | |
4330 | /* new layout: far_copies = 1, near_copies = 2 */ | |
4331 | mddev->new_layout = (1<<8) + 2; | |
4332 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
4333 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
4334 | mddev->raid_disks *= 2; |
4335 | /* make sure it will be not marked as dirty */ | |
4336 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 4337 | mddev->dev_sectors = size; |
dab8b292 TM |
4338 | |
4339 | conf = setup_conf(mddev); | |
02214dc5 | 4340 | if (!IS_ERR(conf)) { |
dafb20fa | 4341 | rdev_for_each(rdev, mddev) |
53a6ab4d | 4342 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 4343 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
4344 | rdev->sectors = size; |
4345 | } | |
02214dc5 KW |
4346 | conf->barrier = 1; |
4347 | } | |
4348 | ||
dab8b292 TM |
4349 | return conf; |
4350 | } | |
4351 | ||
fd01b88c | 4352 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 4353 | { |
e373ab10 | 4354 | struct r0conf *raid0_conf; |
dab8b292 TM |
4355 | |
4356 | /* raid10 can take over: | |
4357 | * raid0 - providing it has only two drives | |
4358 | */ | |
4359 | if (mddev->level == 0) { | |
4360 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
4361 | raid0_conf = mddev->private; |
4362 | if (raid0_conf->nr_strip_zones > 1) { | |
08464e09 N |
4363 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n", |
4364 | mdname(mddev)); | |
dab8b292 TM |
4365 | return ERR_PTR(-EINVAL); |
4366 | } | |
53a6ab4d N |
4367 | return raid10_takeover_raid0(mddev, |
4368 | raid0_conf->strip_zone->zone_end, | |
4369 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
4370 | } |
4371 | return ERR_PTR(-EINVAL); | |
4372 | } | |
4373 | ||
3ea7daa5 N |
4374 | static int raid10_check_reshape(struct mddev *mddev) |
4375 | { | |
4376 | /* Called when there is a request to change | |
4377 | * - layout (to ->new_layout) | |
4378 | * - chunk size (to ->new_chunk_sectors) | |
4379 | * - raid_disks (by delta_disks) | |
4380 | * or when trying to restart a reshape that was ongoing. | |
4381 | * | |
4382 | * We need to validate the request and possibly allocate | |
4383 | * space if that might be an issue later. | |
4384 | * | |
4385 | * Currently we reject any reshape of a 'far' mode array, | |
4386 | * allow chunk size to change if new is generally acceptable, | |
4387 | * allow raid_disks to increase, and allow | |
4388 | * a switch between 'near' mode and 'offset' mode. | |
4389 | */ | |
4390 | struct r10conf *conf = mddev->private; | |
4391 | struct geom geo; | |
4392 | ||
4393 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
4394 | return -EINVAL; | |
4395 | ||
4396 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
4397 | /* mustn't change number of copies */ | |
4398 | return -EINVAL; | |
4399 | if (geo.far_copies > 1 && !geo.far_offset) | |
4400 | /* Cannot switch to 'far' mode */ | |
4401 | return -EINVAL; | |
4402 | ||
4403 | if (mddev->array_sectors & geo.chunk_mask) | |
4404 | /* not factor of array size */ | |
4405 | return -EINVAL; | |
4406 | ||
3ea7daa5 N |
4407 | if (!enough(conf, -1)) |
4408 | return -EINVAL; | |
4409 | ||
4410 | kfree(conf->mirrors_new); | |
4411 | conf->mirrors_new = NULL; | |
4412 | if (mddev->delta_disks > 0) { | |
4413 | /* allocate new 'mirrors' list */ | |
6396bb22 KC |
4414 | conf->mirrors_new = |
4415 | kcalloc(mddev->raid_disks + mddev->delta_disks, | |
4416 | sizeof(struct raid10_info), | |
4417 | GFP_KERNEL); | |
3ea7daa5 N |
4418 | if (!conf->mirrors_new) |
4419 | return -ENOMEM; | |
4420 | } | |
4421 | return 0; | |
4422 | } | |
4423 | ||
4424 | /* | |
4425 | * Need to check if array has failed when deciding whether to: | |
4426 | * - start an array | |
4427 | * - remove non-faulty devices | |
4428 | * - add a spare | |
4429 | * - allow a reshape | |
4430 | * This determination is simple when no reshape is happening. | |
4431 | * However if there is a reshape, we need to carefully check | |
4432 | * both the before and after sections. | |
4433 | * This is because some failed devices may only affect one | |
4434 | * of the two sections, and some non-in_sync devices may | |
4435 | * be insync in the section most affected by failed devices. | |
4436 | */ | |
4437 | static int calc_degraded(struct r10conf *conf) | |
4438 | { | |
4439 | int degraded, degraded2; | |
4440 | int i; | |
4441 | ||
4442 | rcu_read_lock(); | |
4443 | degraded = 0; | |
4444 | /* 'prev' section first */ | |
4445 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
4446 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4447 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4448 | degraded++; | |
4449 | else if (!test_bit(In_sync, &rdev->flags)) | |
4450 | /* When we can reduce the number of devices in | |
4451 | * an array, this might not contribute to | |
4452 | * 'degraded'. It does now. | |
4453 | */ | |
4454 | degraded++; | |
4455 | } | |
4456 | rcu_read_unlock(); | |
4457 | if (conf->geo.raid_disks == conf->prev.raid_disks) | |
4458 | return degraded; | |
4459 | rcu_read_lock(); | |
4460 | degraded2 = 0; | |
4461 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
4462 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4463 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4464 | degraded2++; | |
4465 | else if (!test_bit(In_sync, &rdev->flags)) { | |
4466 | /* If reshape is increasing the number of devices, | |
4467 | * this section has already been recovered, so | |
4468 | * it doesn't contribute to degraded. | |
4469 | * else it does. | |
4470 | */ | |
4471 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
4472 | degraded2++; | |
4473 | } | |
4474 | } | |
4475 | rcu_read_unlock(); | |
4476 | if (degraded2 > degraded) | |
4477 | return degraded2; | |
4478 | return degraded; | |
4479 | } | |
4480 | ||
4481 | static int raid10_start_reshape(struct mddev *mddev) | |
4482 | { | |
4483 | /* A 'reshape' has been requested. This commits | |
4484 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
4485 | * This also checks if there are enough spares and adds them | |
4486 | * to the array. | |
4487 | * We currently require enough spares to make the final | |
4488 | * array non-degraded. We also require that the difference | |
4489 | * between old and new data_offset - on each device - is | |
4490 | * enough that we never risk over-writing. | |
4491 | */ | |
4492 | ||
4493 | unsigned long before_length, after_length; | |
4494 | sector_t min_offset_diff = 0; | |
4495 | int first = 1; | |
4496 | struct geom new; | |
4497 | struct r10conf *conf = mddev->private; | |
4498 | struct md_rdev *rdev; | |
4499 | int spares = 0; | |
bb63a701 | 4500 | int ret; |
3ea7daa5 N |
4501 | |
4502 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4503 | return -EBUSY; | |
4504 | ||
4505 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
4506 | return -EINVAL; | |
4507 | ||
4508 | before_length = ((1 << conf->prev.chunk_shift) * | |
4509 | conf->prev.far_copies); | |
4510 | after_length = ((1 << conf->geo.chunk_shift) * | |
4511 | conf->geo.far_copies); | |
4512 | ||
4513 | rdev_for_each(rdev, mddev) { | |
4514 | if (!test_bit(In_sync, &rdev->flags) | |
4515 | && !test_bit(Faulty, &rdev->flags)) | |
4516 | spares++; | |
4517 | if (rdev->raid_disk >= 0) { | |
4518 | long long diff = (rdev->new_data_offset | |
4519 | - rdev->data_offset); | |
4520 | if (!mddev->reshape_backwards) | |
4521 | diff = -diff; | |
4522 | if (diff < 0) | |
4523 | diff = 0; | |
4524 | if (first || diff < min_offset_diff) | |
4525 | min_offset_diff = diff; | |
b506335e | 4526 | first = 0; |
3ea7daa5 N |
4527 | } |
4528 | } | |
4529 | ||
4530 | if (max(before_length, after_length) > min_offset_diff) | |
4531 | return -EINVAL; | |
4532 | ||
4533 | if (spares < mddev->delta_disks) | |
4534 | return -EINVAL; | |
4535 | ||
4536 | conf->offset_diff = min_offset_diff; | |
4537 | spin_lock_irq(&conf->device_lock); | |
4538 | if (conf->mirrors_new) { | |
4539 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4540 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4541 | smp_mb(); |
c4796e21 | 4542 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4543 | conf->mirrors_old = conf->mirrors; |
4544 | conf->mirrors = conf->mirrors_new; | |
4545 | conf->mirrors_new = NULL; | |
4546 | } | |
4547 | setup_geo(&conf->geo, mddev, geo_start); | |
4548 | smp_mb(); | |
4549 | if (mddev->reshape_backwards) { | |
4550 | sector_t size = raid10_size(mddev, 0, 0); | |
4551 | if (size < mddev->array_sectors) { | |
4552 | spin_unlock_irq(&conf->device_lock); | |
08464e09 N |
4553 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n", |
4554 | mdname(mddev)); | |
3ea7daa5 N |
4555 | return -EINVAL; |
4556 | } | |
4557 | mddev->resync_max_sectors = size; | |
4558 | conf->reshape_progress = size; | |
4559 | } else | |
4560 | conf->reshape_progress = 0; | |
299b0685 | 4561 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4562 | spin_unlock_irq(&conf->device_lock); |
4563 | ||
bb63a701 | 4564 | if (mddev->delta_disks && mddev->bitmap) { |
afd75628 GJ |
4565 | struct mdp_superblock_1 *sb = NULL; |
4566 | sector_t oldsize, newsize; | |
4567 | ||
4568 | oldsize = raid10_size(mddev, 0, 0); | |
4569 | newsize = raid10_size(mddev, 0, conf->geo.raid_disks); | |
4570 | ||
4571 | if (!mddev_is_clustered(mddev)) { | |
4572 | ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
4573 | if (ret) | |
4574 | goto abort; | |
4575 | else | |
4576 | goto out; | |
4577 | } | |
4578 | ||
4579 | rdev_for_each(rdev, mddev) { | |
4580 | if (rdev->raid_disk > -1 && | |
4581 | !test_bit(Faulty, &rdev->flags)) | |
4582 | sb = page_address(rdev->sb_page); | |
4583 | } | |
4584 | ||
4585 | /* | |
4586 | * some node is already performing reshape, and no need to | |
4587 | * call md_bitmap_resize again since it should be called when | |
4588 | * receiving BITMAP_RESIZE msg | |
4589 | */ | |
4590 | if ((sb && (le32_to_cpu(sb->feature_map) & | |
4591 | MD_FEATURE_RESHAPE_ACTIVE)) || (oldsize == newsize)) | |
4592 | goto out; | |
4593 | ||
4594 | ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
bb63a701 N |
4595 | if (ret) |
4596 | goto abort; | |
afd75628 GJ |
4597 | |
4598 | ret = md_cluster_ops->resize_bitmaps(mddev, newsize, oldsize); | |
4599 | if (ret) { | |
4600 | md_bitmap_resize(mddev->bitmap, oldsize, 0, 0); | |
4601 | goto abort; | |
4602 | } | |
bb63a701 | 4603 | } |
afd75628 | 4604 | out: |
3ea7daa5 N |
4605 | if (mddev->delta_disks > 0) { |
4606 | rdev_for_each(rdev, mddev) | |
4607 | if (rdev->raid_disk < 0 && | |
4608 | !test_bit(Faulty, &rdev->flags)) { | |
4609 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4610 | if (rdev->raid_disk >= | |
4611 | conf->prev.raid_disks) | |
4612 | set_bit(In_sync, &rdev->flags); | |
4613 | else | |
4614 | rdev->recovery_offset = 0; | |
4615 | ||
38ffc01f DLM |
4616 | /* Failure here is OK */ |
4617 | sysfs_link_rdev(mddev, rdev); | |
3ea7daa5 N |
4618 | } |
4619 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4620 | && !test_bit(Faulty, &rdev->flags)) { | |
4621 | /* This is a spare that was manually added */ | |
4622 | set_bit(In_sync, &rdev->flags); | |
4623 | } | |
4624 | } | |
4625 | /* When a reshape changes the number of devices, | |
4626 | * ->degraded is measured against the larger of the | |
4627 | * pre and post numbers. | |
4628 | */ | |
4629 | spin_lock_irq(&conf->device_lock); | |
4630 | mddev->degraded = calc_degraded(conf); | |
4631 | spin_unlock_irq(&conf->device_lock); | |
4632 | mddev->raid_disks = conf->geo.raid_disks; | |
4633 | mddev->reshape_position = conf->reshape_progress; | |
2953079c | 4634 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 N |
4635 | |
4636 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4637 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4638 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 N |
4639 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
4640 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4641 | ||
4642 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4643 | "reshape"); | |
4644 | if (!mddev->sync_thread) { | |
bb63a701 N |
4645 | ret = -EAGAIN; |
4646 | goto abort; | |
3ea7daa5 N |
4647 | } |
4648 | conf->reshape_checkpoint = jiffies; | |
4649 | md_wakeup_thread(mddev->sync_thread); | |
4650 | md_new_event(mddev); | |
4651 | return 0; | |
bb63a701 N |
4652 | |
4653 | abort: | |
4654 | mddev->recovery = 0; | |
4655 | spin_lock_irq(&conf->device_lock); | |
4656 | conf->geo = conf->prev; | |
4657 | mddev->raid_disks = conf->geo.raid_disks; | |
4658 | rdev_for_each(rdev, mddev) | |
4659 | rdev->new_data_offset = rdev->data_offset; | |
4660 | smp_wmb(); | |
4661 | conf->reshape_progress = MaxSector; | |
299b0685 | 4662 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4663 | mddev->reshape_position = MaxSector; |
4664 | spin_unlock_irq(&conf->device_lock); | |
4665 | return ret; | |
3ea7daa5 N |
4666 | } |
4667 | ||
4668 | /* Calculate the last device-address that could contain | |
4669 | * any block from the chunk that includes the array-address 's' | |
4670 | * and report the next address. | |
4671 | * i.e. the address returned will be chunk-aligned and after | |
4672 | * any data that is in the chunk containing 's'. | |
4673 | */ | |
4674 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4675 | { | |
4676 | s = (s | geo->chunk_mask) + 1; | |
4677 | s >>= geo->chunk_shift; | |
4678 | s *= geo->near_copies; | |
4679 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4680 | s *= geo->far_copies; | |
4681 | s <<= geo->chunk_shift; | |
4682 | return s; | |
4683 | } | |
4684 | ||
4685 | /* Calculate the first device-address that could contain | |
4686 | * any block from the chunk that includes the array-address 's'. | |
4687 | * This too will be the start of a chunk | |
4688 | */ | |
4689 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4690 | { | |
4691 | s >>= geo->chunk_shift; | |
4692 | s *= geo->near_copies; | |
4693 | sector_div(s, geo->raid_disks); | |
4694 | s *= geo->far_copies; | |
4695 | s <<= geo->chunk_shift; | |
4696 | return s; | |
4697 | } | |
4698 | ||
4699 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4700 | int *skipped) | |
4701 | { | |
4702 | /* We simply copy at most one chunk (smallest of old and new) | |
4703 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4704 | * or we hit a bad block or something. | |
4705 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4706 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4707 | * can record any location. |
4708 | * | |
4709 | * If we will want to write to a location that isn't | |
4710 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4711 | * we need to flush all reshape requests and update the metadata. | |
4712 | * | |
4713 | * When reshaping forwards (e.g. to more devices), we interpret | |
4714 | * 'safe' as the earliest block which might not have been copied | |
4715 | * down yet. We divide this by previous stripe size and multiply | |
4716 | * by previous stripe length to get lowest device offset that we | |
4717 | * cannot write to yet. | |
4718 | * We interpret 'sector_nr' as an address that we want to write to. | |
4719 | * From this we use last_device_address() to find where we might | |
4720 | * write to, and first_device_address on the 'safe' position. | |
4721 | * If this 'next' write position is after the 'safe' position, | |
4722 | * we must update the metadata to increase the 'safe' position. | |
4723 | * | |
4724 | * When reshaping backwards, we round in the opposite direction | |
4725 | * and perform the reverse test: next write position must not be | |
4726 | * less than current safe position. | |
4727 | * | |
4728 | * In all this the minimum difference in data offsets | |
4729 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4730 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4731 | * |
4732 | * We need to prepare all the bios here before we start any IO | |
4733 | * to ensure the size we choose is acceptable to all devices. | |
4734 | * The means one for each copy for write-out and an extra one for | |
4735 | * read-in. | |
4736 | * We store the read-in bio in ->master_bio and the others in | |
4737 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4738 | */ | |
4739 | struct r10conf *conf = mddev->private; | |
4740 | struct r10bio *r10_bio; | |
4741 | sector_t next, safe, last; | |
4742 | int max_sectors; | |
4743 | int nr_sectors; | |
4744 | int s; | |
4745 | struct md_rdev *rdev; | |
4746 | int need_flush = 0; | |
4747 | struct bio *blist; | |
4748 | struct bio *bio, *read_bio; | |
4749 | int sectors_done = 0; | |
f0250618 | 4750 | struct page **pages; |
3ea7daa5 N |
4751 | |
4752 | if (sector_nr == 0) { | |
4753 | /* If restarting in the middle, skip the initial sectors */ | |
4754 | if (mddev->reshape_backwards && | |
4755 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4756 | sector_nr = (raid10_size(mddev, 0, 0) | |
4757 | - conf->reshape_progress); | |
4758 | } else if (!mddev->reshape_backwards && | |
4759 | conf->reshape_progress > 0) | |
4760 | sector_nr = conf->reshape_progress; | |
4761 | if (sector_nr) { | |
4762 | mddev->curr_resync_completed = sector_nr; | |
e1a86dbb | 4763 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
3ea7daa5 N |
4764 | *skipped = 1; |
4765 | return sector_nr; | |
4766 | } | |
4767 | } | |
4768 | ||
4769 | /* We don't use sector_nr to track where we are up to | |
4770 | * as that doesn't work well for ->reshape_backwards. | |
4771 | * So just use ->reshape_progress. | |
4772 | */ | |
4773 | if (mddev->reshape_backwards) { | |
4774 | /* 'next' is the earliest device address that we might | |
4775 | * write to for this chunk in the new layout | |
4776 | */ | |
4777 | next = first_dev_address(conf->reshape_progress - 1, | |
4778 | &conf->geo); | |
4779 | ||
4780 | /* 'safe' is the last device address that we might read from | |
4781 | * in the old layout after a restart | |
4782 | */ | |
4783 | safe = last_dev_address(conf->reshape_safe - 1, | |
4784 | &conf->prev); | |
4785 | ||
4786 | if (next + conf->offset_diff < safe) | |
4787 | need_flush = 1; | |
4788 | ||
4789 | last = conf->reshape_progress - 1; | |
4790 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4791 | & conf->prev.chunk_mask); | |
e287308b ZL |
4792 | if (sector_nr + RESYNC_SECTORS < last) |
4793 | sector_nr = last + 1 - RESYNC_SECTORS; | |
3ea7daa5 N |
4794 | } else { |
4795 | /* 'next' is after the last device address that we | |
4796 | * might write to for this chunk in the new layout | |
4797 | */ | |
4798 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4799 | ||
4800 | /* 'safe' is the earliest device address that we might | |
4801 | * read from in the old layout after a restart | |
4802 | */ | |
4803 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4804 | ||
4805 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4806 | * as that would possibly corrupt data | |
4807 | */ | |
4808 | if (next > safe + conf->offset_diff) | |
4809 | need_flush = 1; | |
4810 | ||
4811 | sector_nr = conf->reshape_progress; | |
4812 | last = sector_nr | (conf->geo.chunk_mask | |
4813 | & conf->prev.chunk_mask); | |
4814 | ||
e287308b ZL |
4815 | if (sector_nr + RESYNC_SECTORS <= last) |
4816 | last = sector_nr + RESYNC_SECTORS - 1; | |
3ea7daa5 N |
4817 | } |
4818 | ||
4819 | if (need_flush || | |
4820 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4821 | /* Need to update reshape_position in metadata */ | |
4822 | wait_barrier(conf); | |
4823 | mddev->reshape_position = conf->reshape_progress; | |
4824 | if (mddev->reshape_backwards) | |
4825 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4826 | - conf->reshape_progress; | |
4827 | else | |
4828 | mddev->curr_resync_completed = conf->reshape_progress; | |
4829 | conf->reshape_checkpoint = jiffies; | |
2953079c | 4830 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 | 4831 | md_wakeup_thread(mddev->thread); |
2953079c | 4832 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
4833 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4834 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4835 | allow_barrier(conf); | |
4836 | return sectors_done; | |
4837 | } | |
3ea7daa5 N |
4838 | conf->reshape_safe = mddev->reshape_position; |
4839 | allow_barrier(conf); | |
4840 | } | |
4841 | ||
1d0ffd26 | 4842 | raise_barrier(conf, 0); |
3ea7daa5 N |
4843 | read_more: |
4844 | /* Now schedule reads for blocks from sector_nr to last */ | |
208410b5 | 4845 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 4846 | r10_bio->state = 0; |
1d0ffd26 | 4847 | raise_barrier(conf, 1); |
3ea7daa5 N |
4848 | atomic_set(&r10_bio->remaining, 0); |
4849 | r10_bio->mddev = mddev; | |
4850 | r10_bio->sector = sector_nr; | |
4851 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4852 | r10_bio->sectors = last - sector_nr + 1; | |
4853 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4854 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4855 | ||
4856 | if (!rdev) { | |
4857 | /* Cannot read from here, so need to record bad blocks | |
4858 | * on all the target devices. | |
4859 | */ | |
4860 | // FIXME | |
afeee514 | 4861 | mempool_free(r10_bio, &conf->r10buf_pool); |
3ea7daa5 N |
4862 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4863 | return sectors_done; | |
4864 | } | |
4865 | ||
a78f18da | 4866 | read_bio = bio_alloc_bioset(GFP_KERNEL, RESYNC_PAGES, &mddev->bio_set); |
3ea7daa5 | 4867 | |
74d46992 | 4868 | bio_set_dev(read_bio, rdev->bdev); |
4f024f37 | 4869 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4870 | + rdev->data_offset); |
4871 | read_bio->bi_private = r10_bio; | |
81fa1520 | 4872 | read_bio->bi_end_io = end_reshape_read; |
796a5cf0 | 4873 | bio_set_op_attrs(read_bio, REQ_OP_READ, 0); |
3ea7daa5 N |
4874 | r10_bio->master_bio = read_bio; |
4875 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4876 | ||
7564beda GJ |
4877 | /* |
4878 | * Broadcast RESYNC message to other nodes, so all nodes would not | |
4879 | * write to the region to avoid conflict. | |
4880 | */ | |
4881 | if (mddev_is_clustered(mddev) && conf->cluster_sync_high <= sector_nr) { | |
4882 | struct mdp_superblock_1 *sb = NULL; | |
4883 | int sb_reshape_pos = 0; | |
4884 | ||
4885 | conf->cluster_sync_low = sector_nr; | |
4886 | conf->cluster_sync_high = sector_nr + CLUSTER_RESYNC_WINDOW_SECTORS; | |
4887 | sb = page_address(rdev->sb_page); | |
4888 | if (sb) { | |
4889 | sb_reshape_pos = le64_to_cpu(sb->reshape_position); | |
4890 | /* | |
4891 | * Set cluster_sync_low again if next address for array | |
4892 | * reshape is less than cluster_sync_low. Since we can't | |
4893 | * update cluster_sync_low until it has finished reshape. | |
4894 | */ | |
4895 | if (sb_reshape_pos < conf->cluster_sync_low) | |
4896 | conf->cluster_sync_low = sb_reshape_pos; | |
4897 | } | |
4898 | ||
4899 | md_cluster_ops->resync_info_update(mddev, conf->cluster_sync_low, | |
4900 | conf->cluster_sync_high); | |
4901 | } | |
4902 | ||
3ea7daa5 N |
4903 | /* Now find the locations in the new layout */ |
4904 | __raid10_find_phys(&conf->geo, r10_bio); | |
4905 | ||
4906 | blist = read_bio; | |
4907 | read_bio->bi_next = NULL; | |
4908 | ||
d094d686 | 4909 | rcu_read_lock(); |
3ea7daa5 N |
4910 | for (s = 0; s < conf->copies*2; s++) { |
4911 | struct bio *b; | |
4912 | int d = r10_bio->devs[s/2].devnum; | |
4913 | struct md_rdev *rdev2; | |
4914 | if (s&1) { | |
d094d686 | 4915 | rdev2 = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4916 | b = r10_bio->devs[s/2].repl_bio; |
4917 | } else { | |
d094d686 | 4918 | rdev2 = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4919 | b = r10_bio->devs[s/2].bio; |
4920 | } | |
4921 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4922 | continue; | |
8be185f2 | 4923 | |
74d46992 | 4924 | bio_set_dev(b, rdev2->bdev); |
4f024f37 KO |
4925 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4926 | rdev2->new_data_offset; | |
3ea7daa5 | 4927 | b->bi_end_io = end_reshape_write; |
796a5cf0 | 4928 | bio_set_op_attrs(b, REQ_OP_WRITE, 0); |
3ea7daa5 | 4929 | b->bi_next = blist; |
3ea7daa5 N |
4930 | blist = b; |
4931 | } | |
4932 | ||
4933 | /* Now add as many pages as possible to all of these bios. */ | |
4934 | ||
4935 | nr_sectors = 0; | |
f0250618 | 4936 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; |
3ea7daa5 | 4937 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { |
f0250618 | 4938 | struct page *page = pages[s / (PAGE_SIZE >> 9)]; |
3ea7daa5 N |
4939 | int len = (max_sectors - s) << 9; |
4940 | if (len > PAGE_SIZE) | |
4941 | len = PAGE_SIZE; | |
4942 | for (bio = blist; bio ; bio = bio->bi_next) { | |
c85ba149 ML |
4943 | /* |
4944 | * won't fail because the vec table is big enough | |
4945 | * to hold all these pages | |
4946 | */ | |
4947 | bio_add_page(bio, page, len, 0); | |
3ea7daa5 N |
4948 | } |
4949 | sector_nr += len >> 9; | |
4950 | nr_sectors += len >> 9; | |
4951 | } | |
d094d686 | 4952 | rcu_read_unlock(); |
3ea7daa5 N |
4953 | r10_bio->sectors = nr_sectors; |
4954 | ||
4955 | /* Now submit the read */ | |
74d46992 | 4956 | md_sync_acct_bio(read_bio, r10_bio->sectors); |
3ea7daa5 N |
4957 | atomic_inc(&r10_bio->remaining); |
4958 | read_bio->bi_next = NULL; | |
ed00aabd | 4959 | submit_bio_noacct(read_bio); |
3ea7daa5 N |
4960 | sectors_done += nr_sectors; |
4961 | if (sector_nr <= last) | |
4962 | goto read_more; | |
4963 | ||
1d0ffd26 XN |
4964 | lower_barrier(conf); |
4965 | ||
3ea7daa5 N |
4966 | /* Now that we have done the whole section we can |
4967 | * update reshape_progress | |
4968 | */ | |
4969 | if (mddev->reshape_backwards) | |
4970 | conf->reshape_progress -= sectors_done; | |
4971 | else | |
4972 | conf->reshape_progress += sectors_done; | |
4973 | ||
4974 | return sectors_done; | |
4975 | } | |
4976 | ||
4977 | static void end_reshape_request(struct r10bio *r10_bio); | |
4978 | static int handle_reshape_read_error(struct mddev *mddev, | |
4979 | struct r10bio *r10_bio); | |
4980 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4981 | { | |
4982 | /* Reshape read completed. Hopefully we have a block | |
4983 | * to write out. | |
4984 | * If we got a read error then we do sync 1-page reads from | |
4985 | * elsewhere until we find the data - or give up. | |
4986 | */ | |
4987 | struct r10conf *conf = mddev->private; | |
4988 | int s; | |
4989 | ||
4990 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4991 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4992 | /* Reshape has been aborted */ | |
4993 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4994 | return; | |
4995 | } | |
4996 | ||
4997 | /* We definitely have the data in the pages, schedule the | |
4998 | * writes. | |
4999 | */ | |
5000 | atomic_set(&r10_bio->remaining, 1); | |
5001 | for (s = 0; s < conf->copies*2; s++) { | |
5002 | struct bio *b; | |
5003 | int d = r10_bio->devs[s/2].devnum; | |
5004 | struct md_rdev *rdev; | |
d094d686 | 5005 | rcu_read_lock(); |
3ea7daa5 | 5006 | if (s&1) { |
d094d686 | 5007 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
5008 | b = r10_bio->devs[s/2].repl_bio; |
5009 | } else { | |
d094d686 | 5010 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
5011 | b = r10_bio->devs[s/2].bio; |
5012 | } | |
d094d686 N |
5013 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
5014 | rcu_read_unlock(); | |
3ea7daa5 | 5015 | continue; |
d094d686 | 5016 | } |
3ea7daa5 | 5017 | atomic_inc(&rdev->nr_pending); |
d094d686 | 5018 | rcu_read_unlock(); |
74d46992 | 5019 | md_sync_acct_bio(b, r10_bio->sectors); |
3ea7daa5 N |
5020 | atomic_inc(&r10_bio->remaining); |
5021 | b->bi_next = NULL; | |
ed00aabd | 5022 | submit_bio_noacct(b); |
3ea7daa5 N |
5023 | } |
5024 | end_reshape_request(r10_bio); | |
5025 | } | |
5026 | ||
5027 | static void end_reshape(struct r10conf *conf) | |
5028 | { | |
5029 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
5030 | return; | |
5031 | ||
5032 | spin_lock_irq(&conf->device_lock); | |
5033 | conf->prev = conf->geo; | |
5034 | md_finish_reshape(conf->mddev); | |
5035 | smp_wmb(); | |
5036 | conf->reshape_progress = MaxSector; | |
299b0685 | 5037 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
5038 | spin_unlock_irq(&conf->device_lock); |
5039 | ||
c2e4cd57 | 5040 | if (conf->mddev->queue) |
16ef5101 | 5041 | raid10_set_io_opt(conf); |
3ea7daa5 N |
5042 | conf->fullsync = 0; |
5043 | } | |
5044 | ||
7564beda GJ |
5045 | static void raid10_update_reshape_pos(struct mddev *mddev) |
5046 | { | |
5047 | struct r10conf *conf = mddev->private; | |
5ebaf80b | 5048 | sector_t lo, hi; |
7564beda | 5049 | |
5ebaf80b GJ |
5050 | md_cluster_ops->resync_info_get(mddev, &lo, &hi); |
5051 | if (((mddev->reshape_position <= hi) && (mddev->reshape_position >= lo)) | |
5052 | || mddev->reshape_position == MaxSector) | |
5053 | conf->reshape_progress = mddev->reshape_position; | |
5054 | else | |
5055 | WARN_ON_ONCE(1); | |
7564beda GJ |
5056 | } |
5057 | ||
3ea7daa5 N |
5058 | static int handle_reshape_read_error(struct mddev *mddev, |
5059 | struct r10bio *r10_bio) | |
5060 | { | |
5061 | /* Use sync reads to get the blocks from somewhere else */ | |
5062 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 5063 | struct r10conf *conf = mddev->private; |
584ed9fa | 5064 | struct r10bio *r10b; |
3ea7daa5 N |
5065 | int slot = 0; |
5066 | int idx = 0; | |
2d06e3b7 ML |
5067 | struct page **pages; |
5068 | ||
8cf05a78 | 5069 | r10b = kmalloc(struct_size(r10b, devs, conf->copies), GFP_NOIO); |
584ed9fa MK |
5070 | if (!r10b) { |
5071 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
5072 | return -ENOMEM; | |
5073 | } | |
5074 | ||
2d06e3b7 ML |
5075 | /* reshape IOs share pages from .devs[0].bio */ |
5076 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; | |
3ea7daa5 | 5077 | |
e0ee7785 N |
5078 | r10b->sector = r10_bio->sector; |
5079 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
5080 | |
5081 | while (sectors) { | |
5082 | int s = sectors; | |
5083 | int success = 0; | |
5084 | int first_slot = slot; | |
5085 | ||
5086 | if (s > (PAGE_SIZE >> 9)) | |
5087 | s = PAGE_SIZE >> 9; | |
5088 | ||
d094d686 | 5089 | rcu_read_lock(); |
3ea7daa5 | 5090 | while (!success) { |
e0ee7785 | 5091 | int d = r10b->devs[slot].devnum; |
d094d686 | 5092 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
5093 | sector_t addr; |
5094 | if (rdev == NULL || | |
5095 | test_bit(Faulty, &rdev->flags) || | |
5096 | !test_bit(In_sync, &rdev->flags)) | |
5097 | goto failed; | |
5098 | ||
e0ee7785 | 5099 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
d094d686 N |
5100 | atomic_inc(&rdev->nr_pending); |
5101 | rcu_read_unlock(); | |
3ea7daa5 N |
5102 | success = sync_page_io(rdev, |
5103 | addr, | |
5104 | s << 9, | |
2d06e3b7 | 5105 | pages[idx], |
796a5cf0 | 5106 | REQ_OP_READ, 0, false); |
d094d686 N |
5107 | rdev_dec_pending(rdev, mddev); |
5108 | rcu_read_lock(); | |
3ea7daa5 N |
5109 | if (success) |
5110 | break; | |
5111 | failed: | |
5112 | slot++; | |
5113 | if (slot >= conf->copies) | |
5114 | slot = 0; | |
5115 | if (slot == first_slot) | |
5116 | break; | |
5117 | } | |
d094d686 | 5118 | rcu_read_unlock(); |
3ea7daa5 N |
5119 | if (!success) { |
5120 | /* couldn't read this block, must give up */ | |
5121 | set_bit(MD_RECOVERY_INTR, | |
5122 | &mddev->recovery); | |
584ed9fa | 5123 | kfree(r10b); |
3ea7daa5 N |
5124 | return -EIO; |
5125 | } | |
5126 | sectors -= s; | |
5127 | idx++; | |
5128 | } | |
584ed9fa | 5129 | kfree(r10b); |
3ea7daa5 N |
5130 | return 0; |
5131 | } | |
5132 | ||
4246a0b6 | 5133 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 5134 | { |
f0250618 | 5135 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
3ea7daa5 N |
5136 | struct mddev *mddev = r10_bio->mddev; |
5137 | struct r10conf *conf = mddev->private; | |
5138 | int d; | |
5139 | int slot; | |
5140 | int repl; | |
5141 | struct md_rdev *rdev = NULL; | |
5142 | ||
5143 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
5144 | if (repl) | |
5145 | rdev = conf->mirrors[d].replacement; | |
5146 | if (!rdev) { | |
5147 | smp_mb(); | |
5148 | rdev = conf->mirrors[d].rdev; | |
5149 | } | |
5150 | ||
4e4cbee9 | 5151 | if (bio->bi_status) { |
3ea7daa5 N |
5152 | /* FIXME should record badblock */ |
5153 | md_error(mddev, rdev); | |
5154 | } | |
5155 | ||
5156 | rdev_dec_pending(rdev, mddev); | |
5157 | end_reshape_request(r10_bio); | |
5158 | } | |
5159 | ||
5160 | static void end_reshape_request(struct r10bio *r10_bio) | |
5161 | { | |
5162 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
5163 | return; | |
5164 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
5165 | bio_put(r10_bio->master_bio); | |
5166 | put_buf(r10_bio); | |
5167 | } | |
5168 | ||
5169 | static void raid10_finish_reshape(struct mddev *mddev) | |
5170 | { | |
5171 | struct r10conf *conf = mddev->private; | |
5172 | ||
5173 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5174 | return; | |
5175 | ||
5176 | if (mddev->delta_disks > 0) { | |
3ea7daa5 N |
5177 | if (mddev->recovery_cp > mddev->resync_max_sectors) { |
5178 | mddev->recovery_cp = mddev->resync_max_sectors; | |
5179 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5180 | } | |
8876391e | 5181 | mddev->resync_max_sectors = mddev->array_sectors; |
63aced61 N |
5182 | } else { |
5183 | int d; | |
d094d686 | 5184 | rcu_read_lock(); |
63aced61 N |
5185 | for (d = conf->geo.raid_disks ; |
5186 | d < conf->geo.raid_disks - mddev->delta_disks; | |
5187 | d++) { | |
d094d686 | 5188 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
63aced61 N |
5189 | if (rdev) |
5190 | clear_bit(In_sync, &rdev->flags); | |
d094d686 | 5191 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
63aced61 N |
5192 | if (rdev) |
5193 | clear_bit(In_sync, &rdev->flags); | |
5194 | } | |
d094d686 | 5195 | rcu_read_unlock(); |
3ea7daa5 N |
5196 | } |
5197 | mddev->layout = mddev->new_layout; | |
5198 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
5199 | mddev->reshape_position = MaxSector; | |
5200 | mddev->delta_disks = 0; | |
5201 | mddev->reshape_backwards = 0; | |
5202 | } | |
5203 | ||
84fc4b56 | 5204 | static struct md_personality raid10_personality = |
1da177e4 LT |
5205 | { |
5206 | .name = "raid10", | |
2604b703 | 5207 | .level = 10, |
1da177e4 | 5208 | .owner = THIS_MODULE, |
849674e4 SL |
5209 | .make_request = raid10_make_request, |
5210 | .run = raid10_run, | |
afa0f557 | 5211 | .free = raid10_free, |
849674e4 SL |
5212 | .status = raid10_status, |
5213 | .error_handler = raid10_error, | |
1da177e4 LT |
5214 | .hot_add_disk = raid10_add_disk, |
5215 | .hot_remove_disk= raid10_remove_disk, | |
5216 | .spare_active = raid10_spare_active, | |
849674e4 | 5217 | .sync_request = raid10_sync_request, |
6cce3b23 | 5218 | .quiesce = raid10_quiesce, |
80c3a6ce | 5219 | .size = raid10_size, |
006a09a0 | 5220 | .resize = raid10_resize, |
dab8b292 | 5221 | .takeover = raid10_takeover, |
3ea7daa5 N |
5222 | .check_reshape = raid10_check_reshape, |
5223 | .start_reshape = raid10_start_reshape, | |
5224 | .finish_reshape = raid10_finish_reshape, | |
7564beda | 5225 | .update_reshape_pos = raid10_update_reshape_pos, |
1da177e4 LT |
5226 | }; |
5227 | ||
5228 | static int __init raid_init(void) | |
5229 | { | |
2604b703 | 5230 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
5231 | } |
5232 | ||
5233 | static void raid_exit(void) | |
5234 | { | |
2604b703 | 5235 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
5236 | } |
5237 | ||
5238 | module_init(raid_init); | |
5239 | module_exit(raid_exit); | |
5240 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5241 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 5242 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 5243 | MODULE_ALIAS("md-raid10"); |
2604b703 | 5244 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
5245 | |
5246 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |