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; |
17c28c2a | 94 | int size = offsetof(struct r10bio, devs[conf->copies]); |
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 | ||
17c28c2a | 241 | for (i = 0; i < conf->copies; 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 N |
299 | |
300 | bio_endio(bio); | |
301 | /* | |
302 | * Wake up any possible resync thread that waits for the device | |
303 | * to go idle. | |
304 | */ | |
305 | allow_barrier(conf); | |
306 | ||
1da177e4 LT |
307 | free_r10bio(r10_bio); |
308 | } | |
309 | ||
310 | /* | |
311 | * Update disk head position estimator based on IRQ completion info. | |
312 | */ | |
9f2c9d12 | 313 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 314 | { |
e879a879 | 315 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
316 | |
317 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
318 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
319 | } | |
320 | ||
778ca018 NK |
321 | /* |
322 | * Find the disk number which triggered given bio | |
323 | */ | |
e879a879 | 324 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 325 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
326 | { |
327 | int slot; | |
69335ef3 | 328 | int repl = 0; |
778ca018 | 329 | |
17c28c2a | 330 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
331 | if (r10_bio->devs[slot].bio == bio) |
332 | break; | |
69335ef3 N |
333 | if (r10_bio->devs[slot].repl_bio == bio) { |
334 | repl = 1; | |
335 | break; | |
336 | } | |
337 | } | |
778ca018 | 338 | |
17c28c2a | 339 | BUG_ON(slot == conf->copies); |
778ca018 NK |
340 | update_head_pos(slot, r10_bio); |
341 | ||
749c55e9 N |
342 | if (slotp) |
343 | *slotp = slot; | |
69335ef3 N |
344 | if (replp) |
345 | *replp = repl; | |
778ca018 NK |
346 | return r10_bio->devs[slot].devnum; |
347 | } | |
348 | ||
4246a0b6 | 349 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 350 | { |
4e4cbee9 | 351 | int uptodate = !bio->bi_status; |
9f2c9d12 | 352 | struct r10bio *r10_bio = bio->bi_private; |
a0e764c5 | 353 | int slot; |
abbf098e | 354 | struct md_rdev *rdev; |
e879a879 | 355 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 356 | |
1da177e4 | 357 | slot = r10_bio->read_slot; |
abbf098e | 358 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
359 | /* |
360 | * this branch is our 'one mirror IO has finished' event handler: | |
361 | */ | |
4443ae10 N |
362 | update_head_pos(slot, r10_bio); |
363 | ||
364 | if (uptodate) { | |
1da177e4 LT |
365 | /* |
366 | * Set R10BIO_Uptodate in our master bio, so that | |
367 | * we will return a good error code to the higher | |
368 | * levels even if IO on some other mirrored buffer fails. | |
369 | * | |
370 | * The 'master' represents the composite IO operation to | |
371 | * user-side. So if something waits for IO, then it will | |
372 | * wait for the 'master' bio. | |
373 | */ | |
374 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
375 | } else { |
376 | /* If all other devices that store this block have | |
377 | * failed, we want to return the error upwards rather | |
378 | * than fail the last device. Here we redefine | |
379 | * "uptodate" to mean "Don't want to retry" | |
380 | */ | |
635f6416 N |
381 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
382 | rdev->raid_disk)) | |
fae8cc5e | 383 | uptodate = 1; |
fae8cc5e N |
384 | } |
385 | if (uptodate) { | |
1da177e4 | 386 | raid_end_bio_io(r10_bio); |
abbf098e | 387 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 388 | } else { |
1da177e4 | 389 | /* |
7c4e06ff | 390 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
391 | */ |
392 | char b[BDEVNAME_SIZE]; | |
08464e09 | 393 | pr_err_ratelimited("md/raid10:%s: %s: rescheduling sector %llu\n", |
8bda470e | 394 | mdname(conf->mddev), |
abbf098e | 395 | bdevname(rdev->bdev, b), |
8bda470e | 396 | (unsigned long long)r10_bio->sector); |
856e08e2 | 397 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
398 | reschedule_retry(r10_bio); |
399 | } | |
1da177e4 LT |
400 | } |
401 | ||
9f2c9d12 | 402 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
403 | { |
404 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
405 | md_bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, |
406 | r10_bio->sectors, | |
407 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
408 | 0); | |
bd870a16 N |
409 | md_write_end(r10_bio->mddev); |
410 | } | |
411 | ||
9f2c9d12 | 412 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
413 | { |
414 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
415 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
416 | reschedule_retry(r10_bio); | |
417 | else { | |
418 | close_write(r10_bio); | |
419 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
420 | reschedule_retry(r10_bio); | |
421 | else | |
422 | raid_end_bio_io(r10_bio); | |
423 | } | |
424 | } | |
425 | } | |
426 | ||
4246a0b6 | 427 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 428 | { |
9f2c9d12 | 429 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 430 | int dev; |
749c55e9 | 431 | int dec_rdev = 1; |
e879a879 | 432 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 433 | int slot, repl; |
4ca40c2c | 434 | struct md_rdev *rdev = NULL; |
1919cbb2 | 435 | struct bio *to_put = NULL; |
579ed34f SL |
436 | bool discard_error; |
437 | ||
4e4cbee9 | 438 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 439 | |
475b0321 | 440 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 441 | |
475b0321 N |
442 | if (repl) |
443 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
444 | if (!rdev) { |
445 | smp_rmb(); | |
446 | repl = 0; | |
475b0321 | 447 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 448 | } |
1da177e4 LT |
449 | /* |
450 | * this branch is our 'one mirror IO has finished' event handler: | |
451 | */ | |
4e4cbee9 | 452 | if (bio->bi_status && !discard_error) { |
475b0321 N |
453 | if (repl) |
454 | /* Never record new bad blocks to replacement, | |
455 | * just fail it. | |
456 | */ | |
457 | md_error(rdev->mddev, rdev); | |
458 | else { | |
459 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
460 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
461 | set_bit(MD_RECOVERY_NEEDED, | |
462 | &rdev->mddev->recovery); | |
1919cbb2 | 463 | |
475b0321 | 464 | dec_rdev = 0; |
1919cbb2 N |
465 | if (test_bit(FailFast, &rdev->flags) && |
466 | (bio->bi_opf & MD_FAILFAST)) { | |
467 | md_error(rdev->mddev, rdev); | |
7cee6d4e YY |
468 | } |
469 | ||
470 | /* | |
471 | * When the device is faulty, it is not necessary to | |
472 | * handle write error. | |
473 | * For failfast, this is the only remaining device, | |
474 | * We need to retry the write without FailFast. | |
475 | */ | |
476 | if (!test_bit(Faulty, &rdev->flags)) | |
1919cbb2 | 477 | set_bit(R10BIO_WriteError, &r10_bio->state); |
7cee6d4e YY |
478 | else { |
479 | r10_bio->devs[slot].bio = NULL; | |
480 | to_put = bio; | |
481 | dec_rdev = 1; | |
482 | } | |
475b0321 | 483 | } |
749c55e9 | 484 | } else { |
1da177e4 LT |
485 | /* |
486 | * Set R10BIO_Uptodate in our master bio, so that | |
487 | * we will return a good error code for to the higher | |
488 | * levels even if IO on some other mirrored buffer fails. | |
489 | * | |
490 | * The 'master' represents the composite IO operation to | |
491 | * user-side. So if something waits for IO, then it will | |
492 | * wait for the 'master' bio. | |
493 | */ | |
749c55e9 N |
494 | sector_t first_bad; |
495 | int bad_sectors; | |
496 | ||
3056e3ae AL |
497 | /* |
498 | * Do not set R10BIO_Uptodate if the current device is | |
499 | * rebuilding or Faulty. This is because we cannot use | |
500 | * such device for properly reading the data back (we could | |
501 | * potentially use it, if the current write would have felt | |
502 | * before rdev->recovery_offset, but for simplicity we don't | |
503 | * check this here. | |
504 | */ | |
505 | if (test_bit(In_sync, &rdev->flags) && | |
506 | !test_bit(Faulty, &rdev->flags)) | |
507 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 508 | |
749c55e9 | 509 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 510 | if (is_badblock(rdev, |
749c55e9 N |
511 | r10_bio->devs[slot].addr, |
512 | r10_bio->sectors, | |
579ed34f | 513 | &first_bad, &bad_sectors) && !discard_error) { |
749c55e9 | 514 | bio_put(bio); |
475b0321 N |
515 | if (repl) |
516 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
517 | else | |
518 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
519 | dec_rdev = 0; |
520 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
521 | } | |
522 | } | |
523 | ||
1da177e4 LT |
524 | /* |
525 | * | |
526 | * Let's see if all mirrored write operations have finished | |
527 | * already. | |
528 | */ | |
19d5f834 | 529 | one_write_done(r10_bio); |
749c55e9 | 530 | if (dec_rdev) |
884162df | 531 | rdev_dec_pending(rdev, conf->mddev); |
1919cbb2 N |
532 | if (to_put) |
533 | bio_put(to_put); | |
1da177e4 LT |
534 | } |
535 | ||
1da177e4 LT |
536 | /* |
537 | * RAID10 layout manager | |
25985edc | 538 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
539 | * parameters: near_copies and far_copies. |
540 | * near_copies * far_copies must be <= raid_disks. | |
541 | * Normally one of these will be 1. | |
542 | * If both are 1, we get raid0. | |
543 | * If near_copies == raid_disks, we get raid1. | |
544 | * | |
25985edc | 545 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
546 | * first chunk, followed by near_copies copies of the next chunk and |
547 | * so on. | |
548 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
549 | * as described above, we start again with a device offset of near_copies. | |
550 | * So we effectively have another copy of the whole array further down all | |
551 | * the drives, but with blocks on different drives. | |
552 | * With this layout, and block is never stored twice on the one device. | |
553 | * | |
554 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 555 | * on each device that it is on. |
1da177e4 LT |
556 | * |
557 | * raid10_find_virt does the reverse mapping, from a device and a | |
558 | * sector offset to a virtual address | |
559 | */ | |
560 | ||
f8c9e74f | 561 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
562 | { |
563 | int n,f; | |
564 | sector_t sector; | |
565 | sector_t chunk; | |
566 | sector_t stripe; | |
567 | int dev; | |
1da177e4 | 568 | int slot = 0; |
9a3152ab JB |
569 | int last_far_set_start, last_far_set_size; |
570 | ||
571 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
572 | last_far_set_start *= geo->far_set_size; | |
573 | ||
574 | last_far_set_size = geo->far_set_size; | |
575 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
576 | |
577 | /* now calculate first sector/dev */ | |
5cf00fcd N |
578 | chunk = r10bio->sector >> geo->chunk_shift; |
579 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 580 | |
5cf00fcd | 581 | chunk *= geo->near_copies; |
1da177e4 | 582 | stripe = chunk; |
5cf00fcd N |
583 | dev = sector_div(stripe, geo->raid_disks); |
584 | if (geo->far_offset) | |
585 | stripe *= geo->far_copies; | |
1da177e4 | 586 | |
5cf00fcd | 587 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
588 | |
589 | /* and calculate all the others */ | |
5cf00fcd | 590 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 591 | int d = dev; |
475901af | 592 | int set; |
1da177e4 | 593 | sector_t s = sector; |
1da177e4 | 594 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 595 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
596 | slot++; |
597 | ||
5cf00fcd | 598 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 599 | set = d / geo->far_set_size; |
5cf00fcd | 600 | d += geo->near_copies; |
475901af | 601 | |
9a3152ab JB |
602 | if ((geo->raid_disks % geo->far_set_size) && |
603 | (d > last_far_set_start)) { | |
604 | d -= last_far_set_start; | |
605 | d %= last_far_set_size; | |
606 | d += last_far_set_start; | |
607 | } else { | |
608 | d %= geo->far_set_size; | |
609 | d += geo->far_set_size * set; | |
610 | } | |
5cf00fcd | 611 | s += geo->stride; |
1da177e4 LT |
612 | r10bio->devs[slot].devnum = d; |
613 | r10bio->devs[slot].addr = s; | |
614 | slot++; | |
615 | } | |
616 | dev++; | |
5cf00fcd | 617 | if (dev >= geo->raid_disks) { |
1da177e4 | 618 | dev = 0; |
5cf00fcd | 619 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
620 | } |
621 | } | |
f8c9e74f N |
622 | } |
623 | ||
624 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
625 | { | |
626 | struct geom *geo = &conf->geo; | |
627 | ||
628 | if (conf->reshape_progress != MaxSector && | |
629 | ((r10bio->sector >= conf->reshape_progress) != | |
630 | conf->mddev->reshape_backwards)) { | |
631 | set_bit(R10BIO_Previous, &r10bio->state); | |
632 | geo = &conf->prev; | |
633 | } else | |
634 | clear_bit(R10BIO_Previous, &r10bio->state); | |
635 | ||
636 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
637 | } |
638 | ||
e879a879 | 639 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
640 | { |
641 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
642 | /* Never use conf->prev as this is only called during resync |
643 | * or recovery, so reshape isn't happening | |
644 | */ | |
5cf00fcd | 645 | struct geom *geo = &conf->geo; |
475901af JB |
646 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
647 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
648 | int last_far_set_start; |
649 | ||
650 | if (geo->raid_disks % geo->far_set_size) { | |
651 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
652 | last_far_set_start *= geo->far_set_size; | |
653 | ||
654 | if (dev >= last_far_set_start) { | |
655 | far_set_size = geo->far_set_size; | |
656 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
657 | far_set_start = last_far_set_start; | |
658 | } | |
659 | } | |
1da177e4 | 660 | |
5cf00fcd N |
661 | offset = sector & geo->chunk_mask; |
662 | if (geo->far_offset) { | |
c93983bf | 663 | int fc; |
5cf00fcd N |
664 | chunk = sector >> geo->chunk_shift; |
665 | fc = sector_div(chunk, geo->far_copies); | |
666 | dev -= fc * geo->near_copies; | |
475901af JB |
667 | if (dev < far_set_start) |
668 | dev += far_set_size; | |
c93983bf | 669 | } else { |
5cf00fcd N |
670 | while (sector >= geo->stride) { |
671 | sector -= geo->stride; | |
475901af JB |
672 | if (dev < (geo->near_copies + far_set_start)) |
673 | dev += far_set_size - geo->near_copies; | |
c93983bf | 674 | else |
5cf00fcd | 675 | dev -= geo->near_copies; |
c93983bf | 676 | } |
5cf00fcd | 677 | chunk = sector >> geo->chunk_shift; |
c93983bf | 678 | } |
5cf00fcd N |
679 | vchunk = chunk * geo->raid_disks + dev; |
680 | sector_div(vchunk, geo->near_copies); | |
681 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
682 | } |
683 | ||
1da177e4 LT |
684 | /* |
685 | * This routine returns the disk from which the requested read should | |
686 | * be done. There is a per-array 'next expected sequential IO' sector | |
687 | * number - if this matches on the next IO then we use the last disk. | |
688 | * There is also a per-disk 'last know head position' sector that is | |
689 | * maintained from IRQ contexts, both the normal and the resync IO | |
690 | * completion handlers update this position correctly. If there is no | |
691 | * perfect sequential match then we pick the disk whose head is closest. | |
692 | * | |
693 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
694 | * because position is mirror, not device based. | |
695 | * | |
696 | * The rdev for the device selected will have nr_pending incremented. | |
697 | */ | |
698 | ||
699 | /* | |
700 | * FIXME: possibly should rethink readbalancing and do it differently | |
701 | * depending on near_copies / far_copies geometry. | |
702 | */ | |
96c3fd1f N |
703 | static struct md_rdev *read_balance(struct r10conf *conf, |
704 | struct r10bio *r10_bio, | |
705 | int *max_sectors) | |
1da177e4 | 706 | { |
af3a2cd6 | 707 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 708 | int disk, slot; |
856e08e2 N |
709 | int sectors = r10_bio->sectors; |
710 | int best_good_sectors; | |
56d99121 | 711 | sector_t new_distance, best_dist; |
e9eeba28 | 712 | struct md_rdev *best_dist_rdev, *best_pending_rdev, *rdev = NULL; |
56d99121 | 713 | int do_balance; |
e9eeba28 GJ |
714 | int best_dist_slot, best_pending_slot; |
715 | bool has_nonrot_disk = false; | |
716 | unsigned int min_pending; | |
5cf00fcd | 717 | struct geom *geo = &conf->geo; |
1da177e4 LT |
718 | |
719 | raid10_find_phys(conf, r10_bio); | |
720 | rcu_read_lock(); | |
e9eeba28 GJ |
721 | best_dist_slot = -1; |
722 | min_pending = UINT_MAX; | |
723 | best_dist_rdev = NULL; | |
724 | best_pending_rdev = NULL; | |
56d99121 | 725 | best_dist = MaxSector; |
856e08e2 | 726 | best_good_sectors = 0; |
56d99121 | 727 | do_balance = 1; |
8d3ca83d | 728 | clear_bit(R10BIO_FailFast, &r10_bio->state); |
1da177e4 LT |
729 | /* |
730 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
731 | * device if no resync is going on (recovery is ok), or below |
732 | * the resync window. We take the first readable disk when | |
733 | * above the resync window. | |
1da177e4 | 734 | */ |
d4098c72 GJ |
735 | if ((conf->mddev->recovery_cp < MaxSector |
736 | && (this_sector + sectors >= conf->next_resync)) || | |
737 | (mddev_is_clustered(conf->mddev) && | |
738 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, | |
739 | this_sector + sectors))) | |
56d99121 | 740 | do_balance = 0; |
1da177e4 | 741 | |
56d99121 | 742 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
743 | sector_t first_bad; |
744 | int bad_sectors; | |
745 | sector_t dev_sector; | |
e9eeba28 GJ |
746 | unsigned int pending; |
747 | bool nonrot; | |
856e08e2 | 748 | |
56d99121 N |
749 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
750 | continue; | |
1da177e4 | 751 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
752 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
753 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
754 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
755 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 | 756 | if (rdev == NULL || |
8ae12666 | 757 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
758 | continue; |
759 | if (!test_bit(In_sync, &rdev->flags) && | |
760 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
761 | continue; |
762 | ||
856e08e2 N |
763 | dev_sector = r10_bio->devs[slot].addr; |
764 | if (is_badblock(rdev, dev_sector, sectors, | |
765 | &first_bad, &bad_sectors)) { | |
766 | if (best_dist < MaxSector) | |
767 | /* Already have a better slot */ | |
768 | continue; | |
769 | if (first_bad <= dev_sector) { | |
770 | /* Cannot read here. If this is the | |
771 | * 'primary' device, then we must not read | |
772 | * beyond 'bad_sectors' from another device. | |
773 | */ | |
774 | bad_sectors -= (dev_sector - first_bad); | |
775 | if (!do_balance && sectors > bad_sectors) | |
776 | sectors = bad_sectors; | |
777 | if (best_good_sectors > sectors) | |
778 | best_good_sectors = sectors; | |
779 | } else { | |
780 | sector_t good_sectors = | |
781 | first_bad - dev_sector; | |
782 | if (good_sectors > best_good_sectors) { | |
783 | best_good_sectors = good_sectors; | |
e9eeba28 GJ |
784 | best_dist_slot = slot; |
785 | best_dist_rdev = rdev; | |
856e08e2 N |
786 | } |
787 | if (!do_balance) | |
788 | /* Must read from here */ | |
789 | break; | |
790 | } | |
791 | continue; | |
792 | } else | |
793 | best_good_sectors = sectors; | |
794 | ||
56d99121 N |
795 | if (!do_balance) |
796 | break; | |
1da177e4 | 797 | |
e9eeba28 GJ |
798 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
799 | has_nonrot_disk |= nonrot; | |
800 | pending = atomic_read(&rdev->nr_pending); | |
801 | if (min_pending > pending && nonrot) { | |
802 | min_pending = pending; | |
803 | best_pending_slot = slot; | |
804 | best_pending_rdev = rdev; | |
805 | } | |
806 | ||
807 | if (best_dist_slot >= 0) | |
8d3ca83d N |
808 | /* At least 2 disks to choose from so failfast is OK */ |
809 | set_bit(R10BIO_FailFast, &r10_bio->state); | |
22dfdf52 N |
810 | /* This optimisation is debatable, and completely destroys |
811 | * sequential read speed for 'far copies' arrays. So only | |
812 | * keep it for 'near' arrays, and review those later. | |
813 | */ | |
e9eeba28 | 814 | if (geo->near_copies > 1 && !pending) |
8d3ca83d | 815 | new_distance = 0; |
8ed3a195 KS |
816 | |
817 | /* for far > 1 always use the lowest address */ | |
8d3ca83d | 818 | else if (geo->far_copies > 1) |
56d99121 | 819 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 820 | else |
56d99121 N |
821 | new_distance = abs(r10_bio->devs[slot].addr - |
822 | conf->mirrors[disk].head_position); | |
e9eeba28 | 823 | |
56d99121 N |
824 | if (new_distance < best_dist) { |
825 | best_dist = new_distance; | |
e9eeba28 GJ |
826 | best_dist_slot = slot; |
827 | best_dist_rdev = rdev; | |
1da177e4 LT |
828 | } |
829 | } | |
abbf098e | 830 | if (slot >= conf->copies) { |
e9eeba28 GJ |
831 | if (has_nonrot_disk) { |
832 | slot = best_pending_slot; | |
833 | rdev = best_pending_rdev; | |
834 | } else { | |
835 | slot = best_dist_slot; | |
836 | rdev = best_dist_rdev; | |
837 | } | |
abbf098e | 838 | } |
1da177e4 | 839 | |
56d99121 | 840 | if (slot >= 0) { |
56d99121 | 841 | atomic_inc(&rdev->nr_pending); |
56d99121 N |
842 | r10_bio->read_slot = slot; |
843 | } else | |
96c3fd1f | 844 | rdev = NULL; |
1da177e4 | 845 | rcu_read_unlock(); |
856e08e2 | 846 | *max_sectors = best_good_sectors; |
1da177e4 | 847 | |
96c3fd1f | 848 | return rdev; |
1da177e4 LT |
849 | } |
850 | ||
e879a879 | 851 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
852 | { |
853 | /* Any writes that have been queued but are awaiting | |
854 | * bitmap updates get flushed here. | |
a35e63ef | 855 | */ |
a35e63ef N |
856 | spin_lock_irq(&conf->device_lock); |
857 | ||
858 | if (conf->pending_bio_list.head) { | |
18022a1b | 859 | struct blk_plug plug; |
a35e63ef | 860 | struct bio *bio; |
18022a1b | 861 | |
a35e63ef | 862 | bio = bio_list_get(&conf->pending_bio_list); |
34db0cd6 | 863 | conf->pending_count = 0; |
a35e63ef | 864 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
865 | |
866 | /* | |
867 | * As this is called in a wait_event() loop (see freeze_array), | |
868 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
869 | * cause a warning when we prepare to wait again. As it is | |
870 | * rare that this path is taken, it is perfectly safe to force | |
871 | * us to go around the wait_event() loop again, so the warning | |
872 | * is a false-positive. Silence the warning by resetting | |
873 | * thread state | |
874 | */ | |
875 | __set_current_state(TASK_RUNNING); | |
876 | ||
18022a1b | 877 | blk_start_plug(&plug); |
a35e63ef N |
878 | /* flush any pending bitmap writes to disk |
879 | * before proceeding w/ I/O */ | |
e64e4018 | 880 | md_bitmap_unplug(conf->mddev->bitmap); |
34db0cd6 | 881 | wake_up(&conf->wait_barrier); |
a35e63ef N |
882 | |
883 | while (bio) { /* submit pending writes */ | |
884 | struct bio *next = bio->bi_next; | |
74d46992 | 885 | struct md_rdev *rdev = (void*)bio->bi_disk; |
a35e63ef | 886 | bio->bi_next = NULL; |
74d46992 | 887 | bio_set_dev(bio, rdev->bdev); |
a9ae93c8 | 888 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 889 | bio_io_error(bio); |
a9ae93c8 | 890 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
74d46992 | 891 | !blk_queue_discard(bio->bi_disk->queue))) |
532a2a3f | 892 | /* Just ignore it */ |
4246a0b6 | 893 | bio_endio(bio); |
532a2a3f | 894 | else |
ed00aabd | 895 | submit_bio_noacct(bio); |
a35e63ef N |
896 | bio = next; |
897 | } | |
18022a1b | 898 | blk_finish_plug(&plug); |
a35e63ef N |
899 | } else |
900 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 901 | } |
7eaceacc | 902 | |
0a27ec96 N |
903 | /* Barriers.... |
904 | * Sometimes we need to suspend IO while we do something else, | |
905 | * either some resync/recovery, or reconfigure the array. | |
906 | * To do this we raise a 'barrier'. | |
907 | * The 'barrier' is a counter that can be raised multiple times | |
908 | * to count how many activities are happening which preclude | |
909 | * normal IO. | |
910 | * We can only raise the barrier if there is no pending IO. | |
911 | * i.e. if nr_pending == 0. | |
912 | * We choose only to raise the barrier if no-one is waiting for the | |
913 | * barrier to go down. This means that as soon as an IO request | |
914 | * is ready, no other operations which require a barrier will start | |
915 | * until the IO request has had a chance. | |
916 | * | |
917 | * So: regular IO calls 'wait_barrier'. When that returns there | |
918 | * is no backgroup IO happening, It must arrange to call | |
919 | * allow_barrier when it has finished its IO. | |
920 | * backgroup IO calls must call raise_barrier. Once that returns | |
921 | * there is no normal IO happeing. It must arrange to call | |
922 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 923 | */ |
1da177e4 | 924 | |
e879a879 | 925 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 926 | { |
6cce3b23 | 927 | BUG_ON(force && !conf->barrier); |
1da177e4 | 928 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 929 | |
6cce3b23 N |
930 | /* Wait until no block IO is waiting (unless 'force') */ |
931 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
eed8c02e | 932 | conf->resync_lock); |
0a27ec96 N |
933 | |
934 | /* block any new IO from starting */ | |
935 | conf->barrier++; | |
936 | ||
c3b328ac | 937 | /* Now wait for all pending IO to complete */ |
0a27ec96 | 938 | wait_event_lock_irq(conf->wait_barrier, |
0e5313e2 | 939 | !atomic_read(&conf->nr_pending) && conf->barrier < RESYNC_DEPTH, |
eed8c02e | 940 | conf->resync_lock); |
0a27ec96 N |
941 | |
942 | spin_unlock_irq(&conf->resync_lock); | |
943 | } | |
944 | ||
e879a879 | 945 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
946 | { |
947 | unsigned long flags; | |
948 | spin_lock_irqsave(&conf->resync_lock, flags); | |
949 | conf->barrier--; | |
950 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
951 | wake_up(&conf->wait_barrier); | |
952 | } | |
953 | ||
e879a879 | 954 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
955 | { |
956 | spin_lock_irq(&conf->resync_lock); | |
957 | if (conf->barrier) { | |
fe630de0 | 958 | struct bio_list *bio_list = current->bio_list; |
0a27ec96 | 959 | conf->nr_waiting++; |
d6b42dcb N |
960 | /* Wait for the barrier to drop. |
961 | * However if there are already pending | |
962 | * requests (preventing the barrier from | |
963 | * rising completely), and the | |
964 | * pre-process bio queue isn't empty, | |
965 | * then don't wait, as we need to empty | |
966 | * that queue to get the nr_pending | |
967 | * count down. | |
968 | */ | |
578b54ad | 969 | raid10_log(conf->mddev, "wait barrier"); |
d6b42dcb N |
970 | wait_event_lock_irq(conf->wait_barrier, |
971 | !conf->barrier || | |
0e5313e2 | 972 | (atomic_read(&conf->nr_pending) && |
fe630de0 VM |
973 | bio_list && |
974 | (!bio_list_empty(&bio_list[0]) || | |
975 | !bio_list_empty(&bio_list[1]))) || | |
976 | /* move on if recovery thread is | |
977 | * blocked by us | |
978 | */ | |
979 | (conf->mddev->thread->tsk == current && | |
980 | test_bit(MD_RECOVERY_RUNNING, | |
981 | &conf->mddev->recovery) && | |
982 | conf->nr_queued > 0), | |
eed8c02e | 983 | conf->resync_lock); |
0a27ec96 | 984 | conf->nr_waiting--; |
0e5313e2 TM |
985 | if (!conf->nr_waiting) |
986 | wake_up(&conf->wait_barrier); | |
1da177e4 | 987 | } |
0e5313e2 | 988 | atomic_inc(&conf->nr_pending); |
1da177e4 LT |
989 | spin_unlock_irq(&conf->resync_lock); |
990 | } | |
991 | ||
e879a879 | 992 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 | 993 | { |
0e5313e2 TM |
994 | if ((atomic_dec_and_test(&conf->nr_pending)) || |
995 | (conf->array_freeze_pending)) | |
996 | wake_up(&conf->wait_barrier); | |
0a27ec96 N |
997 | } |
998 | ||
e2d59925 | 999 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
1000 | { |
1001 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 1002 | * go quiet. |
4443ae10 | 1003 | * We increment barrier and nr_waiting, and then |
e2d59925 | 1004 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
1005 | * This is called in the context of one normal IO request |
1006 | * that has failed. Thus any sync request that might be pending | |
1007 | * will be blocked by nr_pending, and we need to wait for | |
1008 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 1009 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
1010 | * must match the number of pending IOs (nr_pending) before |
1011 | * we continue. | |
4443ae10 N |
1012 | */ |
1013 | spin_lock_irq(&conf->resync_lock); | |
0e5313e2 | 1014 | conf->array_freeze_pending++; |
4443ae10 N |
1015 | conf->barrier++; |
1016 | conf->nr_waiting++; | |
eed8c02e | 1017 | wait_event_lock_irq_cmd(conf->wait_barrier, |
0e5313e2 | 1018 | atomic_read(&conf->nr_pending) == conf->nr_queued+extra, |
eed8c02e LC |
1019 | conf->resync_lock, |
1020 | flush_pending_writes(conf)); | |
c3b328ac | 1021 | |
0e5313e2 | 1022 | conf->array_freeze_pending--; |
4443ae10 N |
1023 | spin_unlock_irq(&conf->resync_lock); |
1024 | } | |
1025 | ||
e879a879 | 1026 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1027 | { |
1028 | /* reverse the effect of the freeze */ | |
1029 | spin_lock_irq(&conf->resync_lock); | |
1030 | conf->barrier--; | |
1031 | conf->nr_waiting--; | |
1032 | wake_up(&conf->wait_barrier); | |
1033 | spin_unlock_irq(&conf->resync_lock); | |
1034 | } | |
1035 | ||
f8c9e74f N |
1036 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1037 | struct md_rdev *rdev) | |
1038 | { | |
1039 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1040 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1041 | return rdev->data_offset; | |
1042 | else | |
1043 | return rdev->new_data_offset; | |
1044 | } | |
1045 | ||
57c67df4 N |
1046 | struct raid10_plug_cb { |
1047 | struct blk_plug_cb cb; | |
1048 | struct bio_list pending; | |
1049 | int pending_cnt; | |
1050 | }; | |
1051 | ||
1052 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1053 | { | |
1054 | struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb, | |
1055 | cb); | |
1056 | struct mddev *mddev = plug->cb.data; | |
1057 | struct r10conf *conf = mddev->private; | |
1058 | struct bio *bio; | |
1059 | ||
874807a8 | 1060 | if (from_schedule || current->bio_list) { |
57c67df4 N |
1061 | spin_lock_irq(&conf->device_lock); |
1062 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1063 | conf->pending_count += plug->pending_cnt; | |
1064 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1065 | wake_up(&conf->wait_barrier); |
57c67df4 N |
1066 | md_wakeup_thread(mddev->thread); |
1067 | kfree(plug); | |
1068 | return; | |
1069 | } | |
1070 | ||
1071 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1072 | bio = bio_list_get(&plug->pending); | |
e64e4018 | 1073 | md_bitmap_unplug(mddev->bitmap); |
57c67df4 N |
1074 | wake_up(&conf->wait_barrier); |
1075 | ||
1076 | while (bio) { /* submit pending writes */ | |
1077 | struct bio *next = bio->bi_next; | |
74d46992 | 1078 | struct md_rdev *rdev = (void*)bio->bi_disk; |
57c67df4 | 1079 | bio->bi_next = NULL; |
74d46992 | 1080 | bio_set_dev(bio, rdev->bdev); |
a9ae93c8 | 1081 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 1082 | bio_io_error(bio); |
a9ae93c8 | 1083 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
74d46992 | 1084 | !blk_queue_discard(bio->bi_disk->queue))) |
32f9f570 | 1085 | /* Just ignore it */ |
4246a0b6 | 1086 | bio_endio(bio); |
32f9f570 | 1087 | else |
ed00aabd | 1088 | submit_bio_noacct(bio); |
57c67df4 N |
1089 | bio = next; |
1090 | } | |
1091 | kfree(plug); | |
1092 | } | |
1093 | ||
caea3c47 GJ |
1094 | /* |
1095 | * 1. Register the new request and wait if the reconstruction thread has put | |
1096 | * up a bar for new requests. Continue immediately if no resync is active | |
1097 | * currently. | |
1098 | * 2. If IO spans the reshape position. Need to wait for reshape to pass. | |
1099 | */ | |
1100 | static void regular_request_wait(struct mddev *mddev, struct r10conf *conf, | |
1101 | struct bio *bio, sector_t sectors) | |
1102 | { | |
1103 | wait_barrier(conf); | |
1104 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
1105 | bio->bi_iter.bi_sector < conf->reshape_progress && | |
1106 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
1107 | raid10_log(conf->mddev, "wait reshape"); | |
1108 | allow_barrier(conf); | |
1109 | wait_event(conf->wait_barrier, | |
1110 | conf->reshape_progress <= bio->bi_iter.bi_sector || | |
1111 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1112 | sectors); | |
1113 | wait_barrier(conf); | |
1114 | } | |
1115 | } | |
1116 | ||
bb5f1ed7 RL |
1117 | static void raid10_read_request(struct mddev *mddev, struct bio *bio, |
1118 | struct r10bio *r10_bio) | |
1da177e4 | 1119 | { |
e879a879 | 1120 | struct r10conf *conf = mddev->private; |
1da177e4 | 1121 | struct bio *read_bio; |
bb5f1ed7 RL |
1122 | const int op = bio_op(bio); |
1123 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
bb5f1ed7 | 1124 | int max_sectors; |
bb5f1ed7 | 1125 | struct md_rdev *rdev; |
545250f2 N |
1126 | char b[BDEVNAME_SIZE]; |
1127 | int slot = r10_bio->read_slot; | |
1128 | struct md_rdev *err_rdev = NULL; | |
1129 | gfp_t gfp = GFP_NOIO; | |
bb5f1ed7 | 1130 | |
93decc56 | 1131 | if (slot >= 0 && r10_bio->devs[slot].rdev) { |
545250f2 N |
1132 | /* |
1133 | * This is an error retry, but we cannot | |
1134 | * safely dereference the rdev in the r10_bio, | |
1135 | * we must use the one in conf. | |
1136 | * If it has already been disconnected (unlikely) | |
1137 | * we lose the device name in error messages. | |
1138 | */ | |
1139 | int disk; | |
1140 | /* | |
1141 | * As we are blocking raid10, it is a little safer to | |
1142 | * use __GFP_HIGH. | |
1143 | */ | |
1144 | gfp = GFP_NOIO | __GFP_HIGH; | |
1145 | ||
1146 | rcu_read_lock(); | |
1147 | disk = r10_bio->devs[slot].devnum; | |
1148 | err_rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
1149 | if (err_rdev) | |
1150 | bdevname(err_rdev->bdev, b); | |
1151 | else { | |
1152 | strcpy(b, "???"); | |
1153 | /* This never gets dereferenced */ | |
1154 | err_rdev = r10_bio->devs[slot].rdev; | |
1155 | } | |
1156 | rcu_read_unlock(); | |
1157 | } | |
bb5f1ed7 | 1158 | |
caea3c47 | 1159 | regular_request_wait(mddev, conf, bio, r10_bio->sectors); |
bb5f1ed7 RL |
1160 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1161 | if (!rdev) { | |
545250f2 N |
1162 | if (err_rdev) { |
1163 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1164 | mdname(mddev), b, | |
1165 | (unsigned long long)r10_bio->sector); | |
1166 | } | |
bb5f1ed7 RL |
1167 | raid_end_bio_io(r10_bio); |
1168 | return; | |
1169 | } | |
545250f2 N |
1170 | if (err_rdev) |
1171 | pr_err_ratelimited("md/raid10:%s: %s: redirecting sector %llu to another mirror\n", | |
1172 | mdname(mddev), | |
1173 | bdevname(rdev->bdev, b), | |
1174 | (unsigned long long)r10_bio->sector); | |
fc9977dd N |
1175 | if (max_sectors < bio_sectors(bio)) { |
1176 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1177 | gfp, &conf->bio_split); |
fc9977dd | 1178 | bio_chain(split, bio); |
e820d55c | 1179 | allow_barrier(conf); |
ed00aabd | 1180 | submit_bio_noacct(bio); |
e820d55c | 1181 | wait_barrier(conf); |
fc9977dd N |
1182 | bio = split; |
1183 | r10_bio->master_bio = bio; | |
1184 | r10_bio->sectors = max_sectors; | |
1185 | } | |
bb5f1ed7 RL |
1186 | slot = r10_bio->read_slot; |
1187 | ||
afeee514 | 1188 | read_bio = bio_clone_fast(bio, gfp, &mddev->bio_set); |
bb5f1ed7 RL |
1189 | |
1190 | r10_bio->devs[slot].bio = read_bio; | |
1191 | r10_bio->devs[slot].rdev = rdev; | |
1192 | ||
1193 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + | |
1194 | choose_data_offset(r10_bio, rdev); | |
74d46992 | 1195 | bio_set_dev(read_bio, rdev->bdev); |
bb5f1ed7 RL |
1196 | read_bio->bi_end_io = raid10_end_read_request; |
1197 | bio_set_op_attrs(read_bio, op, do_sync); | |
1198 | if (test_bit(FailFast, &rdev->flags) && | |
1199 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
1200 | read_bio->bi_opf |= MD_FAILFAST; | |
1201 | read_bio->bi_private = r10_bio; | |
1202 | ||
1203 | if (mddev->gendisk) | |
1c02fca6 | 1204 | trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), |
bb5f1ed7 | 1205 | r10_bio->sector); |
ed00aabd | 1206 | submit_bio_noacct(read_bio); |
bb5f1ed7 RL |
1207 | return; |
1208 | } | |
1209 | ||
27f26a0f GJ |
1210 | static void raid10_write_one_disk(struct mddev *mddev, struct r10bio *r10_bio, |
1211 | struct bio *bio, bool replacement, | |
fc9977dd | 1212 | int n_copy) |
bb5f1ed7 | 1213 | { |
796a5cf0 | 1214 | const int op = bio_op(bio); |
1eff9d32 JA |
1215 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1216 | const unsigned long do_fua = (bio->bi_opf & REQ_FUA); | |
6cce3b23 | 1217 | unsigned long flags; |
57c67df4 N |
1218 | struct blk_plug_cb *cb; |
1219 | struct raid10_plug_cb *plug = NULL; | |
27f26a0f GJ |
1220 | struct r10conf *conf = mddev->private; |
1221 | struct md_rdev *rdev; | |
1222 | int devnum = r10_bio->devs[n_copy].devnum; | |
1223 | struct bio *mbio; | |
1224 | ||
1225 | if (replacement) { | |
1226 | rdev = conf->mirrors[devnum].replacement; | |
1227 | if (rdev == NULL) { | |
1228 | /* Replacement just got moved to main 'rdev' */ | |
1229 | smp_mb(); | |
1230 | rdev = conf->mirrors[devnum].rdev; | |
1231 | } | |
1232 | } else | |
1233 | rdev = conf->mirrors[devnum].rdev; | |
1234 | ||
afeee514 | 1235 | mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
27f26a0f GJ |
1236 | if (replacement) |
1237 | r10_bio->devs[n_copy].repl_bio = mbio; | |
1238 | else | |
1239 | r10_bio->devs[n_copy].bio = mbio; | |
1240 | ||
1241 | mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + | |
1242 | choose_data_offset(r10_bio, rdev)); | |
74d46992 | 1243 | bio_set_dev(mbio, rdev->bdev); |
27f26a0f GJ |
1244 | mbio->bi_end_io = raid10_end_write_request; |
1245 | bio_set_op_attrs(mbio, op, do_sync | do_fua); | |
1246 | if (!replacement && test_bit(FailFast, | |
1247 | &conf->mirrors[devnum].rdev->flags) | |
1248 | && enough(conf, devnum)) | |
1249 | mbio->bi_opf |= MD_FAILFAST; | |
1250 | mbio->bi_private = r10_bio; | |
1251 | ||
1252 | if (conf->mddev->gendisk) | |
1c02fca6 | 1253 | trace_block_bio_remap(mbio, disk_devt(conf->mddev->gendisk), |
27f26a0f GJ |
1254 | r10_bio->sector); |
1255 | /* flush_pending_writes() needs access to the rdev so...*/ | |
74d46992 | 1256 | mbio->bi_disk = (void *)rdev; |
27f26a0f GJ |
1257 | |
1258 | atomic_inc(&r10_bio->remaining); | |
1259 | ||
1260 | cb = blk_check_plugged(raid10_unplug, mddev, sizeof(*plug)); | |
1261 | if (cb) | |
1262 | plug = container_of(cb, struct raid10_plug_cb, cb); | |
1263 | else | |
1264 | plug = NULL; | |
27f26a0f GJ |
1265 | if (plug) { |
1266 | bio_list_add(&plug->pending, mbio); | |
1267 | plug->pending_cnt++; | |
1268 | } else { | |
23b245c0 | 1269 | spin_lock_irqsave(&conf->device_lock, flags); |
27f26a0f GJ |
1270 | bio_list_add(&conf->pending_bio_list, mbio); |
1271 | conf->pending_count++; | |
23b245c0 | 1272 | spin_unlock_irqrestore(&conf->device_lock, flags); |
27f26a0f | 1273 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1274 | } |
27f26a0f GJ |
1275 | } |
1276 | ||
1277 | static void raid10_write_request(struct mddev *mddev, struct bio *bio, | |
1278 | struct r10bio *r10_bio) | |
1279 | { | |
1280 | struct r10conf *conf = mddev->private; | |
1281 | int i; | |
4e2c6567 | 1282 | struct md_rdev *blocked_rdev; |
bb5f1ed7 | 1283 | sector_t sectors; |
d4432c23 | 1284 | int max_sectors; |
1da177e4 | 1285 | |
cb8a7a7e GJ |
1286 | if ((mddev_is_clustered(mddev) && |
1287 | md_cluster_ops->area_resyncing(mddev, WRITE, | |
1288 | bio->bi_iter.bi_sector, | |
1289 | bio_end_sector(bio)))) { | |
1290 | DEFINE_WAIT(w); | |
1291 | for (;;) { | |
1292 | prepare_to_wait(&conf->wait_barrier, | |
1293 | &w, TASK_IDLE); | |
1294 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, | |
1295 | bio->bi_iter.bi_sector, bio_end_sector(bio))) | |
1296 | break; | |
1297 | schedule(); | |
1298 | } | |
1299 | finish_wait(&conf->wait_barrier, &w); | |
1300 | } | |
1301 | ||
fc9977dd | 1302 | sectors = r10_bio->sectors; |
caea3c47 | 1303 | regular_request_wait(mddev, conf, bio, sectors); |
3ea7daa5 | 1304 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
3ea7daa5 | 1305 | (mddev->reshape_backwards |
4f024f37 KO |
1306 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1307 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1308 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1309 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1310 | /* Need to update reshape_position in metadata */ |
1311 | mddev->reshape_position = conf->reshape_progress; | |
2953079c SL |
1312 | set_mask_bits(&mddev->sb_flags, 0, |
1313 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
3ea7daa5 | 1314 | md_wakeup_thread(mddev->thread); |
578b54ad | 1315 | raid10_log(conf->mddev, "wait reshape metadata"); |
3ea7daa5 | 1316 | wait_event(mddev->sb_wait, |
2953079c | 1317 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
3ea7daa5 N |
1318 | |
1319 | conf->reshape_safe = mddev->reshape_position; | |
1320 | } | |
1321 | ||
34db0cd6 N |
1322 | if (conf->pending_count >= max_queued_requests) { |
1323 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1324 | raid10_log(mddev, "wait queued"); |
34db0cd6 N |
1325 | wait_event(conf->wait_barrier, |
1326 | conf->pending_count < max_queued_requests); | |
1327 | } | |
6bfe0b49 | 1328 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1329 | * inc refcount on their rdev. Record them by setting |
1330 | * bios[x] to bio | |
d4432c23 N |
1331 | * If there are known/acknowledged bad blocks on any device |
1332 | * on which we have seen a write error, we want to avoid | |
1333 | * writing to those blocks. This potentially requires several | |
1334 | * writes to write around the bad blocks. Each set of writes | |
fd16f2e8 | 1335 | * gets its own r10_bio with a set of bios attached. |
1da177e4 | 1336 | */ |
c3b328ac | 1337 | |
69335ef3 | 1338 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1339 | raid10_find_phys(conf, r10_bio); |
4e2c6567 SL |
1340 | retry_write: |
1341 | blocked_rdev = NULL; | |
1da177e4 | 1342 | rcu_read_lock(); |
d4432c23 N |
1343 | max_sectors = r10_bio->sectors; |
1344 | ||
1da177e4 LT |
1345 | for (i = 0; i < conf->copies; i++) { |
1346 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1347 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1348 | struct md_rdev *rrdev = rcu_dereference( |
1349 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1350 | if (rdev == rrdev) |
1351 | rrdev = NULL; | |
4e2c6567 SL |
1352 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1353 | atomic_inc(&rdev->nr_pending); | |
1354 | blocked_rdev = rdev; | |
1355 | break; | |
1356 | } | |
1357 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { | |
1358 | atomic_inc(&rrdev->nr_pending); | |
1359 | blocked_rdev = rrdev; | |
1360 | break; | |
1361 | } | |
8ae12666 | 1362 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1363 | rdev = NULL; |
8ae12666 | 1364 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1365 | rrdev = NULL; |
1366 | ||
d4432c23 | 1367 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1368 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1369 | |
1370 | if (!rdev && !rrdev) { | |
6cce3b23 | 1371 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1372 | continue; |
1373 | } | |
e7c0c3fa | 1374 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1375 | sector_t first_bad; |
1376 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1377 | int bad_sectors; | |
1378 | int is_bad; | |
1379 | ||
bb5f1ed7 | 1380 | is_bad = is_badblock(rdev, dev_sector, max_sectors, |
d4432c23 | 1381 | &first_bad, &bad_sectors); |
4e2c6567 SL |
1382 | if (is_bad < 0) { |
1383 | /* Mustn't write here until the bad block | |
1384 | * is acknowledged | |
1385 | */ | |
1386 | atomic_inc(&rdev->nr_pending); | |
1387 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1388 | blocked_rdev = rdev; | |
1389 | break; | |
1390 | } | |
d4432c23 N |
1391 | if (is_bad && first_bad <= dev_sector) { |
1392 | /* Cannot write here at all */ | |
1393 | bad_sectors -= (dev_sector - first_bad); | |
1394 | if (bad_sectors < max_sectors) | |
1395 | /* Mustn't write more than bad_sectors | |
1396 | * to other devices yet | |
1397 | */ | |
1398 | max_sectors = bad_sectors; | |
1399 | /* We don't set R10BIO_Degraded as that | |
1400 | * only applies if the disk is missing, | |
1401 | * so it might be re-added, and we want to | |
1402 | * know to recover this chunk. | |
1403 | * In this case the device is here, and the | |
1404 | * fact that this chunk is not in-sync is | |
1405 | * recorded in the bad block log. | |
1406 | */ | |
1407 | continue; | |
1408 | } | |
1409 | if (is_bad) { | |
1410 | int good_sectors = first_bad - dev_sector; | |
1411 | if (good_sectors < max_sectors) | |
1412 | max_sectors = good_sectors; | |
1413 | } | |
6cce3b23 | 1414 | } |
e7c0c3fa N |
1415 | if (rdev) { |
1416 | r10_bio->devs[i].bio = bio; | |
1417 | atomic_inc(&rdev->nr_pending); | |
1418 | } | |
475b0321 N |
1419 | if (rrdev) { |
1420 | r10_bio->devs[i].repl_bio = bio; | |
1421 | atomic_inc(&rrdev->nr_pending); | |
1422 | } | |
1da177e4 LT |
1423 | } |
1424 | rcu_read_unlock(); | |
1425 | ||
4e2c6567 SL |
1426 | if (unlikely(blocked_rdev)) { |
1427 | /* Have to wait for this device to get unblocked, then retry */ | |
1428 | int j; | |
1429 | int d; | |
1430 | ||
1431 | for (j = 0; j < i; j++) { | |
1432 | if (r10_bio->devs[j].bio) { | |
1433 | d = r10_bio->devs[j].devnum; | |
1434 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1435 | } | |
1436 | if (r10_bio->devs[j].repl_bio) { | |
1437 | struct md_rdev *rdev; | |
1438 | d = r10_bio->devs[j].devnum; | |
1439 | rdev = conf->mirrors[d].replacement; | |
1440 | if (!rdev) { | |
1441 | /* Race with remove_disk */ | |
1442 | smp_mb(); | |
1443 | rdev = conf->mirrors[d].rdev; | |
1444 | } | |
1445 | rdev_dec_pending(rdev, mddev); | |
1446 | } | |
1447 | } | |
1448 | allow_barrier(conf); | |
1449 | raid10_log(conf->mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); | |
1450 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1451 | wait_barrier(conf); | |
1452 | goto retry_write; | |
1453 | } | |
1454 | ||
6b6c8110 | 1455 | if (max_sectors < r10_bio->sectors) |
d4432c23 | 1456 | r10_bio->sectors = max_sectors; |
fc9977dd N |
1457 | |
1458 | if (r10_bio->sectors < bio_sectors(bio)) { | |
1459 | struct bio *split = bio_split(bio, r10_bio->sectors, | |
afeee514 | 1460 | GFP_NOIO, &conf->bio_split); |
fc9977dd | 1461 | bio_chain(split, bio); |
e820d55c | 1462 | allow_barrier(conf); |
ed00aabd | 1463 | submit_bio_noacct(bio); |
e820d55c | 1464 | wait_barrier(conf); |
fc9977dd N |
1465 | bio = split; |
1466 | r10_bio->master_bio = bio; | |
d4432c23 | 1467 | } |
d4432c23 | 1468 | |
4e78064f | 1469 | atomic_set(&r10_bio->remaining, 1); |
e64e4018 | 1470 | md_bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1471 | |
1da177e4 | 1472 | for (i = 0; i < conf->copies; i++) { |
27f26a0f | 1473 | if (r10_bio->devs[i].bio) |
fc9977dd | 1474 | raid10_write_one_disk(mddev, r10_bio, bio, false, i); |
27f26a0f | 1475 | if (r10_bio->devs[i].repl_bio) |
fc9977dd | 1476 | raid10_write_one_disk(mddev, r10_bio, bio, true, i); |
d4432c23 | 1477 | } |
079fa166 | 1478 | one_write_done(r10_bio); |
20d0189b KO |
1479 | } |
1480 | ||
fc9977dd | 1481 | static void __make_request(struct mddev *mddev, struct bio *bio, int sectors) |
bb5f1ed7 RL |
1482 | { |
1483 | struct r10conf *conf = mddev->private; | |
1484 | struct r10bio *r10_bio; | |
1485 | ||
afeee514 | 1486 | r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); |
bb5f1ed7 RL |
1487 | |
1488 | r10_bio->master_bio = bio; | |
fc9977dd | 1489 | r10_bio->sectors = sectors; |
bb5f1ed7 RL |
1490 | |
1491 | r10_bio->mddev = mddev; | |
1492 | r10_bio->sector = bio->bi_iter.bi_sector; | |
1493 | r10_bio->state = 0; | |
93decc56 | 1494 | r10_bio->read_slot = -1; |
17c28c2a | 1495 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * conf->copies); |
bb5f1ed7 RL |
1496 | |
1497 | if (bio_data_dir(bio) == READ) | |
1498 | raid10_read_request(mddev, bio, r10_bio); | |
1499 | else | |
1500 | raid10_write_request(mddev, bio, r10_bio); | |
1501 | } | |
1502 | ||
cc27b0c7 | 1503 | static bool raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1504 | { |
1505 | struct r10conf *conf = mddev->private; | |
1506 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1507 | int chunk_sects = chunk_mask + 1; | |
fc9977dd | 1508 | int sectors = bio_sectors(bio); |
20d0189b | 1509 | |
775d7831 DJ |
1510 | if (unlikely(bio->bi_opf & REQ_PREFLUSH) |
1511 | && md_flush_request(mddev, bio)) | |
cc27b0c7 | 1512 | return true; |
20d0189b | 1513 | |
cc27b0c7 N |
1514 | if (!md_write_start(mddev, bio)) |
1515 | return false; | |
1516 | ||
fc9977dd N |
1517 | /* |
1518 | * If this request crosses a chunk boundary, we need to split | |
1519 | * it. | |
1520 | */ | |
1521 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1522 | sectors > chunk_sects | |
1523 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1524 | || conf->prev.near_copies < | |
1525 | conf->prev.raid_disks))) | |
1526 | sectors = chunk_sects - | |
1527 | (bio->bi_iter.bi_sector & | |
1528 | (chunk_sects - 1)); | |
1529 | __make_request(mddev, bio, sectors); | |
079fa166 N |
1530 | |
1531 | /* In case raid10d snuck in to freeze_array */ | |
1532 | wake_up(&conf->wait_barrier); | |
cc27b0c7 | 1533 | return true; |
1da177e4 LT |
1534 | } |
1535 | ||
849674e4 | 1536 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1537 | { |
e879a879 | 1538 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1539 | int i; |
1540 | ||
5cf00fcd | 1541 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1542 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1543 | if (conf->geo.near_copies > 1) |
1544 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1545 | if (conf->geo.far_copies > 1) { | |
1546 | if (conf->geo.far_offset) | |
1547 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1548 | else |
5cf00fcd | 1549 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1550 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1551 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1552 | } |
5cf00fcd N |
1553 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1554 | conf->geo.raid_disks - mddev->degraded); | |
d44b0a92 N |
1555 | rcu_read_lock(); |
1556 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
1557 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1558 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
1559 | } | |
1560 | rcu_read_unlock(); | |
1da177e4 LT |
1561 | seq_printf(seq, "]"); |
1562 | } | |
1563 | ||
700c7213 N |
1564 | /* check if there are enough drives for |
1565 | * every block to appear on atleast one. | |
1566 | * Don't consider the device numbered 'ignore' | |
1567 | * as we might be about to remove it. | |
1568 | */ | |
635f6416 | 1569 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1570 | { |
1571 | int first = 0; | |
725d6e57 | 1572 | int has_enough = 0; |
635f6416 N |
1573 | int disks, ncopies; |
1574 | if (previous) { | |
1575 | disks = conf->prev.raid_disks; | |
1576 | ncopies = conf->prev.near_copies; | |
1577 | } else { | |
1578 | disks = conf->geo.raid_disks; | |
1579 | ncopies = conf->geo.near_copies; | |
1580 | } | |
700c7213 | 1581 | |
725d6e57 | 1582 | rcu_read_lock(); |
700c7213 N |
1583 | do { |
1584 | int n = conf->copies; | |
1585 | int cnt = 0; | |
80b48124 | 1586 | int this = first; |
700c7213 | 1587 | while (n--) { |
725d6e57 N |
1588 | struct md_rdev *rdev; |
1589 | if (this != ignore && | |
1590 | (rdev = rcu_dereference(conf->mirrors[this].rdev)) && | |
1591 | test_bit(In_sync, &rdev->flags)) | |
700c7213 | 1592 | cnt++; |
635f6416 | 1593 | this = (this+1) % disks; |
700c7213 N |
1594 | } |
1595 | if (cnt == 0) | |
725d6e57 | 1596 | goto out; |
635f6416 | 1597 | first = (first + ncopies) % disks; |
700c7213 | 1598 | } while (first != 0); |
725d6e57 N |
1599 | has_enough = 1; |
1600 | out: | |
1601 | rcu_read_unlock(); | |
1602 | return has_enough; | |
700c7213 N |
1603 | } |
1604 | ||
f8c9e74f N |
1605 | static int enough(struct r10conf *conf, int ignore) |
1606 | { | |
635f6416 N |
1607 | /* when calling 'enough', both 'prev' and 'geo' must |
1608 | * be stable. | |
1609 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1610 | * is held. | |
1611 | */ | |
1612 | return _enough(conf, 0, ignore) && | |
1613 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1614 | } |
1615 | ||
849674e4 | 1616 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1617 | { |
1618 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1619 | struct r10conf *conf = mddev->private; |
635f6416 | 1620 | unsigned long flags; |
1da177e4 LT |
1621 | |
1622 | /* | |
1623 | * If it is not operational, then we have already marked it as dead | |
9a567843 GJ |
1624 | * else if it is the last working disks with "fail_last_dev == false", |
1625 | * ignore the error, let the next level up know. | |
1da177e4 LT |
1626 | * else mark the drive as failed |
1627 | */ | |
635f6416 | 1628 | spin_lock_irqsave(&conf->device_lock, flags); |
9a567843 | 1629 | if (test_bit(In_sync, &rdev->flags) && !mddev->fail_last_dev |
635f6416 | 1630 | && !enough(conf, rdev->raid_disk)) { |
1da177e4 LT |
1631 | /* |
1632 | * Don't fail the drive, just return an IO error. | |
1da177e4 | 1633 | */ |
635f6416 | 1634 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1635 | return; |
635f6416 | 1636 | } |
2446dba0 | 1637 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1638 | mddev->degraded++; |
2446dba0 N |
1639 | /* |
1640 | * If recovery is running, make sure it aborts. | |
1641 | */ | |
1642 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
de393cde | 1643 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1644 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
1645 | set_mask_bits(&mddev->sb_flags, 0, |
1646 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
635f6416 | 1647 | spin_unlock_irqrestore(&conf->device_lock, flags); |
08464e09 N |
1648 | pr_crit("md/raid10:%s: Disk failure on %s, disabling device.\n" |
1649 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
1650 | mdname(mddev), bdevname(rdev->bdev, b), | |
1651 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); | |
1da177e4 LT |
1652 | } |
1653 | ||
e879a879 | 1654 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1655 | { |
1656 | int i; | |
4056ca51 | 1657 | struct md_rdev *rdev; |
1da177e4 | 1658 | |
08464e09 | 1659 | pr_debug("RAID10 conf printout:\n"); |
1da177e4 | 1660 | if (!conf) { |
08464e09 | 1661 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1662 | return; |
1663 | } | |
08464e09 N |
1664 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1665 | conf->geo.raid_disks); | |
1da177e4 | 1666 | |
4056ca51 N |
1667 | /* This is only called with ->reconfix_mutex held, so |
1668 | * rcu protection of rdev is not needed */ | |
5cf00fcd | 1669 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1670 | char b[BDEVNAME_SIZE]; |
4056ca51 N |
1671 | rdev = conf->mirrors[i].rdev; |
1672 | if (rdev) | |
08464e09 N |
1673 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1674 | i, !test_bit(In_sync, &rdev->flags), | |
1675 | !test_bit(Faulty, &rdev->flags), | |
1676 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1677 | } |
1678 | } | |
1679 | ||
e879a879 | 1680 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1681 | { |
0a27ec96 N |
1682 | wait_barrier(conf); |
1683 | allow_barrier(conf); | |
1da177e4 | 1684 | |
afeee514 | 1685 | mempool_exit(&conf->r10buf_pool); |
1da177e4 LT |
1686 | } |
1687 | ||
fd01b88c | 1688 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1689 | { |
1690 | int i; | |
e879a879 | 1691 | struct r10conf *conf = mddev->private; |
dc280d98 | 1692 | struct raid10_info *tmp; |
6b965620 N |
1693 | int count = 0; |
1694 | unsigned long flags; | |
1da177e4 LT |
1695 | |
1696 | /* | |
1697 | * Find all non-in_sync disks within the RAID10 configuration | |
1698 | * and mark them in_sync | |
1699 | */ | |
5cf00fcd | 1700 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1701 | tmp = conf->mirrors + i; |
4ca40c2c N |
1702 | if (tmp->replacement |
1703 | && tmp->replacement->recovery_offset == MaxSector | |
1704 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1705 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1706 | /* Replacement has just become active */ | |
1707 | if (!tmp->rdev | |
1708 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1709 | count++; | |
1710 | if (tmp->rdev) { | |
1711 | /* Replaced device not technically faulty, | |
1712 | * but we need to be sure it gets removed | |
1713 | * and never re-added. | |
1714 | */ | |
1715 | set_bit(Faulty, &tmp->rdev->flags); | |
1716 | sysfs_notify_dirent_safe( | |
1717 | tmp->rdev->sysfs_state); | |
1718 | } | |
1719 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1720 | } else if (tmp->rdev | |
61e4947c | 1721 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
1722 | && !test_bit(Faulty, &tmp->rdev->flags) |
1723 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1724 | count++; |
2863b9eb | 1725 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
1726 | } |
1727 | } | |
6b965620 N |
1728 | spin_lock_irqsave(&conf->device_lock, flags); |
1729 | mddev->degraded -= count; | |
1730 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1731 | |
1732 | print_conf(conf); | |
6b965620 | 1733 | return count; |
1da177e4 LT |
1734 | } |
1735 | ||
fd01b88c | 1736 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1737 | { |
e879a879 | 1738 | struct r10conf *conf = mddev->private; |
199050ea | 1739 | int err = -EEXIST; |
1da177e4 | 1740 | int mirror; |
6c2fce2e | 1741 | int first = 0; |
5cf00fcd | 1742 | int last = conf->geo.raid_disks - 1; |
1da177e4 LT |
1743 | |
1744 | if (mddev->recovery_cp < MaxSector) | |
1745 | /* only hot-add to in-sync arrays, as recovery is | |
1746 | * very different from resync | |
1747 | */ | |
199050ea | 1748 | return -EBUSY; |
635f6416 | 1749 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 1750 | return -EINVAL; |
1da177e4 | 1751 | |
1501efad DW |
1752 | if (md_integrity_add_rdev(rdev, mddev)) |
1753 | return -ENXIO; | |
1754 | ||
a53a6c85 | 1755 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1756 | first = last = rdev->raid_disk; |
1da177e4 | 1757 | |
2c4193df | 1758 | if (rdev->saved_raid_disk >= first && |
9e753ba9 | 1759 | rdev->saved_raid_disk < conf->geo.raid_disks && |
6cce3b23 N |
1760 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1761 | mirror = rdev->saved_raid_disk; | |
1762 | else | |
6c2fce2e | 1763 | mirror = first; |
2bb77736 | 1764 | for ( ; mirror <= last ; mirror++) { |
dc280d98 | 1765 | struct raid10_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1766 | if (p->recovery_disabled == mddev->recovery_disabled) |
1767 | continue; | |
b7044d41 N |
1768 | if (p->rdev) { |
1769 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1770 | p->replacement != NULL) | |
1771 | continue; | |
1772 | clear_bit(In_sync, &rdev->flags); | |
1773 | set_bit(Replacement, &rdev->flags); | |
1774 | rdev->raid_disk = mirror; | |
1775 | err = 0; | |
9092c02d JB |
1776 | if (mddev->gendisk) |
1777 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1778 | rdev->data_offset << 9); | |
b7044d41 N |
1779 | conf->fullsync = 1; |
1780 | rcu_assign_pointer(p->replacement, rdev); | |
1781 | break; | |
1782 | } | |
1da177e4 | 1783 | |
9092c02d JB |
1784 | if (mddev->gendisk) |
1785 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1786 | rdev->data_offset << 9); | |
1da177e4 | 1787 | |
2bb77736 | 1788 | p->head_position = 0; |
d890fa2b | 1789 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1790 | rdev->raid_disk = mirror; |
1791 | err = 0; | |
1792 | if (rdev->saved_raid_disk != mirror) | |
1793 | conf->fullsync = 1; | |
1794 | rcu_assign_pointer(p->rdev, rdev); | |
1795 | break; | |
1796 | } | |
ed30be07 | 1797 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
8b904b5b | 1798 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue); |
532a2a3f | 1799 | |
1da177e4 | 1800 | print_conf(conf); |
199050ea | 1801 | return err; |
1da177e4 LT |
1802 | } |
1803 | ||
b8321b68 | 1804 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1805 | { |
e879a879 | 1806 | struct r10conf *conf = mddev->private; |
1da177e4 | 1807 | int err = 0; |
b8321b68 | 1808 | int number = rdev->raid_disk; |
c8ab903e | 1809 | struct md_rdev **rdevp; |
dc280d98 | 1810 | struct raid10_info *p = conf->mirrors + number; |
1da177e4 LT |
1811 | |
1812 | print_conf(conf); | |
c8ab903e N |
1813 | if (rdev == p->rdev) |
1814 | rdevp = &p->rdev; | |
1815 | else if (rdev == p->replacement) | |
1816 | rdevp = &p->replacement; | |
1817 | else | |
1818 | return 0; | |
1819 | ||
1820 | if (test_bit(In_sync, &rdev->flags) || | |
1821 | atomic_read(&rdev->nr_pending)) { | |
1822 | err = -EBUSY; | |
1823 | goto abort; | |
1824 | } | |
d787be40 | 1825 | /* Only remove non-faulty devices if recovery |
c8ab903e N |
1826 | * is not possible. |
1827 | */ | |
1828 | if (!test_bit(Faulty, &rdev->flags) && | |
1829 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1830 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 1831 | number < conf->geo.raid_disks && |
c8ab903e N |
1832 | enough(conf, -1)) { |
1833 | err = -EBUSY; | |
1834 | goto abort; | |
1da177e4 | 1835 | } |
c8ab903e | 1836 | *rdevp = NULL; |
d787be40 N |
1837 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1838 | synchronize_rcu(); | |
1839 | if (atomic_read(&rdev->nr_pending)) { | |
1840 | /* lost the race, try later */ | |
1841 | err = -EBUSY; | |
1842 | *rdevp = rdev; | |
1843 | goto abort; | |
1844 | } | |
1845 | } | |
1846 | if (p->replacement) { | |
4ca40c2c N |
1847 | /* We must have just cleared 'rdev' */ |
1848 | p->rdev = p->replacement; | |
1849 | clear_bit(Replacement, &p->replacement->flags); | |
1850 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1851 | * but will never see neither -- if they are careful. | |
1852 | */ | |
1853 | p->replacement = NULL; | |
e5bc9c3c | 1854 | } |
4ca40c2c | 1855 | |
e5bc9c3c | 1856 | clear_bit(WantReplacement, &rdev->flags); |
c8ab903e N |
1857 | err = md_integrity_register(mddev); |
1858 | ||
1da177e4 LT |
1859 | abort: |
1860 | ||
1861 | print_conf(conf); | |
1862 | return err; | |
1863 | } | |
1864 | ||
81fa1520 | 1865 | static void __end_sync_read(struct r10bio *r10_bio, struct bio *bio, int d) |
1da177e4 | 1866 | { |
e879a879 | 1867 | struct r10conf *conf = r10_bio->mddev->private; |
0eb3ff12 | 1868 | |
4e4cbee9 | 1869 | if (!bio->bi_status) |
0eb3ff12 | 1870 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
1871 | else |
1872 | /* The write handler will notice the lack of | |
1873 | * R10BIO_Uptodate and record any errors etc | |
1874 | */ | |
4dbcdc75 N |
1875 | atomic_add(r10_bio->sectors, |
1876 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1877 | |
1878 | /* for reconstruct, we always reschedule after a read. | |
1879 | * for resync, only after all reads | |
1880 | */ | |
73d5c38a | 1881 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1882 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1883 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1884 | /* we have read all the blocks, | |
1885 | * do the comparison in process context in raid10d | |
1886 | */ | |
1887 | reschedule_retry(r10_bio); | |
1888 | } | |
1da177e4 LT |
1889 | } |
1890 | ||
81fa1520 ML |
1891 | static void end_sync_read(struct bio *bio) |
1892 | { | |
f0250618 | 1893 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
81fa1520 ML |
1894 | struct r10conf *conf = r10_bio->mddev->private; |
1895 | int d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
1896 | ||
1897 | __end_sync_read(r10_bio, bio, d); | |
1898 | } | |
1899 | ||
1900 | static void end_reshape_read(struct bio *bio) | |
1901 | { | |
f0250618 | 1902 | /* reshape read bio isn't allocated from r10buf_pool */ |
81fa1520 ML |
1903 | struct r10bio *r10_bio = bio->bi_private; |
1904 | ||
1905 | __end_sync_read(r10_bio, bio, r10_bio->read_slot); | |
1906 | } | |
1907 | ||
9f2c9d12 | 1908 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1909 | { |
fd01b88c | 1910 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1911 | |
1da177e4 LT |
1912 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1913 | if (r10_bio->master_bio == NULL) { | |
1914 | /* the primary of several recovery bios */ | |
73d5c38a | 1915 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1916 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1917 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1918 | reschedule_retry(r10_bio); |
1919 | else | |
1920 | put_buf(r10_bio); | |
73d5c38a | 1921 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1922 | break; |
1923 | } else { | |
9f2c9d12 | 1924 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1925 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1926 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1927 | reschedule_retry(r10_bio); |
1928 | else | |
1929 | put_buf(r10_bio); | |
1da177e4 LT |
1930 | r10_bio = r10_bio2; |
1931 | } | |
1932 | } | |
1da177e4 LT |
1933 | } |
1934 | ||
4246a0b6 | 1935 | static void end_sync_write(struct bio *bio) |
5e570289 | 1936 | { |
f0250618 | 1937 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
fd01b88c | 1938 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1939 | struct r10conf *conf = mddev->private; |
5e570289 N |
1940 | int d; |
1941 | sector_t first_bad; | |
1942 | int bad_sectors; | |
1943 | int slot; | |
9ad1aefc | 1944 | int repl; |
4ca40c2c | 1945 | struct md_rdev *rdev = NULL; |
5e570289 | 1946 | |
9ad1aefc N |
1947 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1948 | if (repl) | |
1949 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 1950 | else |
9ad1aefc | 1951 | rdev = conf->mirrors[d].rdev; |
5e570289 | 1952 | |
4e4cbee9 | 1953 | if (bio->bi_status) { |
9ad1aefc N |
1954 | if (repl) |
1955 | md_error(mddev, rdev); | |
1956 | else { | |
1957 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
1958 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1959 | set_bit(MD_RECOVERY_NEEDED, | |
1960 | &rdev->mddev->recovery); | |
9ad1aefc N |
1961 | set_bit(R10BIO_WriteError, &r10_bio->state); |
1962 | } | |
1963 | } else if (is_badblock(rdev, | |
5e570289 N |
1964 | r10_bio->devs[slot].addr, |
1965 | r10_bio->sectors, | |
1966 | &first_bad, &bad_sectors)) | |
1967 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
1968 | ||
9ad1aefc | 1969 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
1970 | |
1971 | end_sync_request(r10_bio); | |
1972 | } | |
1973 | ||
1da177e4 LT |
1974 | /* |
1975 | * Note: sync and recover and handled very differently for raid10 | |
1976 | * This code is for resync. | |
1977 | * For resync, we read through virtual addresses and read all blocks. | |
1978 | * If there is any error, we schedule a write. The lowest numbered | |
1979 | * drive is authoritative. | |
1980 | * However requests come for physical address, so we need to map. | |
1981 | * For every physical address there are raid_disks/copies virtual addresses, | |
1982 | * which is always are least one, but is not necessarly an integer. | |
1983 | * This means that a physical address can span multiple chunks, so we may | |
1984 | * have to submit multiple io requests for a single sync request. | |
1985 | */ | |
1986 | /* | |
1987 | * We check if all blocks are in-sync and only write to blocks that | |
1988 | * aren't in sync | |
1989 | */ | |
9f2c9d12 | 1990 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1991 | { |
e879a879 | 1992 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1993 | int i, first; |
1994 | struct bio *tbio, *fbio; | |
f4380a91 | 1995 | int vcnt; |
cdb76be3 | 1996 | struct page **tpages, **fpages; |
1da177e4 LT |
1997 | |
1998 | atomic_set(&r10_bio->remaining, 1); | |
1999 | ||
2000 | /* find the first device with a block */ | |
2001 | for (i=0; i<conf->copies; i++) | |
4e4cbee9 | 2002 | if (!r10_bio->devs[i].bio->bi_status) |
1da177e4 LT |
2003 | break; |
2004 | ||
2005 | if (i == conf->copies) | |
2006 | goto done; | |
2007 | ||
2008 | first = i; | |
2009 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
2010 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
2011 | fbio->bi_iter.bi_idx = 0; | |
cdb76be3 | 2012 | fpages = get_resync_pages(fbio)->pages; |
1da177e4 | 2013 | |
f4380a91 | 2014 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 2015 | /* now find blocks with errors */ |
0eb3ff12 N |
2016 | for (i=0 ; i < conf->copies ; i++) { |
2017 | int j, d; | |
8d3ca83d | 2018 | struct md_rdev *rdev; |
f0250618 | 2019 | struct resync_pages *rp; |
1da177e4 | 2020 | |
1da177e4 | 2021 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
2022 | |
2023 | if (tbio->bi_end_io != end_sync_read) | |
2024 | continue; | |
2025 | if (i == first) | |
1da177e4 | 2026 | continue; |
cdb76be3 ML |
2027 | |
2028 | tpages = get_resync_pages(tbio)->pages; | |
8d3ca83d N |
2029 | d = r10_bio->devs[i].devnum; |
2030 | rdev = conf->mirrors[d].rdev; | |
4e4cbee9 | 2031 | if (!r10_bio->devs[i].bio->bi_status) { |
0eb3ff12 N |
2032 | /* We know that the bi_io_vec layout is the same for |
2033 | * both 'first' and 'i', so we just compare them. | |
2034 | * All vec entries are PAGE_SIZE; | |
2035 | */ | |
7bb23c49 N |
2036 | int sectors = r10_bio->sectors; |
2037 | for (j = 0; j < vcnt; j++) { | |
2038 | int len = PAGE_SIZE; | |
2039 | if (sectors < (len / 512)) | |
2040 | len = sectors * 512; | |
cdb76be3 ML |
2041 | if (memcmp(page_address(fpages[j]), |
2042 | page_address(tpages[j]), | |
7bb23c49 | 2043 | len)) |
0eb3ff12 | 2044 | break; |
7bb23c49 N |
2045 | sectors -= len/512; |
2046 | } | |
0eb3ff12 N |
2047 | if (j == vcnt) |
2048 | continue; | |
7f7583d4 | 2049 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
2050 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
2051 | /* Don't fix anything. */ | |
2052 | continue; | |
8d3ca83d N |
2053 | } else if (test_bit(FailFast, &rdev->flags)) { |
2054 | /* Just give up on this device */ | |
2055 | md_error(rdev->mddev, rdev); | |
2056 | continue; | |
0eb3ff12 | 2057 | } |
f84ee364 N |
2058 | /* Ok, we need to write this bio, either to correct an |
2059 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
2060 | * First we need to fixup bv_offset, bv_len and |
2061 | * bi_vecs, as the read request might have corrupted these | |
2062 | */ | |
f0250618 | 2063 | rp = get_resync_pages(tbio); |
8be185f2 KO |
2064 | bio_reset(tbio); |
2065 | ||
fb0eb5df ML |
2066 | md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size); |
2067 | ||
f0250618 ML |
2068 | rp->raid_bio = r10_bio; |
2069 | tbio->bi_private = rp; | |
4f024f37 | 2070 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 | 2071 | tbio->bi_end_io = end_sync_write; |
796a5cf0 | 2072 | bio_set_op_attrs(tbio, REQ_OP_WRITE, 0); |
1da177e4 | 2073 | |
c31df25f KO |
2074 | bio_copy_data(tbio, fbio); |
2075 | ||
1da177e4 LT |
2076 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2077 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2078 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2079 | |
1919cbb2 N |
2080 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
2081 | tbio->bi_opf |= MD_FAILFAST; | |
4f024f37 | 2082 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
74d46992 | 2083 | bio_set_dev(tbio, conf->mirrors[d].rdev->bdev); |
ed00aabd | 2084 | submit_bio_noacct(tbio); |
1da177e4 LT |
2085 | } |
2086 | ||
9ad1aefc N |
2087 | /* Now write out to any replacement devices |
2088 | * that are active | |
2089 | */ | |
2090 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2091 | int d; |
9ad1aefc N |
2092 | |
2093 | tbio = r10_bio->devs[i].repl_bio; | |
2094 | if (!tbio || !tbio->bi_end_io) | |
2095 | continue; | |
2096 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2097 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2098 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2099 | d = r10_bio->devs[i].devnum; |
2100 | atomic_inc(&r10_bio->remaining); | |
2101 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2102 | bio_sectors(tbio)); |
ed00aabd | 2103 | submit_bio_noacct(tbio); |
9ad1aefc N |
2104 | } |
2105 | ||
1da177e4 LT |
2106 | done: |
2107 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2108 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2109 | put_buf(r10_bio); | |
2110 | } | |
2111 | } | |
2112 | ||
2113 | /* | |
2114 | * Now for the recovery code. | |
2115 | * Recovery happens across physical sectors. | |
2116 | * We recover all non-is_sync drives by finding the virtual address of | |
2117 | * each, and then choose a working drive that also has that virt address. | |
2118 | * There is a separate r10_bio for each non-in_sync drive. | |
2119 | * Only the first two slots are in use. The first for reading, | |
2120 | * The second for writing. | |
2121 | * | |
2122 | */ | |
9f2c9d12 | 2123 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2124 | { |
2125 | /* We got a read error during recovery. | |
2126 | * We repeat the read in smaller page-sized sections. | |
2127 | * If a read succeeds, write it to the new device or record | |
2128 | * a bad block if we cannot. | |
2129 | * If a read fails, record a bad block on both old and | |
2130 | * new devices. | |
2131 | */ | |
fd01b88c | 2132 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2133 | struct r10conf *conf = mddev->private; |
5e570289 N |
2134 | struct bio *bio = r10_bio->devs[0].bio; |
2135 | sector_t sect = 0; | |
2136 | int sectors = r10_bio->sectors; | |
2137 | int idx = 0; | |
2138 | int dr = r10_bio->devs[0].devnum; | |
2139 | int dw = r10_bio->devs[1].devnum; | |
cdb76be3 | 2140 | struct page **pages = get_resync_pages(bio)->pages; |
5e570289 N |
2141 | |
2142 | while (sectors) { | |
2143 | int s = sectors; | |
3cb03002 | 2144 | struct md_rdev *rdev; |
5e570289 N |
2145 | sector_t addr; |
2146 | int ok; | |
2147 | ||
2148 | if (s > (PAGE_SIZE>>9)) | |
2149 | s = PAGE_SIZE >> 9; | |
2150 | ||
2151 | rdev = conf->mirrors[dr].rdev; | |
2152 | addr = r10_bio->devs[0].addr + sect, | |
2153 | ok = sync_page_io(rdev, | |
2154 | addr, | |
2155 | s << 9, | |
cdb76be3 | 2156 | pages[idx], |
796a5cf0 | 2157 | REQ_OP_READ, 0, false); |
5e570289 N |
2158 | if (ok) { |
2159 | rdev = conf->mirrors[dw].rdev; | |
2160 | addr = r10_bio->devs[1].addr + sect; | |
2161 | ok = sync_page_io(rdev, | |
2162 | addr, | |
2163 | s << 9, | |
cdb76be3 | 2164 | pages[idx], |
796a5cf0 | 2165 | REQ_OP_WRITE, 0, false); |
b7044d41 | 2166 | if (!ok) { |
5e570289 | 2167 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2168 | if (!test_and_set_bit(WantReplacement, |
2169 | &rdev->flags)) | |
2170 | set_bit(MD_RECOVERY_NEEDED, | |
2171 | &rdev->mddev->recovery); | |
2172 | } | |
5e570289 N |
2173 | } |
2174 | if (!ok) { | |
2175 | /* We don't worry if we cannot set a bad block - | |
2176 | * it really is bad so there is no loss in not | |
2177 | * recording it yet | |
2178 | */ | |
2179 | rdev_set_badblocks(rdev, addr, s, 0); | |
2180 | ||
2181 | if (rdev != conf->mirrors[dw].rdev) { | |
2182 | /* need bad block on destination too */ | |
3cb03002 | 2183 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2184 | addr = r10_bio->devs[1].addr + sect; |
2185 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2186 | if (!ok) { | |
2187 | /* just abort the recovery */ | |
08464e09 N |
2188 | pr_notice("md/raid10:%s: recovery aborted due to read error\n", |
2189 | mdname(mddev)); | |
5e570289 N |
2190 | |
2191 | conf->mirrors[dw].recovery_disabled | |
2192 | = mddev->recovery_disabled; | |
2193 | set_bit(MD_RECOVERY_INTR, | |
2194 | &mddev->recovery); | |
2195 | break; | |
2196 | } | |
2197 | } | |
2198 | } | |
2199 | ||
2200 | sectors -= s; | |
2201 | sect += s; | |
2202 | idx++; | |
2203 | } | |
2204 | } | |
1da177e4 | 2205 | |
9f2c9d12 | 2206 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2207 | { |
e879a879 | 2208 | struct r10conf *conf = mddev->private; |
c65060ad | 2209 | int d; |
24afd80d | 2210 | struct bio *wbio, *wbio2; |
1da177e4 | 2211 | |
5e570289 N |
2212 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2213 | fix_recovery_read_error(r10_bio); | |
2214 | end_sync_request(r10_bio); | |
2215 | return; | |
2216 | } | |
2217 | ||
c65060ad NK |
2218 | /* |
2219 | * share the pages with the first bio | |
1da177e4 LT |
2220 | * and submit the write request |
2221 | */ | |
1da177e4 | 2222 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2223 | wbio = r10_bio->devs[1].bio; |
2224 | wbio2 = r10_bio->devs[1].repl_bio; | |
0eb25bb0 | 2225 | /* Need to test wbio2->bi_end_io before we call |
ed00aabd | 2226 | * submit_bio_noacct as if the former is NULL, |
0eb25bb0 N |
2227 | * the latter is free to free wbio2. |
2228 | */ | |
2229 | if (wbio2 && !wbio2->bi_end_io) | |
2230 | wbio2 = NULL; | |
24afd80d N |
2231 | if (wbio->bi_end_io) { |
2232 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2233 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
ed00aabd | 2234 | submit_bio_noacct(wbio); |
24afd80d | 2235 | } |
0eb25bb0 | 2236 | if (wbio2) { |
24afd80d N |
2237 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2238 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2239 | bio_sectors(wbio2)); |
ed00aabd | 2240 | submit_bio_noacct(wbio2); |
24afd80d | 2241 | } |
1da177e4 LT |
2242 | } |
2243 | ||
1e50915f RB |
2244 | /* |
2245 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2246 | * We halve the read error count for every hour that has elapsed | |
2247 | * since the last recorded read error. | |
2248 | * | |
2249 | */ | |
fd01b88c | 2250 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f | 2251 | { |
0e3ef49e | 2252 | long cur_time_mon; |
1e50915f RB |
2253 | unsigned long hours_since_last; |
2254 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2255 | ||
0e3ef49e | 2256 | cur_time_mon = ktime_get_seconds(); |
1e50915f | 2257 | |
0e3ef49e | 2258 | if (rdev->last_read_error == 0) { |
1e50915f RB |
2259 | /* first time we've seen a read error */ |
2260 | rdev->last_read_error = cur_time_mon; | |
2261 | return; | |
2262 | } | |
2263 | ||
0e3ef49e AB |
2264 | hours_since_last = (long)(cur_time_mon - |
2265 | rdev->last_read_error) / 3600; | |
1e50915f RB |
2266 | |
2267 | rdev->last_read_error = cur_time_mon; | |
2268 | ||
2269 | /* | |
2270 | * if hours_since_last is > the number of bits in read_errors | |
2271 | * just set read errors to 0. We do this to avoid | |
2272 | * overflowing the shift of read_errors by hours_since_last. | |
2273 | */ | |
2274 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2275 | atomic_set(&rdev->read_errors, 0); | |
2276 | else | |
2277 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2278 | } | |
2279 | ||
3cb03002 | 2280 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2281 | int sectors, struct page *page, int rw) |
2282 | { | |
2283 | sector_t first_bad; | |
2284 | int bad_sectors; | |
2285 | ||
2286 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2287 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2288 | return -1; | |
796a5cf0 | 2289 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
58c54fcc N |
2290 | /* success */ |
2291 | return 1; | |
b7044d41 | 2292 | if (rw == WRITE) { |
58c54fcc | 2293 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2294 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2295 | set_bit(MD_RECOVERY_NEEDED, | |
2296 | &rdev->mddev->recovery); | |
2297 | } | |
58c54fcc N |
2298 | /* need to record an error - either for the block or the device */ |
2299 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2300 | md_error(rdev->mddev, rdev); | |
2301 | return 0; | |
2302 | } | |
2303 | ||
1da177e4 LT |
2304 | /* |
2305 | * This is a kernel thread which: | |
2306 | * | |
2307 | * 1. Retries failed read operations on working mirrors. | |
2308 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2309 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2310 | */ |
2311 | ||
e879a879 | 2312 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2313 | { |
2314 | int sect = 0; /* Offset from r10_bio->sector */ | |
2315 | int sectors = r10_bio->sectors; | |
13db16d7 | 2316 | struct md_rdev *rdev; |
1e50915f | 2317 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2318 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2319 | |
7c4e06ff N |
2320 | /* still own a reference to this rdev, so it cannot |
2321 | * have been cleared recently. | |
2322 | */ | |
2323 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2324 | |
7c4e06ff N |
2325 | if (test_bit(Faulty, &rdev->flags)) |
2326 | /* drive has already been failed, just ignore any | |
2327 | more fix_read_error() attempts */ | |
2328 | return; | |
1e50915f | 2329 | |
7c4e06ff N |
2330 | check_decay_read_errors(mddev, rdev); |
2331 | atomic_inc(&rdev->read_errors); | |
2332 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2333 | char b[BDEVNAME_SIZE]; | |
2334 | bdevname(rdev->bdev, b); | |
1e50915f | 2335 | |
08464e09 N |
2336 | pr_notice("md/raid10:%s: %s: Raid device exceeded read_error threshold [cur %d:max %d]\n", |
2337 | mdname(mddev), b, | |
2338 | atomic_read(&rdev->read_errors), max_read_errors); | |
2339 | pr_notice("md/raid10:%s: %s: Failing raid device\n", | |
2340 | mdname(mddev), b); | |
d683c8e0 | 2341 | md_error(mddev, rdev); |
fae8cc5e | 2342 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2343 | return; |
1e50915f | 2344 | } |
1e50915f | 2345 | |
6814d536 N |
2346 | while(sectors) { |
2347 | int s = sectors; | |
2348 | int sl = r10_bio->read_slot; | |
2349 | int success = 0; | |
2350 | int start; | |
2351 | ||
2352 | if (s > (PAGE_SIZE>>9)) | |
2353 | s = PAGE_SIZE >> 9; | |
2354 | ||
2355 | rcu_read_lock(); | |
2356 | do { | |
8dbed5ce N |
2357 | sector_t first_bad; |
2358 | int bad_sectors; | |
2359 | ||
0544a21d | 2360 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2361 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2362 | if (rdev && | |
8dbed5ce | 2363 | test_bit(In_sync, &rdev->flags) && |
f5b67ae8 | 2364 | !test_bit(Faulty, &rdev->flags) && |
8dbed5ce N |
2365 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, |
2366 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2367 | atomic_inc(&rdev->nr_pending); |
2368 | rcu_read_unlock(); | |
2b193363 | 2369 | success = sync_page_io(rdev, |
6814d536 | 2370 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2371 | sect, |
6814d536 | 2372 | s<<9, |
796a5cf0 MC |
2373 | conf->tmppage, |
2374 | REQ_OP_READ, 0, false); | |
6814d536 N |
2375 | rdev_dec_pending(rdev, mddev); |
2376 | rcu_read_lock(); | |
2377 | if (success) | |
2378 | break; | |
2379 | } | |
2380 | sl++; | |
2381 | if (sl == conf->copies) | |
2382 | sl = 0; | |
2383 | } while (!success && sl != r10_bio->read_slot); | |
2384 | rcu_read_unlock(); | |
2385 | ||
2386 | if (!success) { | |
58c54fcc N |
2387 | /* Cannot read from anywhere, just mark the block |
2388 | * as bad on the first device to discourage future | |
2389 | * reads. | |
2390 | */ | |
6814d536 | 2391 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2392 | rdev = conf->mirrors[dn].rdev; |
2393 | ||
2394 | if (!rdev_set_badblocks( | |
2395 | rdev, | |
2396 | r10_bio->devs[r10_bio->read_slot].addr | |
2397 | + sect, | |
fae8cc5e | 2398 | s, 0)) { |
58c54fcc | 2399 | md_error(mddev, rdev); |
fae8cc5e N |
2400 | r10_bio->devs[r10_bio->read_slot].bio |
2401 | = IO_BLOCKED; | |
2402 | } | |
6814d536 N |
2403 | break; |
2404 | } | |
2405 | ||
2406 | start = sl; | |
2407 | /* write it back and re-read */ | |
2408 | rcu_read_lock(); | |
2409 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2410 | char b[BDEVNAME_SIZE]; |
0544a21d | 2411 | |
6814d536 N |
2412 | if (sl==0) |
2413 | sl = conf->copies; | |
2414 | sl--; | |
2415 | d = r10_bio->devs[sl].devnum; | |
2416 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2417 | if (!rdev || |
f5b67ae8 | 2418 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2419 | !test_bit(In_sync, &rdev->flags)) |
2420 | continue; | |
2421 | ||
2422 | atomic_inc(&rdev->nr_pending); | |
2423 | rcu_read_unlock(); | |
58c54fcc N |
2424 | if (r10_sync_page_io(rdev, |
2425 | r10_bio->devs[sl].addr + | |
2426 | sect, | |
055d3747 | 2427 | s, conf->tmppage, WRITE) |
1294b9c9 N |
2428 | == 0) { |
2429 | /* Well, this device is dead */ | |
08464e09 N |
2430 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %s)\n", |
2431 | mdname(mddev), s, | |
2432 | (unsigned long long)( | |
2433 | sect + | |
2434 | choose_data_offset(r10_bio, | |
2435 | rdev)), | |
2436 | bdevname(rdev->bdev, b)); | |
2437 | pr_notice("md/raid10:%s: %s: failing drive\n", | |
2438 | mdname(mddev), | |
2439 | bdevname(rdev->bdev, b)); | |
6814d536 | 2440 | } |
1294b9c9 N |
2441 | rdev_dec_pending(rdev, mddev); |
2442 | rcu_read_lock(); | |
6814d536 N |
2443 | } |
2444 | sl = start; | |
2445 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2446 | char b[BDEVNAME_SIZE]; |
0544a21d | 2447 | |
6814d536 N |
2448 | if (sl==0) |
2449 | sl = conf->copies; | |
2450 | sl--; | |
2451 | d = r10_bio->devs[sl].devnum; | |
2452 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2453 | if (!rdev || |
f5b67ae8 | 2454 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2455 | !test_bit(In_sync, &rdev->flags)) |
2456 | continue; | |
6814d536 | 2457 | |
1294b9c9 N |
2458 | atomic_inc(&rdev->nr_pending); |
2459 | rcu_read_unlock(); | |
58c54fcc N |
2460 | switch (r10_sync_page_io(rdev, |
2461 | r10_bio->devs[sl].addr + | |
2462 | sect, | |
055d3747 | 2463 | s, conf->tmppage, |
58c54fcc N |
2464 | READ)) { |
2465 | case 0: | |
1294b9c9 | 2466 | /* Well, this device is dead */ |
08464e09 | 2467 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2468 | mdname(mddev), s, |
2469 | (unsigned long long)( | |
f8c9e74f N |
2470 | sect + |
2471 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 | 2472 | bdevname(rdev->bdev, b)); |
08464e09 | 2473 | pr_notice("md/raid10:%s: %s: failing drive\n", |
1294b9c9 N |
2474 | mdname(mddev), |
2475 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2476 | break; |
2477 | case 1: | |
08464e09 | 2478 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2479 | mdname(mddev), s, |
2480 | (unsigned long long)( | |
f8c9e74f N |
2481 | sect + |
2482 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2483 | bdevname(rdev->bdev, b)); |
2484 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2485 | } |
1294b9c9 N |
2486 | |
2487 | rdev_dec_pending(rdev, mddev); | |
2488 | rcu_read_lock(); | |
6814d536 N |
2489 | } |
2490 | rcu_read_unlock(); | |
2491 | ||
2492 | sectors -= s; | |
2493 | sect += s; | |
2494 | } | |
2495 | } | |
2496 | ||
9f2c9d12 | 2497 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2498 | { |
2499 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2500 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2501 | struct r10conf *conf = mddev->private; |
3cb03002 | 2502 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2503 | /* bio has the data to be written to slot 'i' where |
2504 | * we just recently had a write error. | |
2505 | * We repeatedly clone the bio and trim down to one block, | |
2506 | * then try the write. Where the write fails we record | |
2507 | * a bad block. | |
2508 | * It is conceivable that the bio doesn't exactly align with | |
2509 | * blocks. We must handle this. | |
2510 | * | |
2511 | * We currently own a reference to the rdev. | |
2512 | */ | |
2513 | ||
2514 | int block_sectors; | |
2515 | sector_t sector; | |
2516 | int sectors; | |
2517 | int sect_to_write = r10_bio->sectors; | |
2518 | int ok = 1; | |
2519 | ||
2520 | if (rdev->badblocks.shift < 0) | |
2521 | return 0; | |
2522 | ||
f04ebb0b N |
2523 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2524 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2525 | sector = r10_bio->sector; |
2526 | sectors = ((r10_bio->sector + block_sectors) | |
2527 | & ~(sector_t)(block_sectors - 1)) | |
2528 | - sector; | |
2529 | ||
2530 | while (sect_to_write) { | |
2531 | struct bio *wbio; | |
27028626 | 2532 | sector_t wsector; |
bd870a16 N |
2533 | if (sectors > sect_to_write) |
2534 | sectors = sect_to_write; | |
2535 | /* Write at 'sector' for 'sectors' */ | |
afeee514 | 2536 | wbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
4f024f37 | 2537 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
27028626 TM |
2538 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); |
2539 | wbio->bi_iter.bi_sector = wsector + | |
2540 | choose_data_offset(r10_bio, rdev); | |
74d46992 | 2541 | bio_set_dev(wbio, rdev->bdev); |
796a5cf0 | 2542 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
2543 | |
2544 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 | 2545 | /* Failure! */ |
27028626 | 2546 | ok = rdev_set_badblocks(rdev, wsector, |
bd870a16 N |
2547 | sectors, 0) |
2548 | && ok; | |
2549 | ||
2550 | bio_put(wbio); | |
2551 | sect_to_write -= sectors; | |
2552 | sector += sectors; | |
2553 | sectors = block_sectors; | |
2554 | } | |
2555 | return ok; | |
2556 | } | |
2557 | ||
9f2c9d12 | 2558 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2559 | { |
2560 | int slot = r10_bio->read_slot; | |
560f8e55 | 2561 | struct bio *bio; |
e879a879 | 2562 | struct r10conf *conf = mddev->private; |
abbf098e | 2563 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2564 | |
2565 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2566 | * the block and we can fix it. | |
2567 | * We freeze all other IO, and try reading the block from | |
2568 | * other devices. When we find one, we re-write | |
2569 | * and check it that fixes the read error. | |
2570 | * This is all done synchronously while the array is | |
2571 | * frozen. | |
2572 | */ | |
fae8cc5e | 2573 | bio = r10_bio->devs[slot].bio; |
fae8cc5e N |
2574 | bio_put(bio); |
2575 | r10_bio->devs[slot].bio = NULL; | |
2576 | ||
8d3ca83d N |
2577 | if (mddev->ro) |
2578 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2579 | else if (!test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2580 | freeze_array(conf, 1); |
560f8e55 N |
2581 | fix_read_error(conf, mddev, r10_bio); |
2582 | unfreeze_array(conf); | |
fae8cc5e | 2583 | } else |
8d3ca83d | 2584 | md_error(mddev, rdev); |
fae8cc5e | 2585 | |
abbf098e | 2586 | rdev_dec_pending(rdev, mddev); |
545250f2 N |
2587 | allow_barrier(conf); |
2588 | r10_bio->state = 0; | |
2589 | raid10_read_request(mddev, r10_bio->master_bio, r10_bio); | |
560f8e55 N |
2590 | } |
2591 | ||
e879a879 | 2592 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2593 | { |
2594 | /* Some sort of write request has finished and it | |
2595 | * succeeded in writing where we thought there was a | |
2596 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2597 | * Or possibly if failed and we need to record |
2598 | * a bad block. | |
749c55e9 N |
2599 | */ |
2600 | int m; | |
3cb03002 | 2601 | struct md_rdev *rdev; |
749c55e9 N |
2602 | |
2603 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2604 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2605 | for (m = 0; m < conf->copies; m++) { |
2606 | int dev = r10_bio->devs[m].devnum; | |
2607 | rdev = conf->mirrors[dev].rdev; | |
01a69cab YY |
2608 | if (r10_bio->devs[m].bio == NULL || |
2609 | r10_bio->devs[m].bio->bi_end_io == NULL) | |
1a0b7cd8 | 2610 | continue; |
4e4cbee9 | 2611 | if (!r10_bio->devs[m].bio->bi_status) { |
749c55e9 N |
2612 | rdev_clear_badblocks( |
2613 | rdev, | |
2614 | r10_bio->devs[m].addr, | |
c6563a8c | 2615 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2616 | } else { |
2617 | if (!rdev_set_badblocks( | |
2618 | rdev, | |
2619 | r10_bio->devs[m].addr, | |
2620 | r10_bio->sectors, 0)) | |
2621 | md_error(conf->mddev, rdev); | |
749c55e9 | 2622 | } |
9ad1aefc | 2623 | rdev = conf->mirrors[dev].replacement; |
01a69cab YY |
2624 | if (r10_bio->devs[m].repl_bio == NULL || |
2625 | r10_bio->devs[m].repl_bio->bi_end_io == NULL) | |
9ad1aefc | 2626 | continue; |
4246a0b6 | 2627 | |
4e4cbee9 | 2628 | if (!r10_bio->devs[m].repl_bio->bi_status) { |
9ad1aefc N |
2629 | rdev_clear_badblocks( |
2630 | rdev, | |
2631 | r10_bio->devs[m].addr, | |
c6563a8c | 2632 | r10_bio->sectors, 0); |
9ad1aefc N |
2633 | } else { |
2634 | if (!rdev_set_badblocks( | |
2635 | rdev, | |
2636 | r10_bio->devs[m].addr, | |
2637 | r10_bio->sectors, 0)) | |
2638 | md_error(conf->mddev, rdev); | |
2639 | } | |
1a0b7cd8 | 2640 | } |
749c55e9 N |
2641 | put_buf(r10_bio); |
2642 | } else { | |
95af587e | 2643 | bool fail = false; |
bd870a16 N |
2644 | for (m = 0; m < conf->copies; m++) { |
2645 | int dev = r10_bio->devs[m].devnum; | |
2646 | struct bio *bio = r10_bio->devs[m].bio; | |
2647 | rdev = conf->mirrors[dev].rdev; | |
2648 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2649 | rdev_clear_badblocks( |
2650 | rdev, | |
2651 | r10_bio->devs[m].addr, | |
c6563a8c | 2652 | r10_bio->sectors, 0); |
749c55e9 | 2653 | rdev_dec_pending(rdev, conf->mddev); |
4e4cbee9 | 2654 | } else if (bio != NULL && bio->bi_status) { |
95af587e | 2655 | fail = true; |
bd870a16 N |
2656 | if (!narrow_write_error(r10_bio, m)) { |
2657 | md_error(conf->mddev, rdev); | |
2658 | set_bit(R10BIO_Degraded, | |
2659 | &r10_bio->state); | |
2660 | } | |
2661 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2662 | } |
475b0321 N |
2663 | bio = r10_bio->devs[m].repl_bio; |
2664 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2665 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2666 | rdev_clear_badblocks( |
2667 | rdev, | |
2668 | r10_bio->devs[m].addr, | |
c6563a8c | 2669 | r10_bio->sectors, 0); |
475b0321 N |
2670 | rdev_dec_pending(rdev, conf->mddev); |
2671 | } | |
bd870a16 | 2672 | } |
95af587e N |
2673 | if (fail) { |
2674 | spin_lock_irq(&conf->device_lock); | |
2675 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 2676 | conf->nr_queued++; |
95af587e | 2677 | spin_unlock_irq(&conf->device_lock); |
cf25ae78 GJ |
2678 | /* |
2679 | * In case freeze_array() is waiting for condition | |
2680 | * nr_pending == nr_queued + extra to be true. | |
2681 | */ | |
2682 | wake_up(&conf->wait_barrier); | |
95af587e | 2683 | md_wakeup_thread(conf->mddev->thread); |
c340702c N |
2684 | } else { |
2685 | if (test_bit(R10BIO_WriteError, | |
2686 | &r10_bio->state)) | |
2687 | close_write(r10_bio); | |
95af587e | 2688 | raid_end_bio_io(r10_bio); |
c340702c | 2689 | } |
749c55e9 N |
2690 | } |
2691 | } | |
2692 | ||
4ed8731d | 2693 | static void raid10d(struct md_thread *thread) |
1da177e4 | 2694 | { |
4ed8731d | 2695 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2696 | struct r10bio *r10_bio; |
1da177e4 | 2697 | unsigned long flags; |
e879a879 | 2698 | struct r10conf *conf = mddev->private; |
1da177e4 | 2699 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2700 | struct blk_plug plug; |
1da177e4 LT |
2701 | |
2702 | md_check_recovery(mddev); | |
1da177e4 | 2703 | |
95af587e | 2704 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2705 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
95af587e N |
2706 | LIST_HEAD(tmp); |
2707 | spin_lock_irqsave(&conf->device_lock, flags); | |
2953079c | 2708 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
23ddba80 SL |
2709 | while (!list_empty(&conf->bio_end_io_list)) { |
2710 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2711 | conf->nr_queued--; | |
2712 | } | |
95af587e N |
2713 | } |
2714 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2715 | while (!list_empty(&tmp)) { | |
a452744b MP |
2716 | r10_bio = list_first_entry(&tmp, struct r10bio, |
2717 | retry_list); | |
95af587e | 2718 | list_del(&r10_bio->retry_list); |
c340702c N |
2719 | if (mddev->degraded) |
2720 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
2721 | ||
2722 | if (test_bit(R10BIO_WriteError, | |
2723 | &r10_bio->state)) | |
2724 | close_write(r10_bio); | |
95af587e N |
2725 | raid_end_bio_io(r10_bio); |
2726 | } | |
2727 | } | |
2728 | ||
e1dfa0a2 | 2729 | blk_start_plug(&plug); |
1da177e4 | 2730 | for (;;) { |
6cce3b23 | 2731 | |
0021b7bc | 2732 | flush_pending_writes(conf); |
6cce3b23 | 2733 | |
a35e63ef N |
2734 | spin_lock_irqsave(&conf->device_lock, flags); |
2735 | if (list_empty(head)) { | |
2736 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2737 | break; |
a35e63ef | 2738 | } |
9f2c9d12 | 2739 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2740 | list_del(head->prev); |
4443ae10 | 2741 | conf->nr_queued--; |
1da177e4 LT |
2742 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2743 | ||
2744 | mddev = r10_bio->mddev; | |
070ec55d | 2745 | conf = mddev->private; |
bd870a16 N |
2746 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2747 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 2748 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
2749 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
2750 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 2751 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 2752 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2753 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2754 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2755 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2756 | handle_read_error(mddev, r10_bio); |
fc9977dd N |
2757 | else |
2758 | WARN_ON_ONCE(1); | |
560f8e55 | 2759 | |
1d9d5241 | 2760 | cond_resched(); |
2953079c | 2761 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2762 | md_check_recovery(mddev); |
1da177e4 | 2763 | } |
e1dfa0a2 | 2764 | blk_finish_plug(&plug); |
1da177e4 LT |
2765 | } |
2766 | ||
e879a879 | 2767 | static int init_resync(struct r10conf *conf) |
1da177e4 | 2768 | { |
afeee514 | 2769 | int ret, buffs, i; |
1da177e4 LT |
2770 | |
2771 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 | 2772 | BUG_ON(mempool_initialized(&conf->r10buf_pool)); |
69335ef3 | 2773 | conf->have_replacement = 0; |
5cf00fcd | 2774 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
2775 | if (conf->mirrors[i].replacement) |
2776 | conf->have_replacement = 1; | |
afeee514 KO |
2777 | ret = mempool_init(&conf->r10buf_pool, buffs, |
2778 | r10buf_pool_alloc, r10buf_pool_free, conf); | |
2779 | if (ret) | |
2780 | return ret; | |
1da177e4 LT |
2781 | conf->next_resync = 0; |
2782 | return 0; | |
2783 | } | |
2784 | ||
208410b5 SL |
2785 | static struct r10bio *raid10_alloc_init_r10buf(struct r10conf *conf) |
2786 | { | |
afeee514 | 2787 | struct r10bio *r10bio = mempool_alloc(&conf->r10buf_pool, GFP_NOIO); |
208410b5 SL |
2788 | struct rsync_pages *rp; |
2789 | struct bio *bio; | |
2790 | int nalloc; | |
2791 | int i; | |
2792 | ||
2793 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || | |
2794 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
2795 | nalloc = conf->copies; /* resync */ | |
2796 | else | |
2797 | nalloc = 2; /* recovery */ | |
2798 | ||
2799 | for (i = 0; i < nalloc; i++) { | |
2800 | bio = r10bio->devs[i].bio; | |
2801 | rp = bio->bi_private; | |
2802 | bio_reset(bio); | |
2803 | bio->bi_private = rp; | |
2804 | bio = r10bio->devs[i].repl_bio; | |
2805 | if (bio) { | |
2806 | rp = bio->bi_private; | |
2807 | bio_reset(bio); | |
2808 | bio->bi_private = rp; | |
2809 | } | |
2810 | } | |
2811 | return r10bio; | |
2812 | } | |
2813 | ||
8db87912 GJ |
2814 | /* |
2815 | * Set cluster_sync_high since we need other nodes to add the | |
2816 | * range [cluster_sync_low, cluster_sync_high] to suspend list. | |
2817 | */ | |
2818 | static void raid10_set_cluster_sync_high(struct r10conf *conf) | |
2819 | { | |
2820 | sector_t window_size; | |
2821 | int extra_chunk, chunks; | |
2822 | ||
2823 | /* | |
2824 | * First, here we define "stripe" as a unit which across | |
2825 | * all member devices one time, so we get chunks by use | |
2826 | * raid_disks / near_copies. Otherwise, if near_copies is | |
2827 | * close to raid_disks, then resync window could increases | |
2828 | * linearly with the increase of raid_disks, which means | |
2829 | * we will suspend a really large IO window while it is not | |
2830 | * necessary. If raid_disks is not divisible by near_copies, | |
2831 | * an extra chunk is needed to ensure the whole "stripe" is | |
2832 | * covered. | |
2833 | */ | |
2834 | ||
2835 | chunks = conf->geo.raid_disks / conf->geo.near_copies; | |
2836 | if (conf->geo.raid_disks % conf->geo.near_copies == 0) | |
2837 | extra_chunk = 0; | |
2838 | else | |
2839 | extra_chunk = 1; | |
2840 | window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors; | |
2841 | ||
2842 | /* | |
2843 | * At least use a 32M window to align with raid1's resync window | |
2844 | */ | |
2845 | window_size = (CLUSTER_RESYNC_WINDOW_SECTORS > window_size) ? | |
2846 | CLUSTER_RESYNC_WINDOW_SECTORS : window_size; | |
2847 | ||
2848 | conf->cluster_sync_high = conf->cluster_sync_low + window_size; | |
2849 | } | |
2850 | ||
1da177e4 LT |
2851 | /* |
2852 | * perform a "sync" on one "block" | |
2853 | * | |
2854 | * We need to make sure that no normal I/O request - particularly write | |
2855 | * requests - conflict with active sync requests. | |
2856 | * | |
2857 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2858 | * that can be installed to exclude normal IO requests. | |
2859 | * | |
2860 | * Resync and recovery are handled very differently. | |
2861 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2862 | * | |
2863 | * For resync, we iterate over virtual addresses, read all copies, | |
2864 | * and update if there are differences. If only one copy is live, | |
2865 | * skip it. | |
2866 | * For recovery, we iterate over physical addresses, read a good | |
2867 | * value for each non-in_sync drive, and over-write. | |
2868 | * | |
2869 | * So, for recovery we may have several outstanding complex requests for a | |
2870 | * given address, one for each out-of-sync device. We model this by allocating | |
2871 | * a number of r10_bio structures, one for each out-of-sync device. | |
2872 | * As we setup these structures, we collect all bio's together into a list | |
2873 | * which we then process collectively to add pages, and then process again | |
ed00aabd | 2874 | * to pass to submit_bio_noacct. |
1da177e4 LT |
2875 | * |
2876 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2877 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2878 | * has its remaining count decremented to 0, the whole complex operation | |
2879 | * is complete. | |
2880 | * | |
2881 | */ | |
2882 | ||
849674e4 | 2883 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 2884 | int *skipped) |
1da177e4 | 2885 | { |
e879a879 | 2886 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2887 | struct r10bio *r10_bio; |
1da177e4 LT |
2888 | struct bio *biolist = NULL, *bio; |
2889 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2890 | int i; |
6cce3b23 | 2891 | int max_sync; |
57dab0bd | 2892 | sector_t sync_blocks; |
1da177e4 LT |
2893 | sector_t sectors_skipped = 0; |
2894 | int chunks_skipped = 0; | |
5cf00fcd | 2895 | sector_t chunk_mask = conf->geo.chunk_mask; |
022e510f | 2896 | int page_idx = 0; |
1da177e4 | 2897 | |
afeee514 | 2898 | if (!mempool_initialized(&conf->r10buf_pool)) |
1da177e4 | 2899 | if (init_resync(conf)) |
57afd89f | 2900 | return 0; |
1da177e4 | 2901 | |
7e83ccbe MW |
2902 | /* |
2903 | * Allow skipping a full rebuild for incremental assembly | |
2904 | * of a clean array, like RAID1 does. | |
2905 | */ | |
2906 | if (mddev->bitmap == NULL && | |
2907 | mddev->recovery_cp == MaxSector && | |
13765120 N |
2908 | mddev->reshape_position == MaxSector && |
2909 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 2910 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 2911 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
2912 | conf->fullsync == 0) { |
2913 | *skipped = 1; | |
13765120 | 2914 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
2915 | } |
2916 | ||
1da177e4 | 2917 | skipped: |
58c0fed4 | 2918 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
2919 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
2920 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
2921 | max_sector = mddev->resync_max_sectors; |
2922 | if (sector_nr >= max_sector) { | |
8db87912 GJ |
2923 | conf->cluster_sync_low = 0; |
2924 | conf->cluster_sync_high = 0; | |
2925 | ||
6cce3b23 N |
2926 | /* If we aborted, we need to abort the |
2927 | * sync on the 'current' bitmap chucks (there can | |
2928 | * be several when recovering multiple devices). | |
2929 | * as we may have started syncing it but not finished. | |
2930 | * We can find the current address in | |
2931 | * mddev->curr_resync, but for recovery, | |
2932 | * we need to convert that to several | |
2933 | * virtual addresses. | |
2934 | */ | |
3ea7daa5 N |
2935 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
2936 | end_reshape(conf); | |
b3968552 | 2937 | close_sync(conf); |
3ea7daa5 N |
2938 | return 0; |
2939 | } | |
2940 | ||
6cce3b23 N |
2941 | if (mddev->curr_resync < max_sector) { /* aborted */ |
2942 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
e64e4018 AS |
2943 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
2944 | &sync_blocks, 1); | |
5cf00fcd | 2945 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
2946 | sector_t sect = |
2947 | raid10_find_virt(conf, mddev->curr_resync, i); | |
e64e4018 AS |
2948 | md_bitmap_end_sync(mddev->bitmap, sect, |
2949 | &sync_blocks, 1); | |
6cce3b23 | 2950 | } |
9ad1aefc N |
2951 | } else { |
2952 | /* completed sync */ | |
2953 | if ((!mddev->bitmap || conf->fullsync) | |
2954 | && conf->have_replacement | |
2955 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
2956 | /* Completed a full sync so the replacements | |
2957 | * are now fully recovered. | |
2958 | */ | |
f90145f3 N |
2959 | rcu_read_lock(); |
2960 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
2961 | struct md_rdev *rdev = | |
2962 | rcu_dereference(conf->mirrors[i].replacement); | |
2963 | if (rdev) | |
2964 | rdev->recovery_offset = MaxSector; | |
2965 | } | |
2966 | rcu_read_unlock(); | |
9ad1aefc | 2967 | } |
6cce3b23 | 2968 | conf->fullsync = 0; |
9ad1aefc | 2969 | } |
e64e4018 | 2970 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2971 | close_sync(conf); |
57afd89f | 2972 | *skipped = 1; |
1da177e4 LT |
2973 | return sectors_skipped; |
2974 | } | |
3ea7daa5 N |
2975 | |
2976 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
2977 | return reshape_request(mddev, sector_nr, skipped); | |
2978 | ||
5cf00fcd | 2979 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
2980 | /* if there has been nothing to do on any drive, |
2981 | * then there is nothing to do at all.. | |
2982 | */ | |
57afd89f N |
2983 | *skipped = 1; |
2984 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
2985 | } |
2986 | ||
c6207277 N |
2987 | if (max_sector > mddev->resync_max) |
2988 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
2989 | ||
1da177e4 LT |
2990 | /* make sure whole request will fit in a chunk - if chunks |
2991 | * are meaningful | |
2992 | */ | |
5cf00fcd N |
2993 | if (conf->geo.near_copies < conf->geo.raid_disks && |
2994 | max_sector > (sector_nr | chunk_mask)) | |
2995 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 | 2996 | |
7ac50447 TM |
2997 | /* |
2998 | * If there is non-resync activity waiting for a turn, then let it | |
2999 | * though before starting on this new sync request. | |
3000 | */ | |
3001 | if (conf->nr_waiting) | |
3002 | schedule_timeout_uninterruptible(1); | |
3003 | ||
1da177e4 LT |
3004 | /* Again, very different code for resync and recovery. |
3005 | * Both must result in an r10bio with a list of bios that | |
74d46992 | 3006 | * have bi_end_io, bi_sector, bi_disk set, |
1da177e4 LT |
3007 | * and bi_private set to the r10bio. |
3008 | * For recovery, we may actually create several r10bios | |
3009 | * with 2 bios in each, that correspond to the bios in the main one. | |
3010 | * In this case, the subordinate r10bios link back through a | |
3011 | * borrowed master_bio pointer, and the counter in the master | |
3012 | * includes a ref from each subordinate. | |
3013 | */ | |
3014 | /* First, we decide what to do and set ->bi_end_io | |
3015 | * To end_sync_read if we want to read, and | |
3016 | * end_sync_write if we will want to write. | |
3017 | */ | |
3018 | ||
6cce3b23 | 3019 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
3020 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
3021 | /* recovery... the complicated one */ | |
e875ecea | 3022 | int j; |
1da177e4 LT |
3023 | r10_bio = NULL; |
3024 | ||
5cf00fcd | 3025 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 3026 | int still_degraded; |
9f2c9d12 | 3027 | struct r10bio *rb2; |
ab9d47e9 N |
3028 | sector_t sect; |
3029 | int must_sync; | |
e875ecea | 3030 | int any_working; |
ee37d731 AW |
3031 | int need_recover = 0; |
3032 | int need_replace = 0; | |
dc280d98 | 3033 | struct raid10_info *mirror = &conf->mirrors[i]; |
f90145f3 | 3034 | struct md_rdev *mrdev, *mreplace; |
24afd80d | 3035 | |
f90145f3 N |
3036 | rcu_read_lock(); |
3037 | mrdev = rcu_dereference(mirror->rdev); | |
3038 | mreplace = rcu_dereference(mirror->replacement); | |
3039 | ||
ee37d731 AW |
3040 | if (mrdev != NULL && |
3041 | !test_bit(Faulty, &mrdev->flags) && | |
3042 | !test_bit(In_sync, &mrdev->flags)) | |
3043 | need_recover = 1; | |
3044 | if (mreplace != NULL && | |
3045 | !test_bit(Faulty, &mreplace->flags)) | |
3046 | need_replace = 1; | |
3047 | ||
3048 | if (!need_recover && !need_replace) { | |
f90145f3 | 3049 | rcu_read_unlock(); |
ab9d47e9 | 3050 | continue; |
f90145f3 | 3051 | } |
1da177e4 | 3052 | |
ab9d47e9 N |
3053 | still_degraded = 0; |
3054 | /* want to reconstruct this device */ | |
3055 | rb2 = r10_bio; | |
3056 | sect = raid10_find_virt(conf, sector_nr, i); | |
fc448a18 N |
3057 | if (sect >= mddev->resync_max_sectors) { |
3058 | /* last stripe is not complete - don't | |
3059 | * try to recover this sector. | |
3060 | */ | |
f90145f3 | 3061 | rcu_read_unlock(); |
fc448a18 N |
3062 | continue; |
3063 | } | |
f5b67ae8 N |
3064 | if (mreplace && test_bit(Faulty, &mreplace->flags)) |
3065 | mreplace = NULL; | |
24afd80d N |
3066 | /* Unless we are doing a full sync, or a replacement |
3067 | * we only need to recover the block if it is set in | |
3068 | * the bitmap | |
ab9d47e9 | 3069 | */ |
e64e4018 AS |
3070 | must_sync = md_bitmap_start_sync(mddev->bitmap, sect, |
3071 | &sync_blocks, 1); | |
ab9d47e9 N |
3072 | if (sync_blocks < max_sync) |
3073 | max_sync = sync_blocks; | |
3074 | if (!must_sync && | |
f90145f3 | 3075 | mreplace == NULL && |
ab9d47e9 N |
3076 | !conf->fullsync) { |
3077 | /* yep, skip the sync_blocks here, but don't assume | |
3078 | * that there will never be anything to do here | |
3079 | */ | |
3080 | chunks_skipped = -1; | |
f90145f3 | 3081 | rcu_read_unlock(); |
ab9d47e9 N |
3082 | continue; |
3083 | } | |
f90145f3 N |
3084 | atomic_inc(&mrdev->nr_pending); |
3085 | if (mreplace) | |
3086 | atomic_inc(&mreplace->nr_pending); | |
3087 | rcu_read_unlock(); | |
6cce3b23 | 3088 | |
208410b5 | 3089 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3090 | r10_bio->state = 0; |
ab9d47e9 N |
3091 | raise_barrier(conf, rb2 != NULL); |
3092 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 3093 | |
ab9d47e9 N |
3094 | r10_bio->master_bio = (struct bio*)rb2; |
3095 | if (rb2) | |
3096 | atomic_inc(&rb2->remaining); | |
3097 | r10_bio->mddev = mddev; | |
3098 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
3099 | r10_bio->sector = sect; | |
1da177e4 | 3100 | |
ab9d47e9 N |
3101 | raid10_find_phys(conf, r10_bio); |
3102 | ||
3103 | /* Need to check if the array will still be | |
3104 | * degraded | |
3105 | */ | |
f90145f3 N |
3106 | rcu_read_lock(); |
3107 | for (j = 0; j < conf->geo.raid_disks; j++) { | |
3108 | struct md_rdev *rdev = rcu_dereference( | |
3109 | conf->mirrors[j].rdev); | |
3110 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
ab9d47e9 | 3111 | still_degraded = 1; |
87fc767b | 3112 | break; |
1da177e4 | 3113 | } |
f90145f3 | 3114 | } |
ab9d47e9 | 3115 | |
e64e4018 AS |
3116 | must_sync = md_bitmap_start_sync(mddev->bitmap, sect, |
3117 | &sync_blocks, still_degraded); | |
ab9d47e9 | 3118 | |
e875ecea | 3119 | any_working = 0; |
ab9d47e9 | 3120 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3121 | int k; |
ab9d47e9 | 3122 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3123 | sector_t from_addr, to_addr; |
f90145f3 N |
3124 | struct md_rdev *rdev = |
3125 | rcu_dereference(conf->mirrors[d].rdev); | |
40c356ce N |
3126 | sector_t sector, first_bad; |
3127 | int bad_sectors; | |
f90145f3 N |
3128 | if (!rdev || |
3129 | !test_bit(In_sync, &rdev->flags)) | |
ab9d47e9 N |
3130 | continue; |
3131 | /* This is where we read from */ | |
e875ecea | 3132 | any_working = 1; |
40c356ce N |
3133 | sector = r10_bio->devs[j].addr; |
3134 | ||
3135 | if (is_badblock(rdev, sector, max_sync, | |
3136 | &first_bad, &bad_sectors)) { | |
3137 | if (first_bad > sector) | |
3138 | max_sync = first_bad - sector; | |
3139 | else { | |
3140 | bad_sectors -= (sector | |
3141 | - first_bad); | |
3142 | if (max_sync > bad_sectors) | |
3143 | max_sync = bad_sectors; | |
3144 | continue; | |
3145 | } | |
3146 | } | |
ab9d47e9 N |
3147 | bio = r10_bio->devs[0].bio; |
3148 | bio->bi_next = biolist; | |
3149 | biolist = bio; | |
ab9d47e9 | 3150 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3151 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3152 | if (test_bit(FailFast, &rdev->flags)) |
3153 | bio->bi_opf |= MD_FAILFAST; | |
5e570289 | 3154 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3155 | bio->bi_iter.bi_sector = from_addr + |
3156 | rdev->data_offset; | |
74d46992 | 3157 | bio_set_dev(bio, rdev->bdev); |
24afd80d N |
3158 | atomic_inc(&rdev->nr_pending); |
3159 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3160 | |
3161 | for (k=0; k<conf->copies; k++) | |
3162 | if (r10_bio->devs[k].devnum == i) | |
3163 | break; | |
3164 | BUG_ON(k == conf->copies); | |
5e570289 | 3165 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3166 | r10_bio->devs[0].devnum = d; |
5e570289 | 3167 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3168 | r10_bio->devs[1].devnum = i; |
5e570289 | 3169 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3170 | |
ee37d731 | 3171 | if (need_recover) { |
24afd80d N |
3172 | bio = r10_bio->devs[1].bio; |
3173 | bio->bi_next = biolist; | |
3174 | biolist = bio; | |
24afd80d | 3175 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3176 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3177 | bio->bi_iter.bi_sector = to_addr |
f90145f3 | 3178 | + mrdev->data_offset; |
74d46992 | 3179 | bio_set_dev(bio, mrdev->bdev); |
24afd80d N |
3180 | atomic_inc(&r10_bio->remaining); |
3181 | } else | |
3182 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3183 | ||
3184 | /* and maybe write to replacement */ | |
3185 | bio = r10_bio->devs[1].repl_bio; | |
3186 | if (bio) | |
3187 | bio->bi_end_io = NULL; | |
ee37d731 | 3188 | /* Note: if need_replace, then bio |
24afd80d N |
3189 | * cannot be NULL as r10buf_pool_alloc will |
3190 | * have allocated it. | |
24afd80d | 3191 | */ |
ee37d731 | 3192 | if (!need_replace) |
24afd80d N |
3193 | break; |
3194 | bio->bi_next = biolist; | |
3195 | biolist = bio; | |
24afd80d | 3196 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3197 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3198 | bio->bi_iter.bi_sector = to_addr + |
f90145f3 | 3199 | mreplace->data_offset; |
74d46992 | 3200 | bio_set_dev(bio, mreplace->bdev); |
24afd80d | 3201 | atomic_inc(&r10_bio->remaining); |
ab9d47e9 N |
3202 | break; |
3203 | } | |
f90145f3 | 3204 | rcu_read_unlock(); |
ab9d47e9 | 3205 | if (j == conf->copies) { |
e875ecea N |
3206 | /* Cannot recover, so abort the recovery or |
3207 | * record a bad block */ | |
e875ecea N |
3208 | if (any_working) { |
3209 | /* problem is that there are bad blocks | |
3210 | * on other device(s) | |
3211 | */ | |
3212 | int k; | |
3213 | for (k = 0; k < conf->copies; k++) | |
3214 | if (r10_bio->devs[k].devnum == i) | |
3215 | break; | |
24afd80d | 3216 | if (!test_bit(In_sync, |
f90145f3 | 3217 | &mrdev->flags) |
24afd80d | 3218 | && !rdev_set_badblocks( |
f90145f3 | 3219 | mrdev, |
24afd80d N |
3220 | r10_bio->devs[k].addr, |
3221 | max_sync, 0)) | |
3222 | any_working = 0; | |
f90145f3 | 3223 | if (mreplace && |
24afd80d | 3224 | !rdev_set_badblocks( |
f90145f3 | 3225 | mreplace, |
e875ecea N |
3226 | r10_bio->devs[k].addr, |
3227 | max_sync, 0)) | |
3228 | any_working = 0; | |
3229 | } | |
3230 | if (!any_working) { | |
3231 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3232 | &mddev->recovery)) | |
08464e09 | 3233 | pr_warn("md/raid10:%s: insufficient working devices for recovery.\n", |
e875ecea | 3234 | mdname(mddev)); |
24afd80d | 3235 | mirror->recovery_disabled |
e875ecea N |
3236 | = mddev->recovery_disabled; |
3237 | } | |
e8b84915 N |
3238 | put_buf(r10_bio); |
3239 | if (rb2) | |
3240 | atomic_dec(&rb2->remaining); | |
3241 | r10_bio = rb2; | |
f90145f3 N |
3242 | rdev_dec_pending(mrdev, mddev); |
3243 | if (mreplace) | |
3244 | rdev_dec_pending(mreplace, mddev); | |
ab9d47e9 | 3245 | break; |
1da177e4 | 3246 | } |
f90145f3 N |
3247 | rdev_dec_pending(mrdev, mddev); |
3248 | if (mreplace) | |
3249 | rdev_dec_pending(mreplace, mddev); | |
8d3ca83d N |
3250 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { |
3251 | /* Only want this if there is elsewhere to | |
3252 | * read from. 'j' is currently the first | |
3253 | * readable copy. | |
3254 | */ | |
3255 | int targets = 1; | |
3256 | for (; j < conf->copies; j++) { | |
3257 | int d = r10_bio->devs[j].devnum; | |
3258 | if (conf->mirrors[d].rdev && | |
3259 | test_bit(In_sync, | |
3260 | &conf->mirrors[d].rdev->flags)) | |
3261 | targets++; | |
3262 | } | |
3263 | if (targets == 1) | |
3264 | r10_bio->devs[0].bio->bi_opf | |
3265 | &= ~MD_FAILFAST; | |
3266 | } | |
ab9d47e9 | 3267 | } |
1da177e4 LT |
3268 | if (biolist == NULL) { |
3269 | while (r10_bio) { | |
9f2c9d12 N |
3270 | struct r10bio *rb2 = r10_bio; |
3271 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3272 | rb2->master_bio = NULL; |
3273 | put_buf(rb2); | |
3274 | } | |
3275 | goto giveup; | |
3276 | } | |
3277 | } else { | |
3278 | /* resync. Schedule a read for every block at this virt offset */ | |
3279 | int count = 0; | |
6cce3b23 | 3280 | |
8db87912 GJ |
3281 | /* |
3282 | * Since curr_resync_completed could probably not update in | |
3283 | * time, and we will set cluster_sync_low based on it. | |
3284 | * Let's check against "sector_nr + 2 * RESYNC_SECTORS" for | |
3285 | * safety reason, which ensures curr_resync_completed is | |
3286 | * updated in bitmap_cond_end_sync. | |
3287 | */ | |
e64e4018 AS |
3288 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
3289 | mddev_is_clustered(mddev) && | |
3290 | (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
78200d45 | 3291 | |
e64e4018 AS |
3292 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
3293 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3294 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3295 | &mddev->recovery)) { | |
6cce3b23 N |
3296 | /* We can skip this block */ |
3297 | *skipped = 1; | |
3298 | return sync_blocks + sectors_skipped; | |
3299 | } | |
3300 | if (sync_blocks < max_sync) | |
3301 | max_sync = sync_blocks; | |
208410b5 | 3302 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3303 | r10_bio->state = 0; |
1da177e4 | 3304 | |
1da177e4 LT |
3305 | r10_bio->mddev = mddev; |
3306 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3307 | raise_barrier(conf, 0); |
3308 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3309 | |
3310 | r10_bio->master_bio = NULL; | |
3311 | r10_bio->sector = sector_nr; | |
3312 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3313 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3314 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3315 | |
5cf00fcd | 3316 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3317 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3318 | sector_t first_bad, sector; |
3319 | int bad_sectors; | |
f90145f3 | 3320 | struct md_rdev *rdev; |
40c356ce | 3321 | |
9ad1aefc N |
3322 | if (r10_bio->devs[i].repl_bio) |
3323 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3324 | ||
1da177e4 | 3325 | bio = r10_bio->devs[i].bio; |
4e4cbee9 | 3326 | bio->bi_status = BLK_STS_IOERR; |
f90145f3 N |
3327 | rcu_read_lock(); |
3328 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
3329 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3330 | rcu_read_unlock(); | |
1da177e4 | 3331 | continue; |
f90145f3 | 3332 | } |
40c356ce | 3333 | sector = r10_bio->devs[i].addr; |
f90145f3 | 3334 | if (is_badblock(rdev, sector, max_sync, |
40c356ce N |
3335 | &first_bad, &bad_sectors)) { |
3336 | if (first_bad > sector) | |
3337 | max_sync = first_bad - sector; | |
3338 | else { | |
3339 | bad_sectors -= (sector - first_bad); | |
3340 | if (max_sync > bad_sectors) | |
91502f09 | 3341 | max_sync = bad_sectors; |
f90145f3 | 3342 | rcu_read_unlock(); |
40c356ce N |
3343 | continue; |
3344 | } | |
3345 | } | |
f90145f3 | 3346 | atomic_inc(&rdev->nr_pending); |
1da177e4 LT |
3347 | atomic_inc(&r10_bio->remaining); |
3348 | bio->bi_next = biolist; | |
3349 | biolist = bio; | |
1da177e4 | 3350 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3351 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
1cdd1257 | 3352 | if (test_bit(FailFast, &rdev->flags)) |
8d3ca83d | 3353 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3354 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3355 | bio_set_dev(bio, rdev->bdev); |
1da177e4 | 3356 | count++; |
9ad1aefc | 3357 | |
f90145f3 N |
3358 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3359 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3360 | rcu_read_unlock(); | |
9ad1aefc | 3361 | continue; |
f90145f3 N |
3362 | } |
3363 | atomic_inc(&rdev->nr_pending); | |
9ad1aefc N |
3364 | |
3365 | /* Need to set up for writing to the replacement */ | |
3366 | bio = r10_bio->devs[i].repl_bio; | |
4e4cbee9 | 3367 | bio->bi_status = BLK_STS_IOERR; |
9ad1aefc N |
3368 | |
3369 | sector = r10_bio->devs[i].addr; | |
9ad1aefc N |
3370 | bio->bi_next = biolist; |
3371 | biolist = bio; | |
9ad1aefc | 3372 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3373 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1cdd1257 | 3374 | if (test_bit(FailFast, &rdev->flags)) |
1919cbb2 | 3375 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3376 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3377 | bio_set_dev(bio, rdev->bdev); |
9ad1aefc | 3378 | count++; |
1cdd1257 | 3379 | rcu_read_unlock(); |
1da177e4 LT |
3380 | } |
3381 | ||
3382 | if (count < 2) { | |
3383 | for (i=0; i<conf->copies; i++) { | |
3384 | int d = r10_bio->devs[i].devnum; | |
3385 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3386 | rdev_dec_pending(conf->mirrors[d].rdev, |
3387 | mddev); | |
9ad1aefc N |
3388 | if (r10_bio->devs[i].repl_bio && |
3389 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3390 | rdev_dec_pending( | |
3391 | conf->mirrors[d].replacement, | |
3392 | mddev); | |
1da177e4 LT |
3393 | } |
3394 | put_buf(r10_bio); | |
3395 | biolist = NULL; | |
3396 | goto giveup; | |
3397 | } | |
3398 | } | |
3399 | ||
1da177e4 | 3400 | nr_sectors = 0; |
6cce3b23 N |
3401 | if (sector_nr + max_sync < max_sector) |
3402 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3403 | do { |
3404 | struct page *page; | |
3405 | int len = PAGE_SIZE; | |
1da177e4 LT |
3406 | if (sector_nr + (len>>9) > max_sector) |
3407 | len = (max_sector - sector_nr) << 9; | |
3408 | if (len == 0) | |
3409 | break; | |
3410 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
f0250618 | 3411 | struct resync_pages *rp = get_resync_pages(bio); |
022e510f | 3412 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
3413 | /* |
3414 | * won't fail because the vec table is big enough | |
3415 | * to hold all these pages | |
3416 | */ | |
3417 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
3418 | } |
3419 | nr_sectors += len>>9; | |
3420 | sector_nr += len>>9; | |
022e510f | 3421 | } while (++page_idx < RESYNC_PAGES); |
1da177e4 LT |
3422 | r10_bio->sectors = nr_sectors; |
3423 | ||
8db87912 GJ |
3424 | if (mddev_is_clustered(mddev) && |
3425 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3426 | /* It is resync not recovery */ | |
3427 | if (conf->cluster_sync_high < sector_nr + nr_sectors) { | |
3428 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
3429 | raid10_set_cluster_sync_high(conf); | |
3430 | /* Send resync message */ | |
3431 | md_cluster_ops->resync_info_update(mddev, | |
3432 | conf->cluster_sync_low, | |
3433 | conf->cluster_sync_high); | |
3434 | } | |
3435 | } else if (mddev_is_clustered(mddev)) { | |
3436 | /* This is recovery not resync */ | |
3437 | sector_t sect_va1, sect_va2; | |
3438 | bool broadcast_msg = false; | |
3439 | ||
3440 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3441 | /* | |
3442 | * sector_nr is a device address for recovery, so we | |
3443 | * need translate it to array address before compare | |
3444 | * with cluster_sync_high. | |
3445 | */ | |
3446 | sect_va1 = raid10_find_virt(conf, sector_nr, i); | |
3447 | ||
3448 | if (conf->cluster_sync_high < sect_va1 + nr_sectors) { | |
3449 | broadcast_msg = true; | |
3450 | /* | |
3451 | * curr_resync_completed is similar as | |
3452 | * sector_nr, so make the translation too. | |
3453 | */ | |
3454 | sect_va2 = raid10_find_virt(conf, | |
3455 | mddev->curr_resync_completed, i); | |
3456 | ||
3457 | if (conf->cluster_sync_low == 0 || | |
3458 | conf->cluster_sync_low > sect_va2) | |
3459 | conf->cluster_sync_low = sect_va2; | |
3460 | } | |
3461 | } | |
3462 | if (broadcast_msg) { | |
3463 | raid10_set_cluster_sync_high(conf); | |
3464 | md_cluster_ops->resync_info_update(mddev, | |
3465 | conf->cluster_sync_low, | |
3466 | conf->cluster_sync_high); | |
3467 | } | |
3468 | } | |
3469 | ||
1da177e4 LT |
3470 | while (biolist) { |
3471 | bio = biolist; | |
3472 | biolist = biolist->bi_next; | |
3473 | ||
3474 | bio->bi_next = NULL; | |
f0250618 | 3475 | r10_bio = get_resync_r10bio(bio); |
1da177e4 LT |
3476 | r10_bio->sectors = nr_sectors; |
3477 | ||
3478 | if (bio->bi_end_io == end_sync_read) { | |
74d46992 | 3479 | md_sync_acct_bio(bio, nr_sectors); |
4e4cbee9 | 3480 | bio->bi_status = 0; |
ed00aabd | 3481 | submit_bio_noacct(bio); |
1da177e4 LT |
3482 | } |
3483 | } | |
3484 | ||
57afd89f N |
3485 | if (sectors_skipped) |
3486 | /* pretend they weren't skipped, it makes | |
3487 | * no important difference in this case | |
3488 | */ | |
3489 | md_done_sync(mddev, sectors_skipped, 1); | |
3490 | ||
1da177e4 LT |
3491 | return sectors_skipped + nr_sectors; |
3492 | giveup: | |
3493 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3494 | * drives must be failed or in resync, all drives |
3495 | * have a bad block, so try the next chunk... | |
1da177e4 | 3496 | */ |
09b4068a N |
3497 | if (sector_nr + max_sync < max_sector) |
3498 | max_sector = sector_nr + max_sync; | |
3499 | ||
3500 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3501 | chunks_skipped ++; |
3502 | sector_nr = max_sector; | |
1da177e4 | 3503 | goto skipped; |
1da177e4 LT |
3504 | } |
3505 | ||
80c3a6ce | 3506 | static sector_t |
fd01b88c | 3507 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3508 | { |
3509 | sector_t size; | |
e879a879 | 3510 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3511 | |
3512 | if (!raid_disks) | |
3ea7daa5 N |
3513 | raid_disks = min(conf->geo.raid_disks, |
3514 | conf->prev.raid_disks); | |
80c3a6ce | 3515 | if (!sectors) |
dab8b292 | 3516 | sectors = conf->dev_sectors; |
80c3a6ce | 3517 | |
5cf00fcd N |
3518 | size = sectors >> conf->geo.chunk_shift; |
3519 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3520 | size = size * raid_disks; |
5cf00fcd | 3521 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3522 | |
5cf00fcd | 3523 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3524 | } |
3525 | ||
6508fdbf N |
3526 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3527 | { | |
3528 | /* Calculate the number of sectors-per-device that will | |
3529 | * actually be used, and set conf->dev_sectors and | |
3530 | * conf->stride | |
3531 | */ | |
3532 | ||
5cf00fcd N |
3533 | size = size >> conf->geo.chunk_shift; |
3534 | sector_div(size, conf->geo.far_copies); | |
3535 | size = size * conf->geo.raid_disks; | |
3536 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3537 | /* 'size' is now the number of chunks in the array */ |
3538 | /* calculate "used chunks per device" */ | |
3539 | size = size * conf->copies; | |
3540 | ||
3541 | /* We need to round up when dividing by raid_disks to | |
3542 | * get the stride size. | |
3543 | */ | |
5cf00fcd | 3544 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3545 | |
5cf00fcd | 3546 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3547 | |
5cf00fcd N |
3548 | if (conf->geo.far_offset) |
3549 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3550 | else { |
5cf00fcd N |
3551 | sector_div(size, conf->geo.far_copies); |
3552 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3553 | } |
3554 | } | |
dab8b292 | 3555 | |
deb200d0 N |
3556 | enum geo_type {geo_new, geo_old, geo_start}; |
3557 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3558 | { | |
3559 | int nc, fc, fo; | |
3560 | int layout, chunk, disks; | |
3561 | switch (new) { | |
3562 | case geo_old: | |
3563 | layout = mddev->layout; | |
3564 | chunk = mddev->chunk_sectors; | |
3565 | disks = mddev->raid_disks - mddev->delta_disks; | |
3566 | break; | |
3567 | case geo_new: | |
3568 | layout = mddev->new_layout; | |
3569 | chunk = mddev->new_chunk_sectors; | |
3570 | disks = mddev->raid_disks; | |
3571 | break; | |
3572 | default: /* avoid 'may be unused' warnings */ | |
3573 | case geo_start: /* new when starting reshape - raid_disks not | |
3574 | * updated yet. */ | |
3575 | layout = mddev->new_layout; | |
3576 | chunk = mddev->new_chunk_sectors; | |
3577 | disks = mddev->raid_disks + mddev->delta_disks; | |
3578 | break; | |
3579 | } | |
8bce6d35 | 3580 | if (layout >> 19) |
deb200d0 N |
3581 | return -1; |
3582 | if (chunk < (PAGE_SIZE >> 9) || | |
3583 | !is_power_of_2(chunk)) | |
3584 | return -2; | |
3585 | nc = layout & 255; | |
3586 | fc = (layout >> 8) & 255; | |
3587 | fo = layout & (1<<16); | |
3588 | geo->raid_disks = disks; | |
3589 | geo->near_copies = nc; | |
3590 | geo->far_copies = fc; | |
3591 | geo->far_offset = fo; | |
8bce6d35 N |
3592 | switch (layout >> 17) { |
3593 | case 0: /* original layout. simple but not always optimal */ | |
3594 | geo->far_set_size = disks; | |
3595 | break; | |
3596 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3597 | * actually using this, but leave code here just in case.*/ | |
3598 | geo->far_set_size = disks/fc; | |
3599 | WARN(geo->far_set_size < fc, | |
3600 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3601 | break; | |
3602 | case 2: /* "improved" layout fixed to match documentation */ | |
3603 | geo->far_set_size = fc * nc; | |
3604 | break; | |
3605 | default: /* Not a valid layout */ | |
3606 | return -1; | |
3607 | } | |
deb200d0 N |
3608 | geo->chunk_mask = chunk - 1; |
3609 | geo->chunk_shift = ffz(~chunk); | |
3610 | return nc*fc; | |
3611 | } | |
3612 | ||
e879a879 | 3613 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3614 | { |
e879a879 | 3615 | struct r10conf *conf = NULL; |
dab8b292 | 3616 | int err = -EINVAL; |
deb200d0 N |
3617 | struct geom geo; |
3618 | int copies; | |
3619 | ||
3620 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3621 | |
deb200d0 | 3622 | if (copies == -2) { |
08464e09 N |
3623 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n", |
3624 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3625 | goto out; |
1da177e4 | 3626 | } |
2604b703 | 3627 | |
deb200d0 | 3628 | if (copies < 2 || copies > mddev->raid_disks) { |
08464e09 N |
3629 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
3630 | mdname(mddev), mddev->new_layout); | |
1da177e4 LT |
3631 | goto out; |
3632 | } | |
dab8b292 TM |
3633 | |
3634 | err = -ENOMEM; | |
e879a879 | 3635 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3636 | if (!conf) |
1da177e4 | 3637 | goto out; |
dab8b292 | 3638 | |
3ea7daa5 | 3639 | /* FIXME calc properly */ |
6396bb22 KC |
3640 | conf->mirrors = kcalloc(mddev->raid_disks + max(0, -mddev->delta_disks), |
3641 | sizeof(struct raid10_info), | |
dab8b292 TM |
3642 | GFP_KERNEL); |
3643 | if (!conf->mirrors) | |
3644 | goto out; | |
4443ae10 N |
3645 | |
3646 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3647 | if (!conf->tmppage) | |
dab8b292 TM |
3648 | goto out; |
3649 | ||
deb200d0 N |
3650 | conf->geo = geo; |
3651 | conf->copies = copies; | |
3f677f9c | 3652 | err = mempool_init(&conf->r10bio_pool, NR_RAID_BIOS, r10bio_pool_alloc, |
c7afa803 | 3653 | rbio_pool_free, conf); |
afeee514 | 3654 | if (err) |
dab8b292 TM |
3655 | goto out; |
3656 | ||
afeee514 KO |
3657 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
3658 | if (err) | |
fc9977dd N |
3659 | goto out; |
3660 | ||
6508fdbf | 3661 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3662 | if (mddev->reshape_position == MaxSector) { |
3663 | conf->prev = conf->geo; | |
3664 | conf->reshape_progress = MaxSector; | |
3665 | } else { | |
3666 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3667 | err = -EINVAL; | |
3668 | goto out; | |
3669 | } | |
3670 | conf->reshape_progress = mddev->reshape_position; | |
3671 | if (conf->prev.far_offset) | |
3672 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
3673 | else | |
3674 | /* far_copies must be 1 */ | |
3675 | conf->prev.stride = conf->dev_sectors; | |
3676 | } | |
299b0685 | 3677 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 3678 | spin_lock_init(&conf->device_lock); |
dab8b292 | 3679 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 3680 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 TM |
3681 | |
3682 | spin_lock_init(&conf->resync_lock); | |
3683 | init_waitqueue_head(&conf->wait_barrier); | |
0e5313e2 | 3684 | atomic_set(&conf->nr_pending, 0); |
dab8b292 | 3685 | |
afeee514 | 3686 | err = -ENOMEM; |
0232605d | 3687 | conf->thread = md_register_thread(raid10d, mddev, "raid10"); |
dab8b292 TM |
3688 | if (!conf->thread) |
3689 | goto out; | |
3690 | ||
dab8b292 TM |
3691 | conf->mddev = mddev; |
3692 | return conf; | |
3693 | ||
3694 | out: | |
dab8b292 | 3695 | if (conf) { |
afeee514 | 3696 | mempool_exit(&conf->r10bio_pool); |
dab8b292 TM |
3697 | kfree(conf->mirrors); |
3698 | safe_put_page(conf->tmppage); | |
afeee514 | 3699 | bioset_exit(&conf->bio_split); |
dab8b292 TM |
3700 | kfree(conf); |
3701 | } | |
3702 | return ERR_PTR(err); | |
3703 | } | |
3704 | ||
16ef5101 CH |
3705 | static void raid10_set_io_opt(struct r10conf *conf) |
3706 | { | |
3707 | int raid_disks = conf->geo.raid_disks; | |
3708 | ||
3709 | if (!(conf->geo.raid_disks % conf->geo.near_copies)) | |
3710 | raid_disks /= conf->geo.near_copies; | |
3711 | blk_queue_io_opt(conf->mddev->queue, (conf->mddev->chunk_sectors << 9) * | |
3712 | raid_disks); | |
3713 | } | |
3714 | ||
849674e4 | 3715 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 3716 | { |
e879a879 | 3717 | struct r10conf *conf; |
16ef5101 | 3718 | int i, disk_idx; |
dc280d98 | 3719 | struct raid10_info *disk; |
3cb03002 | 3720 | struct md_rdev *rdev; |
dab8b292 | 3721 | sector_t size; |
3ea7daa5 N |
3722 | sector_t min_offset_diff = 0; |
3723 | int first = 1; | |
532a2a3f | 3724 | bool discard_supported = false; |
dab8b292 | 3725 | |
a415c0f1 N |
3726 | if (mddev_init_writes_pending(mddev) < 0) |
3727 | return -ENOMEM; | |
3728 | ||
dab8b292 TM |
3729 | if (mddev->private == NULL) { |
3730 | conf = setup_conf(mddev); | |
3731 | if (IS_ERR(conf)) | |
3732 | return PTR_ERR(conf); | |
3733 | mddev->private = conf; | |
3734 | } | |
3735 | conf = mddev->private; | |
3736 | if (!conf) | |
3737 | goto out; | |
3738 | ||
8db87912 GJ |
3739 | if (mddev_is_clustered(conf->mddev)) { |
3740 | int fc, fo; | |
3741 | ||
3742 | fc = (mddev->layout >> 8) & 255; | |
3743 | fo = mddev->layout & (1<<16); | |
3744 | if (fc > 1 || fo > 0) { | |
3745 | pr_err("only near layout is supported by clustered" | |
3746 | " raid10\n"); | |
43a52123 | 3747 | goto out_free_conf; |
8db87912 GJ |
3748 | } |
3749 | } | |
3750 | ||
dab8b292 TM |
3751 | mddev->thread = conf->thread; |
3752 | conf->thread = NULL; | |
3753 | ||
cc4d1efd | 3754 | if (mddev->queue) { |
532a2a3f | 3755 | blk_queue_max_discard_sectors(mddev->queue, |
d7cb6be0 | 3756 | mddev->chunk_sectors); |
5026d7a9 | 3757 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 3758 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
16ef5101 CH |
3759 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); |
3760 | raid10_set_io_opt(conf); | |
cc4d1efd | 3761 | } |
8f6c2e4b | 3762 | |
dafb20fa | 3763 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 3764 | long long diff; |
34b343cf | 3765 | |
1da177e4 | 3766 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
3767 | if (disk_idx < 0) |
3768 | continue; | |
3769 | if (disk_idx >= conf->geo.raid_disks && | |
3770 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
3771 | continue; |
3772 | disk = conf->mirrors + disk_idx; | |
3773 | ||
56a2559b N |
3774 | if (test_bit(Replacement, &rdev->flags)) { |
3775 | if (disk->replacement) | |
3776 | goto out_free_conf; | |
3777 | disk->replacement = rdev; | |
3778 | } else { | |
3779 | if (disk->rdev) | |
3780 | goto out_free_conf; | |
3781 | disk->rdev = rdev; | |
3782 | } | |
3ea7daa5 N |
3783 | diff = (rdev->new_data_offset - rdev->data_offset); |
3784 | if (!mddev->reshape_backwards) | |
3785 | diff = -diff; | |
3786 | if (diff < 0) | |
3787 | diff = 0; | |
3788 | if (first || diff < min_offset_diff) | |
3789 | min_offset_diff = diff; | |
56a2559b | 3790 | |
cc4d1efd JB |
3791 | if (mddev->gendisk) |
3792 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
3793 | rdev->data_offset << 9); | |
1da177e4 LT |
3794 | |
3795 | disk->head_position = 0; | |
532a2a3f SL |
3796 | |
3797 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) | |
3798 | discard_supported = true; | |
6f287ca6 | 3799 | first = 0; |
1da177e4 | 3800 | } |
3ea7daa5 | 3801 | |
ed30be07 JB |
3802 | if (mddev->queue) { |
3803 | if (discard_supported) | |
8b904b5b | 3804 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
ed30be07 JB |
3805 | mddev->queue); |
3806 | else | |
8b904b5b | 3807 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
ed30be07 JB |
3808 | mddev->queue); |
3809 | } | |
6d508242 | 3810 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3811 | if (!enough(conf, -1)) { |
08464e09 | 3812 | pr_err("md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3813 | mdname(mddev)); |
1da177e4 LT |
3814 | goto out_free_conf; |
3815 | } | |
3816 | ||
3ea7daa5 N |
3817 | if (conf->reshape_progress != MaxSector) { |
3818 | /* must ensure that shape change is supported */ | |
3819 | if (conf->geo.far_copies != 1 && | |
3820 | conf->geo.far_offset == 0) | |
3821 | goto out_free_conf; | |
3822 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 3823 | conf->prev.far_offset == 0) |
3ea7daa5 N |
3824 | goto out_free_conf; |
3825 | } | |
3826 | ||
1da177e4 | 3827 | mddev->degraded = 0; |
f8c9e74f N |
3828 | for (i = 0; |
3829 | i < conf->geo.raid_disks | |
3830 | || i < conf->prev.raid_disks; | |
3831 | i++) { | |
1da177e4 LT |
3832 | |
3833 | disk = conf->mirrors + i; | |
3834 | ||
56a2559b N |
3835 | if (!disk->rdev && disk->replacement) { |
3836 | /* The replacement is all we have - use it */ | |
3837 | disk->rdev = disk->replacement; | |
3838 | disk->replacement = NULL; | |
3839 | clear_bit(Replacement, &disk->rdev->flags); | |
3840 | } | |
3841 | ||
5fd6c1dc | 3842 | if (!disk->rdev || |
2e333e89 | 3843 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3844 | disk->head_position = 0; |
3845 | mddev->degraded++; | |
0b59bb64 N |
3846 | if (disk->rdev && |
3847 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 3848 | conf->fullsync = 1; |
1da177e4 | 3849 | } |
bda31539 BC |
3850 | |
3851 | if (disk->replacement && | |
3852 | !test_bit(In_sync, &disk->replacement->flags) && | |
3853 | disk->replacement->saved_raid_disk < 0) { | |
3854 | conf->fullsync = 1; | |
3855 | } | |
3856 | ||
d890fa2b | 3857 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3858 | } |
3859 | ||
8c6ac868 | 3860 | if (mddev->recovery_cp != MaxSector) |
08464e09 N |
3861 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", |
3862 | mdname(mddev)); | |
3863 | pr_info("md/raid10:%s: active with %d out of %d devices\n", | |
5cf00fcd N |
3864 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
3865 | conf->geo.raid_disks); | |
1da177e4 LT |
3866 | /* |
3867 | * Ok, everything is just fine now | |
3868 | */ | |
dab8b292 TM |
3869 | mddev->dev_sectors = conf->dev_sectors; |
3870 | size = raid10_size(mddev, 0, 0); | |
3871 | md_set_array_sectors(mddev, size); | |
3872 | mddev->resync_max_sectors = size; | |
46533ff7 | 3873 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
1da177e4 | 3874 | |
a91a2785 MP |
3875 | if (md_integrity_register(mddev)) |
3876 | goto out_free_conf; | |
3877 | ||
3ea7daa5 N |
3878 | if (conf->reshape_progress != MaxSector) { |
3879 | unsigned long before_length, after_length; | |
3880 | ||
3881 | before_length = ((1 << conf->prev.chunk_shift) * | |
3882 | conf->prev.far_copies); | |
3883 | after_length = ((1 << conf->geo.chunk_shift) * | |
3884 | conf->geo.far_copies); | |
3885 | ||
3886 | if (max(before_length, after_length) > min_offset_diff) { | |
3887 | /* This cannot work */ | |
08464e09 | 3888 | pr_warn("md/raid10: offset difference not enough to continue reshape\n"); |
3ea7daa5 N |
3889 | goto out_free_conf; |
3890 | } | |
3891 | conf->offset_diff = min_offset_diff; | |
3892 | ||
3ea7daa5 N |
3893 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
3894 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3895 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3896 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3897 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3898 | "reshape"); | |
e406f12d AP |
3899 | if (!mddev->sync_thread) |
3900 | goto out_free_conf; | |
3ea7daa5 N |
3901 | } |
3902 | ||
1da177e4 LT |
3903 | return 0; |
3904 | ||
3905 | out_free_conf: | |
01f96c0a | 3906 | md_unregister_thread(&mddev->thread); |
afeee514 | 3907 | mempool_exit(&conf->r10bio_pool); |
1345b1d8 | 3908 | safe_put_page(conf->tmppage); |
990a8baf | 3909 | kfree(conf->mirrors); |
1da177e4 LT |
3910 | kfree(conf); |
3911 | mddev->private = NULL; | |
3912 | out: | |
3913 | return -EIO; | |
3914 | } | |
3915 | ||
afa0f557 | 3916 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 3917 | { |
afa0f557 | 3918 | struct r10conf *conf = priv; |
1da177e4 | 3919 | |
afeee514 | 3920 | mempool_exit(&conf->r10bio_pool); |
0fea7ed8 | 3921 | safe_put_page(conf->tmppage); |
990a8baf | 3922 | kfree(conf->mirrors); |
c4796e21 N |
3923 | kfree(conf->mirrors_old); |
3924 | kfree(conf->mirrors_new); | |
afeee514 | 3925 | bioset_exit(&conf->bio_split); |
1da177e4 | 3926 | kfree(conf); |
1da177e4 LT |
3927 | } |
3928 | ||
b03e0ccb | 3929 | static void raid10_quiesce(struct mddev *mddev, int quiesce) |
6cce3b23 | 3930 | { |
e879a879 | 3931 | struct r10conf *conf = mddev->private; |
6cce3b23 | 3932 | |
b03e0ccb | 3933 | if (quiesce) |
6cce3b23 | 3934 | raise_barrier(conf, 0); |
b03e0ccb | 3935 | else |
6cce3b23 | 3936 | lower_barrier(conf); |
6cce3b23 | 3937 | } |
1da177e4 | 3938 | |
006a09a0 N |
3939 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3940 | { | |
3941 | /* Resize of 'far' arrays is not supported. | |
3942 | * For 'near' and 'offset' arrays we can set the | |
3943 | * number of sectors used to be an appropriate multiple | |
3944 | * of the chunk size. | |
3945 | * For 'offset', this is far_copies*chunksize. | |
3946 | * For 'near' the multiplier is the LCM of | |
3947 | * near_copies and raid_disks. | |
3948 | * So if far_copies > 1 && !far_offset, fail. | |
3949 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
3950 | * multiply by chunk_size. Then round to this number. | |
3951 | * This is mostly done by raid10_size() | |
3952 | */ | |
3953 | struct r10conf *conf = mddev->private; | |
3954 | sector_t oldsize, size; | |
3955 | ||
f8c9e74f N |
3956 | if (mddev->reshape_position != MaxSector) |
3957 | return -EBUSY; | |
3958 | ||
5cf00fcd | 3959 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
3960 | return -EINVAL; |
3961 | ||
3962 | oldsize = raid10_size(mddev, 0, 0); | |
3963 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
3964 | if (mddev->external_size && |
3965 | mddev->array_sectors > size) | |
006a09a0 | 3966 | return -EINVAL; |
a4a6125a | 3967 | if (mddev->bitmap) { |
e64e4018 | 3968 | int ret = md_bitmap_resize(mddev->bitmap, size, 0, 0); |
a4a6125a N |
3969 | if (ret) |
3970 | return ret; | |
3971 | } | |
3972 | md_set_array_sectors(mddev, size); | |
006a09a0 N |
3973 | if (sectors > mddev->dev_sectors && |
3974 | mddev->recovery_cp > oldsize) { | |
3975 | mddev->recovery_cp = oldsize; | |
3976 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3977 | } | |
6508fdbf N |
3978 | calc_sectors(conf, sectors); |
3979 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
3980 | mddev->resync_max_sectors = size; |
3981 | return 0; | |
3982 | } | |
3983 | ||
53a6ab4d | 3984 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 3985 | { |
3cb03002 | 3986 | struct md_rdev *rdev; |
e879a879 | 3987 | struct r10conf *conf; |
dab8b292 TM |
3988 | |
3989 | if (mddev->degraded > 0) { | |
08464e09 N |
3990 | pr_warn("md/raid10:%s: Error: degraded raid0!\n", |
3991 | mdname(mddev)); | |
dab8b292 TM |
3992 | return ERR_PTR(-EINVAL); |
3993 | } | |
53a6ab4d | 3994 | sector_div(size, devs); |
dab8b292 | 3995 | |
dab8b292 TM |
3996 | /* Set new parameters */ |
3997 | mddev->new_level = 10; | |
3998 | /* new layout: far_copies = 1, near_copies = 2 */ | |
3999 | mddev->new_layout = (1<<8) + 2; | |
4000 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
4001 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
4002 | mddev->raid_disks *= 2; |
4003 | /* make sure it will be not marked as dirty */ | |
4004 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 4005 | mddev->dev_sectors = size; |
dab8b292 TM |
4006 | |
4007 | conf = setup_conf(mddev); | |
02214dc5 | 4008 | if (!IS_ERR(conf)) { |
dafb20fa | 4009 | rdev_for_each(rdev, mddev) |
53a6ab4d | 4010 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 4011 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
4012 | rdev->sectors = size; |
4013 | } | |
02214dc5 KW |
4014 | conf->barrier = 1; |
4015 | } | |
4016 | ||
dab8b292 TM |
4017 | return conf; |
4018 | } | |
4019 | ||
fd01b88c | 4020 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 4021 | { |
e373ab10 | 4022 | struct r0conf *raid0_conf; |
dab8b292 TM |
4023 | |
4024 | /* raid10 can take over: | |
4025 | * raid0 - providing it has only two drives | |
4026 | */ | |
4027 | if (mddev->level == 0) { | |
4028 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
4029 | raid0_conf = mddev->private; |
4030 | if (raid0_conf->nr_strip_zones > 1) { | |
08464e09 N |
4031 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n", |
4032 | mdname(mddev)); | |
dab8b292 TM |
4033 | return ERR_PTR(-EINVAL); |
4034 | } | |
53a6ab4d N |
4035 | return raid10_takeover_raid0(mddev, |
4036 | raid0_conf->strip_zone->zone_end, | |
4037 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
4038 | } |
4039 | return ERR_PTR(-EINVAL); | |
4040 | } | |
4041 | ||
3ea7daa5 N |
4042 | static int raid10_check_reshape(struct mddev *mddev) |
4043 | { | |
4044 | /* Called when there is a request to change | |
4045 | * - layout (to ->new_layout) | |
4046 | * - chunk size (to ->new_chunk_sectors) | |
4047 | * - raid_disks (by delta_disks) | |
4048 | * or when trying to restart a reshape that was ongoing. | |
4049 | * | |
4050 | * We need to validate the request and possibly allocate | |
4051 | * space if that might be an issue later. | |
4052 | * | |
4053 | * Currently we reject any reshape of a 'far' mode array, | |
4054 | * allow chunk size to change if new is generally acceptable, | |
4055 | * allow raid_disks to increase, and allow | |
4056 | * a switch between 'near' mode and 'offset' mode. | |
4057 | */ | |
4058 | struct r10conf *conf = mddev->private; | |
4059 | struct geom geo; | |
4060 | ||
4061 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
4062 | return -EINVAL; | |
4063 | ||
4064 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
4065 | /* mustn't change number of copies */ | |
4066 | return -EINVAL; | |
4067 | if (geo.far_copies > 1 && !geo.far_offset) | |
4068 | /* Cannot switch to 'far' mode */ | |
4069 | return -EINVAL; | |
4070 | ||
4071 | if (mddev->array_sectors & geo.chunk_mask) | |
4072 | /* not factor of array size */ | |
4073 | return -EINVAL; | |
4074 | ||
3ea7daa5 N |
4075 | if (!enough(conf, -1)) |
4076 | return -EINVAL; | |
4077 | ||
4078 | kfree(conf->mirrors_new); | |
4079 | conf->mirrors_new = NULL; | |
4080 | if (mddev->delta_disks > 0) { | |
4081 | /* allocate new 'mirrors' list */ | |
6396bb22 KC |
4082 | conf->mirrors_new = |
4083 | kcalloc(mddev->raid_disks + mddev->delta_disks, | |
4084 | sizeof(struct raid10_info), | |
4085 | GFP_KERNEL); | |
3ea7daa5 N |
4086 | if (!conf->mirrors_new) |
4087 | return -ENOMEM; | |
4088 | } | |
4089 | return 0; | |
4090 | } | |
4091 | ||
4092 | /* | |
4093 | * Need to check if array has failed when deciding whether to: | |
4094 | * - start an array | |
4095 | * - remove non-faulty devices | |
4096 | * - add a spare | |
4097 | * - allow a reshape | |
4098 | * This determination is simple when no reshape is happening. | |
4099 | * However if there is a reshape, we need to carefully check | |
4100 | * both the before and after sections. | |
4101 | * This is because some failed devices may only affect one | |
4102 | * of the two sections, and some non-in_sync devices may | |
4103 | * be insync in the section most affected by failed devices. | |
4104 | */ | |
4105 | static int calc_degraded(struct r10conf *conf) | |
4106 | { | |
4107 | int degraded, degraded2; | |
4108 | int i; | |
4109 | ||
4110 | rcu_read_lock(); | |
4111 | degraded = 0; | |
4112 | /* 'prev' section first */ | |
4113 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
4114 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4115 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4116 | degraded++; | |
4117 | else if (!test_bit(In_sync, &rdev->flags)) | |
4118 | /* When we can reduce the number of devices in | |
4119 | * an array, this might not contribute to | |
4120 | * 'degraded'. It does now. | |
4121 | */ | |
4122 | degraded++; | |
4123 | } | |
4124 | rcu_read_unlock(); | |
4125 | if (conf->geo.raid_disks == conf->prev.raid_disks) | |
4126 | return degraded; | |
4127 | rcu_read_lock(); | |
4128 | degraded2 = 0; | |
4129 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
4130 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4131 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4132 | degraded2++; | |
4133 | else if (!test_bit(In_sync, &rdev->flags)) { | |
4134 | /* If reshape is increasing the number of devices, | |
4135 | * this section has already been recovered, so | |
4136 | * it doesn't contribute to degraded. | |
4137 | * else it does. | |
4138 | */ | |
4139 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
4140 | degraded2++; | |
4141 | } | |
4142 | } | |
4143 | rcu_read_unlock(); | |
4144 | if (degraded2 > degraded) | |
4145 | return degraded2; | |
4146 | return degraded; | |
4147 | } | |
4148 | ||
4149 | static int raid10_start_reshape(struct mddev *mddev) | |
4150 | { | |
4151 | /* A 'reshape' has been requested. This commits | |
4152 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
4153 | * This also checks if there are enough spares and adds them | |
4154 | * to the array. | |
4155 | * We currently require enough spares to make the final | |
4156 | * array non-degraded. We also require that the difference | |
4157 | * between old and new data_offset - on each device - is | |
4158 | * enough that we never risk over-writing. | |
4159 | */ | |
4160 | ||
4161 | unsigned long before_length, after_length; | |
4162 | sector_t min_offset_diff = 0; | |
4163 | int first = 1; | |
4164 | struct geom new; | |
4165 | struct r10conf *conf = mddev->private; | |
4166 | struct md_rdev *rdev; | |
4167 | int spares = 0; | |
bb63a701 | 4168 | int ret; |
3ea7daa5 N |
4169 | |
4170 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4171 | return -EBUSY; | |
4172 | ||
4173 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
4174 | return -EINVAL; | |
4175 | ||
4176 | before_length = ((1 << conf->prev.chunk_shift) * | |
4177 | conf->prev.far_copies); | |
4178 | after_length = ((1 << conf->geo.chunk_shift) * | |
4179 | conf->geo.far_copies); | |
4180 | ||
4181 | rdev_for_each(rdev, mddev) { | |
4182 | if (!test_bit(In_sync, &rdev->flags) | |
4183 | && !test_bit(Faulty, &rdev->flags)) | |
4184 | spares++; | |
4185 | if (rdev->raid_disk >= 0) { | |
4186 | long long diff = (rdev->new_data_offset | |
4187 | - rdev->data_offset); | |
4188 | if (!mddev->reshape_backwards) | |
4189 | diff = -diff; | |
4190 | if (diff < 0) | |
4191 | diff = 0; | |
4192 | if (first || diff < min_offset_diff) | |
4193 | min_offset_diff = diff; | |
b506335e | 4194 | first = 0; |
3ea7daa5 N |
4195 | } |
4196 | } | |
4197 | ||
4198 | if (max(before_length, after_length) > min_offset_diff) | |
4199 | return -EINVAL; | |
4200 | ||
4201 | if (spares < mddev->delta_disks) | |
4202 | return -EINVAL; | |
4203 | ||
4204 | conf->offset_diff = min_offset_diff; | |
4205 | spin_lock_irq(&conf->device_lock); | |
4206 | if (conf->mirrors_new) { | |
4207 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4208 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4209 | smp_mb(); |
c4796e21 | 4210 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4211 | conf->mirrors_old = conf->mirrors; |
4212 | conf->mirrors = conf->mirrors_new; | |
4213 | conf->mirrors_new = NULL; | |
4214 | } | |
4215 | setup_geo(&conf->geo, mddev, geo_start); | |
4216 | smp_mb(); | |
4217 | if (mddev->reshape_backwards) { | |
4218 | sector_t size = raid10_size(mddev, 0, 0); | |
4219 | if (size < mddev->array_sectors) { | |
4220 | spin_unlock_irq(&conf->device_lock); | |
08464e09 N |
4221 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n", |
4222 | mdname(mddev)); | |
3ea7daa5 N |
4223 | return -EINVAL; |
4224 | } | |
4225 | mddev->resync_max_sectors = size; | |
4226 | conf->reshape_progress = size; | |
4227 | } else | |
4228 | conf->reshape_progress = 0; | |
299b0685 | 4229 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4230 | spin_unlock_irq(&conf->device_lock); |
4231 | ||
bb63a701 | 4232 | if (mddev->delta_disks && mddev->bitmap) { |
afd75628 GJ |
4233 | struct mdp_superblock_1 *sb = NULL; |
4234 | sector_t oldsize, newsize; | |
4235 | ||
4236 | oldsize = raid10_size(mddev, 0, 0); | |
4237 | newsize = raid10_size(mddev, 0, conf->geo.raid_disks); | |
4238 | ||
4239 | if (!mddev_is_clustered(mddev)) { | |
4240 | ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
4241 | if (ret) | |
4242 | goto abort; | |
4243 | else | |
4244 | goto out; | |
4245 | } | |
4246 | ||
4247 | rdev_for_each(rdev, mddev) { | |
4248 | if (rdev->raid_disk > -1 && | |
4249 | !test_bit(Faulty, &rdev->flags)) | |
4250 | sb = page_address(rdev->sb_page); | |
4251 | } | |
4252 | ||
4253 | /* | |
4254 | * some node is already performing reshape, and no need to | |
4255 | * call md_bitmap_resize again since it should be called when | |
4256 | * receiving BITMAP_RESIZE msg | |
4257 | */ | |
4258 | if ((sb && (le32_to_cpu(sb->feature_map) & | |
4259 | MD_FEATURE_RESHAPE_ACTIVE)) || (oldsize == newsize)) | |
4260 | goto out; | |
4261 | ||
4262 | ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
bb63a701 N |
4263 | if (ret) |
4264 | goto abort; | |
afd75628 GJ |
4265 | |
4266 | ret = md_cluster_ops->resize_bitmaps(mddev, newsize, oldsize); | |
4267 | if (ret) { | |
4268 | md_bitmap_resize(mddev->bitmap, oldsize, 0, 0); | |
4269 | goto abort; | |
4270 | } | |
bb63a701 | 4271 | } |
afd75628 | 4272 | out: |
3ea7daa5 N |
4273 | if (mddev->delta_disks > 0) { |
4274 | rdev_for_each(rdev, mddev) | |
4275 | if (rdev->raid_disk < 0 && | |
4276 | !test_bit(Faulty, &rdev->flags)) { | |
4277 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4278 | if (rdev->raid_disk >= | |
4279 | conf->prev.raid_disks) | |
4280 | set_bit(In_sync, &rdev->flags); | |
4281 | else | |
4282 | rdev->recovery_offset = 0; | |
4283 | ||
38ffc01f DLM |
4284 | /* Failure here is OK */ |
4285 | sysfs_link_rdev(mddev, rdev); | |
3ea7daa5 N |
4286 | } |
4287 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4288 | && !test_bit(Faulty, &rdev->flags)) { | |
4289 | /* This is a spare that was manually added */ | |
4290 | set_bit(In_sync, &rdev->flags); | |
4291 | } | |
4292 | } | |
4293 | /* When a reshape changes the number of devices, | |
4294 | * ->degraded is measured against the larger of the | |
4295 | * pre and post numbers. | |
4296 | */ | |
4297 | spin_lock_irq(&conf->device_lock); | |
4298 | mddev->degraded = calc_degraded(conf); | |
4299 | spin_unlock_irq(&conf->device_lock); | |
4300 | mddev->raid_disks = conf->geo.raid_disks; | |
4301 | mddev->reshape_position = conf->reshape_progress; | |
2953079c | 4302 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 N |
4303 | |
4304 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4305 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4306 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 N |
4307 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
4308 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4309 | ||
4310 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4311 | "reshape"); | |
4312 | if (!mddev->sync_thread) { | |
bb63a701 N |
4313 | ret = -EAGAIN; |
4314 | goto abort; | |
3ea7daa5 N |
4315 | } |
4316 | conf->reshape_checkpoint = jiffies; | |
4317 | md_wakeup_thread(mddev->sync_thread); | |
4318 | md_new_event(mddev); | |
4319 | return 0; | |
bb63a701 N |
4320 | |
4321 | abort: | |
4322 | mddev->recovery = 0; | |
4323 | spin_lock_irq(&conf->device_lock); | |
4324 | conf->geo = conf->prev; | |
4325 | mddev->raid_disks = conf->geo.raid_disks; | |
4326 | rdev_for_each(rdev, mddev) | |
4327 | rdev->new_data_offset = rdev->data_offset; | |
4328 | smp_wmb(); | |
4329 | conf->reshape_progress = MaxSector; | |
299b0685 | 4330 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4331 | mddev->reshape_position = MaxSector; |
4332 | spin_unlock_irq(&conf->device_lock); | |
4333 | return ret; | |
3ea7daa5 N |
4334 | } |
4335 | ||
4336 | /* Calculate the last device-address that could contain | |
4337 | * any block from the chunk that includes the array-address 's' | |
4338 | * and report the next address. | |
4339 | * i.e. the address returned will be chunk-aligned and after | |
4340 | * any data that is in the chunk containing 's'. | |
4341 | */ | |
4342 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4343 | { | |
4344 | s = (s | geo->chunk_mask) + 1; | |
4345 | s >>= geo->chunk_shift; | |
4346 | s *= geo->near_copies; | |
4347 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4348 | s *= geo->far_copies; | |
4349 | s <<= geo->chunk_shift; | |
4350 | return s; | |
4351 | } | |
4352 | ||
4353 | /* Calculate the first device-address that could contain | |
4354 | * any block from the chunk that includes the array-address 's'. | |
4355 | * This too will be the start of a chunk | |
4356 | */ | |
4357 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4358 | { | |
4359 | s >>= geo->chunk_shift; | |
4360 | s *= geo->near_copies; | |
4361 | sector_div(s, geo->raid_disks); | |
4362 | s *= geo->far_copies; | |
4363 | s <<= geo->chunk_shift; | |
4364 | return s; | |
4365 | } | |
4366 | ||
4367 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4368 | int *skipped) | |
4369 | { | |
4370 | /* We simply copy at most one chunk (smallest of old and new) | |
4371 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4372 | * or we hit a bad block or something. | |
4373 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4374 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4375 | * can record any location. |
4376 | * | |
4377 | * If we will want to write to a location that isn't | |
4378 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4379 | * we need to flush all reshape requests and update the metadata. | |
4380 | * | |
4381 | * When reshaping forwards (e.g. to more devices), we interpret | |
4382 | * 'safe' as the earliest block which might not have been copied | |
4383 | * down yet. We divide this by previous stripe size and multiply | |
4384 | * by previous stripe length to get lowest device offset that we | |
4385 | * cannot write to yet. | |
4386 | * We interpret 'sector_nr' as an address that we want to write to. | |
4387 | * From this we use last_device_address() to find where we might | |
4388 | * write to, and first_device_address on the 'safe' position. | |
4389 | * If this 'next' write position is after the 'safe' position, | |
4390 | * we must update the metadata to increase the 'safe' position. | |
4391 | * | |
4392 | * When reshaping backwards, we round in the opposite direction | |
4393 | * and perform the reverse test: next write position must not be | |
4394 | * less than current safe position. | |
4395 | * | |
4396 | * In all this the minimum difference in data offsets | |
4397 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4398 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4399 | * |
4400 | * We need to prepare all the bios here before we start any IO | |
4401 | * to ensure the size we choose is acceptable to all devices. | |
4402 | * The means one for each copy for write-out and an extra one for | |
4403 | * read-in. | |
4404 | * We store the read-in bio in ->master_bio and the others in | |
4405 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4406 | */ | |
4407 | struct r10conf *conf = mddev->private; | |
4408 | struct r10bio *r10_bio; | |
4409 | sector_t next, safe, last; | |
4410 | int max_sectors; | |
4411 | int nr_sectors; | |
4412 | int s; | |
4413 | struct md_rdev *rdev; | |
4414 | int need_flush = 0; | |
4415 | struct bio *blist; | |
4416 | struct bio *bio, *read_bio; | |
4417 | int sectors_done = 0; | |
f0250618 | 4418 | struct page **pages; |
3ea7daa5 N |
4419 | |
4420 | if (sector_nr == 0) { | |
4421 | /* If restarting in the middle, skip the initial sectors */ | |
4422 | if (mddev->reshape_backwards && | |
4423 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4424 | sector_nr = (raid10_size(mddev, 0, 0) | |
4425 | - conf->reshape_progress); | |
4426 | } else if (!mddev->reshape_backwards && | |
4427 | conf->reshape_progress > 0) | |
4428 | sector_nr = conf->reshape_progress; | |
4429 | if (sector_nr) { | |
4430 | mddev->curr_resync_completed = sector_nr; | |
e1a86dbb | 4431 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
3ea7daa5 N |
4432 | *skipped = 1; |
4433 | return sector_nr; | |
4434 | } | |
4435 | } | |
4436 | ||
4437 | /* We don't use sector_nr to track where we are up to | |
4438 | * as that doesn't work well for ->reshape_backwards. | |
4439 | * So just use ->reshape_progress. | |
4440 | */ | |
4441 | if (mddev->reshape_backwards) { | |
4442 | /* 'next' is the earliest device address that we might | |
4443 | * write to for this chunk in the new layout | |
4444 | */ | |
4445 | next = first_dev_address(conf->reshape_progress - 1, | |
4446 | &conf->geo); | |
4447 | ||
4448 | /* 'safe' is the last device address that we might read from | |
4449 | * in the old layout after a restart | |
4450 | */ | |
4451 | safe = last_dev_address(conf->reshape_safe - 1, | |
4452 | &conf->prev); | |
4453 | ||
4454 | if (next + conf->offset_diff < safe) | |
4455 | need_flush = 1; | |
4456 | ||
4457 | last = conf->reshape_progress - 1; | |
4458 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4459 | & conf->prev.chunk_mask); | |
e287308b ZL |
4460 | if (sector_nr + RESYNC_SECTORS < last) |
4461 | sector_nr = last + 1 - RESYNC_SECTORS; | |
3ea7daa5 N |
4462 | } else { |
4463 | /* 'next' is after the last device address that we | |
4464 | * might write to for this chunk in the new layout | |
4465 | */ | |
4466 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4467 | ||
4468 | /* 'safe' is the earliest device address that we might | |
4469 | * read from in the old layout after a restart | |
4470 | */ | |
4471 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4472 | ||
4473 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4474 | * as that would possibly corrupt data | |
4475 | */ | |
4476 | if (next > safe + conf->offset_diff) | |
4477 | need_flush = 1; | |
4478 | ||
4479 | sector_nr = conf->reshape_progress; | |
4480 | last = sector_nr | (conf->geo.chunk_mask | |
4481 | & conf->prev.chunk_mask); | |
4482 | ||
e287308b ZL |
4483 | if (sector_nr + RESYNC_SECTORS <= last) |
4484 | last = sector_nr + RESYNC_SECTORS - 1; | |
3ea7daa5 N |
4485 | } |
4486 | ||
4487 | if (need_flush || | |
4488 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4489 | /* Need to update reshape_position in metadata */ | |
4490 | wait_barrier(conf); | |
4491 | mddev->reshape_position = conf->reshape_progress; | |
4492 | if (mddev->reshape_backwards) | |
4493 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4494 | - conf->reshape_progress; | |
4495 | else | |
4496 | mddev->curr_resync_completed = conf->reshape_progress; | |
4497 | conf->reshape_checkpoint = jiffies; | |
2953079c | 4498 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 | 4499 | md_wakeup_thread(mddev->thread); |
2953079c | 4500 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
4501 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4502 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4503 | allow_barrier(conf); | |
4504 | return sectors_done; | |
4505 | } | |
3ea7daa5 N |
4506 | conf->reshape_safe = mddev->reshape_position; |
4507 | allow_barrier(conf); | |
4508 | } | |
4509 | ||
1d0ffd26 | 4510 | raise_barrier(conf, 0); |
3ea7daa5 N |
4511 | read_more: |
4512 | /* Now schedule reads for blocks from sector_nr to last */ | |
208410b5 | 4513 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 4514 | r10_bio->state = 0; |
1d0ffd26 | 4515 | raise_barrier(conf, 1); |
3ea7daa5 N |
4516 | atomic_set(&r10_bio->remaining, 0); |
4517 | r10_bio->mddev = mddev; | |
4518 | r10_bio->sector = sector_nr; | |
4519 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4520 | r10_bio->sectors = last - sector_nr + 1; | |
4521 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4522 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4523 | ||
4524 | if (!rdev) { | |
4525 | /* Cannot read from here, so need to record bad blocks | |
4526 | * on all the target devices. | |
4527 | */ | |
4528 | // FIXME | |
afeee514 | 4529 | mempool_free(r10_bio, &conf->r10buf_pool); |
3ea7daa5 N |
4530 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4531 | return sectors_done; | |
4532 | } | |
4533 | ||
4534 | read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev); | |
4535 | ||
74d46992 | 4536 | bio_set_dev(read_bio, rdev->bdev); |
4f024f37 | 4537 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4538 | + rdev->data_offset); |
4539 | read_bio->bi_private = r10_bio; | |
81fa1520 | 4540 | read_bio->bi_end_io = end_reshape_read; |
796a5cf0 | 4541 | bio_set_op_attrs(read_bio, REQ_OP_READ, 0); |
ce0b0a46 | 4542 | read_bio->bi_flags &= (~0UL << BIO_RESET_BITS); |
4e4cbee9 | 4543 | read_bio->bi_status = 0; |
3ea7daa5 | 4544 | read_bio->bi_vcnt = 0; |
4f024f37 | 4545 | read_bio->bi_iter.bi_size = 0; |
3ea7daa5 N |
4546 | r10_bio->master_bio = read_bio; |
4547 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4548 | ||
7564beda GJ |
4549 | /* |
4550 | * Broadcast RESYNC message to other nodes, so all nodes would not | |
4551 | * write to the region to avoid conflict. | |
4552 | */ | |
4553 | if (mddev_is_clustered(mddev) && conf->cluster_sync_high <= sector_nr) { | |
4554 | struct mdp_superblock_1 *sb = NULL; | |
4555 | int sb_reshape_pos = 0; | |
4556 | ||
4557 | conf->cluster_sync_low = sector_nr; | |
4558 | conf->cluster_sync_high = sector_nr + CLUSTER_RESYNC_WINDOW_SECTORS; | |
4559 | sb = page_address(rdev->sb_page); | |
4560 | if (sb) { | |
4561 | sb_reshape_pos = le64_to_cpu(sb->reshape_position); | |
4562 | /* | |
4563 | * Set cluster_sync_low again if next address for array | |
4564 | * reshape is less than cluster_sync_low. Since we can't | |
4565 | * update cluster_sync_low until it has finished reshape. | |
4566 | */ | |
4567 | if (sb_reshape_pos < conf->cluster_sync_low) | |
4568 | conf->cluster_sync_low = sb_reshape_pos; | |
4569 | } | |
4570 | ||
4571 | md_cluster_ops->resync_info_update(mddev, conf->cluster_sync_low, | |
4572 | conf->cluster_sync_high); | |
4573 | } | |
4574 | ||
3ea7daa5 N |
4575 | /* Now find the locations in the new layout */ |
4576 | __raid10_find_phys(&conf->geo, r10_bio); | |
4577 | ||
4578 | blist = read_bio; | |
4579 | read_bio->bi_next = NULL; | |
4580 | ||
d094d686 | 4581 | rcu_read_lock(); |
3ea7daa5 N |
4582 | for (s = 0; s < conf->copies*2; s++) { |
4583 | struct bio *b; | |
4584 | int d = r10_bio->devs[s/2].devnum; | |
4585 | struct md_rdev *rdev2; | |
4586 | if (s&1) { | |
d094d686 | 4587 | rdev2 = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4588 | b = r10_bio->devs[s/2].repl_bio; |
4589 | } else { | |
d094d686 | 4590 | rdev2 = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4591 | b = r10_bio->devs[s/2].bio; |
4592 | } | |
4593 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4594 | continue; | |
8be185f2 | 4595 | |
74d46992 | 4596 | bio_set_dev(b, rdev2->bdev); |
4f024f37 KO |
4597 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4598 | rdev2->new_data_offset; | |
3ea7daa5 | 4599 | b->bi_end_io = end_reshape_write; |
796a5cf0 | 4600 | bio_set_op_attrs(b, REQ_OP_WRITE, 0); |
3ea7daa5 | 4601 | b->bi_next = blist; |
3ea7daa5 N |
4602 | blist = b; |
4603 | } | |
4604 | ||
4605 | /* Now add as many pages as possible to all of these bios. */ | |
4606 | ||
4607 | nr_sectors = 0; | |
f0250618 | 4608 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; |
3ea7daa5 | 4609 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { |
f0250618 | 4610 | struct page *page = pages[s / (PAGE_SIZE >> 9)]; |
3ea7daa5 N |
4611 | int len = (max_sectors - s) << 9; |
4612 | if (len > PAGE_SIZE) | |
4613 | len = PAGE_SIZE; | |
4614 | for (bio = blist; bio ; bio = bio->bi_next) { | |
c85ba149 ML |
4615 | /* |
4616 | * won't fail because the vec table is big enough | |
4617 | * to hold all these pages | |
4618 | */ | |
4619 | bio_add_page(bio, page, len, 0); | |
3ea7daa5 N |
4620 | } |
4621 | sector_nr += len >> 9; | |
4622 | nr_sectors += len >> 9; | |
4623 | } | |
d094d686 | 4624 | rcu_read_unlock(); |
3ea7daa5 N |
4625 | r10_bio->sectors = nr_sectors; |
4626 | ||
4627 | /* Now submit the read */ | |
74d46992 | 4628 | md_sync_acct_bio(read_bio, r10_bio->sectors); |
3ea7daa5 N |
4629 | atomic_inc(&r10_bio->remaining); |
4630 | read_bio->bi_next = NULL; | |
ed00aabd | 4631 | submit_bio_noacct(read_bio); |
3ea7daa5 N |
4632 | sectors_done += nr_sectors; |
4633 | if (sector_nr <= last) | |
4634 | goto read_more; | |
4635 | ||
1d0ffd26 XN |
4636 | lower_barrier(conf); |
4637 | ||
3ea7daa5 N |
4638 | /* Now that we have done the whole section we can |
4639 | * update reshape_progress | |
4640 | */ | |
4641 | if (mddev->reshape_backwards) | |
4642 | conf->reshape_progress -= sectors_done; | |
4643 | else | |
4644 | conf->reshape_progress += sectors_done; | |
4645 | ||
4646 | return sectors_done; | |
4647 | } | |
4648 | ||
4649 | static void end_reshape_request(struct r10bio *r10_bio); | |
4650 | static int handle_reshape_read_error(struct mddev *mddev, | |
4651 | struct r10bio *r10_bio); | |
4652 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4653 | { | |
4654 | /* Reshape read completed. Hopefully we have a block | |
4655 | * to write out. | |
4656 | * If we got a read error then we do sync 1-page reads from | |
4657 | * elsewhere until we find the data - or give up. | |
4658 | */ | |
4659 | struct r10conf *conf = mddev->private; | |
4660 | int s; | |
4661 | ||
4662 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4663 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4664 | /* Reshape has been aborted */ | |
4665 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4666 | return; | |
4667 | } | |
4668 | ||
4669 | /* We definitely have the data in the pages, schedule the | |
4670 | * writes. | |
4671 | */ | |
4672 | atomic_set(&r10_bio->remaining, 1); | |
4673 | for (s = 0; s < conf->copies*2; s++) { | |
4674 | struct bio *b; | |
4675 | int d = r10_bio->devs[s/2].devnum; | |
4676 | struct md_rdev *rdev; | |
d094d686 | 4677 | rcu_read_lock(); |
3ea7daa5 | 4678 | if (s&1) { |
d094d686 | 4679 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4680 | b = r10_bio->devs[s/2].repl_bio; |
4681 | } else { | |
d094d686 | 4682 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4683 | b = r10_bio->devs[s/2].bio; |
4684 | } | |
d094d686 N |
4685 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
4686 | rcu_read_unlock(); | |
3ea7daa5 | 4687 | continue; |
d094d686 | 4688 | } |
3ea7daa5 | 4689 | atomic_inc(&rdev->nr_pending); |
d094d686 | 4690 | rcu_read_unlock(); |
74d46992 | 4691 | md_sync_acct_bio(b, r10_bio->sectors); |
3ea7daa5 N |
4692 | atomic_inc(&r10_bio->remaining); |
4693 | b->bi_next = NULL; | |
ed00aabd | 4694 | submit_bio_noacct(b); |
3ea7daa5 N |
4695 | } |
4696 | end_reshape_request(r10_bio); | |
4697 | } | |
4698 | ||
4699 | static void end_reshape(struct r10conf *conf) | |
4700 | { | |
4701 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
4702 | return; | |
4703 | ||
4704 | spin_lock_irq(&conf->device_lock); | |
4705 | conf->prev = conf->geo; | |
4706 | md_finish_reshape(conf->mddev); | |
4707 | smp_wmb(); | |
4708 | conf->reshape_progress = MaxSector; | |
299b0685 | 4709 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
4710 | spin_unlock_irq(&conf->device_lock); |
4711 | ||
c2e4cd57 | 4712 | if (conf->mddev->queue) |
16ef5101 | 4713 | raid10_set_io_opt(conf); |
3ea7daa5 N |
4714 | conf->fullsync = 0; |
4715 | } | |
4716 | ||
7564beda GJ |
4717 | static void raid10_update_reshape_pos(struct mddev *mddev) |
4718 | { | |
4719 | struct r10conf *conf = mddev->private; | |
5ebaf80b | 4720 | sector_t lo, hi; |
7564beda | 4721 | |
5ebaf80b GJ |
4722 | md_cluster_ops->resync_info_get(mddev, &lo, &hi); |
4723 | if (((mddev->reshape_position <= hi) && (mddev->reshape_position >= lo)) | |
4724 | || mddev->reshape_position == MaxSector) | |
4725 | conf->reshape_progress = mddev->reshape_position; | |
4726 | else | |
4727 | WARN_ON_ONCE(1); | |
7564beda GJ |
4728 | } |
4729 | ||
3ea7daa5 N |
4730 | static int handle_reshape_read_error(struct mddev *mddev, |
4731 | struct r10bio *r10_bio) | |
4732 | { | |
4733 | /* Use sync reads to get the blocks from somewhere else */ | |
4734 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 4735 | struct r10conf *conf = mddev->private; |
584ed9fa | 4736 | struct r10bio *r10b; |
3ea7daa5 N |
4737 | int slot = 0; |
4738 | int idx = 0; | |
2d06e3b7 ML |
4739 | struct page **pages; |
4740 | ||
8cf05a78 | 4741 | r10b = kmalloc(struct_size(r10b, devs, conf->copies), GFP_NOIO); |
584ed9fa MK |
4742 | if (!r10b) { |
4743 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
4744 | return -ENOMEM; | |
4745 | } | |
4746 | ||
2d06e3b7 ML |
4747 | /* reshape IOs share pages from .devs[0].bio */ |
4748 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; | |
3ea7daa5 | 4749 | |
e0ee7785 N |
4750 | r10b->sector = r10_bio->sector; |
4751 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
4752 | |
4753 | while (sectors) { | |
4754 | int s = sectors; | |
4755 | int success = 0; | |
4756 | int first_slot = slot; | |
4757 | ||
4758 | if (s > (PAGE_SIZE >> 9)) | |
4759 | s = PAGE_SIZE >> 9; | |
4760 | ||
d094d686 | 4761 | rcu_read_lock(); |
3ea7daa5 | 4762 | while (!success) { |
e0ee7785 | 4763 | int d = r10b->devs[slot].devnum; |
d094d686 | 4764 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4765 | sector_t addr; |
4766 | if (rdev == NULL || | |
4767 | test_bit(Faulty, &rdev->flags) || | |
4768 | !test_bit(In_sync, &rdev->flags)) | |
4769 | goto failed; | |
4770 | ||
e0ee7785 | 4771 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
d094d686 N |
4772 | atomic_inc(&rdev->nr_pending); |
4773 | rcu_read_unlock(); | |
3ea7daa5 N |
4774 | success = sync_page_io(rdev, |
4775 | addr, | |
4776 | s << 9, | |
2d06e3b7 | 4777 | pages[idx], |
796a5cf0 | 4778 | REQ_OP_READ, 0, false); |
d094d686 N |
4779 | rdev_dec_pending(rdev, mddev); |
4780 | rcu_read_lock(); | |
3ea7daa5 N |
4781 | if (success) |
4782 | break; | |
4783 | failed: | |
4784 | slot++; | |
4785 | if (slot >= conf->copies) | |
4786 | slot = 0; | |
4787 | if (slot == first_slot) | |
4788 | break; | |
4789 | } | |
d094d686 | 4790 | rcu_read_unlock(); |
3ea7daa5 N |
4791 | if (!success) { |
4792 | /* couldn't read this block, must give up */ | |
4793 | set_bit(MD_RECOVERY_INTR, | |
4794 | &mddev->recovery); | |
584ed9fa | 4795 | kfree(r10b); |
3ea7daa5 N |
4796 | return -EIO; |
4797 | } | |
4798 | sectors -= s; | |
4799 | idx++; | |
4800 | } | |
584ed9fa | 4801 | kfree(r10b); |
3ea7daa5 N |
4802 | return 0; |
4803 | } | |
4804 | ||
4246a0b6 | 4805 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 4806 | { |
f0250618 | 4807 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
3ea7daa5 N |
4808 | struct mddev *mddev = r10_bio->mddev; |
4809 | struct r10conf *conf = mddev->private; | |
4810 | int d; | |
4811 | int slot; | |
4812 | int repl; | |
4813 | struct md_rdev *rdev = NULL; | |
4814 | ||
4815 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
4816 | if (repl) | |
4817 | rdev = conf->mirrors[d].replacement; | |
4818 | if (!rdev) { | |
4819 | smp_mb(); | |
4820 | rdev = conf->mirrors[d].rdev; | |
4821 | } | |
4822 | ||
4e4cbee9 | 4823 | if (bio->bi_status) { |
3ea7daa5 N |
4824 | /* FIXME should record badblock */ |
4825 | md_error(mddev, rdev); | |
4826 | } | |
4827 | ||
4828 | rdev_dec_pending(rdev, mddev); | |
4829 | end_reshape_request(r10_bio); | |
4830 | } | |
4831 | ||
4832 | static void end_reshape_request(struct r10bio *r10_bio) | |
4833 | { | |
4834 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
4835 | return; | |
4836 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
4837 | bio_put(r10_bio->master_bio); | |
4838 | put_buf(r10_bio); | |
4839 | } | |
4840 | ||
4841 | static void raid10_finish_reshape(struct mddev *mddev) | |
4842 | { | |
4843 | struct r10conf *conf = mddev->private; | |
4844 | ||
4845 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
4846 | return; | |
4847 | ||
4848 | if (mddev->delta_disks > 0) { | |
3ea7daa5 N |
4849 | if (mddev->recovery_cp > mddev->resync_max_sectors) { |
4850 | mddev->recovery_cp = mddev->resync_max_sectors; | |
4851 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4852 | } | |
8876391e | 4853 | mddev->resync_max_sectors = mddev->array_sectors; |
63aced61 N |
4854 | } else { |
4855 | int d; | |
d094d686 | 4856 | rcu_read_lock(); |
63aced61 N |
4857 | for (d = conf->geo.raid_disks ; |
4858 | d < conf->geo.raid_disks - mddev->delta_disks; | |
4859 | d++) { | |
d094d686 | 4860 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
63aced61 N |
4861 | if (rdev) |
4862 | clear_bit(In_sync, &rdev->flags); | |
d094d686 | 4863 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
63aced61 N |
4864 | if (rdev) |
4865 | clear_bit(In_sync, &rdev->flags); | |
4866 | } | |
d094d686 | 4867 | rcu_read_unlock(); |
3ea7daa5 N |
4868 | } |
4869 | mddev->layout = mddev->new_layout; | |
4870 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
4871 | mddev->reshape_position = MaxSector; | |
4872 | mddev->delta_disks = 0; | |
4873 | mddev->reshape_backwards = 0; | |
4874 | } | |
4875 | ||
84fc4b56 | 4876 | static struct md_personality raid10_personality = |
1da177e4 LT |
4877 | { |
4878 | .name = "raid10", | |
2604b703 | 4879 | .level = 10, |
1da177e4 | 4880 | .owner = THIS_MODULE, |
849674e4 SL |
4881 | .make_request = raid10_make_request, |
4882 | .run = raid10_run, | |
afa0f557 | 4883 | .free = raid10_free, |
849674e4 SL |
4884 | .status = raid10_status, |
4885 | .error_handler = raid10_error, | |
1da177e4 LT |
4886 | .hot_add_disk = raid10_add_disk, |
4887 | .hot_remove_disk= raid10_remove_disk, | |
4888 | .spare_active = raid10_spare_active, | |
849674e4 | 4889 | .sync_request = raid10_sync_request, |
6cce3b23 | 4890 | .quiesce = raid10_quiesce, |
80c3a6ce | 4891 | .size = raid10_size, |
006a09a0 | 4892 | .resize = raid10_resize, |
dab8b292 | 4893 | .takeover = raid10_takeover, |
3ea7daa5 N |
4894 | .check_reshape = raid10_check_reshape, |
4895 | .start_reshape = raid10_start_reshape, | |
4896 | .finish_reshape = raid10_finish_reshape, | |
7564beda | 4897 | .update_reshape_pos = raid10_update_reshape_pos, |
1da177e4 LT |
4898 | }; |
4899 | ||
4900 | static int __init raid_init(void) | |
4901 | { | |
2604b703 | 4902 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
4903 | } |
4904 | ||
4905 | static void raid_exit(void) | |
4906 | { | |
2604b703 | 4907 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
4908 | } |
4909 | ||
4910 | module_init(raid_init); | |
4911 | module_exit(raid_exit); | |
4912 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 4913 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 4914 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 4915 | MODULE_ALIAS("md-raid10"); |
2604b703 | 4916 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
4917 | |
4918 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |