]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid10.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 2000-2004 Neil Brown | |
5 | * | |
6 | * RAID-10 support for md. | |
7 | * | |
25985edc | 8 | * Base on code in raid1.c. See raid1.c for further copyright information. |
1da177e4 LT |
9 | * |
10 | * | |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
5a0e3ad6 | 21 | #include <linux/slab.h> |
25570727 | 22 | #include <linux/delay.h> |
bff61975 | 23 | #include <linux/blkdev.h> |
056075c7 | 24 | #include <linux/module.h> |
bff61975 | 25 | #include <linux/seq_file.h> |
8bda470e | 26 | #include <linux/ratelimit.h> |
3ea7daa5 | 27 | #include <linux/kthread.h> |
109e3765 | 28 | #include <trace/events/block.h> |
43b2e5d8 | 29 | #include "md.h" |
ef740c37 | 30 | #include "raid10.h" |
dab8b292 | 31 | #include "raid0.h" |
ef740c37 | 32 | #include "bitmap.h" |
1da177e4 LT |
33 | |
34 | /* | |
35 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
36 | * The layout of data is defined by | |
37 | * chunk_size | |
38 | * raid_disks | |
39 | * near_copies (stored in low byte of layout) | |
40 | * far_copies (stored in second byte of layout) | |
c93983bf | 41 | * far_offset (stored in bit 16 of layout ) |
475901af | 42 | * use_far_sets (stored in bit 17 of layout ) |
8bce6d35 | 43 | * use_far_sets_bugfixed (stored in bit 18 of layout ) |
1da177e4 | 44 | * |
475901af JB |
45 | * The data to be stored is divided into chunks using chunksize. Each device |
46 | * is divided into far_copies sections. In each section, chunks are laid out | |
47 | * in a style similar to raid0, but near_copies copies of each chunk is stored | |
48 | * (each on a different drive). The starting device for each section is offset | |
49 | * near_copies from the starting device of the previous section. Thus there | |
50 | * are (near_copies * far_copies) of each chunk, and each is on a different | |
51 | * drive. near_copies and far_copies must be at least one, and their product | |
52 | * is at most raid_disks. | |
c93983bf N |
53 | * |
54 | * If far_offset is true, then the far_copies are handled a bit differently. | |
475901af JB |
55 | * The copies are still in different stripes, but instead of being very far |
56 | * apart on disk, there are adjacent stripes. | |
57 | * | |
58 | * The far and offset algorithms are handled slightly differently if | |
59 | * 'use_far_sets' is true. In this case, the array's devices are grouped into | |
60 | * sets that are (near_copies * far_copies) in size. The far copied stripes | |
61 | * are still shifted by 'near_copies' devices, but this shifting stays confined | |
62 | * to the set rather than the entire array. This is done to improve the number | |
63 | * of device combinations that can fail without causing the array to fail. | |
64 | * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk | |
65 | * on a device): | |
66 | * A B C D A B C D E | |
67 | * ... ... | |
68 | * D A B C E A B C D | |
69 | * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s): | |
70 | * [A B] [C D] [A B] [C D E] | |
71 | * |...| |...| |...| | ... | | |
72 | * [B A] [D C] [B A] [E C D] | |
1da177e4 LT |
73 | */ |
74 | ||
75 | /* | |
76 | * Number of guaranteed r10bios in case of extreme VM load: | |
77 | */ | |
78 | #define NR_RAID10_BIOS 256 | |
79 | ||
473e87ce JB |
80 | /* when we get a read error on a read-only array, we redirect to another |
81 | * device without failing the first device, or trying to over-write to | |
82 | * correct the read error. To keep track of bad blocks on a per-bio | |
83 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
84 | */ | |
85 | #define IO_BLOCKED ((struct bio *)1) | |
86 | /* When we successfully write to a known bad-block, we need to remove the | |
87 | * bad-block marking which must be done from process context. So we record | |
88 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
89 | */ | |
90 | #define IO_MADE_GOOD ((struct bio *)2) | |
91 | ||
92 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
93 | ||
94 | /* When there are this many requests queued to be written by | |
34db0cd6 N |
95 | * the raid10 thread, we become 'congested' to provide back-pressure |
96 | * for writeback. | |
97 | */ | |
98 | static int max_queued_requests = 1024; | |
99 | ||
e879a879 N |
100 | static void allow_barrier(struct r10conf *conf); |
101 | static void lower_barrier(struct r10conf *conf); | |
635f6416 | 102 | static int _enough(struct r10conf *conf, int previous, int ignore); |
3ea7daa5 N |
103 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, |
104 | int *skipped); | |
105 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio); | |
4246a0b6 | 106 | static void end_reshape_write(struct bio *bio); |
3ea7daa5 | 107 | static void end_reshape(struct r10conf *conf); |
0a27ec96 | 108 | |
578b54ad N |
109 | #define raid10_log(md, fmt, args...) \ |
110 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid10 " fmt, ##args); } while (0) | |
111 | ||
dd0fc66f | 112 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 113 | { |
e879a879 | 114 | struct r10conf *conf = data; |
9f2c9d12 | 115 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 116 | |
69335ef3 N |
117 | /* allocate a r10bio with room for raid_disks entries in the |
118 | * bios array */ | |
7eaceacc | 119 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
120 | } |
121 | ||
122 | static void r10bio_pool_free(void *r10_bio, void *data) | |
123 | { | |
124 | kfree(r10_bio); | |
125 | } | |
126 | ||
0310fa21 | 127 | /* Maximum size of each resync request */ |
1da177e4 | 128 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 129 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
130 | /* amount of memory to reserve for resync requests */ |
131 | #define RESYNC_WINDOW (1024*1024) | |
132 | /* maximum number of concurrent requests, memory permitting */ | |
133 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
134 | |
135 | /* | |
136 | * When performing a resync, we need to read and compare, so | |
137 | * we need as many pages are there are copies. | |
138 | * When performing a recovery, we need 2 bios, one for read, | |
139 | * one for write (we recover only one drive per r10buf) | |
140 | * | |
141 | */ | |
dd0fc66f | 142 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 143 | { |
e879a879 | 144 | struct r10conf *conf = data; |
1da177e4 | 145 | struct page *page; |
9f2c9d12 | 146 | struct r10bio *r10_bio; |
1da177e4 LT |
147 | struct bio *bio; |
148 | int i, j; | |
149 | int nalloc; | |
150 | ||
151 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 152 | if (!r10_bio) |
1da177e4 | 153 | return NULL; |
1da177e4 | 154 | |
3ea7daa5 N |
155 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || |
156 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
1da177e4 LT |
157 | nalloc = conf->copies; /* resync */ |
158 | else | |
159 | nalloc = 2; /* recovery */ | |
160 | ||
161 | /* | |
162 | * Allocate bios. | |
163 | */ | |
164 | for (j = nalloc ; j-- ; ) { | |
6746557f | 165 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
166 | if (!bio) |
167 | goto out_free_bio; | |
168 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
169 | if (!conf->have_replacement) |
170 | continue; | |
171 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
172 | if (!bio) | |
173 | goto out_free_bio; | |
174 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
175 | } |
176 | /* | |
177 | * Allocate RESYNC_PAGES data pages and attach them | |
178 | * where needed. | |
179 | */ | |
180 | for (j = 0 ; j < nalloc; j++) { | |
69335ef3 | 181 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
1da177e4 LT |
182 | bio = r10_bio->devs[j].bio; |
183 | for (i = 0; i < RESYNC_PAGES; i++) { | |
3ea7daa5 N |
184 | if (j > 0 && !test_bit(MD_RECOVERY_SYNC, |
185 | &conf->mddev->recovery)) { | |
186 | /* we can share bv_page's during recovery | |
187 | * and reshape */ | |
c65060ad NK |
188 | struct bio *rbio = r10_bio->devs[0].bio; |
189 | page = rbio->bi_io_vec[i].bv_page; | |
190 | get_page(page); | |
191 | } else | |
192 | page = alloc_page(gfp_flags); | |
1da177e4 LT |
193 | if (unlikely(!page)) |
194 | goto out_free_pages; | |
195 | ||
196 | bio->bi_io_vec[i].bv_page = page; | |
69335ef3 N |
197 | if (rbio) |
198 | rbio->bi_io_vec[i].bv_page = page; | |
1da177e4 LT |
199 | } |
200 | } | |
201 | ||
202 | return r10_bio; | |
203 | ||
204 | out_free_pages: | |
205 | for ( ; i > 0 ; i--) | |
1345b1d8 | 206 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
207 | while (j--) |
208 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 209 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
5fdd2cf8 | 210 | j = 0; |
1da177e4 | 211 | out_free_bio: |
5fdd2cf8 | 212 | for ( ; j < nalloc; j++) { |
213 | if (r10_bio->devs[j].bio) | |
214 | bio_put(r10_bio->devs[j].bio); | |
69335ef3 N |
215 | if (r10_bio->devs[j].repl_bio) |
216 | bio_put(r10_bio->devs[j].repl_bio); | |
217 | } | |
1da177e4 LT |
218 | r10bio_pool_free(r10_bio, conf); |
219 | return NULL; | |
220 | } | |
221 | ||
222 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
223 | { | |
224 | int i; | |
e879a879 | 225 | struct r10conf *conf = data; |
9f2c9d12 | 226 | struct r10bio *r10bio = __r10_bio; |
1da177e4 LT |
227 | int j; |
228 | ||
229 | for (j=0; j < conf->copies; j++) { | |
230 | struct bio *bio = r10bio->devs[j].bio; | |
231 | if (bio) { | |
232 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 233 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
234 | bio->bi_io_vec[i].bv_page = NULL; |
235 | } | |
236 | bio_put(bio); | |
237 | } | |
69335ef3 N |
238 | bio = r10bio->devs[j].repl_bio; |
239 | if (bio) | |
240 | bio_put(bio); | |
1da177e4 LT |
241 | } |
242 | r10bio_pool_free(r10bio, conf); | |
243 | } | |
244 | ||
e879a879 | 245 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
246 | { |
247 | int i; | |
248 | ||
249 | for (i = 0; i < conf->copies; i++) { | |
250 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 251 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
252 | bio_put(*bio); |
253 | *bio = NULL; | |
69335ef3 N |
254 | bio = &r10_bio->devs[i].repl_bio; |
255 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
256 | bio_put(*bio); | |
257 | *bio = NULL; | |
1da177e4 LT |
258 | } |
259 | } | |
260 | ||
9f2c9d12 | 261 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 262 | { |
e879a879 | 263 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 264 | |
1da177e4 LT |
265 | put_all_bios(conf, r10_bio); |
266 | mempool_free(r10_bio, conf->r10bio_pool); | |
267 | } | |
268 | ||
9f2c9d12 | 269 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 270 | { |
e879a879 | 271 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
272 | |
273 | mempool_free(r10_bio, conf->r10buf_pool); | |
274 | ||
0a27ec96 | 275 | lower_barrier(conf); |
1da177e4 LT |
276 | } |
277 | ||
9f2c9d12 | 278 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
279 | { |
280 | unsigned long flags; | |
fd01b88c | 281 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 282 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
283 | |
284 | spin_lock_irqsave(&conf->device_lock, flags); | |
285 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 286 | conf->nr_queued ++; |
1da177e4 LT |
287 | spin_unlock_irqrestore(&conf->device_lock, flags); |
288 | ||
388667be AJ |
289 | /* wake up frozen array... */ |
290 | wake_up(&conf->wait_barrier); | |
291 | ||
1da177e4 LT |
292 | md_wakeup_thread(mddev->thread); |
293 | } | |
294 | ||
295 | /* | |
296 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
297 | * operation and are ready to return a success/failure code to the buffer | |
298 | * cache layer. | |
299 | */ | |
9f2c9d12 | 300 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
301 | { |
302 | struct bio *bio = r10_bio->master_bio; | |
856e08e2 | 303 | int done; |
e879a879 | 304 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 305 | |
856e08e2 N |
306 | if (bio->bi_phys_segments) { |
307 | unsigned long flags; | |
308 | spin_lock_irqsave(&conf->device_lock, flags); | |
309 | bio->bi_phys_segments--; | |
310 | done = (bio->bi_phys_segments == 0); | |
311 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
312 | } else | |
313 | done = 1; | |
314 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4246a0b6 | 315 | bio->bi_error = -EIO; |
856e08e2 | 316 | if (done) { |
4246a0b6 | 317 | bio_endio(bio); |
856e08e2 N |
318 | /* |
319 | * Wake up any possible resync thread that waits for the device | |
320 | * to go idle. | |
321 | */ | |
322 | allow_barrier(conf); | |
323 | } | |
1da177e4 LT |
324 | free_r10bio(r10_bio); |
325 | } | |
326 | ||
327 | /* | |
328 | * Update disk head position estimator based on IRQ completion info. | |
329 | */ | |
9f2c9d12 | 330 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 331 | { |
e879a879 | 332 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
333 | |
334 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
335 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
336 | } | |
337 | ||
778ca018 NK |
338 | /* |
339 | * Find the disk number which triggered given bio | |
340 | */ | |
e879a879 | 341 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 342 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
343 | { |
344 | int slot; | |
69335ef3 | 345 | int repl = 0; |
778ca018 | 346 | |
69335ef3 | 347 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
348 | if (r10_bio->devs[slot].bio == bio) |
349 | break; | |
69335ef3 N |
350 | if (r10_bio->devs[slot].repl_bio == bio) { |
351 | repl = 1; | |
352 | break; | |
353 | } | |
354 | } | |
778ca018 NK |
355 | |
356 | BUG_ON(slot == conf->copies); | |
357 | update_head_pos(slot, r10_bio); | |
358 | ||
749c55e9 N |
359 | if (slotp) |
360 | *slotp = slot; | |
69335ef3 N |
361 | if (replp) |
362 | *replp = repl; | |
778ca018 NK |
363 | return r10_bio->devs[slot].devnum; |
364 | } | |
365 | ||
4246a0b6 | 366 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 367 | { |
4246a0b6 | 368 | int uptodate = !bio->bi_error; |
9f2c9d12 | 369 | struct r10bio *r10_bio = bio->bi_private; |
1da177e4 | 370 | int slot, dev; |
abbf098e | 371 | struct md_rdev *rdev; |
e879a879 | 372 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 373 | |
1da177e4 LT |
374 | slot = r10_bio->read_slot; |
375 | dev = r10_bio->devs[slot].devnum; | |
abbf098e | 376 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
377 | /* |
378 | * this branch is our 'one mirror IO has finished' event handler: | |
379 | */ | |
4443ae10 N |
380 | update_head_pos(slot, r10_bio); |
381 | ||
382 | if (uptodate) { | |
1da177e4 LT |
383 | /* |
384 | * Set R10BIO_Uptodate in our master bio, so that | |
385 | * we will return a good error code to the higher | |
386 | * levels even if IO on some other mirrored buffer fails. | |
387 | * | |
388 | * The 'master' represents the composite IO operation to | |
389 | * user-side. So if something waits for IO, then it will | |
390 | * wait for the 'master' bio. | |
391 | */ | |
392 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
393 | } else { |
394 | /* If all other devices that store this block have | |
395 | * failed, we want to return the error upwards rather | |
396 | * than fail the last device. Here we redefine | |
397 | * "uptodate" to mean "Don't want to retry" | |
398 | */ | |
635f6416 N |
399 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
400 | rdev->raid_disk)) | |
fae8cc5e | 401 | uptodate = 1; |
fae8cc5e N |
402 | } |
403 | if (uptodate) { | |
1da177e4 | 404 | raid_end_bio_io(r10_bio); |
abbf098e | 405 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 406 | } else { |
1da177e4 | 407 | /* |
7c4e06ff | 408 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
409 | */ |
410 | char b[BDEVNAME_SIZE]; | |
08464e09 | 411 | pr_err_ratelimited("md/raid10:%s: %s: rescheduling sector %llu\n", |
8bda470e | 412 | mdname(conf->mddev), |
abbf098e | 413 | bdevname(rdev->bdev, b), |
8bda470e | 414 | (unsigned long long)r10_bio->sector); |
856e08e2 | 415 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
416 | reschedule_retry(r10_bio); |
417 | } | |
1da177e4 LT |
418 | } |
419 | ||
9f2c9d12 | 420 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
421 | { |
422 | /* clear the bitmap if all writes complete successfully */ | |
423 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
424 | r10_bio->sectors, | |
425 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
426 | 0); | |
427 | md_write_end(r10_bio->mddev); | |
428 | } | |
429 | ||
9f2c9d12 | 430 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
431 | { |
432 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
433 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
434 | reschedule_retry(r10_bio); | |
435 | else { | |
436 | close_write(r10_bio); | |
437 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
438 | reschedule_retry(r10_bio); | |
439 | else | |
440 | raid_end_bio_io(r10_bio); | |
441 | } | |
442 | } | |
443 | } | |
444 | ||
4246a0b6 | 445 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 446 | { |
9f2c9d12 | 447 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 448 | int dev; |
749c55e9 | 449 | int dec_rdev = 1; |
e879a879 | 450 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 451 | int slot, repl; |
4ca40c2c | 452 | struct md_rdev *rdev = NULL; |
579ed34f SL |
453 | bool discard_error; |
454 | ||
455 | discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; | |
1da177e4 | 456 | |
475b0321 | 457 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 458 | |
475b0321 N |
459 | if (repl) |
460 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
461 | if (!rdev) { |
462 | smp_rmb(); | |
463 | repl = 0; | |
475b0321 | 464 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 465 | } |
1da177e4 LT |
466 | /* |
467 | * this branch is our 'one mirror IO has finished' event handler: | |
468 | */ | |
579ed34f | 469 | if (bio->bi_error && !discard_error) { |
475b0321 N |
470 | if (repl) |
471 | /* Never record new bad blocks to replacement, | |
472 | * just fail it. | |
473 | */ | |
474 | md_error(rdev->mddev, rdev); | |
475 | else { | |
476 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
477 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
478 | set_bit(MD_RECOVERY_NEEDED, | |
479 | &rdev->mddev->recovery); | |
475b0321 N |
480 | set_bit(R10BIO_WriteError, &r10_bio->state); |
481 | dec_rdev = 0; | |
482 | } | |
749c55e9 | 483 | } else { |
1da177e4 LT |
484 | /* |
485 | * Set R10BIO_Uptodate in our master bio, so that | |
486 | * we will return a good error code for to the higher | |
487 | * levels even if IO on some other mirrored buffer fails. | |
488 | * | |
489 | * The 'master' represents the composite IO operation to | |
490 | * user-side. So if something waits for IO, then it will | |
491 | * wait for the 'master' bio. | |
492 | */ | |
749c55e9 N |
493 | sector_t first_bad; |
494 | int bad_sectors; | |
495 | ||
3056e3ae AL |
496 | /* |
497 | * Do not set R10BIO_Uptodate if the current device is | |
498 | * rebuilding or Faulty. This is because we cannot use | |
499 | * such device for properly reading the data back (we could | |
500 | * potentially use it, if the current write would have felt | |
501 | * before rdev->recovery_offset, but for simplicity we don't | |
502 | * check this here. | |
503 | */ | |
504 | if (test_bit(In_sync, &rdev->flags) && | |
505 | !test_bit(Faulty, &rdev->flags)) | |
506 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 507 | |
749c55e9 | 508 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 509 | if (is_badblock(rdev, |
749c55e9 N |
510 | r10_bio->devs[slot].addr, |
511 | r10_bio->sectors, | |
579ed34f | 512 | &first_bad, &bad_sectors) && !discard_error) { |
749c55e9 | 513 | bio_put(bio); |
475b0321 N |
514 | if (repl) |
515 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
516 | else | |
517 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
518 | dec_rdev = 0; |
519 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
520 | } | |
521 | } | |
522 | ||
1da177e4 LT |
523 | /* |
524 | * | |
525 | * Let's see if all mirrored write operations have finished | |
526 | * already. | |
527 | */ | |
19d5f834 | 528 | one_write_done(r10_bio); |
749c55e9 | 529 | if (dec_rdev) |
884162df | 530 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
531 | } |
532 | ||
1da177e4 LT |
533 | /* |
534 | * RAID10 layout manager | |
25985edc | 535 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
536 | * parameters: near_copies and far_copies. |
537 | * near_copies * far_copies must be <= raid_disks. | |
538 | * Normally one of these will be 1. | |
539 | * If both are 1, we get raid0. | |
540 | * If near_copies == raid_disks, we get raid1. | |
541 | * | |
25985edc | 542 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
543 | * first chunk, followed by near_copies copies of the next chunk and |
544 | * so on. | |
545 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
546 | * as described above, we start again with a device offset of near_copies. | |
547 | * So we effectively have another copy of the whole array further down all | |
548 | * the drives, but with blocks on different drives. | |
549 | * With this layout, and block is never stored twice on the one device. | |
550 | * | |
551 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 552 | * on each device that it is on. |
1da177e4 LT |
553 | * |
554 | * raid10_find_virt does the reverse mapping, from a device and a | |
555 | * sector offset to a virtual address | |
556 | */ | |
557 | ||
f8c9e74f | 558 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
559 | { |
560 | int n,f; | |
561 | sector_t sector; | |
562 | sector_t chunk; | |
563 | sector_t stripe; | |
564 | int dev; | |
1da177e4 | 565 | int slot = 0; |
9a3152ab JB |
566 | int last_far_set_start, last_far_set_size; |
567 | ||
568 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
569 | last_far_set_start *= geo->far_set_size; | |
570 | ||
571 | last_far_set_size = geo->far_set_size; | |
572 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
573 | |
574 | /* now calculate first sector/dev */ | |
5cf00fcd N |
575 | chunk = r10bio->sector >> geo->chunk_shift; |
576 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 577 | |
5cf00fcd | 578 | chunk *= geo->near_copies; |
1da177e4 | 579 | stripe = chunk; |
5cf00fcd N |
580 | dev = sector_div(stripe, geo->raid_disks); |
581 | if (geo->far_offset) | |
582 | stripe *= geo->far_copies; | |
1da177e4 | 583 | |
5cf00fcd | 584 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
585 | |
586 | /* and calculate all the others */ | |
5cf00fcd | 587 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 588 | int d = dev; |
475901af | 589 | int set; |
1da177e4 | 590 | sector_t s = sector; |
1da177e4 | 591 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 592 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
593 | slot++; |
594 | ||
5cf00fcd | 595 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 596 | set = d / geo->far_set_size; |
5cf00fcd | 597 | d += geo->near_copies; |
475901af | 598 | |
9a3152ab JB |
599 | if ((geo->raid_disks % geo->far_set_size) && |
600 | (d > last_far_set_start)) { | |
601 | d -= last_far_set_start; | |
602 | d %= last_far_set_size; | |
603 | d += last_far_set_start; | |
604 | } else { | |
605 | d %= geo->far_set_size; | |
606 | d += geo->far_set_size * set; | |
607 | } | |
5cf00fcd | 608 | s += geo->stride; |
1da177e4 LT |
609 | r10bio->devs[slot].devnum = d; |
610 | r10bio->devs[slot].addr = s; | |
611 | slot++; | |
612 | } | |
613 | dev++; | |
5cf00fcd | 614 | if (dev >= geo->raid_disks) { |
1da177e4 | 615 | dev = 0; |
5cf00fcd | 616 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
617 | } |
618 | } | |
f8c9e74f N |
619 | } |
620 | ||
621 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
622 | { | |
623 | struct geom *geo = &conf->geo; | |
624 | ||
625 | if (conf->reshape_progress != MaxSector && | |
626 | ((r10bio->sector >= conf->reshape_progress) != | |
627 | conf->mddev->reshape_backwards)) { | |
628 | set_bit(R10BIO_Previous, &r10bio->state); | |
629 | geo = &conf->prev; | |
630 | } else | |
631 | clear_bit(R10BIO_Previous, &r10bio->state); | |
632 | ||
633 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
634 | } |
635 | ||
e879a879 | 636 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
637 | { |
638 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
639 | /* Never use conf->prev as this is only called during resync |
640 | * or recovery, so reshape isn't happening | |
641 | */ | |
5cf00fcd | 642 | struct geom *geo = &conf->geo; |
475901af JB |
643 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
644 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
645 | int last_far_set_start; |
646 | ||
647 | if (geo->raid_disks % geo->far_set_size) { | |
648 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
649 | last_far_set_start *= geo->far_set_size; | |
650 | ||
651 | if (dev >= last_far_set_start) { | |
652 | far_set_size = geo->far_set_size; | |
653 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
654 | far_set_start = last_far_set_start; | |
655 | } | |
656 | } | |
1da177e4 | 657 | |
5cf00fcd N |
658 | offset = sector & geo->chunk_mask; |
659 | if (geo->far_offset) { | |
c93983bf | 660 | int fc; |
5cf00fcd N |
661 | chunk = sector >> geo->chunk_shift; |
662 | fc = sector_div(chunk, geo->far_copies); | |
663 | dev -= fc * geo->near_copies; | |
475901af JB |
664 | if (dev < far_set_start) |
665 | dev += far_set_size; | |
c93983bf | 666 | } else { |
5cf00fcd N |
667 | while (sector >= geo->stride) { |
668 | sector -= geo->stride; | |
475901af JB |
669 | if (dev < (geo->near_copies + far_set_start)) |
670 | dev += far_set_size - geo->near_copies; | |
c93983bf | 671 | else |
5cf00fcd | 672 | dev -= geo->near_copies; |
c93983bf | 673 | } |
5cf00fcd | 674 | chunk = sector >> geo->chunk_shift; |
c93983bf | 675 | } |
5cf00fcd N |
676 | vchunk = chunk * geo->raid_disks + dev; |
677 | sector_div(vchunk, geo->near_copies); | |
678 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
679 | } |
680 | ||
1da177e4 LT |
681 | /* |
682 | * This routine returns the disk from which the requested read should | |
683 | * be done. There is a per-array 'next expected sequential IO' sector | |
684 | * number - if this matches on the next IO then we use the last disk. | |
685 | * There is also a per-disk 'last know head position' sector that is | |
686 | * maintained from IRQ contexts, both the normal and the resync IO | |
687 | * completion handlers update this position correctly. If there is no | |
688 | * perfect sequential match then we pick the disk whose head is closest. | |
689 | * | |
690 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
691 | * because position is mirror, not device based. | |
692 | * | |
693 | * The rdev for the device selected will have nr_pending incremented. | |
694 | */ | |
695 | ||
696 | /* | |
697 | * FIXME: possibly should rethink readbalancing and do it differently | |
698 | * depending on near_copies / far_copies geometry. | |
699 | */ | |
96c3fd1f N |
700 | static struct md_rdev *read_balance(struct r10conf *conf, |
701 | struct r10bio *r10_bio, | |
702 | int *max_sectors) | |
1da177e4 | 703 | { |
af3a2cd6 | 704 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 705 | int disk, slot; |
856e08e2 N |
706 | int sectors = r10_bio->sectors; |
707 | int best_good_sectors; | |
56d99121 | 708 | sector_t new_distance, best_dist; |
3bbae04b | 709 | struct md_rdev *best_rdev, *rdev = NULL; |
56d99121 N |
710 | int do_balance; |
711 | int best_slot; | |
5cf00fcd | 712 | struct geom *geo = &conf->geo; |
1da177e4 LT |
713 | |
714 | raid10_find_phys(conf, r10_bio); | |
715 | rcu_read_lock(); | |
856e08e2 | 716 | sectors = r10_bio->sectors; |
56d99121 | 717 | best_slot = -1; |
abbf098e | 718 | best_rdev = NULL; |
56d99121 | 719 | best_dist = MaxSector; |
856e08e2 | 720 | best_good_sectors = 0; |
56d99121 | 721 | do_balance = 1; |
8d3ca83d | 722 | clear_bit(R10BIO_FailFast, &r10_bio->state); |
1da177e4 LT |
723 | /* |
724 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
725 | * device if no resync is going on (recovery is ok), or below |
726 | * the resync window. We take the first readable disk when | |
727 | * above the resync window. | |
1da177e4 LT |
728 | */ |
729 | if (conf->mddev->recovery_cp < MaxSector | |
56d99121 N |
730 | && (this_sector + sectors >= conf->next_resync)) |
731 | do_balance = 0; | |
1da177e4 | 732 | |
56d99121 | 733 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
734 | sector_t first_bad; |
735 | int bad_sectors; | |
736 | sector_t dev_sector; | |
737 | ||
56d99121 N |
738 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
739 | continue; | |
1da177e4 | 740 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
741 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
742 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
743 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
744 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 | 745 | if (rdev == NULL || |
8ae12666 | 746 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
747 | continue; |
748 | if (!test_bit(In_sync, &rdev->flags) && | |
749 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
750 | continue; |
751 | ||
856e08e2 N |
752 | dev_sector = r10_bio->devs[slot].addr; |
753 | if (is_badblock(rdev, dev_sector, sectors, | |
754 | &first_bad, &bad_sectors)) { | |
755 | if (best_dist < MaxSector) | |
756 | /* Already have a better slot */ | |
757 | continue; | |
758 | if (first_bad <= dev_sector) { | |
759 | /* Cannot read here. If this is the | |
760 | * 'primary' device, then we must not read | |
761 | * beyond 'bad_sectors' from another device. | |
762 | */ | |
763 | bad_sectors -= (dev_sector - first_bad); | |
764 | if (!do_balance && sectors > bad_sectors) | |
765 | sectors = bad_sectors; | |
766 | if (best_good_sectors > sectors) | |
767 | best_good_sectors = sectors; | |
768 | } else { | |
769 | sector_t good_sectors = | |
770 | first_bad - dev_sector; | |
771 | if (good_sectors > best_good_sectors) { | |
772 | best_good_sectors = good_sectors; | |
773 | best_slot = slot; | |
abbf098e | 774 | best_rdev = rdev; |
856e08e2 N |
775 | } |
776 | if (!do_balance) | |
777 | /* Must read from here */ | |
778 | break; | |
779 | } | |
780 | continue; | |
781 | } else | |
782 | best_good_sectors = sectors; | |
783 | ||
56d99121 N |
784 | if (!do_balance) |
785 | break; | |
1da177e4 | 786 | |
8d3ca83d N |
787 | if (best_slot >= 0) |
788 | /* At least 2 disks to choose from so failfast is OK */ | |
789 | set_bit(R10BIO_FailFast, &r10_bio->state); | |
22dfdf52 N |
790 | /* This optimisation is debatable, and completely destroys |
791 | * sequential read speed for 'far copies' arrays. So only | |
792 | * keep it for 'near' arrays, and review those later. | |
793 | */ | |
5cf00fcd | 794 | if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
8d3ca83d | 795 | new_distance = 0; |
8ed3a195 KS |
796 | |
797 | /* for far > 1 always use the lowest address */ | |
8d3ca83d | 798 | else if (geo->far_copies > 1) |
56d99121 | 799 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 800 | else |
56d99121 N |
801 | new_distance = abs(r10_bio->devs[slot].addr - |
802 | conf->mirrors[disk].head_position); | |
803 | if (new_distance < best_dist) { | |
804 | best_dist = new_distance; | |
805 | best_slot = slot; | |
abbf098e | 806 | best_rdev = rdev; |
1da177e4 LT |
807 | } |
808 | } | |
abbf098e | 809 | if (slot >= conf->copies) { |
56d99121 | 810 | slot = best_slot; |
abbf098e N |
811 | rdev = best_rdev; |
812 | } | |
1da177e4 | 813 | |
56d99121 | 814 | if (slot >= 0) { |
56d99121 | 815 | atomic_inc(&rdev->nr_pending); |
56d99121 N |
816 | r10_bio->read_slot = slot; |
817 | } else | |
96c3fd1f | 818 | rdev = NULL; |
1da177e4 | 819 | rcu_read_unlock(); |
856e08e2 | 820 | *max_sectors = best_good_sectors; |
1da177e4 | 821 | |
96c3fd1f | 822 | return rdev; |
1da177e4 LT |
823 | } |
824 | ||
5c675f83 | 825 | static int raid10_congested(struct mddev *mddev, int bits) |
0d129228 | 826 | { |
e879a879 | 827 | struct r10conf *conf = mddev->private; |
0d129228 N |
828 | int i, ret = 0; |
829 | ||
4452226e | 830 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
831 | conf->pending_count >= max_queued_requests) |
832 | return 1; | |
833 | ||
0d129228 | 834 | rcu_read_lock(); |
f8c9e74f N |
835 | for (i = 0; |
836 | (i < conf->geo.raid_disks || i < conf->prev.raid_disks) | |
837 | && ret == 0; | |
838 | i++) { | |
3cb03002 | 839 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 840 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 841 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
842 | |
843 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
844 | } | |
845 | } | |
846 | rcu_read_unlock(); | |
847 | return ret; | |
848 | } | |
849 | ||
e879a879 | 850 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
851 | { |
852 | /* Any writes that have been queued but are awaiting | |
853 | * bitmap updates get flushed here. | |
a35e63ef | 854 | */ |
a35e63ef N |
855 | spin_lock_irq(&conf->device_lock); |
856 | ||
857 | if (conf->pending_bio_list.head) { | |
858 | struct bio *bio; | |
859 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 860 | conf->pending_count = 0; |
a35e63ef N |
861 | spin_unlock_irq(&conf->device_lock); |
862 | /* flush any pending bitmap writes to disk | |
863 | * before proceeding w/ I/O */ | |
864 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 865 | wake_up(&conf->wait_barrier); |
a35e63ef N |
866 | |
867 | while (bio) { /* submit pending writes */ | |
868 | struct bio *next = bio->bi_next; | |
a9ae93c8 | 869 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 870 | bio->bi_next = NULL; |
a9ae93c8 N |
871 | bio->bi_bdev = rdev->bdev; |
872 | if (test_bit(Faulty, &rdev->flags)) { | |
873 | bio->bi_error = -EIO; | |
874 | bio_endio(bio); | |
875 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
876 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
532a2a3f | 877 | /* Just ignore it */ |
4246a0b6 | 878 | bio_endio(bio); |
532a2a3f SL |
879 | else |
880 | generic_make_request(bio); | |
a35e63ef N |
881 | bio = next; |
882 | } | |
a35e63ef N |
883 | } else |
884 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 885 | } |
7eaceacc | 886 | |
0a27ec96 N |
887 | /* Barriers.... |
888 | * Sometimes we need to suspend IO while we do something else, | |
889 | * either some resync/recovery, or reconfigure the array. | |
890 | * To do this we raise a 'barrier'. | |
891 | * The 'barrier' is a counter that can be raised multiple times | |
892 | * to count how many activities are happening which preclude | |
893 | * normal IO. | |
894 | * We can only raise the barrier if there is no pending IO. | |
895 | * i.e. if nr_pending == 0. | |
896 | * We choose only to raise the barrier if no-one is waiting for the | |
897 | * barrier to go down. This means that as soon as an IO request | |
898 | * is ready, no other operations which require a barrier will start | |
899 | * until the IO request has had a chance. | |
900 | * | |
901 | * So: regular IO calls 'wait_barrier'. When that returns there | |
902 | * is no backgroup IO happening, It must arrange to call | |
903 | * allow_barrier when it has finished its IO. | |
904 | * backgroup IO calls must call raise_barrier. Once that returns | |
905 | * there is no normal IO happeing. It must arrange to call | |
906 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 907 | */ |
1da177e4 | 908 | |
e879a879 | 909 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 910 | { |
6cce3b23 | 911 | BUG_ON(force && !conf->barrier); |
1da177e4 | 912 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 913 | |
6cce3b23 N |
914 | /* Wait until no block IO is waiting (unless 'force') */ |
915 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
eed8c02e | 916 | conf->resync_lock); |
0a27ec96 N |
917 | |
918 | /* block any new IO from starting */ | |
919 | conf->barrier++; | |
920 | ||
c3b328ac | 921 | /* Now wait for all pending IO to complete */ |
0a27ec96 | 922 | wait_event_lock_irq(conf->wait_barrier, |
0e5313e2 | 923 | !atomic_read(&conf->nr_pending) && conf->barrier < RESYNC_DEPTH, |
eed8c02e | 924 | conf->resync_lock); |
0a27ec96 N |
925 | |
926 | spin_unlock_irq(&conf->resync_lock); | |
927 | } | |
928 | ||
e879a879 | 929 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
930 | { |
931 | unsigned long flags; | |
932 | spin_lock_irqsave(&conf->resync_lock, flags); | |
933 | conf->barrier--; | |
934 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
935 | wake_up(&conf->wait_barrier); | |
936 | } | |
937 | ||
e879a879 | 938 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
939 | { |
940 | spin_lock_irq(&conf->resync_lock); | |
941 | if (conf->barrier) { | |
942 | conf->nr_waiting++; | |
d6b42dcb N |
943 | /* Wait for the barrier to drop. |
944 | * However if there are already pending | |
945 | * requests (preventing the barrier from | |
946 | * rising completely), and the | |
947 | * pre-process bio queue isn't empty, | |
948 | * then don't wait, as we need to empty | |
949 | * that queue to get the nr_pending | |
950 | * count down. | |
951 | */ | |
578b54ad | 952 | raid10_log(conf->mddev, "wait barrier"); |
d6b42dcb N |
953 | wait_event_lock_irq(conf->wait_barrier, |
954 | !conf->barrier || | |
0e5313e2 | 955 | (atomic_read(&conf->nr_pending) && |
d6b42dcb N |
956 | current->bio_list && |
957 | !bio_list_empty(current->bio_list)), | |
eed8c02e | 958 | conf->resync_lock); |
0a27ec96 | 959 | conf->nr_waiting--; |
0e5313e2 TM |
960 | if (!conf->nr_waiting) |
961 | wake_up(&conf->wait_barrier); | |
1da177e4 | 962 | } |
0e5313e2 | 963 | atomic_inc(&conf->nr_pending); |
1da177e4 LT |
964 | spin_unlock_irq(&conf->resync_lock); |
965 | } | |
966 | ||
e879a879 | 967 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 | 968 | { |
0e5313e2 TM |
969 | if ((atomic_dec_and_test(&conf->nr_pending)) || |
970 | (conf->array_freeze_pending)) | |
971 | wake_up(&conf->wait_barrier); | |
0a27ec96 N |
972 | } |
973 | ||
e2d59925 | 974 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
975 | { |
976 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 977 | * go quiet. |
4443ae10 | 978 | * We increment barrier and nr_waiting, and then |
e2d59925 | 979 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
980 | * This is called in the context of one normal IO request |
981 | * that has failed. Thus any sync request that might be pending | |
982 | * will be blocked by nr_pending, and we need to wait for | |
983 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 984 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
985 | * must match the number of pending IOs (nr_pending) before |
986 | * we continue. | |
4443ae10 N |
987 | */ |
988 | spin_lock_irq(&conf->resync_lock); | |
0e5313e2 | 989 | conf->array_freeze_pending++; |
4443ae10 N |
990 | conf->barrier++; |
991 | conf->nr_waiting++; | |
eed8c02e | 992 | wait_event_lock_irq_cmd(conf->wait_barrier, |
0e5313e2 | 993 | atomic_read(&conf->nr_pending) == conf->nr_queued+extra, |
eed8c02e LC |
994 | conf->resync_lock, |
995 | flush_pending_writes(conf)); | |
c3b328ac | 996 | |
0e5313e2 | 997 | conf->array_freeze_pending--; |
4443ae10 N |
998 | spin_unlock_irq(&conf->resync_lock); |
999 | } | |
1000 | ||
e879a879 | 1001 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1002 | { |
1003 | /* reverse the effect of the freeze */ | |
1004 | spin_lock_irq(&conf->resync_lock); | |
1005 | conf->barrier--; | |
1006 | conf->nr_waiting--; | |
1007 | wake_up(&conf->wait_barrier); | |
1008 | spin_unlock_irq(&conf->resync_lock); | |
1009 | } | |
1010 | ||
f8c9e74f N |
1011 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1012 | struct md_rdev *rdev) | |
1013 | { | |
1014 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1015 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1016 | return rdev->data_offset; | |
1017 | else | |
1018 | return rdev->new_data_offset; | |
1019 | } | |
1020 | ||
57c67df4 N |
1021 | struct raid10_plug_cb { |
1022 | struct blk_plug_cb cb; | |
1023 | struct bio_list pending; | |
1024 | int pending_cnt; | |
1025 | }; | |
1026 | ||
1027 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1028 | { | |
1029 | struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb, | |
1030 | cb); | |
1031 | struct mddev *mddev = plug->cb.data; | |
1032 | struct r10conf *conf = mddev->private; | |
1033 | struct bio *bio; | |
1034 | ||
874807a8 | 1035 | if (from_schedule || current->bio_list) { |
57c67df4 N |
1036 | spin_lock_irq(&conf->device_lock); |
1037 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1038 | conf->pending_count += plug->pending_cnt; | |
1039 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1040 | wake_up(&conf->wait_barrier); |
57c67df4 N |
1041 | md_wakeup_thread(mddev->thread); |
1042 | kfree(plug); | |
1043 | return; | |
1044 | } | |
1045 | ||
1046 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1047 | bio = bio_list_get(&plug->pending); | |
1048 | bitmap_unplug(mddev->bitmap); | |
1049 | wake_up(&conf->wait_barrier); | |
1050 | ||
1051 | while (bio) { /* submit pending writes */ | |
1052 | struct bio *next = bio->bi_next; | |
a9ae93c8 | 1053 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
57c67df4 | 1054 | bio->bi_next = NULL; |
a9ae93c8 N |
1055 | bio->bi_bdev = rdev->bdev; |
1056 | if (test_bit(Faulty, &rdev->flags)) { | |
1057 | bio->bi_error = -EIO; | |
1058 | bio_endio(bio); | |
1059 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
1060 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
32f9f570 | 1061 | /* Just ignore it */ |
4246a0b6 | 1062 | bio_endio(bio); |
32f9f570 SL |
1063 | else |
1064 | generic_make_request(bio); | |
57c67df4 N |
1065 | bio = next; |
1066 | } | |
1067 | kfree(plug); | |
1068 | } | |
1069 | ||
20d0189b | 1070 | static void __make_request(struct mddev *mddev, struct bio *bio) |
1da177e4 | 1071 | { |
e879a879 | 1072 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 1073 | struct r10bio *r10_bio; |
1da177e4 LT |
1074 | struct bio *read_bio; |
1075 | int i; | |
796a5cf0 | 1076 | const int op = bio_op(bio); |
a362357b | 1077 | const int rw = bio_data_dir(bio); |
1eff9d32 JA |
1078 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1079 | const unsigned long do_fua = (bio->bi_opf & REQ_FUA); | |
6cce3b23 | 1080 | unsigned long flags; |
3cb03002 | 1081 | struct md_rdev *blocked_rdev; |
57c67df4 N |
1082 | struct blk_plug_cb *cb; |
1083 | struct raid10_plug_cb *plug = NULL; | |
d4432c23 N |
1084 | int sectors_handled; |
1085 | int max_sectors; | |
3ea7daa5 | 1086 | int sectors; |
1da177e4 | 1087 | |
9b622e2b TM |
1088 | md_write_start(mddev, bio); |
1089 | ||
cc13b1d1 N |
1090 | /* |
1091 | * Register the new request and wait if the reconstruction | |
1092 | * thread has put up a bar for new requests. | |
1093 | * Continue immediately if no resync is active currently. | |
1094 | */ | |
1095 | wait_barrier(conf); | |
1096 | ||
aa8b57aa | 1097 | sectors = bio_sectors(bio); |
3ea7daa5 | 1098 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
4f024f37 KO |
1099 | bio->bi_iter.bi_sector < conf->reshape_progress && |
1100 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
3ea7daa5 N |
1101 | /* IO spans the reshape position. Need to wait for |
1102 | * reshape to pass | |
1103 | */ | |
578b54ad | 1104 | raid10_log(conf->mddev, "wait reshape"); |
3ea7daa5 N |
1105 | allow_barrier(conf); |
1106 | wait_event(conf->wait_barrier, | |
4f024f37 KO |
1107 | conf->reshape_progress <= bio->bi_iter.bi_sector || |
1108 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1109 | sectors); | |
3ea7daa5 N |
1110 | wait_barrier(conf); |
1111 | } | |
1112 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
1113 | bio_data_dir(bio) == WRITE && | |
1114 | (mddev->reshape_backwards | |
4f024f37 KO |
1115 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1116 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1117 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1118 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1119 | /* Need to update reshape_position in metadata */ |
1120 | mddev->reshape_position = conf->reshape_progress; | |
85ad1d13 GJ |
1121 | set_mask_bits(&mddev->flags, 0, |
1122 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
3ea7daa5 | 1123 | md_wakeup_thread(mddev->thread); |
578b54ad | 1124 | raid10_log(conf->mddev, "wait reshape metadata"); |
3ea7daa5 N |
1125 | wait_event(mddev->sb_wait, |
1126 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1127 | ||
1128 | conf->reshape_safe = mddev->reshape_position; | |
1129 | } | |
1130 | ||
1da177e4 LT |
1131 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); |
1132 | ||
1133 | r10_bio->master_bio = bio; | |
3ea7daa5 | 1134 | r10_bio->sectors = sectors; |
1da177e4 LT |
1135 | |
1136 | r10_bio->mddev = mddev; | |
4f024f37 | 1137 | r10_bio->sector = bio->bi_iter.bi_sector; |
6cce3b23 | 1138 | r10_bio->state = 0; |
1da177e4 | 1139 | |
856e08e2 N |
1140 | /* We might need to issue multiple reads to different |
1141 | * devices if there are bad blocks around, so we keep | |
1142 | * track of the number of reads in bio->bi_phys_segments. | |
1143 | * If this is 0, there is only one r10_bio and no locking | |
1144 | * will be needed when the request completes. If it is | |
1145 | * non-zero, then it is the number of not-completed requests. | |
1146 | */ | |
1147 | bio->bi_phys_segments = 0; | |
b7c44ed9 | 1148 | bio_clear_flag(bio, BIO_SEG_VALID); |
856e08e2 | 1149 | |
a362357b | 1150 | if (rw == READ) { |
1da177e4 LT |
1151 | /* |
1152 | * read balancing logic: | |
1153 | */ | |
96c3fd1f | 1154 | struct md_rdev *rdev; |
856e08e2 N |
1155 | int slot; |
1156 | ||
1157 | read_again: | |
96c3fd1f N |
1158 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1159 | if (!rdev) { | |
1da177e4 | 1160 | raid_end_bio_io(r10_bio); |
5a7bbad2 | 1161 | return; |
1da177e4 | 1162 | } |
96c3fd1f | 1163 | slot = r10_bio->read_slot; |
1da177e4 | 1164 | |
a167f663 | 1165 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1166 | bio_trim(read_bio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1167 | max_sectors); |
1da177e4 LT |
1168 | |
1169 | r10_bio->devs[slot].bio = read_bio; | |
abbf098e | 1170 | r10_bio->devs[slot].rdev = rdev; |
1da177e4 | 1171 | |
4f024f37 | 1172 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + |
f8c9e74f | 1173 | choose_data_offset(r10_bio, rdev); |
96c3fd1f | 1174 | read_bio->bi_bdev = rdev->bdev; |
1da177e4 | 1175 | read_bio->bi_end_io = raid10_end_read_request; |
796a5cf0 | 1176 | bio_set_op_attrs(read_bio, op, do_sync); |
8d3ca83d N |
1177 | if (test_bit(FailFast, &rdev->flags) && |
1178 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
1179 | read_bio->bi_opf |= MD_FAILFAST; | |
1da177e4 LT |
1180 | read_bio->bi_private = r10_bio; |
1181 | ||
109e3765 N |
1182 | if (mddev->gendisk) |
1183 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1184 | read_bio, disk_devt(mddev->gendisk), | |
1185 | r10_bio->sector); | |
856e08e2 N |
1186 | if (max_sectors < r10_bio->sectors) { |
1187 | /* Could not read all from this device, so we will | |
1188 | * need another r10_bio. | |
1189 | */ | |
b50c259e | 1190 | sectors_handled = (r10_bio->sector + max_sectors |
4f024f37 | 1191 | - bio->bi_iter.bi_sector); |
856e08e2 N |
1192 | r10_bio->sectors = max_sectors; |
1193 | spin_lock_irq(&conf->device_lock); | |
1194 | if (bio->bi_phys_segments == 0) | |
1195 | bio->bi_phys_segments = 2; | |
1196 | else | |
1197 | bio->bi_phys_segments++; | |
b50c259e | 1198 | spin_unlock_irq(&conf->device_lock); |
856e08e2 N |
1199 | /* Cannot call generic_make_request directly |
1200 | * as that will be queued in __generic_make_request | |
1201 | * and subsequent mempool_alloc might block | |
1202 | * waiting for it. so hand bio over to raid10d. | |
1203 | */ | |
1204 | reschedule_retry(r10_bio); | |
1205 | ||
1206 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1207 | ||
1208 | r10_bio->master_bio = bio; | |
aa8b57aa | 1209 | r10_bio->sectors = bio_sectors(bio) - sectors_handled; |
856e08e2 N |
1210 | r10_bio->state = 0; |
1211 | r10_bio->mddev = mddev; | |
4f024f37 KO |
1212 | r10_bio->sector = bio->bi_iter.bi_sector + |
1213 | sectors_handled; | |
856e08e2 N |
1214 | goto read_again; |
1215 | } else | |
1216 | generic_make_request(read_bio); | |
5a7bbad2 | 1217 | return; |
1da177e4 LT |
1218 | } |
1219 | ||
1220 | /* | |
1221 | * WRITE: | |
1222 | */ | |
34db0cd6 N |
1223 | if (conf->pending_count >= max_queued_requests) { |
1224 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1225 | raid10_log(mddev, "wait queued"); |
34db0cd6 N |
1226 | wait_event(conf->wait_barrier, |
1227 | conf->pending_count < max_queued_requests); | |
1228 | } | |
6bfe0b49 | 1229 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1230 | * inc refcount on their rdev. Record them by setting |
1231 | * bios[x] to bio | |
d4432c23 N |
1232 | * If there are known/acknowledged bad blocks on any device |
1233 | * on which we have seen a write error, we want to avoid | |
1234 | * writing to those blocks. This potentially requires several | |
1235 | * writes to write around the bad blocks. Each set of writes | |
1236 | * gets its own r10_bio with a set of bios attached. The number | |
1237 | * of r10_bios is recored in bio->bi_phys_segments just as with | |
1238 | * the read case. | |
1da177e4 | 1239 | */ |
c3b328ac | 1240 | |
69335ef3 | 1241 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1242 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1243 | retry_write: |
cb6969e8 | 1244 | blocked_rdev = NULL; |
1da177e4 | 1245 | rcu_read_lock(); |
d4432c23 N |
1246 | max_sectors = r10_bio->sectors; |
1247 | ||
1da177e4 LT |
1248 | for (i = 0; i < conf->copies; i++) { |
1249 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1250 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1251 | struct md_rdev *rrdev = rcu_dereference( |
1252 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1253 | if (rdev == rrdev) |
1254 | rrdev = NULL; | |
6bfe0b49 DW |
1255 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1256 | atomic_inc(&rdev->nr_pending); | |
1257 | blocked_rdev = rdev; | |
1258 | break; | |
1259 | } | |
475b0321 N |
1260 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { |
1261 | atomic_inc(&rrdev->nr_pending); | |
1262 | blocked_rdev = rrdev; | |
1263 | break; | |
1264 | } | |
8ae12666 | 1265 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1266 | rdev = NULL; |
8ae12666 | 1267 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1268 | rrdev = NULL; |
1269 | ||
d4432c23 | 1270 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1271 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1272 | |
1273 | if (!rdev && !rrdev) { | |
6cce3b23 | 1274 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1275 | continue; |
1276 | } | |
e7c0c3fa | 1277 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1278 | sector_t first_bad; |
1279 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1280 | int bad_sectors; | |
1281 | int is_bad; | |
1282 | ||
1283 | is_bad = is_badblock(rdev, dev_sector, | |
1284 | max_sectors, | |
1285 | &first_bad, &bad_sectors); | |
1286 | if (is_bad < 0) { | |
1287 | /* Mustn't write here until the bad block | |
1288 | * is acknowledged | |
1289 | */ | |
1290 | atomic_inc(&rdev->nr_pending); | |
1291 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1292 | blocked_rdev = rdev; | |
1293 | break; | |
1294 | } | |
1295 | if (is_bad && first_bad <= dev_sector) { | |
1296 | /* Cannot write here at all */ | |
1297 | bad_sectors -= (dev_sector - first_bad); | |
1298 | if (bad_sectors < max_sectors) | |
1299 | /* Mustn't write more than bad_sectors | |
1300 | * to other devices yet | |
1301 | */ | |
1302 | max_sectors = bad_sectors; | |
1303 | /* We don't set R10BIO_Degraded as that | |
1304 | * only applies if the disk is missing, | |
1305 | * so it might be re-added, and we want to | |
1306 | * know to recover this chunk. | |
1307 | * In this case the device is here, and the | |
1308 | * fact that this chunk is not in-sync is | |
1309 | * recorded in the bad block log. | |
1310 | */ | |
1311 | continue; | |
1312 | } | |
1313 | if (is_bad) { | |
1314 | int good_sectors = first_bad - dev_sector; | |
1315 | if (good_sectors < max_sectors) | |
1316 | max_sectors = good_sectors; | |
1317 | } | |
6cce3b23 | 1318 | } |
e7c0c3fa N |
1319 | if (rdev) { |
1320 | r10_bio->devs[i].bio = bio; | |
1321 | atomic_inc(&rdev->nr_pending); | |
1322 | } | |
475b0321 N |
1323 | if (rrdev) { |
1324 | r10_bio->devs[i].repl_bio = bio; | |
1325 | atomic_inc(&rrdev->nr_pending); | |
1326 | } | |
1da177e4 LT |
1327 | } |
1328 | rcu_read_unlock(); | |
1329 | ||
6bfe0b49 DW |
1330 | if (unlikely(blocked_rdev)) { |
1331 | /* Have to wait for this device to get unblocked, then retry */ | |
1332 | int j; | |
1333 | int d; | |
1334 | ||
475b0321 | 1335 | for (j = 0; j < i; j++) { |
6bfe0b49 DW |
1336 | if (r10_bio->devs[j].bio) { |
1337 | d = r10_bio->devs[j].devnum; | |
1338 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1339 | } | |
475b0321 | 1340 | if (r10_bio->devs[j].repl_bio) { |
4ca40c2c | 1341 | struct md_rdev *rdev; |
475b0321 | 1342 | d = r10_bio->devs[j].devnum; |
4ca40c2c N |
1343 | rdev = conf->mirrors[d].replacement; |
1344 | if (!rdev) { | |
1345 | /* Race with remove_disk */ | |
1346 | smp_mb(); | |
1347 | rdev = conf->mirrors[d].rdev; | |
1348 | } | |
1349 | rdev_dec_pending(rdev, mddev); | |
475b0321 N |
1350 | } |
1351 | } | |
6bfe0b49 | 1352 | allow_barrier(conf); |
578b54ad | 1353 | raid10_log(conf->mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 DW |
1354 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
1355 | wait_barrier(conf); | |
1356 | goto retry_write; | |
1357 | } | |
1358 | ||
d4432c23 N |
1359 | if (max_sectors < r10_bio->sectors) { |
1360 | /* We are splitting this into multiple parts, so | |
1361 | * we need to prepare for allocating another r10_bio. | |
1362 | */ | |
1363 | r10_bio->sectors = max_sectors; | |
1364 | spin_lock_irq(&conf->device_lock); | |
1365 | if (bio->bi_phys_segments == 0) | |
1366 | bio->bi_phys_segments = 2; | |
1367 | else | |
1368 | bio->bi_phys_segments++; | |
1369 | spin_unlock_irq(&conf->device_lock); | |
1370 | } | |
4f024f37 KO |
1371 | sectors_handled = r10_bio->sector + max_sectors - |
1372 | bio->bi_iter.bi_sector; | |
d4432c23 | 1373 | |
4e78064f | 1374 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1375 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1376 | |
1da177e4 LT |
1377 | for (i = 0; i < conf->copies; i++) { |
1378 | struct bio *mbio; | |
1379 | int d = r10_bio->devs[i].devnum; | |
e7c0c3fa N |
1380 | if (r10_bio->devs[i].bio) { |
1381 | struct md_rdev *rdev = conf->mirrors[d].rdev; | |
1382 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
4f024f37 | 1383 | bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1384 | max_sectors); |
e7c0c3fa N |
1385 | r10_bio->devs[i].bio = mbio; |
1386 | ||
4f024f37 | 1387 | mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr+ |
e7c0c3fa N |
1388 | choose_data_offset(r10_bio, |
1389 | rdev)); | |
109e3765 | 1390 | mbio->bi_bdev = rdev->bdev; |
e7c0c3fa | 1391 | mbio->bi_end_io = raid10_end_write_request; |
288dab8a | 1392 | bio_set_op_attrs(mbio, op, do_sync | do_fua); |
e7c0c3fa N |
1393 | mbio->bi_private = r10_bio; |
1394 | ||
109e3765 N |
1395 | if (conf->mddev->gendisk) |
1396 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1397 | mbio, disk_devt(conf->mddev->gendisk), | |
1398 | r10_bio->sector); | |
1399 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1400 | mbio->bi_bdev = (void*)rdev; | |
1401 | ||
e7c0c3fa N |
1402 | atomic_inc(&r10_bio->remaining); |
1403 | ||
1404 | cb = blk_check_plugged(raid10_unplug, mddev, | |
1405 | sizeof(*plug)); | |
1406 | if (cb) | |
1407 | plug = container_of(cb, struct raid10_plug_cb, | |
1408 | cb); | |
1409 | else | |
1410 | plug = NULL; | |
1411 | spin_lock_irqsave(&conf->device_lock, flags); | |
1412 | if (plug) { | |
1413 | bio_list_add(&plug->pending, mbio); | |
1414 | plug->pending_cnt++; | |
1415 | } else { | |
1416 | bio_list_add(&conf->pending_bio_list, mbio); | |
1417 | conf->pending_count++; | |
1418 | } | |
1419 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1420 | if (!plug) | |
1421 | md_wakeup_thread(mddev->thread); | |
1422 | } | |
57c67df4 | 1423 | |
e7c0c3fa N |
1424 | if (r10_bio->devs[i].repl_bio) { |
1425 | struct md_rdev *rdev = conf->mirrors[d].replacement; | |
1426 | if (rdev == NULL) { | |
1427 | /* Replacement just got moved to main 'rdev' */ | |
1428 | smp_mb(); | |
1429 | rdev = conf->mirrors[d].rdev; | |
1430 | } | |
1431 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
4f024f37 | 1432 | bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1433 | max_sectors); |
e7c0c3fa N |
1434 | r10_bio->devs[i].repl_bio = mbio; |
1435 | ||
4f024f37 | 1436 | mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr + |
e7c0c3fa N |
1437 | choose_data_offset( |
1438 | r10_bio, rdev)); | |
109e3765 | 1439 | mbio->bi_bdev = rdev->bdev; |
e7c0c3fa | 1440 | mbio->bi_end_io = raid10_end_write_request; |
288dab8a | 1441 | bio_set_op_attrs(mbio, op, do_sync | do_fua); |
e7c0c3fa N |
1442 | mbio->bi_private = r10_bio; |
1443 | ||
109e3765 N |
1444 | if (conf->mddev->gendisk) |
1445 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1446 | mbio, disk_devt(conf->mddev->gendisk), | |
1447 | r10_bio->sector); | |
1448 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1449 | mbio->bi_bdev = (void*)rdev; | |
1450 | ||
e7c0c3fa N |
1451 | atomic_inc(&r10_bio->remaining); |
1452 | spin_lock_irqsave(&conf->device_lock, flags); | |
57c67df4 N |
1453 | bio_list_add(&conf->pending_bio_list, mbio); |
1454 | conf->pending_count++; | |
e7c0c3fa N |
1455 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1456 | if (!mddev_check_plugged(mddev)) | |
1457 | md_wakeup_thread(mddev->thread); | |
57c67df4 | 1458 | } |
1da177e4 LT |
1459 | } |
1460 | ||
079fa166 N |
1461 | /* Don't remove the bias on 'remaining' (one_write_done) until |
1462 | * after checking if we need to go around again. | |
1463 | */ | |
a35e63ef | 1464 | |
aa8b57aa | 1465 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1466 | one_write_done(r10_bio); |
5e570289 | 1467 | /* We need another r10_bio. It has already been counted |
d4432c23 N |
1468 | * in bio->bi_phys_segments. |
1469 | */ | |
1470 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1471 | ||
1472 | r10_bio->master_bio = bio; | |
aa8b57aa | 1473 | r10_bio->sectors = bio_sectors(bio) - sectors_handled; |
d4432c23 N |
1474 | |
1475 | r10_bio->mddev = mddev; | |
4f024f37 | 1476 | r10_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
d4432c23 N |
1477 | r10_bio->state = 0; |
1478 | goto retry_write; | |
1479 | } | |
079fa166 | 1480 | one_write_done(r10_bio); |
20d0189b KO |
1481 | } |
1482 | ||
849674e4 | 1483 | static void raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1484 | { |
1485 | struct r10conf *conf = mddev->private; | |
1486 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1487 | int chunk_sects = chunk_mask + 1; | |
1488 | ||
1489 | struct bio *split; | |
1490 | ||
1eff9d32 | 1491 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
20d0189b KO |
1492 | md_flush_request(mddev, bio); |
1493 | return; | |
1494 | } | |
1495 | ||
20d0189b KO |
1496 | do { |
1497 | ||
1498 | /* | |
1499 | * If this request crosses a chunk boundary, we need to split | |
1500 | * it. | |
1501 | */ | |
1502 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1503 | bio_sectors(bio) > chunk_sects | |
1504 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1505 | || conf->prev.near_copies < | |
1506 | conf->prev.raid_disks))) { | |
1507 | split = bio_split(bio, chunk_sects - | |
1508 | (bio->bi_iter.bi_sector & | |
1509 | (chunk_sects - 1)), | |
1510 | GFP_NOIO, fs_bio_set); | |
1511 | bio_chain(split, bio); | |
1512 | } else { | |
1513 | split = bio; | |
1514 | } | |
1515 | ||
1516 | __make_request(mddev, split); | |
1517 | } while (split != bio); | |
079fa166 N |
1518 | |
1519 | /* In case raid10d snuck in to freeze_array */ | |
1520 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1521 | } |
1522 | ||
849674e4 | 1523 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1524 | { |
e879a879 | 1525 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1526 | int i; |
1527 | ||
5cf00fcd | 1528 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1529 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1530 | if (conf->geo.near_copies > 1) |
1531 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1532 | if (conf->geo.far_copies > 1) { | |
1533 | if (conf->geo.far_offset) | |
1534 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1535 | else |
5cf00fcd | 1536 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1537 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1538 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1539 | } |
5cf00fcd N |
1540 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1541 | conf->geo.raid_disks - mddev->degraded); | |
d44b0a92 N |
1542 | rcu_read_lock(); |
1543 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
1544 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1545 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
1546 | } | |
1547 | rcu_read_unlock(); | |
1da177e4 LT |
1548 | seq_printf(seq, "]"); |
1549 | } | |
1550 | ||
700c7213 N |
1551 | /* check if there are enough drives for |
1552 | * every block to appear on atleast one. | |
1553 | * Don't consider the device numbered 'ignore' | |
1554 | * as we might be about to remove it. | |
1555 | */ | |
635f6416 | 1556 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1557 | { |
1558 | int first = 0; | |
725d6e57 | 1559 | int has_enough = 0; |
635f6416 N |
1560 | int disks, ncopies; |
1561 | if (previous) { | |
1562 | disks = conf->prev.raid_disks; | |
1563 | ncopies = conf->prev.near_copies; | |
1564 | } else { | |
1565 | disks = conf->geo.raid_disks; | |
1566 | ncopies = conf->geo.near_copies; | |
1567 | } | |
700c7213 | 1568 | |
725d6e57 | 1569 | rcu_read_lock(); |
700c7213 N |
1570 | do { |
1571 | int n = conf->copies; | |
1572 | int cnt = 0; | |
80b48124 | 1573 | int this = first; |
700c7213 | 1574 | while (n--) { |
725d6e57 N |
1575 | struct md_rdev *rdev; |
1576 | if (this != ignore && | |
1577 | (rdev = rcu_dereference(conf->mirrors[this].rdev)) && | |
1578 | test_bit(In_sync, &rdev->flags)) | |
700c7213 | 1579 | cnt++; |
635f6416 | 1580 | this = (this+1) % disks; |
700c7213 N |
1581 | } |
1582 | if (cnt == 0) | |
725d6e57 | 1583 | goto out; |
635f6416 | 1584 | first = (first + ncopies) % disks; |
700c7213 | 1585 | } while (first != 0); |
725d6e57 N |
1586 | has_enough = 1; |
1587 | out: | |
1588 | rcu_read_unlock(); | |
1589 | return has_enough; | |
700c7213 N |
1590 | } |
1591 | ||
f8c9e74f N |
1592 | static int enough(struct r10conf *conf, int ignore) |
1593 | { | |
635f6416 N |
1594 | /* when calling 'enough', both 'prev' and 'geo' must |
1595 | * be stable. | |
1596 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1597 | * is held. | |
1598 | */ | |
1599 | return _enough(conf, 0, ignore) && | |
1600 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1601 | } |
1602 | ||
849674e4 | 1603 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1604 | { |
1605 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1606 | struct r10conf *conf = mddev->private; |
635f6416 | 1607 | unsigned long flags; |
1da177e4 LT |
1608 | |
1609 | /* | |
1610 | * If it is not operational, then we have already marked it as dead | |
1611 | * else if it is the last working disks, ignore the error, let the | |
1612 | * next level up know. | |
1613 | * else mark the drive as failed | |
1614 | */ | |
635f6416 | 1615 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1616 | if (test_bit(In_sync, &rdev->flags) |
635f6416 | 1617 | && !enough(conf, rdev->raid_disk)) { |
1da177e4 LT |
1618 | /* |
1619 | * Don't fail the drive, just return an IO error. | |
1da177e4 | 1620 | */ |
635f6416 | 1621 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1622 | return; |
635f6416 | 1623 | } |
2446dba0 | 1624 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1625 | mddev->degraded++; |
2446dba0 N |
1626 | /* |
1627 | * If recovery is running, make sure it aborts. | |
1628 | */ | |
1629 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
de393cde | 1630 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1631 | set_bit(Faulty, &rdev->flags); |
85ad1d13 GJ |
1632 | set_mask_bits(&mddev->flags, 0, |
1633 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
635f6416 | 1634 | spin_unlock_irqrestore(&conf->device_lock, flags); |
08464e09 N |
1635 | pr_crit("md/raid10:%s: Disk failure on %s, disabling device.\n" |
1636 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
1637 | mdname(mddev), bdevname(rdev->bdev, b), | |
1638 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); | |
1da177e4 LT |
1639 | } |
1640 | ||
e879a879 | 1641 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1642 | { |
1643 | int i; | |
4056ca51 | 1644 | struct md_rdev *rdev; |
1da177e4 | 1645 | |
08464e09 | 1646 | pr_debug("RAID10 conf printout:\n"); |
1da177e4 | 1647 | if (!conf) { |
08464e09 | 1648 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1649 | return; |
1650 | } | |
08464e09 N |
1651 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1652 | conf->geo.raid_disks); | |
1da177e4 | 1653 | |
4056ca51 N |
1654 | /* This is only called with ->reconfix_mutex held, so |
1655 | * rcu protection of rdev is not needed */ | |
5cf00fcd | 1656 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1657 | char b[BDEVNAME_SIZE]; |
4056ca51 N |
1658 | rdev = conf->mirrors[i].rdev; |
1659 | if (rdev) | |
08464e09 N |
1660 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1661 | i, !test_bit(In_sync, &rdev->flags), | |
1662 | !test_bit(Faulty, &rdev->flags), | |
1663 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1664 | } |
1665 | } | |
1666 | ||
e879a879 | 1667 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1668 | { |
0a27ec96 N |
1669 | wait_barrier(conf); |
1670 | allow_barrier(conf); | |
1da177e4 LT |
1671 | |
1672 | mempool_destroy(conf->r10buf_pool); | |
1673 | conf->r10buf_pool = NULL; | |
1674 | } | |
1675 | ||
fd01b88c | 1676 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1677 | { |
1678 | int i; | |
e879a879 | 1679 | struct r10conf *conf = mddev->private; |
dc280d98 | 1680 | struct raid10_info *tmp; |
6b965620 N |
1681 | int count = 0; |
1682 | unsigned long flags; | |
1da177e4 LT |
1683 | |
1684 | /* | |
1685 | * Find all non-in_sync disks within the RAID10 configuration | |
1686 | * and mark them in_sync | |
1687 | */ | |
5cf00fcd | 1688 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1689 | tmp = conf->mirrors + i; |
4ca40c2c N |
1690 | if (tmp->replacement |
1691 | && tmp->replacement->recovery_offset == MaxSector | |
1692 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1693 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1694 | /* Replacement has just become active */ | |
1695 | if (!tmp->rdev | |
1696 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1697 | count++; | |
1698 | if (tmp->rdev) { | |
1699 | /* Replaced device not technically faulty, | |
1700 | * but we need to be sure it gets removed | |
1701 | * and never re-added. | |
1702 | */ | |
1703 | set_bit(Faulty, &tmp->rdev->flags); | |
1704 | sysfs_notify_dirent_safe( | |
1705 | tmp->rdev->sysfs_state); | |
1706 | } | |
1707 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1708 | } else if (tmp->rdev | |
61e4947c | 1709 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
1710 | && !test_bit(Faulty, &tmp->rdev->flags) |
1711 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1712 | count++; |
2863b9eb | 1713 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
1714 | } |
1715 | } | |
6b965620 N |
1716 | spin_lock_irqsave(&conf->device_lock, flags); |
1717 | mddev->degraded -= count; | |
1718 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1719 | |
1720 | print_conf(conf); | |
6b965620 | 1721 | return count; |
1da177e4 LT |
1722 | } |
1723 | ||
fd01b88c | 1724 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1725 | { |
e879a879 | 1726 | struct r10conf *conf = mddev->private; |
199050ea | 1727 | int err = -EEXIST; |
1da177e4 | 1728 | int mirror; |
6c2fce2e | 1729 | int first = 0; |
5cf00fcd | 1730 | int last = conf->geo.raid_disks - 1; |
1da177e4 LT |
1731 | |
1732 | if (mddev->recovery_cp < MaxSector) | |
1733 | /* only hot-add to in-sync arrays, as recovery is | |
1734 | * very different from resync | |
1735 | */ | |
199050ea | 1736 | return -EBUSY; |
635f6416 | 1737 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 1738 | return -EINVAL; |
1da177e4 | 1739 | |
1501efad DW |
1740 | if (md_integrity_add_rdev(rdev, mddev)) |
1741 | return -ENXIO; | |
1742 | ||
a53a6c85 | 1743 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1744 | first = last = rdev->raid_disk; |
1da177e4 | 1745 | |
2c4193df | 1746 | if (rdev->saved_raid_disk >= first && |
6cce3b23 N |
1747 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1748 | mirror = rdev->saved_raid_disk; | |
1749 | else | |
6c2fce2e | 1750 | mirror = first; |
2bb77736 | 1751 | for ( ; mirror <= last ; mirror++) { |
dc280d98 | 1752 | struct raid10_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1753 | if (p->recovery_disabled == mddev->recovery_disabled) |
1754 | continue; | |
b7044d41 N |
1755 | if (p->rdev) { |
1756 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1757 | p->replacement != NULL) | |
1758 | continue; | |
1759 | clear_bit(In_sync, &rdev->flags); | |
1760 | set_bit(Replacement, &rdev->flags); | |
1761 | rdev->raid_disk = mirror; | |
1762 | err = 0; | |
9092c02d JB |
1763 | if (mddev->gendisk) |
1764 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1765 | rdev->data_offset << 9); | |
b7044d41 N |
1766 | conf->fullsync = 1; |
1767 | rcu_assign_pointer(p->replacement, rdev); | |
1768 | break; | |
1769 | } | |
1da177e4 | 1770 | |
9092c02d JB |
1771 | if (mddev->gendisk) |
1772 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1773 | rdev->data_offset << 9); | |
1da177e4 | 1774 | |
2bb77736 | 1775 | p->head_position = 0; |
d890fa2b | 1776 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1777 | rdev->raid_disk = mirror; |
1778 | err = 0; | |
1779 | if (rdev->saved_raid_disk != mirror) | |
1780 | conf->fullsync = 1; | |
1781 | rcu_assign_pointer(p->rdev, rdev); | |
1782 | break; | |
1783 | } | |
ed30be07 | 1784 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
532a2a3f SL |
1785 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1786 | ||
1da177e4 | 1787 | print_conf(conf); |
199050ea | 1788 | return err; |
1da177e4 LT |
1789 | } |
1790 | ||
b8321b68 | 1791 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1792 | { |
e879a879 | 1793 | struct r10conf *conf = mddev->private; |
1da177e4 | 1794 | int err = 0; |
b8321b68 | 1795 | int number = rdev->raid_disk; |
c8ab903e | 1796 | struct md_rdev **rdevp; |
dc280d98 | 1797 | struct raid10_info *p = conf->mirrors + number; |
1da177e4 LT |
1798 | |
1799 | print_conf(conf); | |
c8ab903e N |
1800 | if (rdev == p->rdev) |
1801 | rdevp = &p->rdev; | |
1802 | else if (rdev == p->replacement) | |
1803 | rdevp = &p->replacement; | |
1804 | else | |
1805 | return 0; | |
1806 | ||
1807 | if (test_bit(In_sync, &rdev->flags) || | |
1808 | atomic_read(&rdev->nr_pending)) { | |
1809 | err = -EBUSY; | |
1810 | goto abort; | |
1811 | } | |
d787be40 | 1812 | /* Only remove non-faulty devices if recovery |
c8ab903e N |
1813 | * is not possible. |
1814 | */ | |
1815 | if (!test_bit(Faulty, &rdev->flags) && | |
1816 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1817 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 1818 | number < conf->geo.raid_disks && |
c8ab903e N |
1819 | enough(conf, -1)) { |
1820 | err = -EBUSY; | |
1821 | goto abort; | |
1da177e4 | 1822 | } |
c8ab903e | 1823 | *rdevp = NULL; |
d787be40 N |
1824 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1825 | synchronize_rcu(); | |
1826 | if (atomic_read(&rdev->nr_pending)) { | |
1827 | /* lost the race, try later */ | |
1828 | err = -EBUSY; | |
1829 | *rdevp = rdev; | |
1830 | goto abort; | |
1831 | } | |
1832 | } | |
1833 | if (p->replacement) { | |
4ca40c2c N |
1834 | /* We must have just cleared 'rdev' */ |
1835 | p->rdev = p->replacement; | |
1836 | clear_bit(Replacement, &p->replacement->flags); | |
1837 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1838 | * but will never see neither -- if they are careful. | |
1839 | */ | |
1840 | p->replacement = NULL; | |
1841 | clear_bit(WantReplacement, &rdev->flags); | |
1842 | } else | |
1843 | /* We might have just remove the Replacement as faulty | |
1844 | * Clear the flag just in case | |
1845 | */ | |
1846 | clear_bit(WantReplacement, &rdev->flags); | |
1847 | ||
c8ab903e N |
1848 | err = md_integrity_register(mddev); |
1849 | ||
1da177e4 LT |
1850 | abort: |
1851 | ||
1852 | print_conf(conf); | |
1853 | return err; | |
1854 | } | |
1855 | ||
4246a0b6 | 1856 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1857 | { |
9f2c9d12 | 1858 | struct r10bio *r10_bio = bio->bi_private; |
e879a879 | 1859 | struct r10conf *conf = r10_bio->mddev->private; |
778ca018 | 1860 | int d; |
1da177e4 | 1861 | |
3ea7daa5 N |
1862 | if (bio == r10_bio->master_bio) { |
1863 | /* this is a reshape read */ | |
1864 | d = r10_bio->read_slot; /* really the read dev */ | |
1865 | } else | |
1866 | d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
0eb3ff12 | 1867 | |
4246a0b6 | 1868 | if (!bio->bi_error) |
0eb3ff12 | 1869 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
1870 | else |
1871 | /* The write handler will notice the lack of | |
1872 | * R10BIO_Uptodate and record any errors etc | |
1873 | */ | |
4dbcdc75 N |
1874 | atomic_add(r10_bio->sectors, |
1875 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1876 | |
1877 | /* for reconstruct, we always reschedule after a read. | |
1878 | * for resync, only after all reads | |
1879 | */ | |
73d5c38a | 1880 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1881 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1882 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1883 | /* we have read all the blocks, | |
1884 | * do the comparison in process context in raid10d | |
1885 | */ | |
1886 | reschedule_retry(r10_bio); | |
1887 | } | |
1da177e4 LT |
1888 | } |
1889 | ||
9f2c9d12 | 1890 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1891 | { |
fd01b88c | 1892 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1893 | |
1da177e4 LT |
1894 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1895 | if (r10_bio->master_bio == NULL) { | |
1896 | /* the primary of several recovery bios */ | |
73d5c38a | 1897 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1898 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1899 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1900 | reschedule_retry(r10_bio); |
1901 | else | |
1902 | put_buf(r10_bio); | |
73d5c38a | 1903 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1904 | break; |
1905 | } else { | |
9f2c9d12 | 1906 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1907 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1908 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1909 | reschedule_retry(r10_bio); |
1910 | else | |
1911 | put_buf(r10_bio); | |
1da177e4 LT |
1912 | r10_bio = r10_bio2; |
1913 | } | |
1914 | } | |
1da177e4 LT |
1915 | } |
1916 | ||
4246a0b6 | 1917 | static void end_sync_write(struct bio *bio) |
5e570289 | 1918 | { |
9f2c9d12 | 1919 | struct r10bio *r10_bio = bio->bi_private; |
fd01b88c | 1920 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1921 | struct r10conf *conf = mddev->private; |
5e570289 N |
1922 | int d; |
1923 | sector_t first_bad; | |
1924 | int bad_sectors; | |
1925 | int slot; | |
9ad1aefc | 1926 | int repl; |
4ca40c2c | 1927 | struct md_rdev *rdev = NULL; |
5e570289 | 1928 | |
9ad1aefc N |
1929 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1930 | if (repl) | |
1931 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 1932 | else |
9ad1aefc | 1933 | rdev = conf->mirrors[d].rdev; |
5e570289 | 1934 | |
4246a0b6 | 1935 | if (bio->bi_error) { |
9ad1aefc N |
1936 | if (repl) |
1937 | md_error(mddev, rdev); | |
1938 | else { | |
1939 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
1940 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1941 | set_bit(MD_RECOVERY_NEEDED, | |
1942 | &rdev->mddev->recovery); | |
9ad1aefc N |
1943 | set_bit(R10BIO_WriteError, &r10_bio->state); |
1944 | } | |
1945 | } else if (is_badblock(rdev, | |
5e570289 N |
1946 | r10_bio->devs[slot].addr, |
1947 | r10_bio->sectors, | |
1948 | &first_bad, &bad_sectors)) | |
1949 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
1950 | ||
9ad1aefc | 1951 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
1952 | |
1953 | end_sync_request(r10_bio); | |
1954 | } | |
1955 | ||
1da177e4 LT |
1956 | /* |
1957 | * Note: sync and recover and handled very differently for raid10 | |
1958 | * This code is for resync. | |
1959 | * For resync, we read through virtual addresses and read all blocks. | |
1960 | * If there is any error, we schedule a write. The lowest numbered | |
1961 | * drive is authoritative. | |
1962 | * However requests come for physical address, so we need to map. | |
1963 | * For every physical address there are raid_disks/copies virtual addresses, | |
1964 | * which is always are least one, but is not necessarly an integer. | |
1965 | * This means that a physical address can span multiple chunks, so we may | |
1966 | * have to submit multiple io requests for a single sync request. | |
1967 | */ | |
1968 | /* | |
1969 | * We check if all blocks are in-sync and only write to blocks that | |
1970 | * aren't in sync | |
1971 | */ | |
9f2c9d12 | 1972 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1973 | { |
e879a879 | 1974 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1975 | int i, first; |
1976 | struct bio *tbio, *fbio; | |
f4380a91 | 1977 | int vcnt; |
1da177e4 LT |
1978 | |
1979 | atomic_set(&r10_bio->remaining, 1); | |
1980 | ||
1981 | /* find the first device with a block */ | |
1982 | for (i=0; i<conf->copies; i++) | |
4246a0b6 | 1983 | if (!r10_bio->devs[i].bio->bi_error) |
1da177e4 LT |
1984 | break; |
1985 | ||
1986 | if (i == conf->copies) | |
1987 | goto done; | |
1988 | ||
1989 | first = i; | |
1990 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
1991 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
1992 | fbio->bi_iter.bi_idx = 0; | |
1da177e4 | 1993 | |
f4380a91 | 1994 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 1995 | /* now find blocks with errors */ |
0eb3ff12 N |
1996 | for (i=0 ; i < conf->copies ; i++) { |
1997 | int j, d; | |
8d3ca83d | 1998 | struct md_rdev *rdev; |
1da177e4 | 1999 | |
1da177e4 | 2000 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
2001 | |
2002 | if (tbio->bi_end_io != end_sync_read) | |
2003 | continue; | |
2004 | if (i == first) | |
1da177e4 | 2005 | continue; |
8d3ca83d N |
2006 | d = r10_bio->devs[i].devnum; |
2007 | rdev = conf->mirrors[d].rdev; | |
4246a0b6 | 2008 | if (!r10_bio->devs[i].bio->bi_error) { |
0eb3ff12 N |
2009 | /* We know that the bi_io_vec layout is the same for |
2010 | * both 'first' and 'i', so we just compare them. | |
2011 | * All vec entries are PAGE_SIZE; | |
2012 | */ | |
7bb23c49 N |
2013 | int sectors = r10_bio->sectors; |
2014 | for (j = 0; j < vcnt; j++) { | |
2015 | int len = PAGE_SIZE; | |
2016 | if (sectors < (len / 512)) | |
2017 | len = sectors * 512; | |
0eb3ff12 N |
2018 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), |
2019 | page_address(tbio->bi_io_vec[j].bv_page), | |
7bb23c49 | 2020 | len)) |
0eb3ff12 | 2021 | break; |
7bb23c49 N |
2022 | sectors -= len/512; |
2023 | } | |
0eb3ff12 N |
2024 | if (j == vcnt) |
2025 | continue; | |
7f7583d4 | 2026 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
2027 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
2028 | /* Don't fix anything. */ | |
2029 | continue; | |
8d3ca83d N |
2030 | } else if (test_bit(FailFast, &rdev->flags)) { |
2031 | /* Just give up on this device */ | |
2032 | md_error(rdev->mddev, rdev); | |
2033 | continue; | |
0eb3ff12 | 2034 | } |
f84ee364 N |
2035 | /* Ok, we need to write this bio, either to correct an |
2036 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
2037 | * First we need to fixup bv_offset, bv_len and |
2038 | * bi_vecs, as the read request might have corrupted these | |
2039 | */ | |
8be185f2 KO |
2040 | bio_reset(tbio); |
2041 | ||
1da177e4 | 2042 | tbio->bi_vcnt = vcnt; |
cc578588 | 2043 | tbio->bi_iter.bi_size = fbio->bi_iter.bi_size; |
1da177e4 | 2044 | tbio->bi_private = r10_bio; |
4f024f37 | 2045 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 | 2046 | tbio->bi_end_io = end_sync_write; |
796a5cf0 | 2047 | bio_set_op_attrs(tbio, REQ_OP_WRITE, 0); |
1da177e4 | 2048 | |
c31df25f KO |
2049 | bio_copy_data(tbio, fbio); |
2050 | ||
1da177e4 LT |
2051 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2052 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2053 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2054 | |
4f024f37 | 2055 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
1da177e4 LT |
2056 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; |
2057 | generic_make_request(tbio); | |
2058 | } | |
2059 | ||
9ad1aefc N |
2060 | /* Now write out to any replacement devices |
2061 | * that are active | |
2062 | */ | |
2063 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2064 | int d; |
9ad1aefc N |
2065 | |
2066 | tbio = r10_bio->devs[i].repl_bio; | |
2067 | if (!tbio || !tbio->bi_end_io) | |
2068 | continue; | |
2069 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2070 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2071 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2072 | d = r10_bio->devs[i].devnum; |
2073 | atomic_inc(&r10_bio->remaining); | |
2074 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2075 | bio_sectors(tbio)); |
9ad1aefc N |
2076 | generic_make_request(tbio); |
2077 | } | |
2078 | ||
1da177e4 LT |
2079 | done: |
2080 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2081 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2082 | put_buf(r10_bio); | |
2083 | } | |
2084 | } | |
2085 | ||
2086 | /* | |
2087 | * Now for the recovery code. | |
2088 | * Recovery happens across physical sectors. | |
2089 | * We recover all non-is_sync drives by finding the virtual address of | |
2090 | * each, and then choose a working drive that also has that virt address. | |
2091 | * There is a separate r10_bio for each non-in_sync drive. | |
2092 | * Only the first two slots are in use. The first for reading, | |
2093 | * The second for writing. | |
2094 | * | |
2095 | */ | |
9f2c9d12 | 2096 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2097 | { |
2098 | /* We got a read error during recovery. | |
2099 | * We repeat the read in smaller page-sized sections. | |
2100 | * If a read succeeds, write it to the new device or record | |
2101 | * a bad block if we cannot. | |
2102 | * If a read fails, record a bad block on both old and | |
2103 | * new devices. | |
2104 | */ | |
fd01b88c | 2105 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2106 | struct r10conf *conf = mddev->private; |
5e570289 N |
2107 | struct bio *bio = r10_bio->devs[0].bio; |
2108 | sector_t sect = 0; | |
2109 | int sectors = r10_bio->sectors; | |
2110 | int idx = 0; | |
2111 | int dr = r10_bio->devs[0].devnum; | |
2112 | int dw = r10_bio->devs[1].devnum; | |
2113 | ||
2114 | while (sectors) { | |
2115 | int s = sectors; | |
3cb03002 | 2116 | struct md_rdev *rdev; |
5e570289 N |
2117 | sector_t addr; |
2118 | int ok; | |
2119 | ||
2120 | if (s > (PAGE_SIZE>>9)) | |
2121 | s = PAGE_SIZE >> 9; | |
2122 | ||
2123 | rdev = conf->mirrors[dr].rdev; | |
2124 | addr = r10_bio->devs[0].addr + sect, | |
2125 | ok = sync_page_io(rdev, | |
2126 | addr, | |
2127 | s << 9, | |
2128 | bio->bi_io_vec[idx].bv_page, | |
796a5cf0 | 2129 | REQ_OP_READ, 0, false); |
5e570289 N |
2130 | if (ok) { |
2131 | rdev = conf->mirrors[dw].rdev; | |
2132 | addr = r10_bio->devs[1].addr + sect; | |
2133 | ok = sync_page_io(rdev, | |
2134 | addr, | |
2135 | s << 9, | |
2136 | bio->bi_io_vec[idx].bv_page, | |
796a5cf0 | 2137 | REQ_OP_WRITE, 0, false); |
b7044d41 | 2138 | if (!ok) { |
5e570289 | 2139 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2140 | if (!test_and_set_bit(WantReplacement, |
2141 | &rdev->flags)) | |
2142 | set_bit(MD_RECOVERY_NEEDED, | |
2143 | &rdev->mddev->recovery); | |
2144 | } | |
5e570289 N |
2145 | } |
2146 | if (!ok) { | |
2147 | /* We don't worry if we cannot set a bad block - | |
2148 | * it really is bad so there is no loss in not | |
2149 | * recording it yet | |
2150 | */ | |
2151 | rdev_set_badblocks(rdev, addr, s, 0); | |
2152 | ||
2153 | if (rdev != conf->mirrors[dw].rdev) { | |
2154 | /* need bad block on destination too */ | |
3cb03002 | 2155 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2156 | addr = r10_bio->devs[1].addr + sect; |
2157 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2158 | if (!ok) { | |
2159 | /* just abort the recovery */ | |
08464e09 N |
2160 | pr_notice("md/raid10:%s: recovery aborted due to read error\n", |
2161 | mdname(mddev)); | |
5e570289 N |
2162 | |
2163 | conf->mirrors[dw].recovery_disabled | |
2164 | = mddev->recovery_disabled; | |
2165 | set_bit(MD_RECOVERY_INTR, | |
2166 | &mddev->recovery); | |
2167 | break; | |
2168 | } | |
2169 | } | |
2170 | } | |
2171 | ||
2172 | sectors -= s; | |
2173 | sect += s; | |
2174 | idx++; | |
2175 | } | |
2176 | } | |
1da177e4 | 2177 | |
9f2c9d12 | 2178 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2179 | { |
e879a879 | 2180 | struct r10conf *conf = mddev->private; |
c65060ad | 2181 | int d; |
24afd80d | 2182 | struct bio *wbio, *wbio2; |
1da177e4 | 2183 | |
5e570289 N |
2184 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2185 | fix_recovery_read_error(r10_bio); | |
2186 | end_sync_request(r10_bio); | |
2187 | return; | |
2188 | } | |
2189 | ||
c65060ad NK |
2190 | /* |
2191 | * share the pages with the first bio | |
1da177e4 LT |
2192 | * and submit the write request |
2193 | */ | |
1da177e4 | 2194 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2195 | wbio = r10_bio->devs[1].bio; |
2196 | wbio2 = r10_bio->devs[1].repl_bio; | |
0eb25bb0 N |
2197 | /* Need to test wbio2->bi_end_io before we call |
2198 | * generic_make_request as if the former is NULL, | |
2199 | * the latter is free to free wbio2. | |
2200 | */ | |
2201 | if (wbio2 && !wbio2->bi_end_io) | |
2202 | wbio2 = NULL; | |
24afd80d N |
2203 | if (wbio->bi_end_io) { |
2204 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2205 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
24afd80d N |
2206 | generic_make_request(wbio); |
2207 | } | |
0eb25bb0 | 2208 | if (wbio2) { |
24afd80d N |
2209 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2210 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2211 | bio_sectors(wbio2)); |
24afd80d N |
2212 | generic_make_request(wbio2); |
2213 | } | |
1da177e4 LT |
2214 | } |
2215 | ||
1e50915f RB |
2216 | /* |
2217 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2218 | * We halve the read error count for every hour that has elapsed | |
2219 | * since the last recorded read error. | |
2220 | * | |
2221 | */ | |
fd01b88c | 2222 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f | 2223 | { |
0e3ef49e | 2224 | long cur_time_mon; |
1e50915f RB |
2225 | unsigned long hours_since_last; |
2226 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2227 | ||
0e3ef49e | 2228 | cur_time_mon = ktime_get_seconds(); |
1e50915f | 2229 | |
0e3ef49e | 2230 | if (rdev->last_read_error == 0) { |
1e50915f RB |
2231 | /* first time we've seen a read error */ |
2232 | rdev->last_read_error = cur_time_mon; | |
2233 | return; | |
2234 | } | |
2235 | ||
0e3ef49e AB |
2236 | hours_since_last = (long)(cur_time_mon - |
2237 | rdev->last_read_error) / 3600; | |
1e50915f RB |
2238 | |
2239 | rdev->last_read_error = cur_time_mon; | |
2240 | ||
2241 | /* | |
2242 | * if hours_since_last is > the number of bits in read_errors | |
2243 | * just set read errors to 0. We do this to avoid | |
2244 | * overflowing the shift of read_errors by hours_since_last. | |
2245 | */ | |
2246 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2247 | atomic_set(&rdev->read_errors, 0); | |
2248 | else | |
2249 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2250 | } | |
2251 | ||
3cb03002 | 2252 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2253 | int sectors, struct page *page, int rw) |
2254 | { | |
2255 | sector_t first_bad; | |
2256 | int bad_sectors; | |
2257 | ||
2258 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2259 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2260 | return -1; | |
796a5cf0 | 2261 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
58c54fcc N |
2262 | /* success */ |
2263 | return 1; | |
b7044d41 | 2264 | if (rw == WRITE) { |
58c54fcc | 2265 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2266 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2267 | set_bit(MD_RECOVERY_NEEDED, | |
2268 | &rdev->mddev->recovery); | |
2269 | } | |
58c54fcc N |
2270 | /* need to record an error - either for the block or the device */ |
2271 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2272 | md_error(rdev->mddev, rdev); | |
2273 | return 0; | |
2274 | } | |
2275 | ||
1da177e4 LT |
2276 | /* |
2277 | * This is a kernel thread which: | |
2278 | * | |
2279 | * 1. Retries failed read operations on working mirrors. | |
2280 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2281 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2282 | */ |
2283 | ||
e879a879 | 2284 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2285 | { |
2286 | int sect = 0; /* Offset from r10_bio->sector */ | |
2287 | int sectors = r10_bio->sectors; | |
3cb03002 | 2288 | struct md_rdev*rdev; |
1e50915f | 2289 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2290 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2291 | |
7c4e06ff N |
2292 | /* still own a reference to this rdev, so it cannot |
2293 | * have been cleared recently. | |
2294 | */ | |
2295 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2296 | |
7c4e06ff N |
2297 | if (test_bit(Faulty, &rdev->flags)) |
2298 | /* drive has already been failed, just ignore any | |
2299 | more fix_read_error() attempts */ | |
2300 | return; | |
1e50915f | 2301 | |
7c4e06ff N |
2302 | check_decay_read_errors(mddev, rdev); |
2303 | atomic_inc(&rdev->read_errors); | |
2304 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2305 | char b[BDEVNAME_SIZE]; | |
2306 | bdevname(rdev->bdev, b); | |
1e50915f | 2307 | |
08464e09 N |
2308 | pr_notice("md/raid10:%s: %s: Raid device exceeded read_error threshold [cur %d:max %d]\n", |
2309 | mdname(mddev), b, | |
2310 | atomic_read(&rdev->read_errors), max_read_errors); | |
2311 | pr_notice("md/raid10:%s: %s: Failing raid device\n", | |
2312 | mdname(mddev), b); | |
d683c8e0 | 2313 | md_error(mddev, rdev); |
fae8cc5e | 2314 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2315 | return; |
1e50915f | 2316 | } |
1e50915f | 2317 | |
6814d536 N |
2318 | while(sectors) { |
2319 | int s = sectors; | |
2320 | int sl = r10_bio->read_slot; | |
2321 | int success = 0; | |
2322 | int start; | |
2323 | ||
2324 | if (s > (PAGE_SIZE>>9)) | |
2325 | s = PAGE_SIZE >> 9; | |
2326 | ||
2327 | rcu_read_lock(); | |
2328 | do { | |
8dbed5ce N |
2329 | sector_t first_bad; |
2330 | int bad_sectors; | |
2331 | ||
0544a21d | 2332 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2333 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2334 | if (rdev && | |
8dbed5ce | 2335 | test_bit(In_sync, &rdev->flags) && |
f5b67ae8 | 2336 | !test_bit(Faulty, &rdev->flags) && |
8dbed5ce N |
2337 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, |
2338 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2339 | atomic_inc(&rdev->nr_pending); |
2340 | rcu_read_unlock(); | |
2b193363 | 2341 | success = sync_page_io(rdev, |
6814d536 | 2342 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2343 | sect, |
6814d536 | 2344 | s<<9, |
796a5cf0 MC |
2345 | conf->tmppage, |
2346 | REQ_OP_READ, 0, false); | |
6814d536 N |
2347 | rdev_dec_pending(rdev, mddev); |
2348 | rcu_read_lock(); | |
2349 | if (success) | |
2350 | break; | |
2351 | } | |
2352 | sl++; | |
2353 | if (sl == conf->copies) | |
2354 | sl = 0; | |
2355 | } while (!success && sl != r10_bio->read_slot); | |
2356 | rcu_read_unlock(); | |
2357 | ||
2358 | if (!success) { | |
58c54fcc N |
2359 | /* Cannot read from anywhere, just mark the block |
2360 | * as bad on the first device to discourage future | |
2361 | * reads. | |
2362 | */ | |
6814d536 | 2363 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2364 | rdev = conf->mirrors[dn].rdev; |
2365 | ||
2366 | if (!rdev_set_badblocks( | |
2367 | rdev, | |
2368 | r10_bio->devs[r10_bio->read_slot].addr | |
2369 | + sect, | |
fae8cc5e | 2370 | s, 0)) { |
58c54fcc | 2371 | md_error(mddev, rdev); |
fae8cc5e N |
2372 | r10_bio->devs[r10_bio->read_slot].bio |
2373 | = IO_BLOCKED; | |
2374 | } | |
6814d536 N |
2375 | break; |
2376 | } | |
2377 | ||
2378 | start = sl; | |
2379 | /* write it back and re-read */ | |
2380 | rcu_read_lock(); | |
2381 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2382 | char b[BDEVNAME_SIZE]; |
0544a21d | 2383 | |
6814d536 N |
2384 | if (sl==0) |
2385 | sl = conf->copies; | |
2386 | sl--; | |
2387 | d = r10_bio->devs[sl].devnum; | |
2388 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2389 | if (!rdev || |
f5b67ae8 | 2390 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2391 | !test_bit(In_sync, &rdev->flags)) |
2392 | continue; | |
2393 | ||
2394 | atomic_inc(&rdev->nr_pending); | |
2395 | rcu_read_unlock(); | |
58c54fcc N |
2396 | if (r10_sync_page_io(rdev, |
2397 | r10_bio->devs[sl].addr + | |
2398 | sect, | |
055d3747 | 2399 | s, conf->tmppage, WRITE) |
1294b9c9 N |
2400 | == 0) { |
2401 | /* Well, this device is dead */ | |
08464e09 N |
2402 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %s)\n", |
2403 | mdname(mddev), s, | |
2404 | (unsigned long long)( | |
2405 | sect + | |
2406 | choose_data_offset(r10_bio, | |
2407 | rdev)), | |
2408 | bdevname(rdev->bdev, b)); | |
2409 | pr_notice("md/raid10:%s: %s: failing drive\n", | |
2410 | mdname(mddev), | |
2411 | bdevname(rdev->bdev, b)); | |
6814d536 | 2412 | } |
1294b9c9 N |
2413 | rdev_dec_pending(rdev, mddev); |
2414 | rcu_read_lock(); | |
6814d536 N |
2415 | } |
2416 | sl = start; | |
2417 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2418 | char b[BDEVNAME_SIZE]; |
0544a21d | 2419 | |
6814d536 N |
2420 | if (sl==0) |
2421 | sl = conf->copies; | |
2422 | sl--; | |
2423 | d = r10_bio->devs[sl].devnum; | |
2424 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2425 | if (!rdev || |
f5b67ae8 | 2426 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2427 | !test_bit(In_sync, &rdev->flags)) |
2428 | continue; | |
6814d536 | 2429 | |
1294b9c9 N |
2430 | atomic_inc(&rdev->nr_pending); |
2431 | rcu_read_unlock(); | |
58c54fcc N |
2432 | switch (r10_sync_page_io(rdev, |
2433 | r10_bio->devs[sl].addr + | |
2434 | sect, | |
055d3747 | 2435 | s, conf->tmppage, |
58c54fcc N |
2436 | READ)) { |
2437 | case 0: | |
1294b9c9 | 2438 | /* Well, this device is dead */ |
08464e09 | 2439 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2440 | mdname(mddev), s, |
2441 | (unsigned long long)( | |
f8c9e74f N |
2442 | sect + |
2443 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 | 2444 | bdevname(rdev->bdev, b)); |
08464e09 | 2445 | pr_notice("md/raid10:%s: %s: failing drive\n", |
1294b9c9 N |
2446 | mdname(mddev), |
2447 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2448 | break; |
2449 | case 1: | |
08464e09 | 2450 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2451 | mdname(mddev), s, |
2452 | (unsigned long long)( | |
f8c9e74f N |
2453 | sect + |
2454 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2455 | bdevname(rdev->bdev, b)); |
2456 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2457 | } |
1294b9c9 N |
2458 | |
2459 | rdev_dec_pending(rdev, mddev); | |
2460 | rcu_read_lock(); | |
6814d536 N |
2461 | } |
2462 | rcu_read_unlock(); | |
2463 | ||
2464 | sectors -= s; | |
2465 | sect += s; | |
2466 | } | |
2467 | } | |
2468 | ||
9f2c9d12 | 2469 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2470 | { |
2471 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2472 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2473 | struct r10conf *conf = mddev->private; |
3cb03002 | 2474 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2475 | /* bio has the data to be written to slot 'i' where |
2476 | * we just recently had a write error. | |
2477 | * We repeatedly clone the bio and trim down to one block, | |
2478 | * then try the write. Where the write fails we record | |
2479 | * a bad block. | |
2480 | * It is conceivable that the bio doesn't exactly align with | |
2481 | * blocks. We must handle this. | |
2482 | * | |
2483 | * We currently own a reference to the rdev. | |
2484 | */ | |
2485 | ||
2486 | int block_sectors; | |
2487 | sector_t sector; | |
2488 | int sectors; | |
2489 | int sect_to_write = r10_bio->sectors; | |
2490 | int ok = 1; | |
2491 | ||
2492 | if (rdev->badblocks.shift < 0) | |
2493 | return 0; | |
2494 | ||
f04ebb0b N |
2495 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2496 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2497 | sector = r10_bio->sector; |
2498 | sectors = ((r10_bio->sector + block_sectors) | |
2499 | & ~(sector_t)(block_sectors - 1)) | |
2500 | - sector; | |
2501 | ||
2502 | while (sect_to_write) { | |
2503 | struct bio *wbio; | |
27028626 | 2504 | sector_t wsector; |
bd870a16 N |
2505 | if (sectors > sect_to_write) |
2506 | sectors = sect_to_write; | |
2507 | /* Write at 'sector' for 'sectors' */ | |
2508 | wbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
4f024f37 | 2509 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
27028626 TM |
2510 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); |
2511 | wbio->bi_iter.bi_sector = wsector + | |
2512 | choose_data_offset(r10_bio, rdev); | |
bd870a16 | 2513 | wbio->bi_bdev = rdev->bdev; |
796a5cf0 | 2514 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
2515 | |
2516 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 | 2517 | /* Failure! */ |
27028626 | 2518 | ok = rdev_set_badblocks(rdev, wsector, |
bd870a16 N |
2519 | sectors, 0) |
2520 | && ok; | |
2521 | ||
2522 | bio_put(wbio); | |
2523 | sect_to_write -= sectors; | |
2524 | sector += sectors; | |
2525 | sectors = block_sectors; | |
2526 | } | |
2527 | return ok; | |
2528 | } | |
2529 | ||
9f2c9d12 | 2530 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2531 | { |
2532 | int slot = r10_bio->read_slot; | |
560f8e55 | 2533 | struct bio *bio; |
e879a879 | 2534 | struct r10conf *conf = mddev->private; |
abbf098e | 2535 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2536 | char b[BDEVNAME_SIZE]; |
2537 | unsigned long do_sync; | |
856e08e2 | 2538 | int max_sectors; |
109e3765 N |
2539 | dev_t bio_dev; |
2540 | sector_t bio_last_sector; | |
560f8e55 N |
2541 | |
2542 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2543 | * the block and we can fix it. | |
2544 | * We freeze all other IO, and try reading the block from | |
2545 | * other devices. When we find one, we re-write | |
2546 | * and check it that fixes the read error. | |
2547 | * This is all done synchronously while the array is | |
2548 | * frozen. | |
2549 | */ | |
fae8cc5e N |
2550 | bio = r10_bio->devs[slot].bio; |
2551 | bdevname(bio->bi_bdev, b); | |
109e3765 N |
2552 | bio_dev = bio->bi_bdev->bd_dev; |
2553 | bio_last_sector = r10_bio->devs[slot].addr + rdev->data_offset + r10_bio->sectors; | |
fae8cc5e N |
2554 | bio_put(bio); |
2555 | r10_bio->devs[slot].bio = NULL; | |
2556 | ||
8d3ca83d N |
2557 | if (mddev->ro) |
2558 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2559 | else if (!test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2560 | freeze_array(conf, 1); |
560f8e55 N |
2561 | fix_read_error(conf, mddev, r10_bio); |
2562 | unfreeze_array(conf); | |
fae8cc5e | 2563 | } else |
8d3ca83d | 2564 | md_error(mddev, rdev); |
fae8cc5e | 2565 | |
abbf098e | 2566 | rdev_dec_pending(rdev, mddev); |
560f8e55 | 2567 | |
7399c31b | 2568 | read_more: |
96c3fd1f N |
2569 | rdev = read_balance(conf, r10_bio, &max_sectors); |
2570 | if (rdev == NULL) { | |
08464e09 N |
2571 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n", |
2572 | mdname(mddev), b, | |
2573 | (unsigned long long)r10_bio->sector); | |
560f8e55 | 2574 | raid_end_bio_io(r10_bio); |
560f8e55 N |
2575 | return; |
2576 | } | |
2577 | ||
1eff9d32 | 2578 | do_sync = (r10_bio->master_bio->bi_opf & REQ_SYNC); |
560f8e55 | 2579 | slot = r10_bio->read_slot; |
08464e09 N |
2580 | pr_err_ratelimited("md/raid10:%s: %s: redirecting sector %llu to another mirror\n", |
2581 | mdname(mddev), | |
2582 | bdevname(rdev->bdev, b), | |
2583 | (unsigned long long)r10_bio->sector); | |
560f8e55 N |
2584 | bio = bio_clone_mddev(r10_bio->master_bio, |
2585 | GFP_NOIO, mddev); | |
4f024f37 | 2586 | bio_trim(bio, r10_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
560f8e55 | 2587 | r10_bio->devs[slot].bio = bio; |
abbf098e | 2588 | r10_bio->devs[slot].rdev = rdev; |
4f024f37 | 2589 | bio->bi_iter.bi_sector = r10_bio->devs[slot].addr |
f8c9e74f | 2590 | + choose_data_offset(r10_bio, rdev); |
560f8e55 | 2591 | bio->bi_bdev = rdev->bdev; |
796a5cf0 | 2592 | bio_set_op_attrs(bio, REQ_OP_READ, do_sync); |
8d3ca83d N |
2593 | if (test_bit(FailFast, &rdev->flags) && |
2594 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
2595 | bio->bi_opf |= MD_FAILFAST; | |
560f8e55 N |
2596 | bio->bi_private = r10_bio; |
2597 | bio->bi_end_io = raid10_end_read_request; | |
109e3765 N |
2598 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), |
2599 | bio, bio_dev, | |
2600 | bio_last_sector - r10_bio->sectors); | |
2601 | ||
7399c31b N |
2602 | if (max_sectors < r10_bio->sectors) { |
2603 | /* Drat - have to split this up more */ | |
2604 | struct bio *mbio = r10_bio->master_bio; | |
2605 | int sectors_handled = | |
2606 | r10_bio->sector + max_sectors | |
4f024f37 | 2607 | - mbio->bi_iter.bi_sector; |
7399c31b N |
2608 | r10_bio->sectors = max_sectors; |
2609 | spin_lock_irq(&conf->device_lock); | |
2610 | if (mbio->bi_phys_segments == 0) | |
2611 | mbio->bi_phys_segments = 2; | |
2612 | else | |
2613 | mbio->bi_phys_segments++; | |
2614 | spin_unlock_irq(&conf->device_lock); | |
2615 | generic_make_request(bio); | |
7399c31b N |
2616 | |
2617 | r10_bio = mempool_alloc(conf->r10bio_pool, | |
2618 | GFP_NOIO); | |
2619 | r10_bio->master_bio = mbio; | |
aa8b57aa | 2620 | r10_bio->sectors = bio_sectors(mbio) - sectors_handled; |
7399c31b N |
2621 | r10_bio->state = 0; |
2622 | set_bit(R10BIO_ReadError, | |
2623 | &r10_bio->state); | |
2624 | r10_bio->mddev = mddev; | |
4f024f37 | 2625 | r10_bio->sector = mbio->bi_iter.bi_sector |
7399c31b N |
2626 | + sectors_handled; |
2627 | ||
2628 | goto read_more; | |
2629 | } else | |
2630 | generic_make_request(bio); | |
560f8e55 N |
2631 | } |
2632 | ||
e879a879 | 2633 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2634 | { |
2635 | /* Some sort of write request has finished and it | |
2636 | * succeeded in writing where we thought there was a | |
2637 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2638 | * Or possibly if failed and we need to record |
2639 | * a bad block. | |
749c55e9 N |
2640 | */ |
2641 | int m; | |
3cb03002 | 2642 | struct md_rdev *rdev; |
749c55e9 N |
2643 | |
2644 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2645 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2646 | for (m = 0; m < conf->copies; m++) { |
2647 | int dev = r10_bio->devs[m].devnum; | |
2648 | rdev = conf->mirrors[dev].rdev; | |
2649 | if (r10_bio->devs[m].bio == NULL) | |
2650 | continue; | |
4246a0b6 | 2651 | if (!r10_bio->devs[m].bio->bi_error) { |
749c55e9 N |
2652 | rdev_clear_badblocks( |
2653 | rdev, | |
2654 | r10_bio->devs[m].addr, | |
c6563a8c | 2655 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2656 | } else { |
2657 | if (!rdev_set_badblocks( | |
2658 | rdev, | |
2659 | r10_bio->devs[m].addr, | |
2660 | r10_bio->sectors, 0)) | |
2661 | md_error(conf->mddev, rdev); | |
749c55e9 | 2662 | } |
9ad1aefc N |
2663 | rdev = conf->mirrors[dev].replacement; |
2664 | if (r10_bio->devs[m].repl_bio == NULL) | |
2665 | continue; | |
4246a0b6 CH |
2666 | |
2667 | if (!r10_bio->devs[m].repl_bio->bi_error) { | |
9ad1aefc N |
2668 | rdev_clear_badblocks( |
2669 | rdev, | |
2670 | r10_bio->devs[m].addr, | |
c6563a8c | 2671 | r10_bio->sectors, 0); |
9ad1aefc N |
2672 | } else { |
2673 | if (!rdev_set_badblocks( | |
2674 | rdev, | |
2675 | r10_bio->devs[m].addr, | |
2676 | r10_bio->sectors, 0)) | |
2677 | md_error(conf->mddev, rdev); | |
2678 | } | |
1a0b7cd8 | 2679 | } |
749c55e9 N |
2680 | put_buf(r10_bio); |
2681 | } else { | |
95af587e | 2682 | bool fail = false; |
bd870a16 N |
2683 | for (m = 0; m < conf->copies; m++) { |
2684 | int dev = r10_bio->devs[m].devnum; | |
2685 | struct bio *bio = r10_bio->devs[m].bio; | |
2686 | rdev = conf->mirrors[dev].rdev; | |
2687 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2688 | rdev_clear_badblocks( |
2689 | rdev, | |
2690 | r10_bio->devs[m].addr, | |
c6563a8c | 2691 | r10_bio->sectors, 0); |
749c55e9 | 2692 | rdev_dec_pending(rdev, conf->mddev); |
4246a0b6 | 2693 | } else if (bio != NULL && bio->bi_error) { |
95af587e | 2694 | fail = true; |
bd870a16 N |
2695 | if (!narrow_write_error(r10_bio, m)) { |
2696 | md_error(conf->mddev, rdev); | |
2697 | set_bit(R10BIO_Degraded, | |
2698 | &r10_bio->state); | |
2699 | } | |
2700 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2701 | } |
475b0321 N |
2702 | bio = r10_bio->devs[m].repl_bio; |
2703 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2704 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2705 | rdev_clear_badblocks( |
2706 | rdev, | |
2707 | r10_bio->devs[m].addr, | |
c6563a8c | 2708 | r10_bio->sectors, 0); |
475b0321 N |
2709 | rdev_dec_pending(rdev, conf->mddev); |
2710 | } | |
bd870a16 | 2711 | } |
95af587e N |
2712 | if (fail) { |
2713 | spin_lock_irq(&conf->device_lock); | |
2714 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 2715 | conf->nr_queued++; |
95af587e N |
2716 | spin_unlock_irq(&conf->device_lock); |
2717 | md_wakeup_thread(conf->mddev->thread); | |
c340702c N |
2718 | } else { |
2719 | if (test_bit(R10BIO_WriteError, | |
2720 | &r10_bio->state)) | |
2721 | close_write(r10_bio); | |
95af587e | 2722 | raid_end_bio_io(r10_bio); |
c340702c | 2723 | } |
749c55e9 N |
2724 | } |
2725 | } | |
2726 | ||
4ed8731d | 2727 | static void raid10d(struct md_thread *thread) |
1da177e4 | 2728 | { |
4ed8731d | 2729 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2730 | struct r10bio *r10_bio; |
1da177e4 | 2731 | unsigned long flags; |
e879a879 | 2732 | struct r10conf *conf = mddev->private; |
1da177e4 | 2733 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2734 | struct blk_plug plug; |
1da177e4 LT |
2735 | |
2736 | md_check_recovery(mddev); | |
1da177e4 | 2737 | |
95af587e N |
2738 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2739 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
2740 | LIST_HEAD(tmp); | |
2741 | spin_lock_irqsave(&conf->device_lock, flags); | |
2742 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
23ddba80 SL |
2743 | while (!list_empty(&conf->bio_end_io_list)) { |
2744 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2745 | conf->nr_queued--; | |
2746 | } | |
95af587e N |
2747 | } |
2748 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2749 | while (!list_empty(&tmp)) { | |
a452744b MP |
2750 | r10_bio = list_first_entry(&tmp, struct r10bio, |
2751 | retry_list); | |
95af587e | 2752 | list_del(&r10_bio->retry_list); |
c340702c N |
2753 | if (mddev->degraded) |
2754 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
2755 | ||
2756 | if (test_bit(R10BIO_WriteError, | |
2757 | &r10_bio->state)) | |
2758 | close_write(r10_bio); | |
95af587e N |
2759 | raid_end_bio_io(r10_bio); |
2760 | } | |
2761 | } | |
2762 | ||
e1dfa0a2 | 2763 | blk_start_plug(&plug); |
1da177e4 | 2764 | for (;;) { |
6cce3b23 | 2765 | |
0021b7bc | 2766 | flush_pending_writes(conf); |
6cce3b23 | 2767 | |
a35e63ef N |
2768 | spin_lock_irqsave(&conf->device_lock, flags); |
2769 | if (list_empty(head)) { | |
2770 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2771 | break; |
a35e63ef | 2772 | } |
9f2c9d12 | 2773 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2774 | list_del(head->prev); |
4443ae10 | 2775 | conf->nr_queued--; |
1da177e4 LT |
2776 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2777 | ||
2778 | mddev = r10_bio->mddev; | |
070ec55d | 2779 | conf = mddev->private; |
bd870a16 N |
2780 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2781 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 2782 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
2783 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
2784 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 2785 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 2786 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2787 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2788 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2789 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2790 | handle_read_error(mddev, r10_bio); |
856e08e2 N |
2791 | else { |
2792 | /* just a partial read to be scheduled from a | |
2793 | * separate context | |
2794 | */ | |
2795 | int slot = r10_bio->read_slot; | |
2796 | generic_make_request(r10_bio->devs[slot].bio); | |
2797 | } | |
560f8e55 | 2798 | |
1d9d5241 | 2799 | cond_resched(); |
de393cde N |
2800 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2801 | md_check_recovery(mddev); | |
1da177e4 | 2802 | } |
e1dfa0a2 | 2803 | blk_finish_plug(&plug); |
1da177e4 LT |
2804 | } |
2805 | ||
e879a879 | 2806 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2807 | { |
2808 | int buffs; | |
69335ef3 | 2809 | int i; |
1da177e4 LT |
2810 | |
2811 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2812 | BUG_ON(conf->r10buf_pool); |
69335ef3 | 2813 | conf->have_replacement = 0; |
5cf00fcd | 2814 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
2815 | if (conf->mirrors[i].replacement) |
2816 | conf->have_replacement = 1; | |
1da177e4 LT |
2817 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2818 | if (!conf->r10buf_pool) | |
2819 | return -ENOMEM; | |
2820 | conf->next_resync = 0; | |
2821 | return 0; | |
2822 | } | |
2823 | ||
2824 | /* | |
2825 | * perform a "sync" on one "block" | |
2826 | * | |
2827 | * We need to make sure that no normal I/O request - particularly write | |
2828 | * requests - conflict with active sync requests. | |
2829 | * | |
2830 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2831 | * that can be installed to exclude normal IO requests. | |
2832 | * | |
2833 | * Resync and recovery are handled very differently. | |
2834 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2835 | * | |
2836 | * For resync, we iterate over virtual addresses, read all copies, | |
2837 | * and update if there are differences. If only one copy is live, | |
2838 | * skip it. | |
2839 | * For recovery, we iterate over physical addresses, read a good | |
2840 | * value for each non-in_sync drive, and over-write. | |
2841 | * | |
2842 | * So, for recovery we may have several outstanding complex requests for a | |
2843 | * given address, one for each out-of-sync device. We model this by allocating | |
2844 | * a number of r10_bio structures, one for each out-of-sync device. | |
2845 | * As we setup these structures, we collect all bio's together into a list | |
2846 | * which we then process collectively to add pages, and then process again | |
2847 | * to pass to generic_make_request. | |
2848 | * | |
2849 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2850 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2851 | * has its remaining count decremented to 0, the whole complex operation | |
2852 | * is complete. | |
2853 | * | |
2854 | */ | |
2855 | ||
849674e4 | 2856 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 2857 | int *skipped) |
1da177e4 | 2858 | { |
e879a879 | 2859 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2860 | struct r10bio *r10_bio; |
1da177e4 LT |
2861 | struct bio *biolist = NULL, *bio; |
2862 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2863 | int i; |
6cce3b23 | 2864 | int max_sync; |
57dab0bd | 2865 | sector_t sync_blocks; |
1da177e4 LT |
2866 | sector_t sectors_skipped = 0; |
2867 | int chunks_skipped = 0; | |
5cf00fcd | 2868 | sector_t chunk_mask = conf->geo.chunk_mask; |
1da177e4 LT |
2869 | |
2870 | if (!conf->r10buf_pool) | |
2871 | if (init_resync(conf)) | |
57afd89f | 2872 | return 0; |
1da177e4 | 2873 | |
7e83ccbe MW |
2874 | /* |
2875 | * Allow skipping a full rebuild for incremental assembly | |
2876 | * of a clean array, like RAID1 does. | |
2877 | */ | |
2878 | if (mddev->bitmap == NULL && | |
2879 | mddev->recovery_cp == MaxSector && | |
13765120 N |
2880 | mddev->reshape_position == MaxSector && |
2881 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 2882 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 2883 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
2884 | conf->fullsync == 0) { |
2885 | *skipped = 1; | |
13765120 | 2886 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
2887 | } |
2888 | ||
1da177e4 | 2889 | skipped: |
58c0fed4 | 2890 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
2891 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
2892 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
2893 | max_sector = mddev->resync_max_sectors; |
2894 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
2895 | /* If we aborted, we need to abort the |
2896 | * sync on the 'current' bitmap chucks (there can | |
2897 | * be several when recovering multiple devices). | |
2898 | * as we may have started syncing it but not finished. | |
2899 | * We can find the current address in | |
2900 | * mddev->curr_resync, but for recovery, | |
2901 | * we need to convert that to several | |
2902 | * virtual addresses. | |
2903 | */ | |
3ea7daa5 N |
2904 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
2905 | end_reshape(conf); | |
b3968552 | 2906 | close_sync(conf); |
3ea7daa5 N |
2907 | return 0; |
2908 | } | |
2909 | ||
6cce3b23 N |
2910 | if (mddev->curr_resync < max_sector) { /* aborted */ |
2911 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2912 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
2913 | &sync_blocks, 1); | |
5cf00fcd | 2914 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
2915 | sector_t sect = |
2916 | raid10_find_virt(conf, mddev->curr_resync, i); | |
2917 | bitmap_end_sync(mddev->bitmap, sect, | |
2918 | &sync_blocks, 1); | |
2919 | } | |
9ad1aefc N |
2920 | } else { |
2921 | /* completed sync */ | |
2922 | if ((!mddev->bitmap || conf->fullsync) | |
2923 | && conf->have_replacement | |
2924 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
2925 | /* Completed a full sync so the replacements | |
2926 | * are now fully recovered. | |
2927 | */ | |
f90145f3 N |
2928 | rcu_read_lock(); |
2929 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
2930 | struct md_rdev *rdev = | |
2931 | rcu_dereference(conf->mirrors[i].replacement); | |
2932 | if (rdev) | |
2933 | rdev->recovery_offset = MaxSector; | |
2934 | } | |
2935 | rcu_read_unlock(); | |
9ad1aefc | 2936 | } |
6cce3b23 | 2937 | conf->fullsync = 0; |
9ad1aefc | 2938 | } |
6cce3b23 | 2939 | bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2940 | close_sync(conf); |
57afd89f | 2941 | *skipped = 1; |
1da177e4 LT |
2942 | return sectors_skipped; |
2943 | } | |
3ea7daa5 N |
2944 | |
2945 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
2946 | return reshape_request(mddev, sector_nr, skipped); | |
2947 | ||
5cf00fcd | 2948 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
2949 | /* if there has been nothing to do on any drive, |
2950 | * then there is nothing to do at all.. | |
2951 | */ | |
57afd89f N |
2952 | *skipped = 1; |
2953 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
2954 | } |
2955 | ||
c6207277 N |
2956 | if (max_sector > mddev->resync_max) |
2957 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
2958 | ||
1da177e4 LT |
2959 | /* make sure whole request will fit in a chunk - if chunks |
2960 | * are meaningful | |
2961 | */ | |
5cf00fcd N |
2962 | if (conf->geo.near_copies < conf->geo.raid_disks && |
2963 | max_sector > (sector_nr | chunk_mask)) | |
2964 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 | 2965 | |
7ac50447 TM |
2966 | /* |
2967 | * If there is non-resync activity waiting for a turn, then let it | |
2968 | * though before starting on this new sync request. | |
2969 | */ | |
2970 | if (conf->nr_waiting) | |
2971 | schedule_timeout_uninterruptible(1); | |
2972 | ||
1da177e4 LT |
2973 | /* Again, very different code for resync and recovery. |
2974 | * Both must result in an r10bio with a list of bios that | |
2975 | * have bi_end_io, bi_sector, bi_bdev set, | |
2976 | * and bi_private set to the r10bio. | |
2977 | * For recovery, we may actually create several r10bios | |
2978 | * with 2 bios in each, that correspond to the bios in the main one. | |
2979 | * In this case, the subordinate r10bios link back through a | |
2980 | * borrowed master_bio pointer, and the counter in the master | |
2981 | * includes a ref from each subordinate. | |
2982 | */ | |
2983 | /* First, we decide what to do and set ->bi_end_io | |
2984 | * To end_sync_read if we want to read, and | |
2985 | * end_sync_write if we will want to write. | |
2986 | */ | |
2987 | ||
6cce3b23 | 2988 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
2989 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
2990 | /* recovery... the complicated one */ | |
e875ecea | 2991 | int j; |
1da177e4 LT |
2992 | r10_bio = NULL; |
2993 | ||
5cf00fcd | 2994 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 2995 | int still_degraded; |
9f2c9d12 | 2996 | struct r10bio *rb2; |
ab9d47e9 N |
2997 | sector_t sect; |
2998 | int must_sync; | |
e875ecea | 2999 | int any_working; |
dc280d98 | 3000 | struct raid10_info *mirror = &conf->mirrors[i]; |
f90145f3 | 3001 | struct md_rdev *mrdev, *mreplace; |
24afd80d | 3002 | |
f90145f3 N |
3003 | rcu_read_lock(); |
3004 | mrdev = rcu_dereference(mirror->rdev); | |
3005 | mreplace = rcu_dereference(mirror->replacement); | |
3006 | ||
3007 | if ((mrdev == NULL || | |
f5b67ae8 | 3008 | test_bit(Faulty, &mrdev->flags) || |
f90145f3 N |
3009 | test_bit(In_sync, &mrdev->flags)) && |
3010 | (mreplace == NULL || | |
3011 | test_bit(Faulty, &mreplace->flags))) { | |
3012 | rcu_read_unlock(); | |
ab9d47e9 | 3013 | continue; |
f90145f3 | 3014 | } |
1da177e4 | 3015 | |
ab9d47e9 N |
3016 | still_degraded = 0; |
3017 | /* want to reconstruct this device */ | |
3018 | rb2 = r10_bio; | |
3019 | sect = raid10_find_virt(conf, sector_nr, i); | |
fc448a18 N |
3020 | if (sect >= mddev->resync_max_sectors) { |
3021 | /* last stripe is not complete - don't | |
3022 | * try to recover this sector. | |
3023 | */ | |
f90145f3 | 3024 | rcu_read_unlock(); |
fc448a18 N |
3025 | continue; |
3026 | } | |
f5b67ae8 N |
3027 | if (mreplace && test_bit(Faulty, &mreplace->flags)) |
3028 | mreplace = NULL; | |
24afd80d N |
3029 | /* Unless we are doing a full sync, or a replacement |
3030 | * we only need to recover the block if it is set in | |
3031 | * the bitmap | |
ab9d47e9 N |
3032 | */ |
3033 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3034 | &sync_blocks, 1); | |
3035 | if (sync_blocks < max_sync) | |
3036 | max_sync = sync_blocks; | |
3037 | if (!must_sync && | |
f90145f3 | 3038 | mreplace == NULL && |
ab9d47e9 N |
3039 | !conf->fullsync) { |
3040 | /* yep, skip the sync_blocks here, but don't assume | |
3041 | * that there will never be anything to do here | |
3042 | */ | |
3043 | chunks_skipped = -1; | |
f90145f3 | 3044 | rcu_read_unlock(); |
ab9d47e9 N |
3045 | continue; |
3046 | } | |
f90145f3 N |
3047 | atomic_inc(&mrdev->nr_pending); |
3048 | if (mreplace) | |
3049 | atomic_inc(&mreplace->nr_pending); | |
3050 | rcu_read_unlock(); | |
6cce3b23 | 3051 | |
ab9d47e9 | 3052 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 3053 | r10_bio->state = 0; |
ab9d47e9 N |
3054 | raise_barrier(conf, rb2 != NULL); |
3055 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 3056 | |
ab9d47e9 N |
3057 | r10_bio->master_bio = (struct bio*)rb2; |
3058 | if (rb2) | |
3059 | atomic_inc(&rb2->remaining); | |
3060 | r10_bio->mddev = mddev; | |
3061 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
3062 | r10_bio->sector = sect; | |
1da177e4 | 3063 | |
ab9d47e9 N |
3064 | raid10_find_phys(conf, r10_bio); |
3065 | ||
3066 | /* Need to check if the array will still be | |
3067 | * degraded | |
3068 | */ | |
f90145f3 N |
3069 | rcu_read_lock(); |
3070 | for (j = 0; j < conf->geo.raid_disks; j++) { | |
3071 | struct md_rdev *rdev = rcu_dereference( | |
3072 | conf->mirrors[j].rdev); | |
3073 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
ab9d47e9 | 3074 | still_degraded = 1; |
87fc767b | 3075 | break; |
1da177e4 | 3076 | } |
f90145f3 | 3077 | } |
ab9d47e9 N |
3078 | |
3079 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3080 | &sync_blocks, still_degraded); | |
3081 | ||
e875ecea | 3082 | any_working = 0; |
ab9d47e9 | 3083 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3084 | int k; |
ab9d47e9 | 3085 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3086 | sector_t from_addr, to_addr; |
f90145f3 N |
3087 | struct md_rdev *rdev = |
3088 | rcu_dereference(conf->mirrors[d].rdev); | |
40c356ce N |
3089 | sector_t sector, first_bad; |
3090 | int bad_sectors; | |
f90145f3 N |
3091 | if (!rdev || |
3092 | !test_bit(In_sync, &rdev->flags)) | |
ab9d47e9 N |
3093 | continue; |
3094 | /* This is where we read from */ | |
e875ecea | 3095 | any_working = 1; |
40c356ce N |
3096 | sector = r10_bio->devs[j].addr; |
3097 | ||
3098 | if (is_badblock(rdev, sector, max_sync, | |
3099 | &first_bad, &bad_sectors)) { | |
3100 | if (first_bad > sector) | |
3101 | max_sync = first_bad - sector; | |
3102 | else { | |
3103 | bad_sectors -= (sector | |
3104 | - first_bad); | |
3105 | if (max_sync > bad_sectors) | |
3106 | max_sync = bad_sectors; | |
3107 | continue; | |
3108 | } | |
3109 | } | |
ab9d47e9 | 3110 | bio = r10_bio->devs[0].bio; |
8be185f2 | 3111 | bio_reset(bio); |
ab9d47e9 N |
3112 | bio->bi_next = biolist; |
3113 | biolist = bio; | |
3114 | bio->bi_private = r10_bio; | |
3115 | bio->bi_end_io = end_sync_read; | |
796a5cf0 | 3116 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3117 | if (test_bit(FailFast, &rdev->flags)) |
3118 | bio->bi_opf |= MD_FAILFAST; | |
5e570289 | 3119 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3120 | bio->bi_iter.bi_sector = from_addr + |
3121 | rdev->data_offset; | |
24afd80d N |
3122 | bio->bi_bdev = rdev->bdev; |
3123 | atomic_inc(&rdev->nr_pending); | |
3124 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3125 | |
3126 | for (k=0; k<conf->copies; k++) | |
3127 | if (r10_bio->devs[k].devnum == i) | |
3128 | break; | |
3129 | BUG_ON(k == conf->copies); | |
5e570289 | 3130 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3131 | r10_bio->devs[0].devnum = d; |
5e570289 | 3132 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3133 | r10_bio->devs[1].devnum = i; |
5e570289 | 3134 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3135 | |
f90145f3 | 3136 | if (!test_bit(In_sync, &mrdev->flags)) { |
24afd80d | 3137 | bio = r10_bio->devs[1].bio; |
8be185f2 | 3138 | bio_reset(bio); |
24afd80d N |
3139 | bio->bi_next = biolist; |
3140 | biolist = bio; | |
3141 | bio->bi_private = r10_bio; | |
3142 | bio->bi_end_io = end_sync_write; | |
796a5cf0 | 3143 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3144 | bio->bi_iter.bi_sector = to_addr |
f90145f3 N |
3145 | + mrdev->data_offset; |
3146 | bio->bi_bdev = mrdev->bdev; | |
24afd80d N |
3147 | atomic_inc(&r10_bio->remaining); |
3148 | } else | |
3149 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3150 | ||
3151 | /* and maybe write to replacement */ | |
3152 | bio = r10_bio->devs[1].repl_bio; | |
3153 | if (bio) | |
3154 | bio->bi_end_io = NULL; | |
f90145f3 | 3155 | /* Note: if mreplace != NULL, then bio |
24afd80d N |
3156 | * cannot be NULL as r10buf_pool_alloc will |
3157 | * have allocated it. | |
3158 | * So the second test here is pointless. | |
3159 | * But it keeps semantic-checkers happy, and | |
3160 | * this comment keeps human reviewers | |
3161 | * happy. | |
3162 | */ | |
f90145f3 N |
3163 | if (mreplace == NULL || bio == NULL || |
3164 | test_bit(Faulty, &mreplace->flags)) | |
24afd80d | 3165 | break; |
8be185f2 | 3166 | bio_reset(bio); |
24afd80d N |
3167 | bio->bi_next = biolist; |
3168 | biolist = bio; | |
3169 | bio->bi_private = r10_bio; | |
3170 | bio->bi_end_io = end_sync_write; | |
796a5cf0 | 3171 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3172 | bio->bi_iter.bi_sector = to_addr + |
f90145f3 N |
3173 | mreplace->data_offset; |
3174 | bio->bi_bdev = mreplace->bdev; | |
24afd80d | 3175 | atomic_inc(&r10_bio->remaining); |
ab9d47e9 N |
3176 | break; |
3177 | } | |
f90145f3 | 3178 | rcu_read_unlock(); |
ab9d47e9 | 3179 | if (j == conf->copies) { |
e875ecea N |
3180 | /* Cannot recover, so abort the recovery or |
3181 | * record a bad block */ | |
e875ecea N |
3182 | if (any_working) { |
3183 | /* problem is that there are bad blocks | |
3184 | * on other device(s) | |
3185 | */ | |
3186 | int k; | |
3187 | for (k = 0; k < conf->copies; k++) | |
3188 | if (r10_bio->devs[k].devnum == i) | |
3189 | break; | |
24afd80d | 3190 | if (!test_bit(In_sync, |
f90145f3 | 3191 | &mrdev->flags) |
24afd80d | 3192 | && !rdev_set_badblocks( |
f90145f3 | 3193 | mrdev, |
24afd80d N |
3194 | r10_bio->devs[k].addr, |
3195 | max_sync, 0)) | |
3196 | any_working = 0; | |
f90145f3 | 3197 | if (mreplace && |
24afd80d | 3198 | !rdev_set_badblocks( |
f90145f3 | 3199 | mreplace, |
e875ecea N |
3200 | r10_bio->devs[k].addr, |
3201 | max_sync, 0)) | |
3202 | any_working = 0; | |
3203 | } | |
3204 | if (!any_working) { | |
3205 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3206 | &mddev->recovery)) | |
08464e09 | 3207 | pr_warn("md/raid10:%s: insufficient working devices for recovery.\n", |
e875ecea | 3208 | mdname(mddev)); |
24afd80d | 3209 | mirror->recovery_disabled |
e875ecea N |
3210 | = mddev->recovery_disabled; |
3211 | } | |
e8b84915 N |
3212 | put_buf(r10_bio); |
3213 | if (rb2) | |
3214 | atomic_dec(&rb2->remaining); | |
3215 | r10_bio = rb2; | |
f90145f3 N |
3216 | rdev_dec_pending(mrdev, mddev); |
3217 | if (mreplace) | |
3218 | rdev_dec_pending(mreplace, mddev); | |
ab9d47e9 | 3219 | break; |
1da177e4 | 3220 | } |
f90145f3 N |
3221 | rdev_dec_pending(mrdev, mddev); |
3222 | if (mreplace) | |
3223 | rdev_dec_pending(mreplace, mddev); | |
8d3ca83d N |
3224 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { |
3225 | /* Only want this if there is elsewhere to | |
3226 | * read from. 'j' is currently the first | |
3227 | * readable copy. | |
3228 | */ | |
3229 | int targets = 1; | |
3230 | for (; j < conf->copies; j++) { | |
3231 | int d = r10_bio->devs[j].devnum; | |
3232 | if (conf->mirrors[d].rdev && | |
3233 | test_bit(In_sync, | |
3234 | &conf->mirrors[d].rdev->flags)) | |
3235 | targets++; | |
3236 | } | |
3237 | if (targets == 1) | |
3238 | r10_bio->devs[0].bio->bi_opf | |
3239 | &= ~MD_FAILFAST; | |
3240 | } | |
ab9d47e9 | 3241 | } |
1da177e4 LT |
3242 | if (biolist == NULL) { |
3243 | while (r10_bio) { | |
9f2c9d12 N |
3244 | struct r10bio *rb2 = r10_bio; |
3245 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3246 | rb2->master_bio = NULL; |
3247 | put_buf(rb2); | |
3248 | } | |
3249 | goto giveup; | |
3250 | } | |
3251 | } else { | |
3252 | /* resync. Schedule a read for every block at this virt offset */ | |
3253 | int count = 0; | |
6cce3b23 | 3254 | |
c40f341f | 3255 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, 0); |
78200d45 | 3256 | |
6cce3b23 N |
3257 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
3258 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3259 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3260 | &mddev->recovery)) { | |
6cce3b23 N |
3261 | /* We can skip this block */ |
3262 | *skipped = 1; | |
3263 | return sync_blocks + sectors_skipped; | |
3264 | } | |
3265 | if (sync_blocks < max_sync) | |
3266 | max_sync = sync_blocks; | |
1da177e4 | 3267 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 3268 | r10_bio->state = 0; |
1da177e4 | 3269 | |
1da177e4 LT |
3270 | r10_bio->mddev = mddev; |
3271 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3272 | raise_barrier(conf, 0); |
3273 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3274 | |
3275 | r10_bio->master_bio = NULL; | |
3276 | r10_bio->sector = sector_nr; | |
3277 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3278 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3279 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3280 | |
5cf00fcd | 3281 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3282 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3283 | sector_t first_bad, sector; |
3284 | int bad_sectors; | |
f90145f3 | 3285 | struct md_rdev *rdev; |
40c356ce | 3286 | |
9ad1aefc N |
3287 | if (r10_bio->devs[i].repl_bio) |
3288 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3289 | ||
1da177e4 | 3290 | bio = r10_bio->devs[i].bio; |
8be185f2 | 3291 | bio_reset(bio); |
4246a0b6 | 3292 | bio->bi_error = -EIO; |
f90145f3 N |
3293 | rcu_read_lock(); |
3294 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
3295 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3296 | rcu_read_unlock(); | |
1da177e4 | 3297 | continue; |
f90145f3 | 3298 | } |
40c356ce | 3299 | sector = r10_bio->devs[i].addr; |
f90145f3 | 3300 | if (is_badblock(rdev, sector, max_sync, |
40c356ce N |
3301 | &first_bad, &bad_sectors)) { |
3302 | if (first_bad > sector) | |
3303 | max_sync = first_bad - sector; | |
3304 | else { | |
3305 | bad_sectors -= (sector - first_bad); | |
3306 | if (max_sync > bad_sectors) | |
91502f09 | 3307 | max_sync = bad_sectors; |
f90145f3 | 3308 | rcu_read_unlock(); |
40c356ce N |
3309 | continue; |
3310 | } | |
3311 | } | |
f90145f3 | 3312 | atomic_inc(&rdev->nr_pending); |
1da177e4 LT |
3313 | atomic_inc(&r10_bio->remaining); |
3314 | bio->bi_next = biolist; | |
3315 | biolist = bio; | |
3316 | bio->bi_private = r10_bio; | |
3317 | bio->bi_end_io = end_sync_read; | |
796a5cf0 | 3318 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3319 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
3320 | bio->bi_opf |= MD_FAILFAST; | |
f90145f3 N |
3321 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
3322 | bio->bi_bdev = rdev->bdev; | |
1da177e4 | 3323 | count++; |
9ad1aefc | 3324 | |
f90145f3 N |
3325 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3326 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3327 | rcu_read_unlock(); | |
9ad1aefc | 3328 | continue; |
f90145f3 N |
3329 | } |
3330 | atomic_inc(&rdev->nr_pending); | |
3331 | rcu_read_unlock(); | |
9ad1aefc N |
3332 | |
3333 | /* Need to set up for writing to the replacement */ | |
3334 | bio = r10_bio->devs[i].repl_bio; | |
8be185f2 | 3335 | bio_reset(bio); |
4246a0b6 | 3336 | bio->bi_error = -EIO; |
9ad1aefc N |
3337 | |
3338 | sector = r10_bio->devs[i].addr; | |
9ad1aefc N |
3339 | bio->bi_next = biolist; |
3340 | biolist = bio; | |
3341 | bio->bi_private = r10_bio; | |
3342 | bio->bi_end_io = end_sync_write; | |
796a5cf0 | 3343 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
f90145f3 N |
3344 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
3345 | bio->bi_bdev = rdev->bdev; | |
9ad1aefc | 3346 | count++; |
1da177e4 LT |
3347 | } |
3348 | ||
3349 | if (count < 2) { | |
3350 | for (i=0; i<conf->copies; i++) { | |
3351 | int d = r10_bio->devs[i].devnum; | |
3352 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3353 | rdev_dec_pending(conf->mirrors[d].rdev, |
3354 | mddev); | |
9ad1aefc N |
3355 | if (r10_bio->devs[i].repl_bio && |
3356 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3357 | rdev_dec_pending( | |
3358 | conf->mirrors[d].replacement, | |
3359 | mddev); | |
1da177e4 LT |
3360 | } |
3361 | put_buf(r10_bio); | |
3362 | biolist = NULL; | |
3363 | goto giveup; | |
3364 | } | |
3365 | } | |
3366 | ||
1da177e4 | 3367 | nr_sectors = 0; |
6cce3b23 N |
3368 | if (sector_nr + max_sync < max_sector) |
3369 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3370 | do { |
3371 | struct page *page; | |
3372 | int len = PAGE_SIZE; | |
1da177e4 LT |
3373 | if (sector_nr + (len>>9) > max_sector) |
3374 | len = (max_sector - sector_nr) << 9; | |
3375 | if (len == 0) | |
3376 | break; | |
3377 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
ab9d47e9 | 3378 | struct bio *bio2; |
1da177e4 | 3379 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
ab9d47e9 N |
3380 | if (bio_add_page(bio, page, len, 0)) |
3381 | continue; | |
3382 | ||
3383 | /* stop here */ | |
3384 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
3385 | for (bio2 = biolist; | |
3386 | bio2 && bio2 != bio; | |
3387 | bio2 = bio2->bi_next) { | |
3388 | /* remove last page from this bio */ | |
3389 | bio2->bi_vcnt--; | |
4f024f37 | 3390 | bio2->bi_iter.bi_size -= len; |
b7c44ed9 | 3391 | bio_clear_flag(bio2, BIO_SEG_VALID); |
1da177e4 | 3392 | } |
ab9d47e9 | 3393 | goto bio_full; |
1da177e4 LT |
3394 | } |
3395 | nr_sectors += len>>9; | |
3396 | sector_nr += len>>9; | |
3397 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
3398 | bio_full: | |
3399 | r10_bio->sectors = nr_sectors; | |
3400 | ||
3401 | while (biolist) { | |
3402 | bio = biolist; | |
3403 | biolist = biolist->bi_next; | |
3404 | ||
3405 | bio->bi_next = NULL; | |
3406 | r10_bio = bio->bi_private; | |
3407 | r10_bio->sectors = nr_sectors; | |
3408 | ||
3409 | if (bio->bi_end_io == end_sync_read) { | |
3410 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
4246a0b6 | 3411 | bio->bi_error = 0; |
1da177e4 LT |
3412 | generic_make_request(bio); |
3413 | } | |
3414 | } | |
3415 | ||
57afd89f N |
3416 | if (sectors_skipped) |
3417 | /* pretend they weren't skipped, it makes | |
3418 | * no important difference in this case | |
3419 | */ | |
3420 | md_done_sync(mddev, sectors_skipped, 1); | |
3421 | ||
1da177e4 LT |
3422 | return sectors_skipped + nr_sectors; |
3423 | giveup: | |
3424 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3425 | * drives must be failed or in resync, all drives |
3426 | * have a bad block, so try the next chunk... | |
1da177e4 | 3427 | */ |
09b4068a N |
3428 | if (sector_nr + max_sync < max_sector) |
3429 | max_sector = sector_nr + max_sync; | |
3430 | ||
3431 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3432 | chunks_skipped ++; |
3433 | sector_nr = max_sector; | |
1da177e4 | 3434 | goto skipped; |
1da177e4 LT |
3435 | } |
3436 | ||
80c3a6ce | 3437 | static sector_t |
fd01b88c | 3438 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3439 | { |
3440 | sector_t size; | |
e879a879 | 3441 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3442 | |
3443 | if (!raid_disks) | |
3ea7daa5 N |
3444 | raid_disks = min(conf->geo.raid_disks, |
3445 | conf->prev.raid_disks); | |
80c3a6ce | 3446 | if (!sectors) |
dab8b292 | 3447 | sectors = conf->dev_sectors; |
80c3a6ce | 3448 | |
5cf00fcd N |
3449 | size = sectors >> conf->geo.chunk_shift; |
3450 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3451 | size = size * raid_disks; |
5cf00fcd | 3452 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3453 | |
5cf00fcd | 3454 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3455 | } |
3456 | ||
6508fdbf N |
3457 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3458 | { | |
3459 | /* Calculate the number of sectors-per-device that will | |
3460 | * actually be used, and set conf->dev_sectors and | |
3461 | * conf->stride | |
3462 | */ | |
3463 | ||
5cf00fcd N |
3464 | size = size >> conf->geo.chunk_shift; |
3465 | sector_div(size, conf->geo.far_copies); | |
3466 | size = size * conf->geo.raid_disks; | |
3467 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3468 | /* 'size' is now the number of chunks in the array */ |
3469 | /* calculate "used chunks per device" */ | |
3470 | size = size * conf->copies; | |
3471 | ||
3472 | /* We need to round up when dividing by raid_disks to | |
3473 | * get the stride size. | |
3474 | */ | |
5cf00fcd | 3475 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3476 | |
5cf00fcd | 3477 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3478 | |
5cf00fcd N |
3479 | if (conf->geo.far_offset) |
3480 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3481 | else { |
5cf00fcd N |
3482 | sector_div(size, conf->geo.far_copies); |
3483 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3484 | } |
3485 | } | |
dab8b292 | 3486 | |
deb200d0 N |
3487 | enum geo_type {geo_new, geo_old, geo_start}; |
3488 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3489 | { | |
3490 | int nc, fc, fo; | |
3491 | int layout, chunk, disks; | |
3492 | switch (new) { | |
3493 | case geo_old: | |
3494 | layout = mddev->layout; | |
3495 | chunk = mddev->chunk_sectors; | |
3496 | disks = mddev->raid_disks - mddev->delta_disks; | |
3497 | break; | |
3498 | case geo_new: | |
3499 | layout = mddev->new_layout; | |
3500 | chunk = mddev->new_chunk_sectors; | |
3501 | disks = mddev->raid_disks; | |
3502 | break; | |
3503 | default: /* avoid 'may be unused' warnings */ | |
3504 | case geo_start: /* new when starting reshape - raid_disks not | |
3505 | * updated yet. */ | |
3506 | layout = mddev->new_layout; | |
3507 | chunk = mddev->new_chunk_sectors; | |
3508 | disks = mddev->raid_disks + mddev->delta_disks; | |
3509 | break; | |
3510 | } | |
8bce6d35 | 3511 | if (layout >> 19) |
deb200d0 N |
3512 | return -1; |
3513 | if (chunk < (PAGE_SIZE >> 9) || | |
3514 | !is_power_of_2(chunk)) | |
3515 | return -2; | |
3516 | nc = layout & 255; | |
3517 | fc = (layout >> 8) & 255; | |
3518 | fo = layout & (1<<16); | |
3519 | geo->raid_disks = disks; | |
3520 | geo->near_copies = nc; | |
3521 | geo->far_copies = fc; | |
3522 | geo->far_offset = fo; | |
8bce6d35 N |
3523 | switch (layout >> 17) { |
3524 | case 0: /* original layout. simple but not always optimal */ | |
3525 | geo->far_set_size = disks; | |
3526 | break; | |
3527 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3528 | * actually using this, but leave code here just in case.*/ | |
3529 | geo->far_set_size = disks/fc; | |
3530 | WARN(geo->far_set_size < fc, | |
3531 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3532 | break; | |
3533 | case 2: /* "improved" layout fixed to match documentation */ | |
3534 | geo->far_set_size = fc * nc; | |
3535 | break; | |
3536 | default: /* Not a valid layout */ | |
3537 | return -1; | |
3538 | } | |
deb200d0 N |
3539 | geo->chunk_mask = chunk - 1; |
3540 | geo->chunk_shift = ffz(~chunk); | |
3541 | return nc*fc; | |
3542 | } | |
3543 | ||
e879a879 | 3544 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3545 | { |
e879a879 | 3546 | struct r10conf *conf = NULL; |
dab8b292 | 3547 | int err = -EINVAL; |
deb200d0 N |
3548 | struct geom geo; |
3549 | int copies; | |
3550 | ||
3551 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3552 | |
deb200d0 | 3553 | if (copies == -2) { |
08464e09 N |
3554 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n", |
3555 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3556 | goto out; |
1da177e4 | 3557 | } |
2604b703 | 3558 | |
deb200d0 | 3559 | if (copies < 2 || copies > mddev->raid_disks) { |
08464e09 N |
3560 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
3561 | mdname(mddev), mddev->new_layout); | |
1da177e4 LT |
3562 | goto out; |
3563 | } | |
dab8b292 TM |
3564 | |
3565 | err = -ENOMEM; | |
e879a879 | 3566 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3567 | if (!conf) |
1da177e4 | 3568 | goto out; |
dab8b292 | 3569 | |
3ea7daa5 | 3570 | /* FIXME calc properly */ |
dc280d98 | 3571 | conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks + |
78eaa0d4 | 3572 | max(0,-mddev->delta_disks)), |
dab8b292 TM |
3573 | GFP_KERNEL); |
3574 | if (!conf->mirrors) | |
3575 | goto out; | |
4443ae10 N |
3576 | |
3577 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3578 | if (!conf->tmppage) | |
dab8b292 TM |
3579 | goto out; |
3580 | ||
deb200d0 N |
3581 | conf->geo = geo; |
3582 | conf->copies = copies; | |
dab8b292 TM |
3583 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, |
3584 | r10bio_pool_free, conf); | |
3585 | if (!conf->r10bio_pool) | |
3586 | goto out; | |
3587 | ||
6508fdbf | 3588 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3589 | if (mddev->reshape_position == MaxSector) { |
3590 | conf->prev = conf->geo; | |
3591 | conf->reshape_progress = MaxSector; | |
3592 | } else { | |
3593 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3594 | err = -EINVAL; | |
3595 | goto out; | |
3596 | } | |
3597 | conf->reshape_progress = mddev->reshape_position; | |
3598 | if (conf->prev.far_offset) | |
3599 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
3600 | else | |
3601 | /* far_copies must be 1 */ | |
3602 | conf->prev.stride = conf->dev_sectors; | |
3603 | } | |
299b0685 | 3604 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 3605 | spin_lock_init(&conf->device_lock); |
dab8b292 | 3606 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 3607 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 TM |
3608 | |
3609 | spin_lock_init(&conf->resync_lock); | |
3610 | init_waitqueue_head(&conf->wait_barrier); | |
0e5313e2 | 3611 | atomic_set(&conf->nr_pending, 0); |
dab8b292 | 3612 | |
0232605d | 3613 | conf->thread = md_register_thread(raid10d, mddev, "raid10"); |
dab8b292 TM |
3614 | if (!conf->thread) |
3615 | goto out; | |
3616 | ||
dab8b292 TM |
3617 | conf->mddev = mddev; |
3618 | return conf; | |
3619 | ||
3620 | out: | |
dab8b292 | 3621 | if (conf) { |
644df1a8 | 3622 | mempool_destroy(conf->r10bio_pool); |
dab8b292 TM |
3623 | kfree(conf->mirrors); |
3624 | safe_put_page(conf->tmppage); | |
3625 | kfree(conf); | |
3626 | } | |
3627 | return ERR_PTR(err); | |
3628 | } | |
3629 | ||
849674e4 | 3630 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 3631 | { |
e879a879 | 3632 | struct r10conf *conf; |
dab8b292 | 3633 | int i, disk_idx, chunk_size; |
dc280d98 | 3634 | struct raid10_info *disk; |
3cb03002 | 3635 | struct md_rdev *rdev; |
dab8b292 | 3636 | sector_t size; |
3ea7daa5 N |
3637 | sector_t min_offset_diff = 0; |
3638 | int first = 1; | |
532a2a3f | 3639 | bool discard_supported = false; |
dab8b292 TM |
3640 | |
3641 | if (mddev->private == NULL) { | |
3642 | conf = setup_conf(mddev); | |
3643 | if (IS_ERR(conf)) | |
3644 | return PTR_ERR(conf); | |
3645 | mddev->private = conf; | |
3646 | } | |
3647 | conf = mddev->private; | |
3648 | if (!conf) | |
3649 | goto out; | |
3650 | ||
dab8b292 TM |
3651 | mddev->thread = conf->thread; |
3652 | conf->thread = NULL; | |
3653 | ||
8f6c2e4b | 3654 | chunk_size = mddev->chunk_sectors << 9; |
cc4d1efd | 3655 | if (mddev->queue) { |
532a2a3f SL |
3656 | blk_queue_max_discard_sectors(mddev->queue, |
3657 | mddev->chunk_sectors); | |
5026d7a9 | 3658 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
cc4d1efd JB |
3659 | blk_queue_io_min(mddev->queue, chunk_size); |
3660 | if (conf->geo.raid_disks % conf->geo.near_copies) | |
3661 | blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks); | |
3662 | else | |
3663 | blk_queue_io_opt(mddev->queue, chunk_size * | |
3664 | (conf->geo.raid_disks / conf->geo.near_copies)); | |
3665 | } | |
8f6c2e4b | 3666 | |
dafb20fa | 3667 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 3668 | long long diff; |
aba336bd | 3669 | struct request_queue *q; |
34b343cf | 3670 | |
1da177e4 | 3671 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
3672 | if (disk_idx < 0) |
3673 | continue; | |
3674 | if (disk_idx >= conf->geo.raid_disks && | |
3675 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
3676 | continue; |
3677 | disk = conf->mirrors + disk_idx; | |
3678 | ||
56a2559b N |
3679 | if (test_bit(Replacement, &rdev->flags)) { |
3680 | if (disk->replacement) | |
3681 | goto out_free_conf; | |
3682 | disk->replacement = rdev; | |
3683 | } else { | |
3684 | if (disk->rdev) | |
3685 | goto out_free_conf; | |
3686 | disk->rdev = rdev; | |
3687 | } | |
aba336bd | 3688 | q = bdev_get_queue(rdev->bdev); |
3ea7daa5 N |
3689 | diff = (rdev->new_data_offset - rdev->data_offset); |
3690 | if (!mddev->reshape_backwards) | |
3691 | diff = -diff; | |
3692 | if (diff < 0) | |
3693 | diff = 0; | |
3694 | if (first || diff < min_offset_diff) | |
3695 | min_offset_diff = diff; | |
56a2559b | 3696 | |
cc4d1efd JB |
3697 | if (mddev->gendisk) |
3698 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
3699 | rdev->data_offset << 9); | |
1da177e4 LT |
3700 | |
3701 | disk->head_position = 0; | |
532a2a3f SL |
3702 | |
3703 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) | |
3704 | discard_supported = true; | |
1da177e4 | 3705 | } |
3ea7daa5 | 3706 | |
ed30be07 JB |
3707 | if (mddev->queue) { |
3708 | if (discard_supported) | |
3709 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3710 | mddev->queue); | |
3711 | else | |
3712 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3713 | mddev->queue); | |
3714 | } | |
6d508242 | 3715 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3716 | if (!enough(conf, -1)) { |
08464e09 | 3717 | pr_err("md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3718 | mdname(mddev)); |
1da177e4 LT |
3719 | goto out_free_conf; |
3720 | } | |
3721 | ||
3ea7daa5 N |
3722 | if (conf->reshape_progress != MaxSector) { |
3723 | /* must ensure that shape change is supported */ | |
3724 | if (conf->geo.far_copies != 1 && | |
3725 | conf->geo.far_offset == 0) | |
3726 | goto out_free_conf; | |
3727 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 3728 | conf->prev.far_offset == 0) |
3ea7daa5 N |
3729 | goto out_free_conf; |
3730 | } | |
3731 | ||
1da177e4 | 3732 | mddev->degraded = 0; |
f8c9e74f N |
3733 | for (i = 0; |
3734 | i < conf->geo.raid_disks | |
3735 | || i < conf->prev.raid_disks; | |
3736 | i++) { | |
1da177e4 LT |
3737 | |
3738 | disk = conf->mirrors + i; | |
3739 | ||
56a2559b N |
3740 | if (!disk->rdev && disk->replacement) { |
3741 | /* The replacement is all we have - use it */ | |
3742 | disk->rdev = disk->replacement; | |
3743 | disk->replacement = NULL; | |
3744 | clear_bit(Replacement, &disk->rdev->flags); | |
3745 | } | |
3746 | ||
5fd6c1dc | 3747 | if (!disk->rdev || |
2e333e89 | 3748 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3749 | disk->head_position = 0; |
3750 | mddev->degraded++; | |
0b59bb64 N |
3751 | if (disk->rdev && |
3752 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 3753 | conf->fullsync = 1; |
1da177e4 | 3754 | } |
d890fa2b | 3755 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3756 | } |
3757 | ||
8c6ac868 | 3758 | if (mddev->recovery_cp != MaxSector) |
08464e09 N |
3759 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", |
3760 | mdname(mddev)); | |
3761 | pr_info("md/raid10:%s: active with %d out of %d devices\n", | |
5cf00fcd N |
3762 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
3763 | conf->geo.raid_disks); | |
1da177e4 LT |
3764 | /* |
3765 | * Ok, everything is just fine now | |
3766 | */ | |
dab8b292 TM |
3767 | mddev->dev_sectors = conf->dev_sectors; |
3768 | size = raid10_size(mddev, 0, 0); | |
3769 | md_set_array_sectors(mddev, size); | |
3770 | mddev->resync_max_sectors = size; | |
46533ff7 | 3771 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
1da177e4 | 3772 | |
cc4d1efd | 3773 | if (mddev->queue) { |
5cf00fcd | 3774 | int stripe = conf->geo.raid_disks * |
9d8f0363 | 3775 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
cc4d1efd JB |
3776 | |
3777 | /* Calculate max read-ahead size. | |
3778 | * We need to readahead at least twice a whole stripe.... | |
3779 | * maybe... | |
3780 | */ | |
5cf00fcd | 3781 | stripe /= conf->geo.near_copies; |
3ea7daa5 N |
3782 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
3783 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
1da177e4 LT |
3784 | } |
3785 | ||
a91a2785 MP |
3786 | if (md_integrity_register(mddev)) |
3787 | goto out_free_conf; | |
3788 | ||
3ea7daa5 N |
3789 | if (conf->reshape_progress != MaxSector) { |
3790 | unsigned long before_length, after_length; | |
3791 | ||
3792 | before_length = ((1 << conf->prev.chunk_shift) * | |
3793 | conf->prev.far_copies); | |
3794 | after_length = ((1 << conf->geo.chunk_shift) * | |
3795 | conf->geo.far_copies); | |
3796 | ||
3797 | if (max(before_length, after_length) > min_offset_diff) { | |
3798 | /* This cannot work */ | |
08464e09 | 3799 | pr_warn("md/raid10: offset difference not enough to continue reshape\n"); |
3ea7daa5 N |
3800 | goto out_free_conf; |
3801 | } | |
3802 | conf->offset_diff = min_offset_diff; | |
3803 | ||
3ea7daa5 N |
3804 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
3805 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3806 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3807 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3808 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3809 | "reshape"); | |
3810 | } | |
3811 | ||
1da177e4 LT |
3812 | return 0; |
3813 | ||
3814 | out_free_conf: | |
01f96c0a | 3815 | md_unregister_thread(&mddev->thread); |
644df1a8 | 3816 | mempool_destroy(conf->r10bio_pool); |
1345b1d8 | 3817 | safe_put_page(conf->tmppage); |
990a8baf | 3818 | kfree(conf->mirrors); |
1da177e4 LT |
3819 | kfree(conf); |
3820 | mddev->private = NULL; | |
3821 | out: | |
3822 | return -EIO; | |
3823 | } | |
3824 | ||
afa0f557 | 3825 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 3826 | { |
afa0f557 | 3827 | struct r10conf *conf = priv; |
1da177e4 | 3828 | |
644df1a8 | 3829 | mempool_destroy(conf->r10bio_pool); |
0fea7ed8 | 3830 | safe_put_page(conf->tmppage); |
990a8baf | 3831 | kfree(conf->mirrors); |
c4796e21 N |
3832 | kfree(conf->mirrors_old); |
3833 | kfree(conf->mirrors_new); | |
1da177e4 | 3834 | kfree(conf); |
1da177e4 LT |
3835 | } |
3836 | ||
fd01b88c | 3837 | static void raid10_quiesce(struct mddev *mddev, int state) |
6cce3b23 | 3838 | { |
e879a879 | 3839 | struct r10conf *conf = mddev->private; |
6cce3b23 N |
3840 | |
3841 | switch(state) { | |
3842 | case 1: | |
3843 | raise_barrier(conf, 0); | |
3844 | break; | |
3845 | case 0: | |
3846 | lower_barrier(conf); | |
3847 | break; | |
3848 | } | |
6cce3b23 | 3849 | } |
1da177e4 | 3850 | |
006a09a0 N |
3851 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3852 | { | |
3853 | /* Resize of 'far' arrays is not supported. | |
3854 | * For 'near' and 'offset' arrays we can set the | |
3855 | * number of sectors used to be an appropriate multiple | |
3856 | * of the chunk size. | |
3857 | * For 'offset', this is far_copies*chunksize. | |
3858 | * For 'near' the multiplier is the LCM of | |
3859 | * near_copies and raid_disks. | |
3860 | * So if far_copies > 1 && !far_offset, fail. | |
3861 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
3862 | * multiply by chunk_size. Then round to this number. | |
3863 | * This is mostly done by raid10_size() | |
3864 | */ | |
3865 | struct r10conf *conf = mddev->private; | |
3866 | sector_t oldsize, size; | |
3867 | ||
f8c9e74f N |
3868 | if (mddev->reshape_position != MaxSector) |
3869 | return -EBUSY; | |
3870 | ||
5cf00fcd | 3871 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
3872 | return -EINVAL; |
3873 | ||
3874 | oldsize = raid10_size(mddev, 0, 0); | |
3875 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
3876 | if (mddev->external_size && |
3877 | mddev->array_sectors > size) | |
006a09a0 | 3878 | return -EINVAL; |
a4a6125a N |
3879 | if (mddev->bitmap) { |
3880 | int ret = bitmap_resize(mddev->bitmap, size, 0, 0); | |
3881 | if (ret) | |
3882 | return ret; | |
3883 | } | |
3884 | md_set_array_sectors(mddev, size); | |
859644f0 HM |
3885 | if (mddev->queue) { |
3886 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
3887 | revalidate_disk(mddev->gendisk); | |
3888 | } | |
006a09a0 N |
3889 | if (sectors > mddev->dev_sectors && |
3890 | mddev->recovery_cp > oldsize) { | |
3891 | mddev->recovery_cp = oldsize; | |
3892 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3893 | } | |
6508fdbf N |
3894 | calc_sectors(conf, sectors); |
3895 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
3896 | mddev->resync_max_sectors = size; |
3897 | return 0; | |
3898 | } | |
3899 | ||
53a6ab4d | 3900 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 3901 | { |
3cb03002 | 3902 | struct md_rdev *rdev; |
e879a879 | 3903 | struct r10conf *conf; |
dab8b292 TM |
3904 | |
3905 | if (mddev->degraded > 0) { | |
08464e09 N |
3906 | pr_warn("md/raid10:%s: Error: degraded raid0!\n", |
3907 | mdname(mddev)); | |
dab8b292 TM |
3908 | return ERR_PTR(-EINVAL); |
3909 | } | |
53a6ab4d | 3910 | sector_div(size, devs); |
dab8b292 | 3911 | |
dab8b292 TM |
3912 | /* Set new parameters */ |
3913 | mddev->new_level = 10; | |
3914 | /* new layout: far_copies = 1, near_copies = 2 */ | |
3915 | mddev->new_layout = (1<<8) + 2; | |
3916 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
3917 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
3918 | mddev->raid_disks *= 2; |
3919 | /* make sure it will be not marked as dirty */ | |
3920 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 3921 | mddev->dev_sectors = size; |
dab8b292 TM |
3922 | |
3923 | conf = setup_conf(mddev); | |
02214dc5 | 3924 | if (!IS_ERR(conf)) { |
dafb20fa | 3925 | rdev_for_each(rdev, mddev) |
53a6ab4d | 3926 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 3927 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
3928 | rdev->sectors = size; |
3929 | } | |
02214dc5 KW |
3930 | conf->barrier = 1; |
3931 | } | |
3932 | ||
dab8b292 TM |
3933 | return conf; |
3934 | } | |
3935 | ||
fd01b88c | 3936 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 3937 | { |
e373ab10 | 3938 | struct r0conf *raid0_conf; |
dab8b292 TM |
3939 | |
3940 | /* raid10 can take over: | |
3941 | * raid0 - providing it has only two drives | |
3942 | */ | |
3943 | if (mddev->level == 0) { | |
3944 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
3945 | raid0_conf = mddev->private; |
3946 | if (raid0_conf->nr_strip_zones > 1) { | |
08464e09 N |
3947 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n", |
3948 | mdname(mddev)); | |
dab8b292 TM |
3949 | return ERR_PTR(-EINVAL); |
3950 | } | |
53a6ab4d N |
3951 | return raid10_takeover_raid0(mddev, |
3952 | raid0_conf->strip_zone->zone_end, | |
3953 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
3954 | } |
3955 | return ERR_PTR(-EINVAL); | |
3956 | } | |
3957 | ||
3ea7daa5 N |
3958 | static int raid10_check_reshape(struct mddev *mddev) |
3959 | { | |
3960 | /* Called when there is a request to change | |
3961 | * - layout (to ->new_layout) | |
3962 | * - chunk size (to ->new_chunk_sectors) | |
3963 | * - raid_disks (by delta_disks) | |
3964 | * or when trying to restart a reshape that was ongoing. | |
3965 | * | |
3966 | * We need to validate the request and possibly allocate | |
3967 | * space if that might be an issue later. | |
3968 | * | |
3969 | * Currently we reject any reshape of a 'far' mode array, | |
3970 | * allow chunk size to change if new is generally acceptable, | |
3971 | * allow raid_disks to increase, and allow | |
3972 | * a switch between 'near' mode and 'offset' mode. | |
3973 | */ | |
3974 | struct r10conf *conf = mddev->private; | |
3975 | struct geom geo; | |
3976 | ||
3977 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
3978 | return -EINVAL; | |
3979 | ||
3980 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
3981 | /* mustn't change number of copies */ | |
3982 | return -EINVAL; | |
3983 | if (geo.far_copies > 1 && !geo.far_offset) | |
3984 | /* Cannot switch to 'far' mode */ | |
3985 | return -EINVAL; | |
3986 | ||
3987 | if (mddev->array_sectors & geo.chunk_mask) | |
3988 | /* not factor of array size */ | |
3989 | return -EINVAL; | |
3990 | ||
3ea7daa5 N |
3991 | if (!enough(conf, -1)) |
3992 | return -EINVAL; | |
3993 | ||
3994 | kfree(conf->mirrors_new); | |
3995 | conf->mirrors_new = NULL; | |
3996 | if (mddev->delta_disks > 0) { | |
3997 | /* allocate new 'mirrors' list */ | |
3998 | conf->mirrors_new = kzalloc( | |
dc280d98 | 3999 | sizeof(struct raid10_info) |
3ea7daa5 N |
4000 | *(mddev->raid_disks + |
4001 | mddev->delta_disks), | |
4002 | GFP_KERNEL); | |
4003 | if (!conf->mirrors_new) | |
4004 | return -ENOMEM; | |
4005 | } | |
4006 | return 0; | |
4007 | } | |
4008 | ||
4009 | /* | |
4010 | * Need to check if array has failed when deciding whether to: | |
4011 | * - start an array | |
4012 | * - remove non-faulty devices | |
4013 | * - add a spare | |
4014 | * - allow a reshape | |
4015 | * This determination is simple when no reshape is happening. | |
4016 | * However if there is a reshape, we need to carefully check | |
4017 | * both the before and after sections. | |
4018 | * This is because some failed devices may only affect one | |
4019 | * of the two sections, and some non-in_sync devices may | |
4020 | * be insync in the section most affected by failed devices. | |
4021 | */ | |
4022 | static int calc_degraded(struct r10conf *conf) | |
4023 | { | |
4024 | int degraded, degraded2; | |
4025 | int i; | |
4026 | ||
4027 | rcu_read_lock(); | |
4028 | degraded = 0; | |
4029 | /* 'prev' section first */ | |
4030 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
4031 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4032 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4033 | degraded++; | |
4034 | else if (!test_bit(In_sync, &rdev->flags)) | |
4035 | /* When we can reduce the number of devices in | |
4036 | * an array, this might not contribute to | |
4037 | * 'degraded'. It does now. | |
4038 | */ | |
4039 | degraded++; | |
4040 | } | |
4041 | rcu_read_unlock(); | |
4042 | if (conf->geo.raid_disks == conf->prev.raid_disks) | |
4043 | return degraded; | |
4044 | rcu_read_lock(); | |
4045 | degraded2 = 0; | |
4046 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
4047 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4048 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4049 | degraded2++; | |
4050 | else if (!test_bit(In_sync, &rdev->flags)) { | |
4051 | /* If reshape is increasing the number of devices, | |
4052 | * this section has already been recovered, so | |
4053 | * it doesn't contribute to degraded. | |
4054 | * else it does. | |
4055 | */ | |
4056 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
4057 | degraded2++; | |
4058 | } | |
4059 | } | |
4060 | rcu_read_unlock(); | |
4061 | if (degraded2 > degraded) | |
4062 | return degraded2; | |
4063 | return degraded; | |
4064 | } | |
4065 | ||
4066 | static int raid10_start_reshape(struct mddev *mddev) | |
4067 | { | |
4068 | /* A 'reshape' has been requested. This commits | |
4069 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
4070 | * This also checks if there are enough spares and adds them | |
4071 | * to the array. | |
4072 | * We currently require enough spares to make the final | |
4073 | * array non-degraded. We also require that the difference | |
4074 | * between old and new data_offset - on each device - is | |
4075 | * enough that we never risk over-writing. | |
4076 | */ | |
4077 | ||
4078 | unsigned long before_length, after_length; | |
4079 | sector_t min_offset_diff = 0; | |
4080 | int first = 1; | |
4081 | struct geom new; | |
4082 | struct r10conf *conf = mddev->private; | |
4083 | struct md_rdev *rdev; | |
4084 | int spares = 0; | |
bb63a701 | 4085 | int ret; |
3ea7daa5 N |
4086 | |
4087 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4088 | return -EBUSY; | |
4089 | ||
4090 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
4091 | return -EINVAL; | |
4092 | ||
4093 | before_length = ((1 << conf->prev.chunk_shift) * | |
4094 | conf->prev.far_copies); | |
4095 | after_length = ((1 << conf->geo.chunk_shift) * | |
4096 | conf->geo.far_copies); | |
4097 | ||
4098 | rdev_for_each(rdev, mddev) { | |
4099 | if (!test_bit(In_sync, &rdev->flags) | |
4100 | && !test_bit(Faulty, &rdev->flags)) | |
4101 | spares++; | |
4102 | if (rdev->raid_disk >= 0) { | |
4103 | long long diff = (rdev->new_data_offset | |
4104 | - rdev->data_offset); | |
4105 | if (!mddev->reshape_backwards) | |
4106 | diff = -diff; | |
4107 | if (diff < 0) | |
4108 | diff = 0; | |
4109 | if (first || diff < min_offset_diff) | |
4110 | min_offset_diff = diff; | |
4111 | } | |
4112 | } | |
4113 | ||
4114 | if (max(before_length, after_length) > min_offset_diff) | |
4115 | return -EINVAL; | |
4116 | ||
4117 | if (spares < mddev->delta_disks) | |
4118 | return -EINVAL; | |
4119 | ||
4120 | conf->offset_diff = min_offset_diff; | |
4121 | spin_lock_irq(&conf->device_lock); | |
4122 | if (conf->mirrors_new) { | |
4123 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4124 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4125 | smp_mb(); |
c4796e21 | 4126 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4127 | conf->mirrors_old = conf->mirrors; |
4128 | conf->mirrors = conf->mirrors_new; | |
4129 | conf->mirrors_new = NULL; | |
4130 | } | |
4131 | setup_geo(&conf->geo, mddev, geo_start); | |
4132 | smp_mb(); | |
4133 | if (mddev->reshape_backwards) { | |
4134 | sector_t size = raid10_size(mddev, 0, 0); | |
4135 | if (size < mddev->array_sectors) { | |
4136 | spin_unlock_irq(&conf->device_lock); | |
08464e09 N |
4137 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n", |
4138 | mdname(mddev)); | |
3ea7daa5 N |
4139 | return -EINVAL; |
4140 | } | |
4141 | mddev->resync_max_sectors = size; | |
4142 | conf->reshape_progress = size; | |
4143 | } else | |
4144 | conf->reshape_progress = 0; | |
299b0685 | 4145 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4146 | spin_unlock_irq(&conf->device_lock); |
4147 | ||
bb63a701 N |
4148 | if (mddev->delta_disks && mddev->bitmap) { |
4149 | ret = bitmap_resize(mddev->bitmap, | |
4150 | raid10_size(mddev, 0, | |
4151 | conf->geo.raid_disks), | |
4152 | 0, 0); | |
4153 | if (ret) | |
4154 | goto abort; | |
4155 | } | |
3ea7daa5 N |
4156 | if (mddev->delta_disks > 0) { |
4157 | rdev_for_each(rdev, mddev) | |
4158 | if (rdev->raid_disk < 0 && | |
4159 | !test_bit(Faulty, &rdev->flags)) { | |
4160 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4161 | if (rdev->raid_disk >= | |
4162 | conf->prev.raid_disks) | |
4163 | set_bit(In_sync, &rdev->flags); | |
4164 | else | |
4165 | rdev->recovery_offset = 0; | |
4166 | ||
4167 | if (sysfs_link_rdev(mddev, rdev)) | |
4168 | /* Failure here is OK */; | |
4169 | } | |
4170 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4171 | && !test_bit(Faulty, &rdev->flags)) { | |
4172 | /* This is a spare that was manually added */ | |
4173 | set_bit(In_sync, &rdev->flags); | |
4174 | } | |
4175 | } | |
4176 | /* When a reshape changes the number of devices, | |
4177 | * ->degraded is measured against the larger of the | |
4178 | * pre and post numbers. | |
4179 | */ | |
4180 | spin_lock_irq(&conf->device_lock); | |
4181 | mddev->degraded = calc_degraded(conf); | |
4182 | spin_unlock_irq(&conf->device_lock); | |
4183 | mddev->raid_disks = conf->geo.raid_disks; | |
4184 | mddev->reshape_position = conf->reshape_progress; | |
4185 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
4186 | ||
4187 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4188 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4189 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 N |
4190 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
4191 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4192 | ||
4193 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4194 | "reshape"); | |
4195 | if (!mddev->sync_thread) { | |
bb63a701 N |
4196 | ret = -EAGAIN; |
4197 | goto abort; | |
3ea7daa5 N |
4198 | } |
4199 | conf->reshape_checkpoint = jiffies; | |
4200 | md_wakeup_thread(mddev->sync_thread); | |
4201 | md_new_event(mddev); | |
4202 | return 0; | |
bb63a701 N |
4203 | |
4204 | abort: | |
4205 | mddev->recovery = 0; | |
4206 | spin_lock_irq(&conf->device_lock); | |
4207 | conf->geo = conf->prev; | |
4208 | mddev->raid_disks = conf->geo.raid_disks; | |
4209 | rdev_for_each(rdev, mddev) | |
4210 | rdev->new_data_offset = rdev->data_offset; | |
4211 | smp_wmb(); | |
4212 | conf->reshape_progress = MaxSector; | |
299b0685 | 4213 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4214 | mddev->reshape_position = MaxSector; |
4215 | spin_unlock_irq(&conf->device_lock); | |
4216 | return ret; | |
3ea7daa5 N |
4217 | } |
4218 | ||
4219 | /* Calculate the last device-address that could contain | |
4220 | * any block from the chunk that includes the array-address 's' | |
4221 | * and report the next address. | |
4222 | * i.e. the address returned will be chunk-aligned and after | |
4223 | * any data that is in the chunk containing 's'. | |
4224 | */ | |
4225 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4226 | { | |
4227 | s = (s | geo->chunk_mask) + 1; | |
4228 | s >>= geo->chunk_shift; | |
4229 | s *= geo->near_copies; | |
4230 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4231 | s *= geo->far_copies; | |
4232 | s <<= geo->chunk_shift; | |
4233 | return s; | |
4234 | } | |
4235 | ||
4236 | /* Calculate the first device-address that could contain | |
4237 | * any block from the chunk that includes the array-address 's'. | |
4238 | * This too will be the start of a chunk | |
4239 | */ | |
4240 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4241 | { | |
4242 | s >>= geo->chunk_shift; | |
4243 | s *= geo->near_copies; | |
4244 | sector_div(s, geo->raid_disks); | |
4245 | s *= geo->far_copies; | |
4246 | s <<= geo->chunk_shift; | |
4247 | return s; | |
4248 | } | |
4249 | ||
4250 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4251 | int *skipped) | |
4252 | { | |
4253 | /* We simply copy at most one chunk (smallest of old and new) | |
4254 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4255 | * or we hit a bad block or something. | |
4256 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4257 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4258 | * can record any location. |
4259 | * | |
4260 | * If we will want to write to a location that isn't | |
4261 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4262 | * we need to flush all reshape requests and update the metadata. | |
4263 | * | |
4264 | * When reshaping forwards (e.g. to more devices), we interpret | |
4265 | * 'safe' as the earliest block which might not have been copied | |
4266 | * down yet. We divide this by previous stripe size and multiply | |
4267 | * by previous stripe length to get lowest device offset that we | |
4268 | * cannot write to yet. | |
4269 | * We interpret 'sector_nr' as an address that we want to write to. | |
4270 | * From this we use last_device_address() to find where we might | |
4271 | * write to, and first_device_address on the 'safe' position. | |
4272 | * If this 'next' write position is after the 'safe' position, | |
4273 | * we must update the metadata to increase the 'safe' position. | |
4274 | * | |
4275 | * When reshaping backwards, we round in the opposite direction | |
4276 | * and perform the reverse test: next write position must not be | |
4277 | * less than current safe position. | |
4278 | * | |
4279 | * In all this the minimum difference in data offsets | |
4280 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4281 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4282 | * |
4283 | * We need to prepare all the bios here before we start any IO | |
4284 | * to ensure the size we choose is acceptable to all devices. | |
4285 | * The means one for each copy for write-out and an extra one for | |
4286 | * read-in. | |
4287 | * We store the read-in bio in ->master_bio and the others in | |
4288 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4289 | */ | |
4290 | struct r10conf *conf = mddev->private; | |
4291 | struct r10bio *r10_bio; | |
4292 | sector_t next, safe, last; | |
4293 | int max_sectors; | |
4294 | int nr_sectors; | |
4295 | int s; | |
4296 | struct md_rdev *rdev; | |
4297 | int need_flush = 0; | |
4298 | struct bio *blist; | |
4299 | struct bio *bio, *read_bio; | |
4300 | int sectors_done = 0; | |
4301 | ||
4302 | if (sector_nr == 0) { | |
4303 | /* If restarting in the middle, skip the initial sectors */ | |
4304 | if (mddev->reshape_backwards && | |
4305 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4306 | sector_nr = (raid10_size(mddev, 0, 0) | |
4307 | - conf->reshape_progress); | |
4308 | } else if (!mddev->reshape_backwards && | |
4309 | conf->reshape_progress > 0) | |
4310 | sector_nr = conf->reshape_progress; | |
4311 | if (sector_nr) { | |
4312 | mddev->curr_resync_completed = sector_nr; | |
4313 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
4314 | *skipped = 1; | |
4315 | return sector_nr; | |
4316 | } | |
4317 | } | |
4318 | ||
4319 | /* We don't use sector_nr to track where we are up to | |
4320 | * as that doesn't work well for ->reshape_backwards. | |
4321 | * So just use ->reshape_progress. | |
4322 | */ | |
4323 | if (mddev->reshape_backwards) { | |
4324 | /* 'next' is the earliest device address that we might | |
4325 | * write to for this chunk in the new layout | |
4326 | */ | |
4327 | next = first_dev_address(conf->reshape_progress - 1, | |
4328 | &conf->geo); | |
4329 | ||
4330 | /* 'safe' is the last device address that we might read from | |
4331 | * in the old layout after a restart | |
4332 | */ | |
4333 | safe = last_dev_address(conf->reshape_safe - 1, | |
4334 | &conf->prev); | |
4335 | ||
4336 | if (next + conf->offset_diff < safe) | |
4337 | need_flush = 1; | |
4338 | ||
4339 | last = conf->reshape_progress - 1; | |
4340 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4341 | & conf->prev.chunk_mask); | |
4342 | if (sector_nr + RESYNC_BLOCK_SIZE/512 < last) | |
4343 | sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512; | |
4344 | } else { | |
4345 | /* 'next' is after the last device address that we | |
4346 | * might write to for this chunk in the new layout | |
4347 | */ | |
4348 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4349 | ||
4350 | /* 'safe' is the earliest device address that we might | |
4351 | * read from in the old layout after a restart | |
4352 | */ | |
4353 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4354 | ||
4355 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4356 | * as that would possibly corrupt data | |
4357 | */ | |
4358 | if (next > safe + conf->offset_diff) | |
4359 | need_flush = 1; | |
4360 | ||
4361 | sector_nr = conf->reshape_progress; | |
4362 | last = sector_nr | (conf->geo.chunk_mask | |
4363 | & conf->prev.chunk_mask); | |
4364 | ||
4365 | if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last) | |
4366 | last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1; | |
4367 | } | |
4368 | ||
4369 | if (need_flush || | |
4370 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4371 | /* Need to update reshape_position in metadata */ | |
4372 | wait_barrier(conf); | |
4373 | mddev->reshape_position = conf->reshape_progress; | |
4374 | if (mddev->reshape_backwards) | |
4375 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4376 | - conf->reshape_progress; | |
4377 | else | |
4378 | mddev->curr_resync_completed = conf->reshape_progress; | |
4379 | conf->reshape_checkpoint = jiffies; | |
4380 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
4381 | md_wakeup_thread(mddev->thread); | |
4382 | wait_event(mddev->sb_wait, mddev->flags == 0 || | |
c91abf5a N |
4383 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4384 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4385 | allow_barrier(conf); | |
4386 | return sectors_done; | |
4387 | } | |
3ea7daa5 N |
4388 | conf->reshape_safe = mddev->reshape_position; |
4389 | allow_barrier(conf); | |
4390 | } | |
4391 | ||
4392 | read_more: | |
4393 | /* Now schedule reads for blocks from sector_nr to last */ | |
4394 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); | |
cb8b12b5 | 4395 | r10_bio->state = 0; |
3ea7daa5 N |
4396 | raise_barrier(conf, sectors_done != 0); |
4397 | atomic_set(&r10_bio->remaining, 0); | |
4398 | r10_bio->mddev = mddev; | |
4399 | r10_bio->sector = sector_nr; | |
4400 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4401 | r10_bio->sectors = last - sector_nr + 1; | |
4402 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4403 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4404 | ||
4405 | if (!rdev) { | |
4406 | /* Cannot read from here, so need to record bad blocks | |
4407 | * on all the target devices. | |
4408 | */ | |
4409 | // FIXME | |
e337aead | 4410 | mempool_free(r10_bio, conf->r10buf_pool); |
3ea7daa5 N |
4411 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4412 | return sectors_done; | |
4413 | } | |
4414 | ||
4415 | read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev); | |
4416 | ||
4417 | read_bio->bi_bdev = rdev->bdev; | |
4f024f37 | 4418 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4419 | + rdev->data_offset); |
4420 | read_bio->bi_private = r10_bio; | |
4421 | read_bio->bi_end_io = end_sync_read; | |
796a5cf0 | 4422 | bio_set_op_attrs(read_bio, REQ_OP_READ, 0); |
ce0b0a46 | 4423 | read_bio->bi_flags &= (~0UL << BIO_RESET_BITS); |
4246a0b6 | 4424 | read_bio->bi_error = 0; |
3ea7daa5 | 4425 | read_bio->bi_vcnt = 0; |
4f024f37 | 4426 | read_bio->bi_iter.bi_size = 0; |
3ea7daa5 N |
4427 | r10_bio->master_bio = read_bio; |
4428 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4429 | ||
4430 | /* Now find the locations in the new layout */ | |
4431 | __raid10_find_phys(&conf->geo, r10_bio); | |
4432 | ||
4433 | blist = read_bio; | |
4434 | read_bio->bi_next = NULL; | |
4435 | ||
d094d686 | 4436 | rcu_read_lock(); |
3ea7daa5 N |
4437 | for (s = 0; s < conf->copies*2; s++) { |
4438 | struct bio *b; | |
4439 | int d = r10_bio->devs[s/2].devnum; | |
4440 | struct md_rdev *rdev2; | |
4441 | if (s&1) { | |
d094d686 | 4442 | rdev2 = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4443 | b = r10_bio->devs[s/2].repl_bio; |
4444 | } else { | |
d094d686 | 4445 | rdev2 = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4446 | b = r10_bio->devs[s/2].bio; |
4447 | } | |
4448 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4449 | continue; | |
8be185f2 KO |
4450 | |
4451 | bio_reset(b); | |
3ea7daa5 | 4452 | b->bi_bdev = rdev2->bdev; |
4f024f37 KO |
4453 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4454 | rdev2->new_data_offset; | |
3ea7daa5 N |
4455 | b->bi_private = r10_bio; |
4456 | b->bi_end_io = end_reshape_write; | |
796a5cf0 | 4457 | bio_set_op_attrs(b, REQ_OP_WRITE, 0); |
3ea7daa5 | 4458 | b->bi_next = blist; |
3ea7daa5 N |
4459 | blist = b; |
4460 | } | |
4461 | ||
4462 | /* Now add as many pages as possible to all of these bios. */ | |
4463 | ||
4464 | nr_sectors = 0; | |
4465 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { | |
4466 | struct page *page = r10_bio->devs[0].bio->bi_io_vec[s/(PAGE_SIZE>>9)].bv_page; | |
4467 | int len = (max_sectors - s) << 9; | |
4468 | if (len > PAGE_SIZE) | |
4469 | len = PAGE_SIZE; | |
4470 | for (bio = blist; bio ; bio = bio->bi_next) { | |
4471 | struct bio *bio2; | |
4472 | if (bio_add_page(bio, page, len, 0)) | |
4473 | continue; | |
4474 | ||
4475 | /* Didn't fit, must stop */ | |
4476 | for (bio2 = blist; | |
4477 | bio2 && bio2 != bio; | |
4478 | bio2 = bio2->bi_next) { | |
4479 | /* Remove last page from this bio */ | |
4480 | bio2->bi_vcnt--; | |
4f024f37 | 4481 | bio2->bi_iter.bi_size -= len; |
b7c44ed9 | 4482 | bio_clear_flag(bio2, BIO_SEG_VALID); |
3ea7daa5 N |
4483 | } |
4484 | goto bio_full; | |
4485 | } | |
4486 | sector_nr += len >> 9; | |
4487 | nr_sectors += len >> 9; | |
4488 | } | |
4489 | bio_full: | |
d094d686 | 4490 | rcu_read_unlock(); |
3ea7daa5 N |
4491 | r10_bio->sectors = nr_sectors; |
4492 | ||
4493 | /* Now submit the read */ | |
4494 | md_sync_acct(read_bio->bi_bdev, r10_bio->sectors); | |
4495 | atomic_inc(&r10_bio->remaining); | |
4496 | read_bio->bi_next = NULL; | |
4497 | generic_make_request(read_bio); | |
4498 | sector_nr += nr_sectors; | |
4499 | sectors_done += nr_sectors; | |
4500 | if (sector_nr <= last) | |
4501 | goto read_more; | |
4502 | ||
4503 | /* Now that we have done the whole section we can | |
4504 | * update reshape_progress | |
4505 | */ | |
4506 | if (mddev->reshape_backwards) | |
4507 | conf->reshape_progress -= sectors_done; | |
4508 | else | |
4509 | conf->reshape_progress += sectors_done; | |
4510 | ||
4511 | return sectors_done; | |
4512 | } | |
4513 | ||
4514 | static void end_reshape_request(struct r10bio *r10_bio); | |
4515 | static int handle_reshape_read_error(struct mddev *mddev, | |
4516 | struct r10bio *r10_bio); | |
4517 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4518 | { | |
4519 | /* Reshape read completed. Hopefully we have a block | |
4520 | * to write out. | |
4521 | * If we got a read error then we do sync 1-page reads from | |
4522 | * elsewhere until we find the data - or give up. | |
4523 | */ | |
4524 | struct r10conf *conf = mddev->private; | |
4525 | int s; | |
4526 | ||
4527 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4528 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4529 | /* Reshape has been aborted */ | |
4530 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4531 | return; | |
4532 | } | |
4533 | ||
4534 | /* We definitely have the data in the pages, schedule the | |
4535 | * writes. | |
4536 | */ | |
4537 | atomic_set(&r10_bio->remaining, 1); | |
4538 | for (s = 0; s < conf->copies*2; s++) { | |
4539 | struct bio *b; | |
4540 | int d = r10_bio->devs[s/2].devnum; | |
4541 | struct md_rdev *rdev; | |
d094d686 | 4542 | rcu_read_lock(); |
3ea7daa5 | 4543 | if (s&1) { |
d094d686 | 4544 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4545 | b = r10_bio->devs[s/2].repl_bio; |
4546 | } else { | |
d094d686 | 4547 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4548 | b = r10_bio->devs[s/2].bio; |
4549 | } | |
d094d686 N |
4550 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
4551 | rcu_read_unlock(); | |
3ea7daa5 | 4552 | continue; |
d094d686 | 4553 | } |
3ea7daa5 | 4554 | atomic_inc(&rdev->nr_pending); |
d094d686 | 4555 | rcu_read_unlock(); |
3ea7daa5 N |
4556 | md_sync_acct(b->bi_bdev, r10_bio->sectors); |
4557 | atomic_inc(&r10_bio->remaining); | |
4558 | b->bi_next = NULL; | |
4559 | generic_make_request(b); | |
4560 | } | |
4561 | end_reshape_request(r10_bio); | |
4562 | } | |
4563 | ||
4564 | static void end_reshape(struct r10conf *conf) | |
4565 | { | |
4566 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
4567 | return; | |
4568 | ||
4569 | spin_lock_irq(&conf->device_lock); | |
4570 | conf->prev = conf->geo; | |
4571 | md_finish_reshape(conf->mddev); | |
4572 | smp_wmb(); | |
4573 | conf->reshape_progress = MaxSector; | |
299b0685 | 4574 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
4575 | spin_unlock_irq(&conf->device_lock); |
4576 | ||
4577 | /* read-ahead size must cover two whole stripes, which is | |
4578 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4579 | */ | |
4580 | if (conf->mddev->queue) { | |
4581 | int stripe = conf->geo.raid_disks * | |
4582 | ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE); | |
4583 | stripe /= conf->geo.near_copies; | |
4584 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4585 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4586 | } | |
4587 | conf->fullsync = 0; | |
4588 | } | |
4589 | ||
3ea7daa5 N |
4590 | static int handle_reshape_read_error(struct mddev *mddev, |
4591 | struct r10bio *r10_bio) | |
4592 | { | |
4593 | /* Use sync reads to get the blocks from somewhere else */ | |
4594 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 4595 | struct r10conf *conf = mddev->private; |
e0ee7785 N |
4596 | struct { |
4597 | struct r10bio r10_bio; | |
4598 | struct r10dev devs[conf->copies]; | |
4599 | } on_stack; | |
4600 | struct r10bio *r10b = &on_stack.r10_bio; | |
3ea7daa5 N |
4601 | int slot = 0; |
4602 | int idx = 0; | |
4603 | struct bio_vec *bvec = r10_bio->master_bio->bi_io_vec; | |
4604 | ||
e0ee7785 N |
4605 | r10b->sector = r10_bio->sector; |
4606 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
4607 | |
4608 | while (sectors) { | |
4609 | int s = sectors; | |
4610 | int success = 0; | |
4611 | int first_slot = slot; | |
4612 | ||
4613 | if (s > (PAGE_SIZE >> 9)) | |
4614 | s = PAGE_SIZE >> 9; | |
4615 | ||
d094d686 | 4616 | rcu_read_lock(); |
3ea7daa5 | 4617 | while (!success) { |
e0ee7785 | 4618 | int d = r10b->devs[slot].devnum; |
d094d686 | 4619 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4620 | sector_t addr; |
4621 | if (rdev == NULL || | |
4622 | test_bit(Faulty, &rdev->flags) || | |
4623 | !test_bit(In_sync, &rdev->flags)) | |
4624 | goto failed; | |
4625 | ||
e0ee7785 | 4626 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
d094d686 N |
4627 | atomic_inc(&rdev->nr_pending); |
4628 | rcu_read_unlock(); | |
3ea7daa5 N |
4629 | success = sync_page_io(rdev, |
4630 | addr, | |
4631 | s << 9, | |
4632 | bvec[idx].bv_page, | |
796a5cf0 | 4633 | REQ_OP_READ, 0, false); |
d094d686 N |
4634 | rdev_dec_pending(rdev, mddev); |
4635 | rcu_read_lock(); | |
3ea7daa5 N |
4636 | if (success) |
4637 | break; | |
4638 | failed: | |
4639 | slot++; | |
4640 | if (slot >= conf->copies) | |
4641 | slot = 0; | |
4642 | if (slot == first_slot) | |
4643 | break; | |
4644 | } | |
d094d686 | 4645 | rcu_read_unlock(); |
3ea7daa5 N |
4646 | if (!success) { |
4647 | /* couldn't read this block, must give up */ | |
4648 | set_bit(MD_RECOVERY_INTR, | |
4649 | &mddev->recovery); | |
4650 | return -EIO; | |
4651 | } | |
4652 | sectors -= s; | |
4653 | idx++; | |
4654 | } | |
4655 | return 0; | |
4656 | } | |
4657 | ||
4246a0b6 | 4658 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 4659 | { |
3ea7daa5 N |
4660 | struct r10bio *r10_bio = bio->bi_private; |
4661 | struct mddev *mddev = r10_bio->mddev; | |
4662 | struct r10conf *conf = mddev->private; | |
4663 | int d; | |
4664 | int slot; | |
4665 | int repl; | |
4666 | struct md_rdev *rdev = NULL; | |
4667 | ||
4668 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
4669 | if (repl) | |
4670 | rdev = conf->mirrors[d].replacement; | |
4671 | if (!rdev) { | |
4672 | smp_mb(); | |
4673 | rdev = conf->mirrors[d].rdev; | |
4674 | } | |
4675 | ||
4246a0b6 | 4676 | if (bio->bi_error) { |
3ea7daa5 N |
4677 | /* FIXME should record badblock */ |
4678 | md_error(mddev, rdev); | |
4679 | } | |
4680 | ||
4681 | rdev_dec_pending(rdev, mddev); | |
4682 | end_reshape_request(r10_bio); | |
4683 | } | |
4684 | ||
4685 | static void end_reshape_request(struct r10bio *r10_bio) | |
4686 | { | |
4687 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
4688 | return; | |
4689 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
4690 | bio_put(r10_bio->master_bio); | |
4691 | put_buf(r10_bio); | |
4692 | } | |
4693 | ||
4694 | static void raid10_finish_reshape(struct mddev *mddev) | |
4695 | { | |
4696 | struct r10conf *conf = mddev->private; | |
4697 | ||
4698 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
4699 | return; | |
4700 | ||
4701 | if (mddev->delta_disks > 0) { | |
4702 | sector_t size = raid10_size(mddev, 0, 0); | |
4703 | md_set_array_sectors(mddev, size); | |
4704 | if (mddev->recovery_cp > mddev->resync_max_sectors) { | |
4705 | mddev->recovery_cp = mddev->resync_max_sectors; | |
4706 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4707 | } | |
4708 | mddev->resync_max_sectors = size; | |
859644f0 HM |
4709 | if (mddev->queue) { |
4710 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4711 | revalidate_disk(mddev->gendisk); | |
4712 | } | |
63aced61 N |
4713 | } else { |
4714 | int d; | |
d094d686 | 4715 | rcu_read_lock(); |
63aced61 N |
4716 | for (d = conf->geo.raid_disks ; |
4717 | d < conf->geo.raid_disks - mddev->delta_disks; | |
4718 | d++) { | |
d094d686 | 4719 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
63aced61 N |
4720 | if (rdev) |
4721 | clear_bit(In_sync, &rdev->flags); | |
d094d686 | 4722 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
63aced61 N |
4723 | if (rdev) |
4724 | clear_bit(In_sync, &rdev->flags); | |
4725 | } | |
d094d686 | 4726 | rcu_read_unlock(); |
3ea7daa5 N |
4727 | } |
4728 | mddev->layout = mddev->new_layout; | |
4729 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
4730 | mddev->reshape_position = MaxSector; | |
4731 | mddev->delta_disks = 0; | |
4732 | mddev->reshape_backwards = 0; | |
4733 | } | |
4734 | ||
84fc4b56 | 4735 | static struct md_personality raid10_personality = |
1da177e4 LT |
4736 | { |
4737 | .name = "raid10", | |
2604b703 | 4738 | .level = 10, |
1da177e4 | 4739 | .owner = THIS_MODULE, |
849674e4 SL |
4740 | .make_request = raid10_make_request, |
4741 | .run = raid10_run, | |
afa0f557 | 4742 | .free = raid10_free, |
849674e4 SL |
4743 | .status = raid10_status, |
4744 | .error_handler = raid10_error, | |
1da177e4 LT |
4745 | .hot_add_disk = raid10_add_disk, |
4746 | .hot_remove_disk= raid10_remove_disk, | |
4747 | .spare_active = raid10_spare_active, | |
849674e4 | 4748 | .sync_request = raid10_sync_request, |
6cce3b23 | 4749 | .quiesce = raid10_quiesce, |
80c3a6ce | 4750 | .size = raid10_size, |
006a09a0 | 4751 | .resize = raid10_resize, |
dab8b292 | 4752 | .takeover = raid10_takeover, |
3ea7daa5 N |
4753 | .check_reshape = raid10_check_reshape, |
4754 | .start_reshape = raid10_start_reshape, | |
4755 | .finish_reshape = raid10_finish_reshape, | |
5c675f83 | 4756 | .congested = raid10_congested, |
1da177e4 LT |
4757 | }; |
4758 | ||
4759 | static int __init raid_init(void) | |
4760 | { | |
2604b703 | 4761 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
4762 | } |
4763 | ||
4764 | static void raid_exit(void) | |
4765 | { | |
2604b703 | 4766 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
4767 | } |
4768 | ||
4769 | module_init(raid_init); | |
4770 | module_exit(raid_exit); | |
4771 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 4772 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 4773 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 4774 | MODULE_ALIAS("md-raid10"); |
2604b703 | 4775 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
4776 | |
4777 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |