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> |
43b2e5d8 | 28 | #include "md.h" |
ef740c37 | 29 | #include "raid10.h" |
dab8b292 | 30 | #include "raid0.h" |
ef740c37 | 31 | #include "bitmap.h" |
1da177e4 LT |
32 | |
33 | /* | |
34 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
35 | * The layout of data is defined by | |
36 | * chunk_size | |
37 | * raid_disks | |
38 | * near_copies (stored in low byte of layout) | |
39 | * far_copies (stored in second byte of layout) | |
c93983bf | 40 | * far_offset (stored in bit 16 of layout ) |
475901af | 41 | * use_far_sets (stored in bit 17 of layout ) |
8bce6d35 | 42 | * use_far_sets_bugfixed (stored in bit 18 of layout ) |
1da177e4 | 43 | * |
475901af JB |
44 | * The data to be stored is divided into chunks using chunksize. Each device |
45 | * is divided into far_copies sections. In each section, chunks are laid out | |
46 | * in a style similar to raid0, but near_copies copies of each chunk is stored | |
47 | * (each on a different drive). The starting device for each section is offset | |
48 | * near_copies from the starting device of the previous section. Thus there | |
49 | * are (near_copies * far_copies) of each chunk, and each is on a different | |
50 | * drive. near_copies and far_copies must be at least one, and their product | |
51 | * is at most raid_disks. | |
c93983bf N |
52 | * |
53 | * If far_offset is true, then the far_copies are handled a bit differently. | |
475901af JB |
54 | * The copies are still in different stripes, but instead of being very far |
55 | * apart on disk, there are adjacent stripes. | |
56 | * | |
57 | * The far and offset algorithms are handled slightly differently if | |
58 | * 'use_far_sets' is true. In this case, the array's devices are grouped into | |
59 | * sets that are (near_copies * far_copies) in size. The far copied stripes | |
60 | * are still shifted by 'near_copies' devices, but this shifting stays confined | |
61 | * to the set rather than the entire array. This is done to improve the number | |
62 | * of device combinations that can fail without causing the array to fail. | |
63 | * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk | |
64 | * on a device): | |
65 | * A B C D A B C D E | |
66 | * ... ... | |
67 | * D A B C E A B C D | |
68 | * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s): | |
69 | * [A B] [C D] [A B] [C D E] | |
70 | * |...| |...| |...| | ... | | |
71 | * [B A] [D C] [B A] [E C D] | |
1da177e4 LT |
72 | */ |
73 | ||
74 | /* | |
75 | * Number of guaranteed r10bios in case of extreme VM load: | |
76 | */ | |
77 | #define NR_RAID10_BIOS 256 | |
78 | ||
473e87ce JB |
79 | /* when we get a read error on a read-only array, we redirect to another |
80 | * device without failing the first device, or trying to over-write to | |
81 | * correct the read error. To keep track of bad blocks on a per-bio | |
82 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
83 | */ | |
84 | #define IO_BLOCKED ((struct bio *)1) | |
85 | /* When we successfully write to a known bad-block, we need to remove the | |
86 | * bad-block marking which must be done from process context. So we record | |
87 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
88 | */ | |
89 | #define IO_MADE_GOOD ((struct bio *)2) | |
90 | ||
91 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
92 | ||
93 | /* When there are this many requests queued to be written by | |
34db0cd6 N |
94 | * the raid10 thread, we become 'congested' to provide back-pressure |
95 | * for writeback. | |
96 | */ | |
97 | static int max_queued_requests = 1024; | |
98 | ||
e879a879 N |
99 | static void allow_barrier(struct r10conf *conf); |
100 | static void lower_barrier(struct r10conf *conf); | |
635f6416 | 101 | static int _enough(struct r10conf *conf, int previous, int ignore); |
3ea7daa5 N |
102 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, |
103 | int *skipped); | |
104 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio); | |
4246a0b6 | 105 | static void end_reshape_write(struct bio *bio); |
3ea7daa5 | 106 | static void end_reshape(struct r10conf *conf); |
0a27ec96 | 107 | |
dd0fc66f | 108 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 109 | { |
e879a879 | 110 | struct r10conf *conf = data; |
9f2c9d12 | 111 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 112 | |
69335ef3 N |
113 | /* allocate a r10bio with room for raid_disks entries in the |
114 | * bios array */ | |
7eaceacc | 115 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
116 | } |
117 | ||
118 | static void r10bio_pool_free(void *r10_bio, void *data) | |
119 | { | |
120 | kfree(r10_bio); | |
121 | } | |
122 | ||
0310fa21 | 123 | /* Maximum size of each resync request */ |
1da177e4 | 124 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 125 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
126 | /* amount of memory to reserve for resync requests */ |
127 | #define RESYNC_WINDOW (1024*1024) | |
128 | /* maximum number of concurrent requests, memory permitting */ | |
129 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
130 | |
131 | /* | |
132 | * When performing a resync, we need to read and compare, so | |
133 | * we need as many pages are there are copies. | |
134 | * When performing a recovery, we need 2 bios, one for read, | |
135 | * one for write (we recover only one drive per r10buf) | |
136 | * | |
137 | */ | |
dd0fc66f | 138 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 139 | { |
e879a879 | 140 | struct r10conf *conf = data; |
1da177e4 | 141 | struct page *page; |
9f2c9d12 | 142 | struct r10bio *r10_bio; |
1da177e4 LT |
143 | struct bio *bio; |
144 | int i, j; | |
145 | int nalloc; | |
146 | ||
147 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 148 | if (!r10_bio) |
1da177e4 | 149 | return NULL; |
1da177e4 | 150 | |
3ea7daa5 N |
151 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || |
152 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
1da177e4 LT |
153 | nalloc = conf->copies; /* resync */ |
154 | else | |
155 | nalloc = 2; /* recovery */ | |
156 | ||
157 | /* | |
158 | * Allocate bios. | |
159 | */ | |
160 | for (j = nalloc ; j-- ; ) { | |
6746557f | 161 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
162 | if (!bio) |
163 | goto out_free_bio; | |
164 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
165 | if (!conf->have_replacement) |
166 | continue; | |
167 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
168 | if (!bio) | |
169 | goto out_free_bio; | |
170 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
171 | } |
172 | /* | |
173 | * Allocate RESYNC_PAGES data pages and attach them | |
174 | * where needed. | |
175 | */ | |
176 | for (j = 0 ; j < nalloc; j++) { | |
69335ef3 | 177 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
1da177e4 LT |
178 | bio = r10_bio->devs[j].bio; |
179 | for (i = 0; i < RESYNC_PAGES; i++) { | |
3ea7daa5 N |
180 | if (j > 0 && !test_bit(MD_RECOVERY_SYNC, |
181 | &conf->mddev->recovery)) { | |
182 | /* we can share bv_page's during recovery | |
183 | * and reshape */ | |
c65060ad NK |
184 | struct bio *rbio = r10_bio->devs[0].bio; |
185 | page = rbio->bi_io_vec[i].bv_page; | |
186 | get_page(page); | |
187 | } else | |
188 | page = alloc_page(gfp_flags); | |
1da177e4 LT |
189 | if (unlikely(!page)) |
190 | goto out_free_pages; | |
191 | ||
192 | bio->bi_io_vec[i].bv_page = page; | |
69335ef3 N |
193 | if (rbio) |
194 | rbio->bi_io_vec[i].bv_page = page; | |
1da177e4 LT |
195 | } |
196 | } | |
197 | ||
198 | return r10_bio; | |
199 | ||
200 | out_free_pages: | |
201 | for ( ; i > 0 ; i--) | |
1345b1d8 | 202 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
203 | while (j--) |
204 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 205 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
5fdd2cf8 | 206 | j = 0; |
1da177e4 | 207 | out_free_bio: |
5fdd2cf8 | 208 | for ( ; j < nalloc; j++) { |
209 | if (r10_bio->devs[j].bio) | |
210 | bio_put(r10_bio->devs[j].bio); | |
69335ef3 N |
211 | if (r10_bio->devs[j].repl_bio) |
212 | bio_put(r10_bio->devs[j].repl_bio); | |
213 | } | |
1da177e4 LT |
214 | r10bio_pool_free(r10_bio, conf); |
215 | return NULL; | |
216 | } | |
217 | ||
218 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
219 | { | |
220 | int i; | |
e879a879 | 221 | struct r10conf *conf = data; |
9f2c9d12 | 222 | struct r10bio *r10bio = __r10_bio; |
1da177e4 LT |
223 | int j; |
224 | ||
225 | for (j=0; j < conf->copies; j++) { | |
226 | struct bio *bio = r10bio->devs[j].bio; | |
227 | if (bio) { | |
228 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 229 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
230 | bio->bi_io_vec[i].bv_page = NULL; |
231 | } | |
232 | bio_put(bio); | |
233 | } | |
69335ef3 N |
234 | bio = r10bio->devs[j].repl_bio; |
235 | if (bio) | |
236 | bio_put(bio); | |
1da177e4 LT |
237 | } |
238 | r10bio_pool_free(r10bio, conf); | |
239 | } | |
240 | ||
e879a879 | 241 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
242 | { |
243 | int i; | |
244 | ||
245 | for (i = 0; i < conf->copies; i++) { | |
246 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 247 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
248 | bio_put(*bio); |
249 | *bio = NULL; | |
69335ef3 N |
250 | bio = &r10_bio->devs[i].repl_bio; |
251 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
252 | bio_put(*bio); | |
253 | *bio = NULL; | |
1da177e4 LT |
254 | } |
255 | } | |
256 | ||
9f2c9d12 | 257 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 258 | { |
e879a879 | 259 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 260 | |
1da177e4 LT |
261 | put_all_bios(conf, r10_bio); |
262 | mempool_free(r10_bio, conf->r10bio_pool); | |
263 | } | |
264 | ||
9f2c9d12 | 265 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 266 | { |
e879a879 | 267 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
268 | |
269 | mempool_free(r10_bio, conf->r10buf_pool); | |
270 | ||
0a27ec96 | 271 | lower_barrier(conf); |
1da177e4 LT |
272 | } |
273 | ||
9f2c9d12 | 274 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
275 | { |
276 | unsigned long flags; | |
fd01b88c | 277 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 278 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
279 | |
280 | spin_lock_irqsave(&conf->device_lock, flags); | |
281 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 282 | conf->nr_queued ++; |
1da177e4 LT |
283 | spin_unlock_irqrestore(&conf->device_lock, flags); |
284 | ||
388667be AJ |
285 | /* wake up frozen array... */ |
286 | wake_up(&conf->wait_barrier); | |
287 | ||
1da177e4 LT |
288 | md_wakeup_thread(mddev->thread); |
289 | } | |
290 | ||
291 | /* | |
292 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
293 | * operation and are ready to return a success/failure code to the buffer | |
294 | * cache layer. | |
295 | */ | |
9f2c9d12 | 296 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
297 | { |
298 | struct bio *bio = r10_bio->master_bio; | |
856e08e2 | 299 | int done; |
e879a879 | 300 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 301 | |
856e08e2 N |
302 | if (bio->bi_phys_segments) { |
303 | unsigned long flags; | |
304 | spin_lock_irqsave(&conf->device_lock, flags); | |
305 | bio->bi_phys_segments--; | |
306 | done = (bio->bi_phys_segments == 0); | |
307 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
308 | } else | |
309 | done = 1; | |
310 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4246a0b6 | 311 | bio->bi_error = -EIO; |
856e08e2 | 312 | if (done) { |
4246a0b6 | 313 | bio_endio(bio); |
856e08e2 N |
314 | /* |
315 | * Wake up any possible resync thread that waits for the device | |
316 | * to go idle. | |
317 | */ | |
318 | allow_barrier(conf); | |
319 | } | |
1da177e4 LT |
320 | free_r10bio(r10_bio); |
321 | } | |
322 | ||
323 | /* | |
324 | * Update disk head position estimator based on IRQ completion info. | |
325 | */ | |
9f2c9d12 | 326 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 327 | { |
e879a879 | 328 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
329 | |
330 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
331 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
332 | } | |
333 | ||
778ca018 NK |
334 | /* |
335 | * Find the disk number which triggered given bio | |
336 | */ | |
e879a879 | 337 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 338 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
339 | { |
340 | int slot; | |
69335ef3 | 341 | int repl = 0; |
778ca018 | 342 | |
69335ef3 | 343 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
344 | if (r10_bio->devs[slot].bio == bio) |
345 | break; | |
69335ef3 N |
346 | if (r10_bio->devs[slot].repl_bio == bio) { |
347 | repl = 1; | |
348 | break; | |
349 | } | |
350 | } | |
778ca018 NK |
351 | |
352 | BUG_ON(slot == conf->copies); | |
353 | update_head_pos(slot, r10_bio); | |
354 | ||
749c55e9 N |
355 | if (slotp) |
356 | *slotp = slot; | |
69335ef3 N |
357 | if (replp) |
358 | *replp = repl; | |
778ca018 NK |
359 | return r10_bio->devs[slot].devnum; |
360 | } | |
361 | ||
4246a0b6 | 362 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 363 | { |
4246a0b6 | 364 | int uptodate = !bio->bi_error; |
9f2c9d12 | 365 | struct r10bio *r10_bio = bio->bi_private; |
1da177e4 | 366 | int slot, dev; |
abbf098e | 367 | struct md_rdev *rdev; |
e879a879 | 368 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 369 | |
1da177e4 LT |
370 | slot = r10_bio->read_slot; |
371 | dev = r10_bio->devs[slot].devnum; | |
abbf098e | 372 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
373 | /* |
374 | * this branch is our 'one mirror IO has finished' event handler: | |
375 | */ | |
4443ae10 N |
376 | update_head_pos(slot, r10_bio); |
377 | ||
378 | if (uptodate) { | |
1da177e4 LT |
379 | /* |
380 | * Set R10BIO_Uptodate in our master bio, so that | |
381 | * we will return a good error code to the higher | |
382 | * levels even if IO on some other mirrored buffer fails. | |
383 | * | |
384 | * The 'master' represents the composite IO operation to | |
385 | * user-side. So if something waits for IO, then it will | |
386 | * wait for the 'master' bio. | |
387 | */ | |
388 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
389 | } else { |
390 | /* If all other devices that store this block have | |
391 | * failed, we want to return the error upwards rather | |
392 | * than fail the last device. Here we redefine | |
393 | * "uptodate" to mean "Don't want to retry" | |
394 | */ | |
635f6416 N |
395 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
396 | rdev->raid_disk)) | |
fae8cc5e | 397 | uptodate = 1; |
fae8cc5e N |
398 | } |
399 | if (uptodate) { | |
1da177e4 | 400 | raid_end_bio_io(r10_bio); |
abbf098e | 401 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 402 | } else { |
1da177e4 | 403 | /* |
7c4e06ff | 404 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
405 | */ |
406 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
407 | printk_ratelimited(KERN_ERR |
408 | "md/raid10:%s: %s: rescheduling sector %llu\n", | |
409 | mdname(conf->mddev), | |
abbf098e | 410 | bdevname(rdev->bdev, b), |
8bda470e | 411 | (unsigned long long)r10_bio->sector); |
856e08e2 | 412 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
413 | reschedule_retry(r10_bio); |
414 | } | |
1da177e4 LT |
415 | } |
416 | ||
9f2c9d12 | 417 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
418 | { |
419 | /* clear the bitmap if all writes complete successfully */ | |
420 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
421 | r10_bio->sectors, | |
422 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
423 | 0); | |
424 | md_write_end(r10_bio->mddev); | |
425 | } | |
426 | ||
9f2c9d12 | 427 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
428 | { |
429 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
430 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
431 | reschedule_retry(r10_bio); | |
432 | else { | |
433 | close_write(r10_bio); | |
434 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
435 | reschedule_retry(r10_bio); | |
436 | else | |
437 | raid_end_bio_io(r10_bio); | |
438 | } | |
439 | } | |
440 | } | |
441 | ||
4246a0b6 | 442 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 443 | { |
9f2c9d12 | 444 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 445 | int dev; |
749c55e9 | 446 | int dec_rdev = 1; |
e879a879 | 447 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 448 | int slot, repl; |
4ca40c2c | 449 | struct md_rdev *rdev = NULL; |
1da177e4 | 450 | |
475b0321 | 451 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 452 | |
475b0321 N |
453 | if (repl) |
454 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
455 | if (!rdev) { |
456 | smp_rmb(); | |
457 | repl = 0; | |
475b0321 | 458 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 459 | } |
1da177e4 LT |
460 | /* |
461 | * this branch is our 'one mirror IO has finished' event handler: | |
462 | */ | |
4246a0b6 | 463 | if (bio->bi_error) { |
475b0321 N |
464 | if (repl) |
465 | /* Never record new bad blocks to replacement, | |
466 | * just fail it. | |
467 | */ | |
468 | md_error(rdev->mddev, rdev); | |
469 | else { | |
470 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
471 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
472 | set_bit(MD_RECOVERY_NEEDED, | |
473 | &rdev->mddev->recovery); | |
475b0321 N |
474 | set_bit(R10BIO_WriteError, &r10_bio->state); |
475 | dec_rdev = 0; | |
476 | } | |
749c55e9 | 477 | } else { |
1da177e4 LT |
478 | /* |
479 | * Set R10BIO_Uptodate in our master bio, so that | |
480 | * we will return a good error code for to the higher | |
481 | * levels even if IO on some other mirrored buffer fails. | |
482 | * | |
483 | * The 'master' represents the composite IO operation to | |
484 | * user-side. So if something waits for IO, then it will | |
485 | * wait for the 'master' bio. | |
486 | */ | |
749c55e9 N |
487 | sector_t first_bad; |
488 | int bad_sectors; | |
489 | ||
3056e3ae AL |
490 | /* |
491 | * Do not set R10BIO_Uptodate if the current device is | |
492 | * rebuilding or Faulty. This is because we cannot use | |
493 | * such device for properly reading the data back (we could | |
494 | * potentially use it, if the current write would have felt | |
495 | * before rdev->recovery_offset, but for simplicity we don't | |
496 | * check this here. | |
497 | */ | |
498 | if (test_bit(In_sync, &rdev->flags) && | |
499 | !test_bit(Faulty, &rdev->flags)) | |
500 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 501 | |
749c55e9 | 502 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 503 | if (is_badblock(rdev, |
749c55e9 N |
504 | r10_bio->devs[slot].addr, |
505 | r10_bio->sectors, | |
506 | &first_bad, &bad_sectors)) { | |
507 | bio_put(bio); | |
475b0321 N |
508 | if (repl) |
509 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
510 | else | |
511 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
512 | dec_rdev = 0; |
513 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
514 | } | |
515 | } | |
516 | ||
1da177e4 LT |
517 | /* |
518 | * | |
519 | * Let's see if all mirrored write operations have finished | |
520 | * already. | |
521 | */ | |
19d5f834 | 522 | one_write_done(r10_bio); |
749c55e9 | 523 | if (dec_rdev) |
884162df | 524 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
525 | } |
526 | ||
1da177e4 LT |
527 | /* |
528 | * RAID10 layout manager | |
25985edc | 529 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
530 | * parameters: near_copies and far_copies. |
531 | * near_copies * far_copies must be <= raid_disks. | |
532 | * Normally one of these will be 1. | |
533 | * If both are 1, we get raid0. | |
534 | * If near_copies == raid_disks, we get raid1. | |
535 | * | |
25985edc | 536 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
537 | * first chunk, followed by near_copies copies of the next chunk and |
538 | * so on. | |
539 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
540 | * as described above, we start again with a device offset of near_copies. | |
541 | * So we effectively have another copy of the whole array further down all | |
542 | * the drives, but with blocks on different drives. | |
543 | * With this layout, and block is never stored twice on the one device. | |
544 | * | |
545 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 546 | * on each device that it is on. |
1da177e4 LT |
547 | * |
548 | * raid10_find_virt does the reverse mapping, from a device and a | |
549 | * sector offset to a virtual address | |
550 | */ | |
551 | ||
f8c9e74f | 552 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
553 | { |
554 | int n,f; | |
555 | sector_t sector; | |
556 | sector_t chunk; | |
557 | sector_t stripe; | |
558 | int dev; | |
1da177e4 | 559 | int slot = 0; |
9a3152ab JB |
560 | int last_far_set_start, last_far_set_size; |
561 | ||
562 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
563 | last_far_set_start *= geo->far_set_size; | |
564 | ||
565 | last_far_set_size = geo->far_set_size; | |
566 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
567 | |
568 | /* now calculate first sector/dev */ | |
5cf00fcd N |
569 | chunk = r10bio->sector >> geo->chunk_shift; |
570 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 571 | |
5cf00fcd | 572 | chunk *= geo->near_copies; |
1da177e4 | 573 | stripe = chunk; |
5cf00fcd N |
574 | dev = sector_div(stripe, geo->raid_disks); |
575 | if (geo->far_offset) | |
576 | stripe *= geo->far_copies; | |
1da177e4 | 577 | |
5cf00fcd | 578 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
579 | |
580 | /* and calculate all the others */ | |
5cf00fcd | 581 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 582 | int d = dev; |
475901af | 583 | int set; |
1da177e4 | 584 | sector_t s = sector; |
1da177e4 | 585 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 586 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
587 | slot++; |
588 | ||
5cf00fcd | 589 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 590 | set = d / geo->far_set_size; |
5cf00fcd | 591 | d += geo->near_copies; |
475901af | 592 | |
9a3152ab JB |
593 | if ((geo->raid_disks % geo->far_set_size) && |
594 | (d > last_far_set_start)) { | |
595 | d -= last_far_set_start; | |
596 | d %= last_far_set_size; | |
597 | d += last_far_set_start; | |
598 | } else { | |
599 | d %= geo->far_set_size; | |
600 | d += geo->far_set_size * set; | |
601 | } | |
5cf00fcd | 602 | s += geo->stride; |
1da177e4 LT |
603 | r10bio->devs[slot].devnum = d; |
604 | r10bio->devs[slot].addr = s; | |
605 | slot++; | |
606 | } | |
607 | dev++; | |
5cf00fcd | 608 | if (dev >= geo->raid_disks) { |
1da177e4 | 609 | dev = 0; |
5cf00fcd | 610 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
611 | } |
612 | } | |
f8c9e74f N |
613 | } |
614 | ||
615 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
616 | { | |
617 | struct geom *geo = &conf->geo; | |
618 | ||
619 | if (conf->reshape_progress != MaxSector && | |
620 | ((r10bio->sector >= conf->reshape_progress) != | |
621 | conf->mddev->reshape_backwards)) { | |
622 | set_bit(R10BIO_Previous, &r10bio->state); | |
623 | geo = &conf->prev; | |
624 | } else | |
625 | clear_bit(R10BIO_Previous, &r10bio->state); | |
626 | ||
627 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
628 | } |
629 | ||
e879a879 | 630 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
631 | { |
632 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
633 | /* Never use conf->prev as this is only called during resync |
634 | * or recovery, so reshape isn't happening | |
635 | */ | |
5cf00fcd | 636 | struct geom *geo = &conf->geo; |
475901af JB |
637 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
638 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
639 | int last_far_set_start; |
640 | ||
641 | if (geo->raid_disks % geo->far_set_size) { | |
642 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
643 | last_far_set_start *= geo->far_set_size; | |
644 | ||
645 | if (dev >= last_far_set_start) { | |
646 | far_set_size = geo->far_set_size; | |
647 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
648 | far_set_start = last_far_set_start; | |
649 | } | |
650 | } | |
1da177e4 | 651 | |
5cf00fcd N |
652 | offset = sector & geo->chunk_mask; |
653 | if (geo->far_offset) { | |
c93983bf | 654 | int fc; |
5cf00fcd N |
655 | chunk = sector >> geo->chunk_shift; |
656 | fc = sector_div(chunk, geo->far_copies); | |
657 | dev -= fc * geo->near_copies; | |
475901af JB |
658 | if (dev < far_set_start) |
659 | dev += far_set_size; | |
c93983bf | 660 | } else { |
5cf00fcd N |
661 | while (sector >= geo->stride) { |
662 | sector -= geo->stride; | |
475901af JB |
663 | if (dev < (geo->near_copies + far_set_start)) |
664 | dev += far_set_size - geo->near_copies; | |
c93983bf | 665 | else |
5cf00fcd | 666 | dev -= geo->near_copies; |
c93983bf | 667 | } |
5cf00fcd | 668 | chunk = sector >> geo->chunk_shift; |
c93983bf | 669 | } |
5cf00fcd N |
670 | vchunk = chunk * geo->raid_disks + dev; |
671 | sector_div(vchunk, geo->near_copies); | |
672 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
673 | } |
674 | ||
1da177e4 LT |
675 | /* |
676 | * This routine returns the disk from which the requested read should | |
677 | * be done. There is a per-array 'next expected sequential IO' sector | |
678 | * number - if this matches on the next IO then we use the last disk. | |
679 | * There is also a per-disk 'last know head position' sector that is | |
680 | * maintained from IRQ contexts, both the normal and the resync IO | |
681 | * completion handlers update this position correctly. If there is no | |
682 | * perfect sequential match then we pick the disk whose head is closest. | |
683 | * | |
684 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
685 | * because position is mirror, not device based. | |
686 | * | |
687 | * The rdev for the device selected will have nr_pending incremented. | |
688 | */ | |
689 | ||
690 | /* | |
691 | * FIXME: possibly should rethink readbalancing and do it differently | |
692 | * depending on near_copies / far_copies geometry. | |
693 | */ | |
96c3fd1f N |
694 | static struct md_rdev *read_balance(struct r10conf *conf, |
695 | struct r10bio *r10_bio, | |
696 | int *max_sectors) | |
1da177e4 | 697 | { |
af3a2cd6 | 698 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 699 | int disk, slot; |
856e08e2 N |
700 | int sectors = r10_bio->sectors; |
701 | int best_good_sectors; | |
56d99121 | 702 | sector_t new_distance, best_dist; |
3bbae04b | 703 | struct md_rdev *best_rdev, *rdev = NULL; |
56d99121 N |
704 | int do_balance; |
705 | int best_slot; | |
5cf00fcd | 706 | struct geom *geo = &conf->geo; |
1da177e4 LT |
707 | |
708 | raid10_find_phys(conf, r10_bio); | |
709 | rcu_read_lock(); | |
56d99121 | 710 | retry: |
856e08e2 | 711 | sectors = r10_bio->sectors; |
56d99121 | 712 | best_slot = -1; |
abbf098e | 713 | best_rdev = NULL; |
56d99121 | 714 | best_dist = MaxSector; |
856e08e2 | 715 | best_good_sectors = 0; |
56d99121 | 716 | do_balance = 1; |
1da177e4 LT |
717 | /* |
718 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
719 | * device if no resync is going on (recovery is ok), or below |
720 | * the resync window. We take the first readable disk when | |
721 | * above the resync window. | |
1da177e4 LT |
722 | */ |
723 | if (conf->mddev->recovery_cp < MaxSector | |
56d99121 N |
724 | && (this_sector + sectors >= conf->next_resync)) |
725 | do_balance = 0; | |
1da177e4 | 726 | |
56d99121 | 727 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
728 | sector_t first_bad; |
729 | int bad_sectors; | |
730 | sector_t dev_sector; | |
731 | ||
56d99121 N |
732 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
733 | continue; | |
1da177e4 | 734 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
735 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
736 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
737 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
738 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 | 739 | if (rdev == NULL || |
8ae12666 | 740 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
741 | continue; |
742 | if (!test_bit(In_sync, &rdev->flags) && | |
743 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
744 | continue; |
745 | ||
856e08e2 N |
746 | dev_sector = r10_bio->devs[slot].addr; |
747 | if (is_badblock(rdev, dev_sector, sectors, | |
748 | &first_bad, &bad_sectors)) { | |
749 | if (best_dist < MaxSector) | |
750 | /* Already have a better slot */ | |
751 | continue; | |
752 | if (first_bad <= dev_sector) { | |
753 | /* Cannot read here. If this is the | |
754 | * 'primary' device, then we must not read | |
755 | * beyond 'bad_sectors' from another device. | |
756 | */ | |
757 | bad_sectors -= (dev_sector - first_bad); | |
758 | if (!do_balance && sectors > bad_sectors) | |
759 | sectors = bad_sectors; | |
760 | if (best_good_sectors > sectors) | |
761 | best_good_sectors = sectors; | |
762 | } else { | |
763 | sector_t good_sectors = | |
764 | first_bad - dev_sector; | |
765 | if (good_sectors > best_good_sectors) { | |
766 | best_good_sectors = good_sectors; | |
767 | best_slot = slot; | |
abbf098e | 768 | best_rdev = rdev; |
856e08e2 N |
769 | } |
770 | if (!do_balance) | |
771 | /* Must read from here */ | |
772 | break; | |
773 | } | |
774 | continue; | |
775 | } else | |
776 | best_good_sectors = sectors; | |
777 | ||
56d99121 N |
778 | if (!do_balance) |
779 | break; | |
1da177e4 | 780 | |
22dfdf52 N |
781 | /* This optimisation is debatable, and completely destroys |
782 | * sequential read speed for 'far copies' arrays. So only | |
783 | * keep it for 'near' arrays, and review those later. | |
784 | */ | |
5cf00fcd | 785 | if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
1da177e4 | 786 | break; |
8ed3a195 KS |
787 | |
788 | /* for far > 1 always use the lowest address */ | |
5cf00fcd | 789 | if (geo->far_copies > 1) |
56d99121 | 790 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 791 | else |
56d99121 N |
792 | new_distance = abs(r10_bio->devs[slot].addr - |
793 | conf->mirrors[disk].head_position); | |
794 | if (new_distance < best_dist) { | |
795 | best_dist = new_distance; | |
796 | best_slot = slot; | |
abbf098e | 797 | best_rdev = rdev; |
1da177e4 LT |
798 | } |
799 | } | |
abbf098e | 800 | if (slot >= conf->copies) { |
56d99121 | 801 | slot = best_slot; |
abbf098e N |
802 | rdev = best_rdev; |
803 | } | |
1da177e4 | 804 | |
56d99121 | 805 | if (slot >= 0) { |
56d99121 N |
806 | atomic_inc(&rdev->nr_pending); |
807 | if (test_bit(Faulty, &rdev->flags)) { | |
808 | /* Cannot risk returning a device that failed | |
809 | * before we inc'ed nr_pending | |
810 | */ | |
811 | rdev_dec_pending(rdev, conf->mddev); | |
812 | goto retry; | |
813 | } | |
814 | r10_bio->read_slot = slot; | |
815 | } else | |
96c3fd1f | 816 | rdev = NULL; |
1da177e4 | 817 | rcu_read_unlock(); |
856e08e2 | 818 | *max_sectors = best_good_sectors; |
1da177e4 | 819 | |
96c3fd1f | 820 | return rdev; |
1da177e4 LT |
821 | } |
822 | ||
5c675f83 | 823 | static int raid10_congested(struct mddev *mddev, int bits) |
0d129228 | 824 | { |
e879a879 | 825 | struct r10conf *conf = mddev->private; |
0d129228 N |
826 | int i, ret = 0; |
827 | ||
4452226e | 828 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
829 | conf->pending_count >= max_queued_requests) |
830 | return 1; | |
831 | ||
0d129228 | 832 | rcu_read_lock(); |
f8c9e74f N |
833 | for (i = 0; |
834 | (i < conf->geo.raid_disks || i < conf->prev.raid_disks) | |
835 | && ret == 0; | |
836 | i++) { | |
3cb03002 | 837 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 838 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 839 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
840 | |
841 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
842 | } | |
843 | } | |
844 | rcu_read_unlock(); | |
845 | return ret; | |
846 | } | |
847 | ||
e879a879 | 848 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
849 | { |
850 | /* Any writes that have been queued but are awaiting | |
851 | * bitmap updates get flushed here. | |
a35e63ef | 852 | */ |
a35e63ef N |
853 | spin_lock_irq(&conf->device_lock); |
854 | ||
855 | if (conf->pending_bio_list.head) { | |
856 | struct bio *bio; | |
857 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 858 | conf->pending_count = 0; |
a35e63ef N |
859 | spin_unlock_irq(&conf->device_lock); |
860 | /* flush any pending bitmap writes to disk | |
861 | * before proceeding w/ I/O */ | |
862 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 863 | wake_up(&conf->wait_barrier); |
a35e63ef N |
864 | |
865 | while (bio) { /* submit pending writes */ | |
866 | struct bio *next = bio->bi_next; | |
867 | bio->bi_next = NULL; | |
532a2a3f SL |
868 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
869 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
870 | /* Just ignore it */ | |
4246a0b6 | 871 | bio_endio(bio); |
532a2a3f SL |
872 | else |
873 | generic_make_request(bio); | |
a35e63ef N |
874 | bio = next; |
875 | } | |
a35e63ef N |
876 | } else |
877 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 878 | } |
7eaceacc | 879 | |
0a27ec96 N |
880 | /* Barriers.... |
881 | * Sometimes we need to suspend IO while we do something else, | |
882 | * either some resync/recovery, or reconfigure the array. | |
883 | * To do this we raise a 'barrier'. | |
884 | * The 'barrier' is a counter that can be raised multiple times | |
885 | * to count how many activities are happening which preclude | |
886 | * normal IO. | |
887 | * We can only raise the barrier if there is no pending IO. | |
888 | * i.e. if nr_pending == 0. | |
889 | * We choose only to raise the barrier if no-one is waiting for the | |
890 | * barrier to go down. This means that as soon as an IO request | |
891 | * is ready, no other operations which require a barrier will start | |
892 | * until the IO request has had a chance. | |
893 | * | |
894 | * So: regular IO calls 'wait_barrier'. When that returns there | |
895 | * is no backgroup IO happening, It must arrange to call | |
896 | * allow_barrier when it has finished its IO. | |
897 | * backgroup IO calls must call raise_barrier. Once that returns | |
898 | * there is no normal IO happeing. It must arrange to call | |
899 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 900 | */ |
1da177e4 | 901 | |
e879a879 | 902 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 903 | { |
6cce3b23 | 904 | BUG_ON(force && !conf->barrier); |
1da177e4 | 905 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 906 | |
6cce3b23 N |
907 | /* Wait until no block IO is waiting (unless 'force') */ |
908 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
eed8c02e | 909 | conf->resync_lock); |
0a27ec96 N |
910 | |
911 | /* block any new IO from starting */ | |
912 | conf->barrier++; | |
913 | ||
c3b328ac | 914 | /* Now wait for all pending IO to complete */ |
0a27ec96 N |
915 | wait_event_lock_irq(conf->wait_barrier, |
916 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
eed8c02e | 917 | conf->resync_lock); |
0a27ec96 N |
918 | |
919 | spin_unlock_irq(&conf->resync_lock); | |
920 | } | |
921 | ||
e879a879 | 922 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
923 | { |
924 | unsigned long flags; | |
925 | spin_lock_irqsave(&conf->resync_lock, flags); | |
926 | conf->barrier--; | |
927 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
928 | wake_up(&conf->wait_barrier); | |
929 | } | |
930 | ||
e879a879 | 931 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
932 | { |
933 | spin_lock_irq(&conf->resync_lock); | |
934 | if (conf->barrier) { | |
935 | conf->nr_waiting++; | |
d6b42dcb N |
936 | /* Wait for the barrier to drop. |
937 | * However if there are already pending | |
938 | * requests (preventing the barrier from | |
939 | * rising completely), and the | |
940 | * pre-process bio queue isn't empty, | |
941 | * then don't wait, as we need to empty | |
942 | * that queue to get the nr_pending | |
943 | * count down. | |
944 | */ | |
945 | wait_event_lock_irq(conf->wait_barrier, | |
946 | !conf->barrier || | |
947 | (conf->nr_pending && | |
948 | current->bio_list && | |
949 | !bio_list_empty(current->bio_list)), | |
eed8c02e | 950 | conf->resync_lock); |
0a27ec96 | 951 | conf->nr_waiting--; |
1da177e4 | 952 | } |
0a27ec96 | 953 | conf->nr_pending++; |
1da177e4 LT |
954 | spin_unlock_irq(&conf->resync_lock); |
955 | } | |
956 | ||
e879a879 | 957 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 N |
958 | { |
959 | unsigned long flags; | |
960 | spin_lock_irqsave(&conf->resync_lock, flags); | |
961 | conf->nr_pending--; | |
962 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
963 | wake_up(&conf->wait_barrier); | |
964 | } | |
965 | ||
e2d59925 | 966 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
967 | { |
968 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 969 | * go quiet. |
4443ae10 | 970 | * We increment barrier and nr_waiting, and then |
e2d59925 | 971 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
972 | * This is called in the context of one normal IO request |
973 | * that has failed. Thus any sync request that might be pending | |
974 | * will be blocked by nr_pending, and we need to wait for | |
975 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 976 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
977 | * must match the number of pending IOs (nr_pending) before |
978 | * we continue. | |
4443ae10 N |
979 | */ |
980 | spin_lock_irq(&conf->resync_lock); | |
981 | conf->barrier++; | |
982 | conf->nr_waiting++; | |
eed8c02e | 983 | wait_event_lock_irq_cmd(conf->wait_barrier, |
e2d59925 | 984 | conf->nr_pending == conf->nr_queued+extra, |
eed8c02e LC |
985 | conf->resync_lock, |
986 | flush_pending_writes(conf)); | |
c3b328ac | 987 | |
4443ae10 N |
988 | spin_unlock_irq(&conf->resync_lock); |
989 | } | |
990 | ||
e879a879 | 991 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
992 | { |
993 | /* reverse the effect of the freeze */ | |
994 | spin_lock_irq(&conf->resync_lock); | |
995 | conf->barrier--; | |
996 | conf->nr_waiting--; | |
997 | wake_up(&conf->wait_barrier); | |
998 | spin_unlock_irq(&conf->resync_lock); | |
999 | } | |
1000 | ||
f8c9e74f N |
1001 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1002 | struct md_rdev *rdev) | |
1003 | { | |
1004 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1005 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1006 | return rdev->data_offset; | |
1007 | else | |
1008 | return rdev->new_data_offset; | |
1009 | } | |
1010 | ||
57c67df4 N |
1011 | struct raid10_plug_cb { |
1012 | struct blk_plug_cb cb; | |
1013 | struct bio_list pending; | |
1014 | int pending_cnt; | |
1015 | }; | |
1016 | ||
1017 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1018 | { | |
1019 | struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb, | |
1020 | cb); | |
1021 | struct mddev *mddev = plug->cb.data; | |
1022 | struct r10conf *conf = mddev->private; | |
1023 | struct bio *bio; | |
1024 | ||
874807a8 | 1025 | if (from_schedule || current->bio_list) { |
57c67df4 N |
1026 | spin_lock_irq(&conf->device_lock); |
1027 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1028 | conf->pending_count += plug->pending_cnt; | |
1029 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1030 | wake_up(&conf->wait_barrier); |
57c67df4 N |
1031 | md_wakeup_thread(mddev->thread); |
1032 | kfree(plug); | |
1033 | return; | |
1034 | } | |
1035 | ||
1036 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1037 | bio = bio_list_get(&plug->pending); | |
1038 | bitmap_unplug(mddev->bitmap); | |
1039 | wake_up(&conf->wait_barrier); | |
1040 | ||
1041 | while (bio) { /* submit pending writes */ | |
1042 | struct bio *next = bio->bi_next; | |
1043 | bio->bi_next = NULL; | |
32f9f570 SL |
1044 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
1045 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
1046 | /* Just ignore it */ | |
4246a0b6 | 1047 | bio_endio(bio); |
32f9f570 SL |
1048 | else |
1049 | generic_make_request(bio); | |
57c67df4 N |
1050 | bio = next; |
1051 | } | |
1052 | kfree(plug); | |
1053 | } | |
1054 | ||
20d0189b | 1055 | static void __make_request(struct mddev *mddev, struct bio *bio) |
1da177e4 | 1056 | { |
e879a879 | 1057 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 1058 | struct r10bio *r10_bio; |
1da177e4 LT |
1059 | struct bio *read_bio; |
1060 | int i; | |
a362357b | 1061 | const int rw = bio_data_dir(bio); |
2c7d46ec | 1062 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 1063 | const unsigned long do_fua = (bio->bi_rw & REQ_FUA); |
532a2a3f SL |
1064 | const unsigned long do_discard = (bio->bi_rw |
1065 | & (REQ_DISCARD | REQ_SECURE)); | |
c8dc9c65 | 1066 | const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME); |
6cce3b23 | 1067 | unsigned long flags; |
3cb03002 | 1068 | struct md_rdev *blocked_rdev; |
57c67df4 N |
1069 | struct blk_plug_cb *cb; |
1070 | struct raid10_plug_cb *plug = NULL; | |
d4432c23 N |
1071 | int sectors_handled; |
1072 | int max_sectors; | |
3ea7daa5 | 1073 | int sectors; |
1da177e4 | 1074 | |
cc13b1d1 N |
1075 | /* |
1076 | * Register the new request and wait if the reconstruction | |
1077 | * thread has put up a bar for new requests. | |
1078 | * Continue immediately if no resync is active currently. | |
1079 | */ | |
1080 | wait_barrier(conf); | |
1081 | ||
aa8b57aa | 1082 | sectors = bio_sectors(bio); |
3ea7daa5 | 1083 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
4f024f37 KO |
1084 | bio->bi_iter.bi_sector < conf->reshape_progress && |
1085 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
3ea7daa5 N |
1086 | /* IO spans the reshape position. Need to wait for |
1087 | * reshape to pass | |
1088 | */ | |
1089 | allow_barrier(conf); | |
1090 | wait_event(conf->wait_barrier, | |
4f024f37 KO |
1091 | conf->reshape_progress <= bio->bi_iter.bi_sector || |
1092 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1093 | sectors); | |
3ea7daa5 N |
1094 | wait_barrier(conf); |
1095 | } | |
1096 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
1097 | bio_data_dir(bio) == WRITE && | |
1098 | (mddev->reshape_backwards | |
4f024f37 KO |
1099 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1100 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1101 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1102 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1103 | /* Need to update reshape_position in metadata */ |
1104 | mddev->reshape_position = conf->reshape_progress; | |
85ad1d13 GJ |
1105 | set_mask_bits(&mddev->flags, 0, |
1106 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
3ea7daa5 N |
1107 | md_wakeup_thread(mddev->thread); |
1108 | wait_event(mddev->sb_wait, | |
1109 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)); | |
1110 | ||
1111 | conf->reshape_safe = mddev->reshape_position; | |
1112 | } | |
1113 | ||
1da177e4 LT |
1114 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); |
1115 | ||
1116 | r10_bio->master_bio = bio; | |
3ea7daa5 | 1117 | r10_bio->sectors = sectors; |
1da177e4 LT |
1118 | |
1119 | r10_bio->mddev = mddev; | |
4f024f37 | 1120 | r10_bio->sector = bio->bi_iter.bi_sector; |
6cce3b23 | 1121 | r10_bio->state = 0; |
1da177e4 | 1122 | |
856e08e2 N |
1123 | /* We might need to issue multiple reads to different |
1124 | * devices if there are bad blocks around, so we keep | |
1125 | * track of the number of reads in bio->bi_phys_segments. | |
1126 | * If this is 0, there is only one r10_bio and no locking | |
1127 | * will be needed when the request completes. If it is | |
1128 | * non-zero, then it is the number of not-completed requests. | |
1129 | */ | |
1130 | bio->bi_phys_segments = 0; | |
b7c44ed9 | 1131 | bio_clear_flag(bio, BIO_SEG_VALID); |
856e08e2 | 1132 | |
a362357b | 1133 | if (rw == READ) { |
1da177e4 LT |
1134 | /* |
1135 | * read balancing logic: | |
1136 | */ | |
96c3fd1f | 1137 | struct md_rdev *rdev; |
856e08e2 N |
1138 | int slot; |
1139 | ||
1140 | read_again: | |
96c3fd1f N |
1141 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1142 | if (!rdev) { | |
1da177e4 | 1143 | raid_end_bio_io(r10_bio); |
5a7bbad2 | 1144 | return; |
1da177e4 | 1145 | } |
96c3fd1f | 1146 | slot = r10_bio->read_slot; |
1da177e4 | 1147 | |
a167f663 | 1148 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1149 | bio_trim(read_bio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1150 | max_sectors); |
1da177e4 LT |
1151 | |
1152 | r10_bio->devs[slot].bio = read_bio; | |
abbf098e | 1153 | r10_bio->devs[slot].rdev = rdev; |
1da177e4 | 1154 | |
4f024f37 | 1155 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + |
f8c9e74f | 1156 | choose_data_offset(r10_bio, rdev); |
96c3fd1f | 1157 | read_bio->bi_bdev = rdev->bdev; |
1da177e4 | 1158 | read_bio->bi_end_io = raid10_end_read_request; |
7b6d91da | 1159 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1160 | read_bio->bi_private = r10_bio; |
1161 | ||
856e08e2 N |
1162 | if (max_sectors < r10_bio->sectors) { |
1163 | /* Could not read all from this device, so we will | |
1164 | * need another r10_bio. | |
1165 | */ | |
b50c259e | 1166 | sectors_handled = (r10_bio->sector + max_sectors |
4f024f37 | 1167 | - bio->bi_iter.bi_sector); |
856e08e2 N |
1168 | r10_bio->sectors = max_sectors; |
1169 | spin_lock_irq(&conf->device_lock); | |
1170 | if (bio->bi_phys_segments == 0) | |
1171 | bio->bi_phys_segments = 2; | |
1172 | else | |
1173 | bio->bi_phys_segments++; | |
b50c259e | 1174 | spin_unlock_irq(&conf->device_lock); |
856e08e2 N |
1175 | /* Cannot call generic_make_request directly |
1176 | * as that will be queued in __generic_make_request | |
1177 | * and subsequent mempool_alloc might block | |
1178 | * waiting for it. so hand bio over to raid10d. | |
1179 | */ | |
1180 | reschedule_retry(r10_bio); | |
1181 | ||
1182 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1183 | ||
1184 | r10_bio->master_bio = bio; | |
aa8b57aa | 1185 | r10_bio->sectors = bio_sectors(bio) - sectors_handled; |
856e08e2 N |
1186 | r10_bio->state = 0; |
1187 | r10_bio->mddev = mddev; | |
4f024f37 KO |
1188 | r10_bio->sector = bio->bi_iter.bi_sector + |
1189 | sectors_handled; | |
856e08e2 N |
1190 | goto read_again; |
1191 | } else | |
1192 | generic_make_request(read_bio); | |
5a7bbad2 | 1193 | return; |
1da177e4 LT |
1194 | } |
1195 | ||
1196 | /* | |
1197 | * WRITE: | |
1198 | */ | |
34db0cd6 N |
1199 | if (conf->pending_count >= max_queued_requests) { |
1200 | md_wakeup_thread(mddev->thread); | |
1201 | wait_event(conf->wait_barrier, | |
1202 | conf->pending_count < max_queued_requests); | |
1203 | } | |
6bfe0b49 | 1204 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1205 | * inc refcount on their rdev. Record them by setting |
1206 | * bios[x] to bio | |
d4432c23 N |
1207 | * If there are known/acknowledged bad blocks on any device |
1208 | * on which we have seen a write error, we want to avoid | |
1209 | * writing to those blocks. This potentially requires several | |
1210 | * writes to write around the bad blocks. Each set of writes | |
1211 | * gets its own r10_bio with a set of bios attached. The number | |
1212 | * of r10_bios is recored in bio->bi_phys_segments just as with | |
1213 | * the read case. | |
1da177e4 | 1214 | */ |
c3b328ac | 1215 | |
69335ef3 | 1216 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1217 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1218 | retry_write: |
cb6969e8 | 1219 | blocked_rdev = NULL; |
1da177e4 | 1220 | rcu_read_lock(); |
d4432c23 N |
1221 | max_sectors = r10_bio->sectors; |
1222 | ||
1da177e4 LT |
1223 | for (i = 0; i < conf->copies; i++) { |
1224 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1225 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1226 | struct md_rdev *rrdev = rcu_dereference( |
1227 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1228 | if (rdev == rrdev) |
1229 | rrdev = NULL; | |
6bfe0b49 DW |
1230 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1231 | atomic_inc(&rdev->nr_pending); | |
1232 | blocked_rdev = rdev; | |
1233 | break; | |
1234 | } | |
475b0321 N |
1235 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { |
1236 | atomic_inc(&rrdev->nr_pending); | |
1237 | blocked_rdev = rrdev; | |
1238 | break; | |
1239 | } | |
8ae12666 | 1240 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1241 | rdev = NULL; |
8ae12666 | 1242 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1243 | rrdev = NULL; |
1244 | ||
d4432c23 | 1245 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1246 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1247 | |
1248 | if (!rdev && !rrdev) { | |
6cce3b23 | 1249 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1250 | continue; |
1251 | } | |
e7c0c3fa | 1252 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1253 | sector_t first_bad; |
1254 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1255 | int bad_sectors; | |
1256 | int is_bad; | |
1257 | ||
1258 | is_bad = is_badblock(rdev, dev_sector, | |
1259 | max_sectors, | |
1260 | &first_bad, &bad_sectors); | |
1261 | if (is_bad < 0) { | |
1262 | /* Mustn't write here until the bad block | |
1263 | * is acknowledged | |
1264 | */ | |
1265 | atomic_inc(&rdev->nr_pending); | |
1266 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1267 | blocked_rdev = rdev; | |
1268 | break; | |
1269 | } | |
1270 | if (is_bad && first_bad <= dev_sector) { | |
1271 | /* Cannot write here at all */ | |
1272 | bad_sectors -= (dev_sector - first_bad); | |
1273 | if (bad_sectors < max_sectors) | |
1274 | /* Mustn't write more than bad_sectors | |
1275 | * to other devices yet | |
1276 | */ | |
1277 | max_sectors = bad_sectors; | |
1278 | /* We don't set R10BIO_Degraded as that | |
1279 | * only applies if the disk is missing, | |
1280 | * so it might be re-added, and we want to | |
1281 | * know to recover this chunk. | |
1282 | * In this case the device is here, and the | |
1283 | * fact that this chunk is not in-sync is | |
1284 | * recorded in the bad block log. | |
1285 | */ | |
1286 | continue; | |
1287 | } | |
1288 | if (is_bad) { | |
1289 | int good_sectors = first_bad - dev_sector; | |
1290 | if (good_sectors < max_sectors) | |
1291 | max_sectors = good_sectors; | |
1292 | } | |
6cce3b23 | 1293 | } |
e7c0c3fa N |
1294 | if (rdev) { |
1295 | r10_bio->devs[i].bio = bio; | |
1296 | atomic_inc(&rdev->nr_pending); | |
1297 | } | |
475b0321 N |
1298 | if (rrdev) { |
1299 | r10_bio->devs[i].repl_bio = bio; | |
1300 | atomic_inc(&rrdev->nr_pending); | |
1301 | } | |
1da177e4 LT |
1302 | } |
1303 | rcu_read_unlock(); | |
1304 | ||
6bfe0b49 DW |
1305 | if (unlikely(blocked_rdev)) { |
1306 | /* Have to wait for this device to get unblocked, then retry */ | |
1307 | int j; | |
1308 | int d; | |
1309 | ||
475b0321 | 1310 | for (j = 0; j < i; j++) { |
6bfe0b49 DW |
1311 | if (r10_bio->devs[j].bio) { |
1312 | d = r10_bio->devs[j].devnum; | |
1313 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1314 | } | |
475b0321 | 1315 | if (r10_bio->devs[j].repl_bio) { |
4ca40c2c | 1316 | struct md_rdev *rdev; |
475b0321 | 1317 | d = r10_bio->devs[j].devnum; |
4ca40c2c N |
1318 | rdev = conf->mirrors[d].replacement; |
1319 | if (!rdev) { | |
1320 | /* Race with remove_disk */ | |
1321 | smp_mb(); | |
1322 | rdev = conf->mirrors[d].rdev; | |
1323 | } | |
1324 | rdev_dec_pending(rdev, mddev); | |
475b0321 N |
1325 | } |
1326 | } | |
6bfe0b49 DW |
1327 | allow_barrier(conf); |
1328 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1329 | wait_barrier(conf); | |
1330 | goto retry_write; | |
1331 | } | |
1332 | ||
d4432c23 N |
1333 | if (max_sectors < r10_bio->sectors) { |
1334 | /* We are splitting this into multiple parts, so | |
1335 | * we need to prepare for allocating another r10_bio. | |
1336 | */ | |
1337 | r10_bio->sectors = max_sectors; | |
1338 | spin_lock_irq(&conf->device_lock); | |
1339 | if (bio->bi_phys_segments == 0) | |
1340 | bio->bi_phys_segments = 2; | |
1341 | else | |
1342 | bio->bi_phys_segments++; | |
1343 | spin_unlock_irq(&conf->device_lock); | |
1344 | } | |
4f024f37 KO |
1345 | sectors_handled = r10_bio->sector + max_sectors - |
1346 | bio->bi_iter.bi_sector; | |
d4432c23 | 1347 | |
4e78064f | 1348 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1349 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1350 | |
1da177e4 LT |
1351 | for (i = 0; i < conf->copies; i++) { |
1352 | struct bio *mbio; | |
1353 | int d = r10_bio->devs[i].devnum; | |
e7c0c3fa N |
1354 | if (r10_bio->devs[i].bio) { |
1355 | struct md_rdev *rdev = conf->mirrors[d].rdev; | |
1356 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
4f024f37 | 1357 | bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1358 | max_sectors); |
e7c0c3fa N |
1359 | r10_bio->devs[i].bio = mbio; |
1360 | ||
4f024f37 | 1361 | mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr+ |
e7c0c3fa N |
1362 | choose_data_offset(r10_bio, |
1363 | rdev)); | |
1364 | mbio->bi_bdev = rdev->bdev; | |
1365 | mbio->bi_end_io = raid10_end_write_request; | |
c8dc9c65 JL |
1366 | mbio->bi_rw = |
1367 | WRITE | do_sync | do_fua | do_discard | do_same; | |
e7c0c3fa N |
1368 | mbio->bi_private = r10_bio; |
1369 | ||
1370 | atomic_inc(&r10_bio->remaining); | |
1371 | ||
1372 | cb = blk_check_plugged(raid10_unplug, mddev, | |
1373 | sizeof(*plug)); | |
1374 | if (cb) | |
1375 | plug = container_of(cb, struct raid10_plug_cb, | |
1376 | cb); | |
1377 | else | |
1378 | plug = NULL; | |
1379 | spin_lock_irqsave(&conf->device_lock, flags); | |
1380 | if (plug) { | |
1381 | bio_list_add(&plug->pending, mbio); | |
1382 | plug->pending_cnt++; | |
1383 | } else { | |
1384 | bio_list_add(&conf->pending_bio_list, mbio); | |
1385 | conf->pending_count++; | |
1386 | } | |
1387 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1388 | if (!plug) | |
1389 | md_wakeup_thread(mddev->thread); | |
1390 | } | |
57c67df4 | 1391 | |
e7c0c3fa N |
1392 | if (r10_bio->devs[i].repl_bio) { |
1393 | struct md_rdev *rdev = conf->mirrors[d].replacement; | |
1394 | if (rdev == NULL) { | |
1395 | /* Replacement just got moved to main 'rdev' */ | |
1396 | smp_mb(); | |
1397 | rdev = conf->mirrors[d].rdev; | |
1398 | } | |
1399 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
4f024f37 | 1400 | bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1401 | max_sectors); |
e7c0c3fa N |
1402 | r10_bio->devs[i].repl_bio = mbio; |
1403 | ||
4f024f37 | 1404 | mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr + |
e7c0c3fa N |
1405 | choose_data_offset( |
1406 | r10_bio, rdev)); | |
1407 | mbio->bi_bdev = rdev->bdev; | |
1408 | mbio->bi_end_io = raid10_end_write_request; | |
c8dc9c65 JL |
1409 | mbio->bi_rw = |
1410 | WRITE | do_sync | do_fua | do_discard | do_same; | |
e7c0c3fa N |
1411 | mbio->bi_private = r10_bio; |
1412 | ||
1413 | atomic_inc(&r10_bio->remaining); | |
1414 | spin_lock_irqsave(&conf->device_lock, flags); | |
57c67df4 N |
1415 | bio_list_add(&conf->pending_bio_list, mbio); |
1416 | conf->pending_count++; | |
e7c0c3fa N |
1417 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1418 | if (!mddev_check_plugged(mddev)) | |
1419 | md_wakeup_thread(mddev->thread); | |
57c67df4 | 1420 | } |
1da177e4 LT |
1421 | } |
1422 | ||
079fa166 N |
1423 | /* Don't remove the bias on 'remaining' (one_write_done) until |
1424 | * after checking if we need to go around again. | |
1425 | */ | |
a35e63ef | 1426 | |
aa8b57aa | 1427 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1428 | one_write_done(r10_bio); |
5e570289 | 1429 | /* We need another r10_bio. It has already been counted |
d4432c23 N |
1430 | * in bio->bi_phys_segments. |
1431 | */ | |
1432 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1433 | ||
1434 | r10_bio->master_bio = bio; | |
aa8b57aa | 1435 | r10_bio->sectors = bio_sectors(bio) - sectors_handled; |
d4432c23 N |
1436 | |
1437 | r10_bio->mddev = mddev; | |
4f024f37 | 1438 | r10_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
d4432c23 N |
1439 | r10_bio->state = 0; |
1440 | goto retry_write; | |
1441 | } | |
079fa166 | 1442 | one_write_done(r10_bio); |
20d0189b KO |
1443 | } |
1444 | ||
849674e4 | 1445 | static void raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1446 | { |
1447 | struct r10conf *conf = mddev->private; | |
1448 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1449 | int chunk_sects = chunk_mask + 1; | |
1450 | ||
1451 | struct bio *split; | |
1452 | ||
1453 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { | |
1454 | md_flush_request(mddev, bio); | |
1455 | return; | |
1456 | } | |
1457 | ||
1458 | md_write_start(mddev, bio); | |
1459 | ||
20d0189b KO |
1460 | do { |
1461 | ||
1462 | /* | |
1463 | * If this request crosses a chunk boundary, we need to split | |
1464 | * it. | |
1465 | */ | |
1466 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1467 | bio_sectors(bio) > chunk_sects | |
1468 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1469 | || conf->prev.near_copies < | |
1470 | conf->prev.raid_disks))) { | |
1471 | split = bio_split(bio, chunk_sects - | |
1472 | (bio->bi_iter.bi_sector & | |
1473 | (chunk_sects - 1)), | |
1474 | GFP_NOIO, fs_bio_set); | |
1475 | bio_chain(split, bio); | |
1476 | } else { | |
1477 | split = bio; | |
1478 | } | |
1479 | ||
1480 | __make_request(mddev, split); | |
1481 | } while (split != bio); | |
079fa166 N |
1482 | |
1483 | /* In case raid10d snuck in to freeze_array */ | |
1484 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1485 | } |
1486 | ||
849674e4 | 1487 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1488 | { |
e879a879 | 1489 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1490 | int i; |
1491 | ||
5cf00fcd | 1492 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1493 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1494 | if (conf->geo.near_copies > 1) |
1495 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1496 | if (conf->geo.far_copies > 1) { | |
1497 | if (conf->geo.far_offset) | |
1498 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1499 | else |
5cf00fcd | 1500 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1501 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1502 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1503 | } |
5cf00fcd N |
1504 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1505 | conf->geo.raid_disks - mddev->degraded); | |
1506 | for (i = 0; i < conf->geo.raid_disks; i++) | |
1da177e4 LT |
1507 | seq_printf(seq, "%s", |
1508 | conf->mirrors[i].rdev && | |
b2d444d7 | 1509 | test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
1510 | seq_printf(seq, "]"); |
1511 | } | |
1512 | ||
700c7213 N |
1513 | /* check if there are enough drives for |
1514 | * every block to appear on atleast one. | |
1515 | * Don't consider the device numbered 'ignore' | |
1516 | * as we might be about to remove it. | |
1517 | */ | |
635f6416 | 1518 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1519 | { |
1520 | int first = 0; | |
725d6e57 | 1521 | int has_enough = 0; |
635f6416 N |
1522 | int disks, ncopies; |
1523 | if (previous) { | |
1524 | disks = conf->prev.raid_disks; | |
1525 | ncopies = conf->prev.near_copies; | |
1526 | } else { | |
1527 | disks = conf->geo.raid_disks; | |
1528 | ncopies = conf->geo.near_copies; | |
1529 | } | |
700c7213 | 1530 | |
725d6e57 | 1531 | rcu_read_lock(); |
700c7213 N |
1532 | do { |
1533 | int n = conf->copies; | |
1534 | int cnt = 0; | |
80b48124 | 1535 | int this = first; |
700c7213 | 1536 | while (n--) { |
725d6e57 N |
1537 | struct md_rdev *rdev; |
1538 | if (this != ignore && | |
1539 | (rdev = rcu_dereference(conf->mirrors[this].rdev)) && | |
1540 | test_bit(In_sync, &rdev->flags)) | |
700c7213 | 1541 | cnt++; |
635f6416 | 1542 | this = (this+1) % disks; |
700c7213 N |
1543 | } |
1544 | if (cnt == 0) | |
725d6e57 | 1545 | goto out; |
635f6416 | 1546 | first = (first + ncopies) % disks; |
700c7213 | 1547 | } while (first != 0); |
725d6e57 N |
1548 | has_enough = 1; |
1549 | out: | |
1550 | rcu_read_unlock(); | |
1551 | return has_enough; | |
700c7213 N |
1552 | } |
1553 | ||
f8c9e74f N |
1554 | static int enough(struct r10conf *conf, int ignore) |
1555 | { | |
635f6416 N |
1556 | /* when calling 'enough', both 'prev' and 'geo' must |
1557 | * be stable. | |
1558 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1559 | * is held. | |
1560 | */ | |
1561 | return _enough(conf, 0, ignore) && | |
1562 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1563 | } |
1564 | ||
849674e4 | 1565 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1566 | { |
1567 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1568 | struct r10conf *conf = mddev->private; |
635f6416 | 1569 | unsigned long flags; |
1da177e4 LT |
1570 | |
1571 | /* | |
1572 | * If it is not operational, then we have already marked it as dead | |
1573 | * else if it is the last working disks, ignore the error, let the | |
1574 | * next level up know. | |
1575 | * else mark the drive as failed | |
1576 | */ | |
635f6416 | 1577 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1578 | if (test_bit(In_sync, &rdev->flags) |
635f6416 | 1579 | && !enough(conf, rdev->raid_disk)) { |
1da177e4 LT |
1580 | /* |
1581 | * Don't fail the drive, just return an IO error. | |
1da177e4 | 1582 | */ |
635f6416 | 1583 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1584 | return; |
635f6416 | 1585 | } |
2446dba0 | 1586 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1587 | mddev->degraded++; |
2446dba0 N |
1588 | /* |
1589 | * If recovery is running, make sure it aborts. | |
1590 | */ | |
1591 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
de393cde | 1592 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1593 | set_bit(Faulty, &rdev->flags); |
85ad1d13 GJ |
1594 | set_mask_bits(&mddev->flags, 0, |
1595 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
635f6416 | 1596 | spin_unlock_irqrestore(&conf->device_lock, flags); |
067032bc JP |
1597 | printk(KERN_ALERT |
1598 | "md/raid10:%s: Disk failure on %s, disabling device.\n" | |
1599 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
128595ed | 1600 | mdname(mddev), bdevname(rdev->bdev, b), |
5cf00fcd | 1601 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); |
1da177e4 LT |
1602 | } |
1603 | ||
e879a879 | 1604 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1605 | { |
1606 | int i; | |
dc280d98 | 1607 | struct raid10_info *tmp; |
1da177e4 | 1608 | |
128595ed | 1609 | printk(KERN_DEBUG "RAID10 conf printout:\n"); |
1da177e4 | 1610 | if (!conf) { |
128595ed | 1611 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1612 | return; |
1613 | } | |
5cf00fcd N |
1614 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1615 | conf->geo.raid_disks); | |
1da177e4 | 1616 | |
5cf00fcd | 1617 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 LT |
1618 | char b[BDEVNAME_SIZE]; |
1619 | tmp = conf->mirrors + i; | |
1620 | if (tmp->rdev) | |
128595ed | 1621 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
b2d444d7 N |
1622 | i, !test_bit(In_sync, &tmp->rdev->flags), |
1623 | !test_bit(Faulty, &tmp->rdev->flags), | |
1da177e4 LT |
1624 | bdevname(tmp->rdev->bdev,b)); |
1625 | } | |
1626 | } | |
1627 | ||
e879a879 | 1628 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1629 | { |
0a27ec96 N |
1630 | wait_barrier(conf); |
1631 | allow_barrier(conf); | |
1da177e4 LT |
1632 | |
1633 | mempool_destroy(conf->r10buf_pool); | |
1634 | conf->r10buf_pool = NULL; | |
1635 | } | |
1636 | ||
fd01b88c | 1637 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1638 | { |
1639 | int i; | |
e879a879 | 1640 | struct r10conf *conf = mddev->private; |
dc280d98 | 1641 | struct raid10_info *tmp; |
6b965620 N |
1642 | int count = 0; |
1643 | unsigned long flags; | |
1da177e4 LT |
1644 | |
1645 | /* | |
1646 | * Find all non-in_sync disks within the RAID10 configuration | |
1647 | * and mark them in_sync | |
1648 | */ | |
5cf00fcd | 1649 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1650 | tmp = conf->mirrors + i; |
4ca40c2c N |
1651 | if (tmp->replacement |
1652 | && tmp->replacement->recovery_offset == MaxSector | |
1653 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1654 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1655 | /* Replacement has just become active */ | |
1656 | if (!tmp->rdev | |
1657 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1658 | count++; | |
1659 | if (tmp->rdev) { | |
1660 | /* Replaced device not technically faulty, | |
1661 | * but we need to be sure it gets removed | |
1662 | * and never re-added. | |
1663 | */ | |
1664 | set_bit(Faulty, &tmp->rdev->flags); | |
1665 | sysfs_notify_dirent_safe( | |
1666 | tmp->rdev->sysfs_state); | |
1667 | } | |
1668 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1669 | } else if (tmp->rdev | |
61e4947c | 1670 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
1671 | && !test_bit(Faulty, &tmp->rdev->flags) |
1672 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1673 | count++; |
2863b9eb | 1674 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
1675 | } |
1676 | } | |
6b965620 N |
1677 | spin_lock_irqsave(&conf->device_lock, flags); |
1678 | mddev->degraded -= count; | |
1679 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1680 | |
1681 | print_conf(conf); | |
6b965620 | 1682 | return count; |
1da177e4 LT |
1683 | } |
1684 | ||
fd01b88c | 1685 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1686 | { |
e879a879 | 1687 | struct r10conf *conf = mddev->private; |
199050ea | 1688 | int err = -EEXIST; |
1da177e4 | 1689 | int mirror; |
6c2fce2e | 1690 | int first = 0; |
5cf00fcd | 1691 | int last = conf->geo.raid_disks - 1; |
1da177e4 LT |
1692 | |
1693 | if (mddev->recovery_cp < MaxSector) | |
1694 | /* only hot-add to in-sync arrays, as recovery is | |
1695 | * very different from resync | |
1696 | */ | |
199050ea | 1697 | return -EBUSY; |
635f6416 | 1698 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 1699 | return -EINVAL; |
1da177e4 | 1700 | |
1501efad DW |
1701 | if (md_integrity_add_rdev(rdev, mddev)) |
1702 | return -ENXIO; | |
1703 | ||
a53a6c85 | 1704 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1705 | first = last = rdev->raid_disk; |
1da177e4 | 1706 | |
2c4193df | 1707 | if (rdev->saved_raid_disk >= first && |
6cce3b23 N |
1708 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1709 | mirror = rdev->saved_raid_disk; | |
1710 | else | |
6c2fce2e | 1711 | mirror = first; |
2bb77736 | 1712 | for ( ; mirror <= last ; mirror++) { |
dc280d98 | 1713 | struct raid10_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1714 | if (p->recovery_disabled == mddev->recovery_disabled) |
1715 | continue; | |
b7044d41 N |
1716 | if (p->rdev) { |
1717 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1718 | p->replacement != NULL) | |
1719 | continue; | |
1720 | clear_bit(In_sync, &rdev->flags); | |
1721 | set_bit(Replacement, &rdev->flags); | |
1722 | rdev->raid_disk = mirror; | |
1723 | err = 0; | |
9092c02d JB |
1724 | if (mddev->gendisk) |
1725 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1726 | rdev->data_offset << 9); | |
b7044d41 N |
1727 | conf->fullsync = 1; |
1728 | rcu_assign_pointer(p->replacement, rdev); | |
1729 | break; | |
1730 | } | |
1da177e4 | 1731 | |
9092c02d JB |
1732 | if (mddev->gendisk) |
1733 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1734 | rdev->data_offset << 9); | |
1da177e4 | 1735 | |
2bb77736 | 1736 | p->head_position = 0; |
d890fa2b | 1737 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1738 | rdev->raid_disk = mirror; |
1739 | err = 0; | |
1740 | if (rdev->saved_raid_disk != mirror) | |
1741 | conf->fullsync = 1; | |
1742 | rcu_assign_pointer(p->rdev, rdev); | |
1743 | break; | |
1744 | } | |
ed30be07 | 1745 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
532a2a3f SL |
1746 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1747 | ||
1da177e4 | 1748 | print_conf(conf); |
199050ea | 1749 | return err; |
1da177e4 LT |
1750 | } |
1751 | ||
b8321b68 | 1752 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1753 | { |
e879a879 | 1754 | struct r10conf *conf = mddev->private; |
1da177e4 | 1755 | int err = 0; |
b8321b68 | 1756 | int number = rdev->raid_disk; |
c8ab903e | 1757 | struct md_rdev **rdevp; |
dc280d98 | 1758 | struct raid10_info *p = conf->mirrors + number; |
1da177e4 LT |
1759 | |
1760 | print_conf(conf); | |
c8ab903e N |
1761 | if (rdev == p->rdev) |
1762 | rdevp = &p->rdev; | |
1763 | else if (rdev == p->replacement) | |
1764 | rdevp = &p->replacement; | |
1765 | else | |
1766 | return 0; | |
1767 | ||
1768 | if (test_bit(In_sync, &rdev->flags) || | |
1769 | atomic_read(&rdev->nr_pending)) { | |
1770 | err = -EBUSY; | |
1771 | goto abort; | |
1772 | } | |
1773 | /* Only remove faulty devices if recovery | |
1774 | * is not possible. | |
1775 | */ | |
1776 | if (!test_bit(Faulty, &rdev->flags) && | |
1777 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1778 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 1779 | number < conf->geo.raid_disks && |
c8ab903e N |
1780 | enough(conf, -1)) { |
1781 | err = -EBUSY; | |
1782 | goto abort; | |
1da177e4 | 1783 | } |
c8ab903e N |
1784 | *rdevp = NULL; |
1785 | synchronize_rcu(); | |
1786 | if (atomic_read(&rdev->nr_pending)) { | |
1787 | /* lost the race, try later */ | |
1788 | err = -EBUSY; | |
1789 | *rdevp = rdev; | |
1790 | goto abort; | |
4ca40c2c N |
1791 | } else if (p->replacement) { |
1792 | /* We must have just cleared 'rdev' */ | |
1793 | p->rdev = p->replacement; | |
1794 | clear_bit(Replacement, &p->replacement->flags); | |
1795 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1796 | * but will never see neither -- if they are careful. | |
1797 | */ | |
1798 | p->replacement = NULL; | |
1799 | clear_bit(WantReplacement, &rdev->flags); | |
1800 | } else | |
1801 | /* We might have just remove the Replacement as faulty | |
1802 | * Clear the flag just in case | |
1803 | */ | |
1804 | clear_bit(WantReplacement, &rdev->flags); | |
1805 | ||
c8ab903e N |
1806 | err = md_integrity_register(mddev); |
1807 | ||
1da177e4 LT |
1808 | abort: |
1809 | ||
1810 | print_conf(conf); | |
1811 | return err; | |
1812 | } | |
1813 | ||
4246a0b6 | 1814 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1815 | { |
9f2c9d12 | 1816 | struct r10bio *r10_bio = bio->bi_private; |
e879a879 | 1817 | struct r10conf *conf = r10_bio->mddev->private; |
778ca018 | 1818 | int d; |
1da177e4 | 1819 | |
3ea7daa5 N |
1820 | if (bio == r10_bio->master_bio) { |
1821 | /* this is a reshape read */ | |
1822 | d = r10_bio->read_slot; /* really the read dev */ | |
1823 | } else | |
1824 | d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
0eb3ff12 | 1825 | |
4246a0b6 | 1826 | if (!bio->bi_error) |
0eb3ff12 | 1827 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
1828 | else |
1829 | /* The write handler will notice the lack of | |
1830 | * R10BIO_Uptodate and record any errors etc | |
1831 | */ | |
4dbcdc75 N |
1832 | atomic_add(r10_bio->sectors, |
1833 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1834 | |
1835 | /* for reconstruct, we always reschedule after a read. | |
1836 | * for resync, only after all reads | |
1837 | */ | |
73d5c38a | 1838 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1839 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1840 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1841 | /* we have read all the blocks, | |
1842 | * do the comparison in process context in raid10d | |
1843 | */ | |
1844 | reschedule_retry(r10_bio); | |
1845 | } | |
1da177e4 LT |
1846 | } |
1847 | ||
9f2c9d12 | 1848 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1849 | { |
fd01b88c | 1850 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1851 | |
1da177e4 LT |
1852 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1853 | if (r10_bio->master_bio == NULL) { | |
1854 | /* the primary of several recovery bios */ | |
73d5c38a | 1855 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1856 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1857 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1858 | reschedule_retry(r10_bio); |
1859 | else | |
1860 | put_buf(r10_bio); | |
73d5c38a | 1861 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1862 | break; |
1863 | } else { | |
9f2c9d12 | 1864 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1865 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1866 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1867 | reschedule_retry(r10_bio); |
1868 | else | |
1869 | put_buf(r10_bio); | |
1da177e4 LT |
1870 | r10_bio = r10_bio2; |
1871 | } | |
1872 | } | |
1da177e4 LT |
1873 | } |
1874 | ||
4246a0b6 | 1875 | static void end_sync_write(struct bio *bio) |
5e570289 | 1876 | { |
9f2c9d12 | 1877 | struct r10bio *r10_bio = bio->bi_private; |
fd01b88c | 1878 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1879 | struct r10conf *conf = mddev->private; |
5e570289 N |
1880 | int d; |
1881 | sector_t first_bad; | |
1882 | int bad_sectors; | |
1883 | int slot; | |
9ad1aefc | 1884 | int repl; |
4ca40c2c | 1885 | struct md_rdev *rdev = NULL; |
5e570289 | 1886 | |
9ad1aefc N |
1887 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1888 | if (repl) | |
1889 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 1890 | else |
9ad1aefc | 1891 | rdev = conf->mirrors[d].rdev; |
5e570289 | 1892 | |
4246a0b6 | 1893 | if (bio->bi_error) { |
9ad1aefc N |
1894 | if (repl) |
1895 | md_error(mddev, rdev); | |
1896 | else { | |
1897 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
1898 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1899 | set_bit(MD_RECOVERY_NEEDED, | |
1900 | &rdev->mddev->recovery); | |
9ad1aefc N |
1901 | set_bit(R10BIO_WriteError, &r10_bio->state); |
1902 | } | |
1903 | } else if (is_badblock(rdev, | |
5e570289 N |
1904 | r10_bio->devs[slot].addr, |
1905 | r10_bio->sectors, | |
1906 | &first_bad, &bad_sectors)) | |
1907 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
1908 | ||
9ad1aefc | 1909 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
1910 | |
1911 | end_sync_request(r10_bio); | |
1912 | } | |
1913 | ||
1da177e4 LT |
1914 | /* |
1915 | * Note: sync and recover and handled very differently for raid10 | |
1916 | * This code is for resync. | |
1917 | * For resync, we read through virtual addresses and read all blocks. | |
1918 | * If there is any error, we schedule a write. The lowest numbered | |
1919 | * drive is authoritative. | |
1920 | * However requests come for physical address, so we need to map. | |
1921 | * For every physical address there are raid_disks/copies virtual addresses, | |
1922 | * which is always are least one, but is not necessarly an integer. | |
1923 | * This means that a physical address can span multiple chunks, so we may | |
1924 | * have to submit multiple io requests for a single sync request. | |
1925 | */ | |
1926 | /* | |
1927 | * We check if all blocks are in-sync and only write to blocks that | |
1928 | * aren't in sync | |
1929 | */ | |
9f2c9d12 | 1930 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1931 | { |
e879a879 | 1932 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1933 | int i, first; |
1934 | struct bio *tbio, *fbio; | |
f4380a91 | 1935 | int vcnt; |
1da177e4 LT |
1936 | |
1937 | atomic_set(&r10_bio->remaining, 1); | |
1938 | ||
1939 | /* find the first device with a block */ | |
1940 | for (i=0; i<conf->copies; i++) | |
4246a0b6 | 1941 | if (!r10_bio->devs[i].bio->bi_error) |
1da177e4 LT |
1942 | break; |
1943 | ||
1944 | if (i == conf->copies) | |
1945 | goto done; | |
1946 | ||
1947 | first = i; | |
1948 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
1949 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
1950 | fbio->bi_iter.bi_idx = 0; | |
1da177e4 | 1951 | |
f4380a91 | 1952 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 1953 | /* now find blocks with errors */ |
0eb3ff12 N |
1954 | for (i=0 ; i < conf->copies ; i++) { |
1955 | int j, d; | |
1da177e4 | 1956 | |
1da177e4 | 1957 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
1958 | |
1959 | if (tbio->bi_end_io != end_sync_read) | |
1960 | continue; | |
1961 | if (i == first) | |
1da177e4 | 1962 | continue; |
4246a0b6 | 1963 | if (!r10_bio->devs[i].bio->bi_error) { |
0eb3ff12 N |
1964 | /* We know that the bi_io_vec layout is the same for |
1965 | * both 'first' and 'i', so we just compare them. | |
1966 | * All vec entries are PAGE_SIZE; | |
1967 | */ | |
7bb23c49 N |
1968 | int sectors = r10_bio->sectors; |
1969 | for (j = 0; j < vcnt; j++) { | |
1970 | int len = PAGE_SIZE; | |
1971 | if (sectors < (len / 512)) | |
1972 | len = sectors * 512; | |
0eb3ff12 N |
1973 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), |
1974 | page_address(tbio->bi_io_vec[j].bv_page), | |
7bb23c49 | 1975 | len)) |
0eb3ff12 | 1976 | break; |
7bb23c49 N |
1977 | sectors -= len/512; |
1978 | } | |
0eb3ff12 N |
1979 | if (j == vcnt) |
1980 | continue; | |
7f7583d4 | 1981 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
1982 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
1983 | /* Don't fix anything. */ | |
1984 | continue; | |
0eb3ff12 | 1985 | } |
f84ee364 N |
1986 | /* Ok, we need to write this bio, either to correct an |
1987 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
1988 | * First we need to fixup bv_offset, bv_len and |
1989 | * bi_vecs, as the read request might have corrupted these | |
1990 | */ | |
8be185f2 KO |
1991 | bio_reset(tbio); |
1992 | ||
1da177e4 | 1993 | tbio->bi_vcnt = vcnt; |
cc578588 | 1994 | tbio->bi_iter.bi_size = fbio->bi_iter.bi_size; |
1da177e4 LT |
1995 | tbio->bi_rw = WRITE; |
1996 | tbio->bi_private = r10_bio; | |
4f024f37 | 1997 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 LT |
1998 | tbio->bi_end_io = end_sync_write; |
1999 | ||
c31df25f KO |
2000 | bio_copy_data(tbio, fbio); |
2001 | ||
1da177e4 LT |
2002 | d = r10_bio->devs[i].devnum; |
2003 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
2004 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2005 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2006 | |
4f024f37 | 2007 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
1da177e4 LT |
2008 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; |
2009 | generic_make_request(tbio); | |
2010 | } | |
2011 | ||
9ad1aefc N |
2012 | /* Now write out to any replacement devices |
2013 | * that are active | |
2014 | */ | |
2015 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2016 | int d; |
9ad1aefc N |
2017 | |
2018 | tbio = r10_bio->devs[i].repl_bio; | |
2019 | if (!tbio || !tbio->bi_end_io) | |
2020 | continue; | |
2021 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2022 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2023 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2024 | d = r10_bio->devs[i].devnum; |
2025 | atomic_inc(&r10_bio->remaining); | |
2026 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2027 | bio_sectors(tbio)); |
9ad1aefc N |
2028 | generic_make_request(tbio); |
2029 | } | |
2030 | ||
1da177e4 LT |
2031 | done: |
2032 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2033 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2034 | put_buf(r10_bio); | |
2035 | } | |
2036 | } | |
2037 | ||
2038 | /* | |
2039 | * Now for the recovery code. | |
2040 | * Recovery happens across physical sectors. | |
2041 | * We recover all non-is_sync drives by finding the virtual address of | |
2042 | * each, and then choose a working drive that also has that virt address. | |
2043 | * There is a separate r10_bio for each non-in_sync drive. | |
2044 | * Only the first two slots are in use. The first for reading, | |
2045 | * The second for writing. | |
2046 | * | |
2047 | */ | |
9f2c9d12 | 2048 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2049 | { |
2050 | /* We got a read error during recovery. | |
2051 | * We repeat the read in smaller page-sized sections. | |
2052 | * If a read succeeds, write it to the new device or record | |
2053 | * a bad block if we cannot. | |
2054 | * If a read fails, record a bad block on both old and | |
2055 | * new devices. | |
2056 | */ | |
fd01b88c | 2057 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2058 | struct r10conf *conf = mddev->private; |
5e570289 N |
2059 | struct bio *bio = r10_bio->devs[0].bio; |
2060 | sector_t sect = 0; | |
2061 | int sectors = r10_bio->sectors; | |
2062 | int idx = 0; | |
2063 | int dr = r10_bio->devs[0].devnum; | |
2064 | int dw = r10_bio->devs[1].devnum; | |
2065 | ||
2066 | while (sectors) { | |
2067 | int s = sectors; | |
3cb03002 | 2068 | struct md_rdev *rdev; |
5e570289 N |
2069 | sector_t addr; |
2070 | int ok; | |
2071 | ||
2072 | if (s > (PAGE_SIZE>>9)) | |
2073 | s = PAGE_SIZE >> 9; | |
2074 | ||
2075 | rdev = conf->mirrors[dr].rdev; | |
2076 | addr = r10_bio->devs[0].addr + sect, | |
2077 | ok = sync_page_io(rdev, | |
2078 | addr, | |
2079 | s << 9, | |
2080 | bio->bi_io_vec[idx].bv_page, | |
2081 | READ, false); | |
2082 | if (ok) { | |
2083 | rdev = conf->mirrors[dw].rdev; | |
2084 | addr = r10_bio->devs[1].addr + sect; | |
2085 | ok = sync_page_io(rdev, | |
2086 | addr, | |
2087 | s << 9, | |
2088 | bio->bi_io_vec[idx].bv_page, | |
2089 | WRITE, false); | |
b7044d41 | 2090 | if (!ok) { |
5e570289 | 2091 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2092 | if (!test_and_set_bit(WantReplacement, |
2093 | &rdev->flags)) | |
2094 | set_bit(MD_RECOVERY_NEEDED, | |
2095 | &rdev->mddev->recovery); | |
2096 | } | |
5e570289 N |
2097 | } |
2098 | if (!ok) { | |
2099 | /* We don't worry if we cannot set a bad block - | |
2100 | * it really is bad so there is no loss in not | |
2101 | * recording it yet | |
2102 | */ | |
2103 | rdev_set_badblocks(rdev, addr, s, 0); | |
2104 | ||
2105 | if (rdev != conf->mirrors[dw].rdev) { | |
2106 | /* need bad block on destination too */ | |
3cb03002 | 2107 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2108 | addr = r10_bio->devs[1].addr + sect; |
2109 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2110 | if (!ok) { | |
2111 | /* just abort the recovery */ | |
2112 | printk(KERN_NOTICE | |
2113 | "md/raid10:%s: recovery aborted" | |
2114 | " due to read error\n", | |
2115 | mdname(mddev)); | |
2116 | ||
2117 | conf->mirrors[dw].recovery_disabled | |
2118 | = mddev->recovery_disabled; | |
2119 | set_bit(MD_RECOVERY_INTR, | |
2120 | &mddev->recovery); | |
2121 | break; | |
2122 | } | |
2123 | } | |
2124 | } | |
2125 | ||
2126 | sectors -= s; | |
2127 | sect += s; | |
2128 | idx++; | |
2129 | } | |
2130 | } | |
1da177e4 | 2131 | |
9f2c9d12 | 2132 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2133 | { |
e879a879 | 2134 | struct r10conf *conf = mddev->private; |
c65060ad | 2135 | int d; |
24afd80d | 2136 | struct bio *wbio, *wbio2; |
1da177e4 | 2137 | |
5e570289 N |
2138 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2139 | fix_recovery_read_error(r10_bio); | |
2140 | end_sync_request(r10_bio); | |
2141 | return; | |
2142 | } | |
2143 | ||
c65060ad NK |
2144 | /* |
2145 | * share the pages with the first bio | |
1da177e4 LT |
2146 | * and submit the write request |
2147 | */ | |
1da177e4 | 2148 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2149 | wbio = r10_bio->devs[1].bio; |
2150 | wbio2 = r10_bio->devs[1].repl_bio; | |
0eb25bb0 N |
2151 | /* Need to test wbio2->bi_end_io before we call |
2152 | * generic_make_request as if the former is NULL, | |
2153 | * the latter is free to free wbio2. | |
2154 | */ | |
2155 | if (wbio2 && !wbio2->bi_end_io) | |
2156 | wbio2 = NULL; | |
24afd80d N |
2157 | if (wbio->bi_end_io) { |
2158 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2159 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
24afd80d N |
2160 | generic_make_request(wbio); |
2161 | } | |
0eb25bb0 | 2162 | if (wbio2) { |
24afd80d N |
2163 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2164 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2165 | bio_sectors(wbio2)); |
24afd80d N |
2166 | generic_make_request(wbio2); |
2167 | } | |
1da177e4 LT |
2168 | } |
2169 | ||
1e50915f RB |
2170 | /* |
2171 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2172 | * We halve the read error count for every hour that has elapsed | |
2173 | * since the last recorded read error. | |
2174 | * | |
2175 | */ | |
fd01b88c | 2176 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f RB |
2177 | { |
2178 | struct timespec cur_time_mon; | |
2179 | unsigned long hours_since_last; | |
2180 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2181 | ||
2182 | ktime_get_ts(&cur_time_mon); | |
2183 | ||
2184 | if (rdev->last_read_error.tv_sec == 0 && | |
2185 | rdev->last_read_error.tv_nsec == 0) { | |
2186 | /* first time we've seen a read error */ | |
2187 | rdev->last_read_error = cur_time_mon; | |
2188 | return; | |
2189 | } | |
2190 | ||
2191 | hours_since_last = (cur_time_mon.tv_sec - | |
2192 | rdev->last_read_error.tv_sec) / 3600; | |
2193 | ||
2194 | rdev->last_read_error = cur_time_mon; | |
2195 | ||
2196 | /* | |
2197 | * if hours_since_last is > the number of bits in read_errors | |
2198 | * just set read errors to 0. We do this to avoid | |
2199 | * overflowing the shift of read_errors by hours_since_last. | |
2200 | */ | |
2201 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2202 | atomic_set(&rdev->read_errors, 0); | |
2203 | else | |
2204 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2205 | } | |
2206 | ||
3cb03002 | 2207 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2208 | int sectors, struct page *page, int rw) |
2209 | { | |
2210 | sector_t first_bad; | |
2211 | int bad_sectors; | |
2212 | ||
2213 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2214 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2215 | return -1; | |
2216 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
2217 | /* success */ | |
2218 | return 1; | |
b7044d41 | 2219 | if (rw == WRITE) { |
58c54fcc | 2220 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2221 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2222 | set_bit(MD_RECOVERY_NEEDED, | |
2223 | &rdev->mddev->recovery); | |
2224 | } | |
58c54fcc N |
2225 | /* need to record an error - either for the block or the device */ |
2226 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2227 | md_error(rdev->mddev, rdev); | |
2228 | return 0; | |
2229 | } | |
2230 | ||
1da177e4 LT |
2231 | /* |
2232 | * This is a kernel thread which: | |
2233 | * | |
2234 | * 1. Retries failed read operations on working mirrors. | |
2235 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2236 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2237 | */ |
2238 | ||
e879a879 | 2239 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2240 | { |
2241 | int sect = 0; /* Offset from r10_bio->sector */ | |
2242 | int sectors = r10_bio->sectors; | |
3cb03002 | 2243 | struct md_rdev*rdev; |
1e50915f | 2244 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2245 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2246 | |
7c4e06ff N |
2247 | /* still own a reference to this rdev, so it cannot |
2248 | * have been cleared recently. | |
2249 | */ | |
2250 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2251 | |
7c4e06ff N |
2252 | if (test_bit(Faulty, &rdev->flags)) |
2253 | /* drive has already been failed, just ignore any | |
2254 | more fix_read_error() attempts */ | |
2255 | return; | |
1e50915f | 2256 | |
7c4e06ff N |
2257 | check_decay_read_errors(mddev, rdev); |
2258 | atomic_inc(&rdev->read_errors); | |
2259 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2260 | char b[BDEVNAME_SIZE]; | |
2261 | bdevname(rdev->bdev, b); | |
1e50915f | 2262 | |
7c4e06ff N |
2263 | printk(KERN_NOTICE |
2264 | "md/raid10:%s: %s: Raid device exceeded " | |
2265 | "read_error threshold [cur %d:max %d]\n", | |
2266 | mdname(mddev), b, | |
2267 | atomic_read(&rdev->read_errors), max_read_errors); | |
2268 | printk(KERN_NOTICE | |
2269 | "md/raid10:%s: %s: Failing raid device\n", | |
2270 | mdname(mddev), b); | |
2271 | md_error(mddev, conf->mirrors[d].rdev); | |
fae8cc5e | 2272 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2273 | return; |
1e50915f | 2274 | } |
1e50915f | 2275 | |
6814d536 N |
2276 | while(sectors) { |
2277 | int s = sectors; | |
2278 | int sl = r10_bio->read_slot; | |
2279 | int success = 0; | |
2280 | int start; | |
2281 | ||
2282 | if (s > (PAGE_SIZE>>9)) | |
2283 | s = PAGE_SIZE >> 9; | |
2284 | ||
2285 | rcu_read_lock(); | |
2286 | do { | |
8dbed5ce N |
2287 | sector_t first_bad; |
2288 | int bad_sectors; | |
2289 | ||
0544a21d | 2290 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2291 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2292 | if (rdev && | |
8dbed5ce N |
2293 | test_bit(In_sync, &rdev->flags) && |
2294 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, | |
2295 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2296 | atomic_inc(&rdev->nr_pending); |
2297 | rcu_read_unlock(); | |
2b193363 | 2298 | success = sync_page_io(rdev, |
6814d536 | 2299 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2300 | sect, |
6814d536 | 2301 | s<<9, |
ccebd4c4 | 2302 | conf->tmppage, READ, false); |
6814d536 N |
2303 | rdev_dec_pending(rdev, mddev); |
2304 | rcu_read_lock(); | |
2305 | if (success) | |
2306 | break; | |
2307 | } | |
2308 | sl++; | |
2309 | if (sl == conf->copies) | |
2310 | sl = 0; | |
2311 | } while (!success && sl != r10_bio->read_slot); | |
2312 | rcu_read_unlock(); | |
2313 | ||
2314 | if (!success) { | |
58c54fcc N |
2315 | /* Cannot read from anywhere, just mark the block |
2316 | * as bad on the first device to discourage future | |
2317 | * reads. | |
2318 | */ | |
6814d536 | 2319 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2320 | rdev = conf->mirrors[dn].rdev; |
2321 | ||
2322 | if (!rdev_set_badblocks( | |
2323 | rdev, | |
2324 | r10_bio->devs[r10_bio->read_slot].addr | |
2325 | + sect, | |
fae8cc5e | 2326 | s, 0)) { |
58c54fcc | 2327 | md_error(mddev, rdev); |
fae8cc5e N |
2328 | r10_bio->devs[r10_bio->read_slot].bio |
2329 | = IO_BLOCKED; | |
2330 | } | |
6814d536 N |
2331 | break; |
2332 | } | |
2333 | ||
2334 | start = sl; | |
2335 | /* write it back and re-read */ | |
2336 | rcu_read_lock(); | |
2337 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2338 | char b[BDEVNAME_SIZE]; |
0544a21d | 2339 | |
6814d536 N |
2340 | if (sl==0) |
2341 | sl = conf->copies; | |
2342 | sl--; | |
2343 | d = r10_bio->devs[sl].devnum; | |
2344 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 N |
2345 | if (!rdev || |
2346 | !test_bit(In_sync, &rdev->flags)) | |
2347 | continue; | |
2348 | ||
2349 | atomic_inc(&rdev->nr_pending); | |
2350 | rcu_read_unlock(); | |
58c54fcc N |
2351 | if (r10_sync_page_io(rdev, |
2352 | r10_bio->devs[sl].addr + | |
2353 | sect, | |
055d3747 | 2354 | s, conf->tmppage, WRITE) |
1294b9c9 N |
2355 | == 0) { |
2356 | /* Well, this device is dead */ | |
2357 | printk(KERN_NOTICE | |
2358 | "md/raid10:%s: read correction " | |
2359 | "write failed" | |
2360 | " (%d sectors at %llu on %s)\n", | |
2361 | mdname(mddev), s, | |
2362 | (unsigned long long)( | |
f8c9e74f N |
2363 | sect + |
2364 | choose_data_offset(r10_bio, | |
2365 | rdev)), | |
1294b9c9 N |
2366 | bdevname(rdev->bdev, b)); |
2367 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
2368 | "drive\n", | |
2369 | mdname(mddev), | |
2370 | bdevname(rdev->bdev, b)); | |
6814d536 | 2371 | } |
1294b9c9 N |
2372 | rdev_dec_pending(rdev, mddev); |
2373 | rcu_read_lock(); | |
6814d536 N |
2374 | } |
2375 | sl = start; | |
2376 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2377 | char b[BDEVNAME_SIZE]; |
0544a21d | 2378 | |
6814d536 N |
2379 | if (sl==0) |
2380 | sl = conf->copies; | |
2381 | sl--; | |
2382 | d = r10_bio->devs[sl].devnum; | |
2383 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 N |
2384 | if (!rdev || |
2385 | !test_bit(In_sync, &rdev->flags)) | |
2386 | continue; | |
6814d536 | 2387 | |
1294b9c9 N |
2388 | atomic_inc(&rdev->nr_pending); |
2389 | rcu_read_unlock(); | |
58c54fcc N |
2390 | switch (r10_sync_page_io(rdev, |
2391 | r10_bio->devs[sl].addr + | |
2392 | sect, | |
055d3747 | 2393 | s, conf->tmppage, |
58c54fcc N |
2394 | READ)) { |
2395 | case 0: | |
1294b9c9 N |
2396 | /* Well, this device is dead */ |
2397 | printk(KERN_NOTICE | |
2398 | "md/raid10:%s: unable to read back " | |
2399 | "corrected sectors" | |
2400 | " (%d sectors at %llu on %s)\n", | |
2401 | mdname(mddev), s, | |
2402 | (unsigned long long)( | |
f8c9e74f N |
2403 | sect + |
2404 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2405 | bdevname(rdev->bdev, b)); |
2406 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
2407 | "drive\n", | |
2408 | mdname(mddev), | |
2409 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2410 | break; |
2411 | case 1: | |
1294b9c9 N |
2412 | printk(KERN_INFO |
2413 | "md/raid10:%s: read error corrected" | |
2414 | " (%d sectors at %llu on %s)\n", | |
2415 | mdname(mddev), s, | |
2416 | (unsigned long long)( | |
f8c9e74f N |
2417 | sect + |
2418 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2419 | bdevname(rdev->bdev, b)); |
2420 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2421 | } |
1294b9c9 N |
2422 | |
2423 | rdev_dec_pending(rdev, mddev); | |
2424 | rcu_read_lock(); | |
6814d536 N |
2425 | } |
2426 | rcu_read_unlock(); | |
2427 | ||
2428 | sectors -= s; | |
2429 | sect += s; | |
2430 | } | |
2431 | } | |
2432 | ||
9f2c9d12 | 2433 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2434 | { |
2435 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2436 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2437 | struct r10conf *conf = mddev->private; |
3cb03002 | 2438 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2439 | /* bio has the data to be written to slot 'i' where |
2440 | * we just recently had a write error. | |
2441 | * We repeatedly clone the bio and trim down to one block, | |
2442 | * then try the write. Where the write fails we record | |
2443 | * a bad block. | |
2444 | * It is conceivable that the bio doesn't exactly align with | |
2445 | * blocks. We must handle this. | |
2446 | * | |
2447 | * We currently own a reference to the rdev. | |
2448 | */ | |
2449 | ||
2450 | int block_sectors; | |
2451 | sector_t sector; | |
2452 | int sectors; | |
2453 | int sect_to_write = r10_bio->sectors; | |
2454 | int ok = 1; | |
2455 | ||
2456 | if (rdev->badblocks.shift < 0) | |
2457 | return 0; | |
2458 | ||
f04ebb0b N |
2459 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2460 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2461 | sector = r10_bio->sector; |
2462 | sectors = ((r10_bio->sector + block_sectors) | |
2463 | & ~(sector_t)(block_sectors - 1)) | |
2464 | - sector; | |
2465 | ||
2466 | while (sect_to_write) { | |
2467 | struct bio *wbio; | |
2468 | if (sectors > sect_to_write) | |
2469 | sectors = sect_to_write; | |
2470 | /* Write at 'sector' for 'sectors' */ | |
2471 | wbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
4f024f37 KO |
2472 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
2473 | wbio->bi_iter.bi_sector = (r10_bio->devs[i].addr+ | |
f8c9e74f | 2474 | choose_data_offset(r10_bio, rdev) + |
bd870a16 N |
2475 | (sector - r10_bio->sector)); |
2476 | wbio->bi_bdev = rdev->bdev; | |
4e49ea4a MC |
2477 | wbio->bi_rw = WRITE; |
2478 | ||
2479 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 N |
2480 | /* Failure! */ |
2481 | ok = rdev_set_badblocks(rdev, sector, | |
2482 | sectors, 0) | |
2483 | && ok; | |
2484 | ||
2485 | bio_put(wbio); | |
2486 | sect_to_write -= sectors; | |
2487 | sector += sectors; | |
2488 | sectors = block_sectors; | |
2489 | } | |
2490 | return ok; | |
2491 | } | |
2492 | ||
9f2c9d12 | 2493 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2494 | { |
2495 | int slot = r10_bio->read_slot; | |
560f8e55 | 2496 | struct bio *bio; |
e879a879 | 2497 | struct r10conf *conf = mddev->private; |
abbf098e | 2498 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2499 | char b[BDEVNAME_SIZE]; |
2500 | unsigned long do_sync; | |
856e08e2 | 2501 | int max_sectors; |
560f8e55 N |
2502 | |
2503 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2504 | * the block and we can fix it. | |
2505 | * We freeze all other IO, and try reading the block from | |
2506 | * other devices. When we find one, we re-write | |
2507 | * and check it that fixes the read error. | |
2508 | * This is all done synchronously while the array is | |
2509 | * frozen. | |
2510 | */ | |
fae8cc5e N |
2511 | bio = r10_bio->devs[slot].bio; |
2512 | bdevname(bio->bi_bdev, b); | |
2513 | bio_put(bio); | |
2514 | r10_bio->devs[slot].bio = NULL; | |
2515 | ||
560f8e55 | 2516 | if (mddev->ro == 0) { |
e2d59925 | 2517 | freeze_array(conf, 1); |
560f8e55 N |
2518 | fix_read_error(conf, mddev, r10_bio); |
2519 | unfreeze_array(conf); | |
fae8cc5e N |
2520 | } else |
2521 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2522 | ||
abbf098e | 2523 | rdev_dec_pending(rdev, mddev); |
560f8e55 | 2524 | |
7399c31b | 2525 | read_more: |
96c3fd1f N |
2526 | rdev = read_balance(conf, r10_bio, &max_sectors); |
2527 | if (rdev == NULL) { | |
560f8e55 N |
2528 | printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O" |
2529 | " read error for block %llu\n", | |
7399c31b | 2530 | mdname(mddev), b, |
560f8e55 N |
2531 | (unsigned long long)r10_bio->sector); |
2532 | raid_end_bio_io(r10_bio); | |
560f8e55 N |
2533 | return; |
2534 | } | |
2535 | ||
2536 | do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC); | |
560f8e55 | 2537 | slot = r10_bio->read_slot; |
560f8e55 N |
2538 | printk_ratelimited( |
2539 | KERN_ERR | |
055d3747 | 2540 | "md/raid10:%s: %s: redirecting " |
560f8e55 N |
2541 | "sector %llu to another mirror\n", |
2542 | mdname(mddev), | |
2543 | bdevname(rdev->bdev, b), | |
2544 | (unsigned long long)r10_bio->sector); | |
2545 | bio = bio_clone_mddev(r10_bio->master_bio, | |
2546 | GFP_NOIO, mddev); | |
4f024f37 | 2547 | bio_trim(bio, r10_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
560f8e55 | 2548 | r10_bio->devs[slot].bio = bio; |
abbf098e | 2549 | r10_bio->devs[slot].rdev = rdev; |
4f024f37 | 2550 | bio->bi_iter.bi_sector = r10_bio->devs[slot].addr |
f8c9e74f | 2551 | + choose_data_offset(r10_bio, rdev); |
560f8e55 N |
2552 | bio->bi_bdev = rdev->bdev; |
2553 | bio->bi_rw = READ | do_sync; | |
2554 | bio->bi_private = r10_bio; | |
2555 | bio->bi_end_io = raid10_end_read_request; | |
7399c31b N |
2556 | if (max_sectors < r10_bio->sectors) { |
2557 | /* Drat - have to split this up more */ | |
2558 | struct bio *mbio = r10_bio->master_bio; | |
2559 | int sectors_handled = | |
2560 | r10_bio->sector + max_sectors | |
4f024f37 | 2561 | - mbio->bi_iter.bi_sector; |
7399c31b N |
2562 | r10_bio->sectors = max_sectors; |
2563 | spin_lock_irq(&conf->device_lock); | |
2564 | if (mbio->bi_phys_segments == 0) | |
2565 | mbio->bi_phys_segments = 2; | |
2566 | else | |
2567 | mbio->bi_phys_segments++; | |
2568 | spin_unlock_irq(&conf->device_lock); | |
2569 | generic_make_request(bio); | |
7399c31b N |
2570 | |
2571 | r10_bio = mempool_alloc(conf->r10bio_pool, | |
2572 | GFP_NOIO); | |
2573 | r10_bio->master_bio = mbio; | |
aa8b57aa | 2574 | r10_bio->sectors = bio_sectors(mbio) - sectors_handled; |
7399c31b N |
2575 | r10_bio->state = 0; |
2576 | set_bit(R10BIO_ReadError, | |
2577 | &r10_bio->state); | |
2578 | r10_bio->mddev = mddev; | |
4f024f37 | 2579 | r10_bio->sector = mbio->bi_iter.bi_sector |
7399c31b N |
2580 | + sectors_handled; |
2581 | ||
2582 | goto read_more; | |
2583 | } else | |
2584 | generic_make_request(bio); | |
560f8e55 N |
2585 | } |
2586 | ||
e879a879 | 2587 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2588 | { |
2589 | /* Some sort of write request has finished and it | |
2590 | * succeeded in writing where we thought there was a | |
2591 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2592 | * Or possibly if failed and we need to record |
2593 | * a bad block. | |
749c55e9 N |
2594 | */ |
2595 | int m; | |
3cb03002 | 2596 | struct md_rdev *rdev; |
749c55e9 N |
2597 | |
2598 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2599 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2600 | for (m = 0; m < conf->copies; m++) { |
2601 | int dev = r10_bio->devs[m].devnum; | |
2602 | rdev = conf->mirrors[dev].rdev; | |
2603 | if (r10_bio->devs[m].bio == NULL) | |
2604 | continue; | |
4246a0b6 | 2605 | if (!r10_bio->devs[m].bio->bi_error) { |
749c55e9 N |
2606 | rdev_clear_badblocks( |
2607 | rdev, | |
2608 | r10_bio->devs[m].addr, | |
c6563a8c | 2609 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2610 | } else { |
2611 | if (!rdev_set_badblocks( | |
2612 | rdev, | |
2613 | r10_bio->devs[m].addr, | |
2614 | r10_bio->sectors, 0)) | |
2615 | md_error(conf->mddev, rdev); | |
749c55e9 | 2616 | } |
9ad1aefc N |
2617 | rdev = conf->mirrors[dev].replacement; |
2618 | if (r10_bio->devs[m].repl_bio == NULL) | |
2619 | continue; | |
4246a0b6 CH |
2620 | |
2621 | if (!r10_bio->devs[m].repl_bio->bi_error) { | |
9ad1aefc N |
2622 | rdev_clear_badblocks( |
2623 | rdev, | |
2624 | r10_bio->devs[m].addr, | |
c6563a8c | 2625 | r10_bio->sectors, 0); |
9ad1aefc N |
2626 | } else { |
2627 | if (!rdev_set_badblocks( | |
2628 | rdev, | |
2629 | r10_bio->devs[m].addr, | |
2630 | r10_bio->sectors, 0)) | |
2631 | md_error(conf->mddev, rdev); | |
2632 | } | |
1a0b7cd8 | 2633 | } |
749c55e9 N |
2634 | put_buf(r10_bio); |
2635 | } else { | |
95af587e | 2636 | bool fail = false; |
bd870a16 N |
2637 | for (m = 0; m < conf->copies; m++) { |
2638 | int dev = r10_bio->devs[m].devnum; | |
2639 | struct bio *bio = r10_bio->devs[m].bio; | |
2640 | rdev = conf->mirrors[dev].rdev; | |
2641 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2642 | rdev_clear_badblocks( |
2643 | rdev, | |
2644 | r10_bio->devs[m].addr, | |
c6563a8c | 2645 | r10_bio->sectors, 0); |
749c55e9 | 2646 | rdev_dec_pending(rdev, conf->mddev); |
4246a0b6 | 2647 | } else if (bio != NULL && bio->bi_error) { |
95af587e | 2648 | fail = true; |
bd870a16 N |
2649 | if (!narrow_write_error(r10_bio, m)) { |
2650 | md_error(conf->mddev, rdev); | |
2651 | set_bit(R10BIO_Degraded, | |
2652 | &r10_bio->state); | |
2653 | } | |
2654 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2655 | } |
475b0321 N |
2656 | bio = r10_bio->devs[m].repl_bio; |
2657 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2658 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2659 | rdev_clear_badblocks( |
2660 | rdev, | |
2661 | r10_bio->devs[m].addr, | |
c6563a8c | 2662 | r10_bio->sectors, 0); |
475b0321 N |
2663 | rdev_dec_pending(rdev, conf->mddev); |
2664 | } | |
bd870a16 | 2665 | } |
95af587e N |
2666 | if (fail) { |
2667 | spin_lock_irq(&conf->device_lock); | |
2668 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 2669 | conf->nr_queued++; |
95af587e N |
2670 | spin_unlock_irq(&conf->device_lock); |
2671 | md_wakeup_thread(conf->mddev->thread); | |
c340702c N |
2672 | } else { |
2673 | if (test_bit(R10BIO_WriteError, | |
2674 | &r10_bio->state)) | |
2675 | close_write(r10_bio); | |
95af587e | 2676 | raid_end_bio_io(r10_bio); |
c340702c | 2677 | } |
749c55e9 N |
2678 | } |
2679 | } | |
2680 | ||
4ed8731d | 2681 | static void raid10d(struct md_thread *thread) |
1da177e4 | 2682 | { |
4ed8731d | 2683 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2684 | struct r10bio *r10_bio; |
1da177e4 | 2685 | unsigned long flags; |
e879a879 | 2686 | struct r10conf *conf = mddev->private; |
1da177e4 | 2687 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2688 | struct blk_plug plug; |
1da177e4 LT |
2689 | |
2690 | md_check_recovery(mddev); | |
1da177e4 | 2691 | |
95af587e N |
2692 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2693 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
2694 | LIST_HEAD(tmp); | |
2695 | spin_lock_irqsave(&conf->device_lock, flags); | |
2696 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
23ddba80 SL |
2697 | while (!list_empty(&conf->bio_end_io_list)) { |
2698 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2699 | conf->nr_queued--; | |
2700 | } | |
95af587e N |
2701 | } |
2702 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2703 | while (!list_empty(&tmp)) { | |
a452744b MP |
2704 | r10_bio = list_first_entry(&tmp, struct r10bio, |
2705 | retry_list); | |
95af587e | 2706 | list_del(&r10_bio->retry_list); |
c340702c N |
2707 | if (mddev->degraded) |
2708 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
2709 | ||
2710 | if (test_bit(R10BIO_WriteError, | |
2711 | &r10_bio->state)) | |
2712 | close_write(r10_bio); | |
95af587e N |
2713 | raid_end_bio_io(r10_bio); |
2714 | } | |
2715 | } | |
2716 | ||
e1dfa0a2 | 2717 | blk_start_plug(&plug); |
1da177e4 | 2718 | for (;;) { |
6cce3b23 | 2719 | |
0021b7bc | 2720 | flush_pending_writes(conf); |
6cce3b23 | 2721 | |
a35e63ef N |
2722 | spin_lock_irqsave(&conf->device_lock, flags); |
2723 | if (list_empty(head)) { | |
2724 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2725 | break; |
a35e63ef | 2726 | } |
9f2c9d12 | 2727 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2728 | list_del(head->prev); |
4443ae10 | 2729 | conf->nr_queued--; |
1da177e4 LT |
2730 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2731 | ||
2732 | mddev = r10_bio->mddev; | |
070ec55d | 2733 | conf = mddev->private; |
bd870a16 N |
2734 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2735 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 2736 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
2737 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
2738 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 2739 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 2740 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2741 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2742 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2743 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2744 | handle_read_error(mddev, r10_bio); |
856e08e2 N |
2745 | else { |
2746 | /* just a partial read to be scheduled from a | |
2747 | * separate context | |
2748 | */ | |
2749 | int slot = r10_bio->read_slot; | |
2750 | generic_make_request(r10_bio->devs[slot].bio); | |
2751 | } | |
560f8e55 | 2752 | |
1d9d5241 | 2753 | cond_resched(); |
de393cde N |
2754 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2755 | md_check_recovery(mddev); | |
1da177e4 | 2756 | } |
e1dfa0a2 | 2757 | blk_finish_plug(&plug); |
1da177e4 LT |
2758 | } |
2759 | ||
e879a879 | 2760 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2761 | { |
2762 | int buffs; | |
69335ef3 | 2763 | int i; |
1da177e4 LT |
2764 | |
2765 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2766 | BUG_ON(conf->r10buf_pool); |
69335ef3 | 2767 | conf->have_replacement = 0; |
5cf00fcd | 2768 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
2769 | if (conf->mirrors[i].replacement) |
2770 | conf->have_replacement = 1; | |
1da177e4 LT |
2771 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2772 | if (!conf->r10buf_pool) | |
2773 | return -ENOMEM; | |
2774 | conf->next_resync = 0; | |
2775 | return 0; | |
2776 | } | |
2777 | ||
2778 | /* | |
2779 | * perform a "sync" on one "block" | |
2780 | * | |
2781 | * We need to make sure that no normal I/O request - particularly write | |
2782 | * requests - conflict with active sync requests. | |
2783 | * | |
2784 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2785 | * that can be installed to exclude normal IO requests. | |
2786 | * | |
2787 | * Resync and recovery are handled very differently. | |
2788 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2789 | * | |
2790 | * For resync, we iterate over virtual addresses, read all copies, | |
2791 | * and update if there are differences. If only one copy is live, | |
2792 | * skip it. | |
2793 | * For recovery, we iterate over physical addresses, read a good | |
2794 | * value for each non-in_sync drive, and over-write. | |
2795 | * | |
2796 | * So, for recovery we may have several outstanding complex requests for a | |
2797 | * given address, one for each out-of-sync device. We model this by allocating | |
2798 | * a number of r10_bio structures, one for each out-of-sync device. | |
2799 | * As we setup these structures, we collect all bio's together into a list | |
2800 | * which we then process collectively to add pages, and then process again | |
2801 | * to pass to generic_make_request. | |
2802 | * | |
2803 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2804 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2805 | * has its remaining count decremented to 0, the whole complex operation | |
2806 | * is complete. | |
2807 | * | |
2808 | */ | |
2809 | ||
849674e4 | 2810 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 2811 | int *skipped) |
1da177e4 | 2812 | { |
e879a879 | 2813 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2814 | struct r10bio *r10_bio; |
1da177e4 LT |
2815 | struct bio *biolist = NULL, *bio; |
2816 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2817 | int i; |
6cce3b23 | 2818 | int max_sync; |
57dab0bd | 2819 | sector_t sync_blocks; |
1da177e4 LT |
2820 | sector_t sectors_skipped = 0; |
2821 | int chunks_skipped = 0; | |
5cf00fcd | 2822 | sector_t chunk_mask = conf->geo.chunk_mask; |
1da177e4 LT |
2823 | |
2824 | if (!conf->r10buf_pool) | |
2825 | if (init_resync(conf)) | |
57afd89f | 2826 | return 0; |
1da177e4 | 2827 | |
7e83ccbe MW |
2828 | /* |
2829 | * Allow skipping a full rebuild for incremental assembly | |
2830 | * of a clean array, like RAID1 does. | |
2831 | */ | |
2832 | if (mddev->bitmap == NULL && | |
2833 | mddev->recovery_cp == MaxSector && | |
13765120 N |
2834 | mddev->reshape_position == MaxSector && |
2835 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 2836 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 2837 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
2838 | conf->fullsync == 0) { |
2839 | *skipped = 1; | |
13765120 | 2840 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
2841 | } |
2842 | ||
1da177e4 | 2843 | skipped: |
58c0fed4 | 2844 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
2845 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
2846 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
2847 | max_sector = mddev->resync_max_sectors; |
2848 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
2849 | /* If we aborted, we need to abort the |
2850 | * sync on the 'current' bitmap chucks (there can | |
2851 | * be several when recovering multiple devices). | |
2852 | * as we may have started syncing it but not finished. | |
2853 | * We can find the current address in | |
2854 | * mddev->curr_resync, but for recovery, | |
2855 | * we need to convert that to several | |
2856 | * virtual addresses. | |
2857 | */ | |
3ea7daa5 N |
2858 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
2859 | end_reshape(conf); | |
b3968552 | 2860 | close_sync(conf); |
3ea7daa5 N |
2861 | return 0; |
2862 | } | |
2863 | ||
6cce3b23 N |
2864 | if (mddev->curr_resync < max_sector) { /* aborted */ |
2865 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2866 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
2867 | &sync_blocks, 1); | |
5cf00fcd | 2868 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
2869 | sector_t sect = |
2870 | raid10_find_virt(conf, mddev->curr_resync, i); | |
2871 | bitmap_end_sync(mddev->bitmap, sect, | |
2872 | &sync_blocks, 1); | |
2873 | } | |
9ad1aefc N |
2874 | } else { |
2875 | /* completed sync */ | |
2876 | if ((!mddev->bitmap || conf->fullsync) | |
2877 | && conf->have_replacement | |
2878 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
2879 | /* Completed a full sync so the replacements | |
2880 | * are now fully recovered. | |
2881 | */ | |
5cf00fcd | 2882 | for (i = 0; i < conf->geo.raid_disks; i++) |
9ad1aefc N |
2883 | if (conf->mirrors[i].replacement) |
2884 | conf->mirrors[i].replacement | |
2885 | ->recovery_offset | |
2886 | = MaxSector; | |
2887 | } | |
6cce3b23 | 2888 | conf->fullsync = 0; |
9ad1aefc | 2889 | } |
6cce3b23 | 2890 | bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2891 | close_sync(conf); |
57afd89f | 2892 | *skipped = 1; |
1da177e4 LT |
2893 | return sectors_skipped; |
2894 | } | |
3ea7daa5 N |
2895 | |
2896 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
2897 | return reshape_request(mddev, sector_nr, skipped); | |
2898 | ||
5cf00fcd | 2899 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
2900 | /* if there has been nothing to do on any drive, |
2901 | * then there is nothing to do at all.. | |
2902 | */ | |
57afd89f N |
2903 | *skipped = 1; |
2904 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
2905 | } |
2906 | ||
c6207277 N |
2907 | if (max_sector > mddev->resync_max) |
2908 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
2909 | ||
1da177e4 LT |
2910 | /* make sure whole request will fit in a chunk - if chunks |
2911 | * are meaningful | |
2912 | */ | |
5cf00fcd N |
2913 | if (conf->geo.near_copies < conf->geo.raid_disks && |
2914 | max_sector > (sector_nr | chunk_mask)) | |
2915 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 LT |
2916 | |
2917 | /* Again, very different code for resync and recovery. | |
2918 | * Both must result in an r10bio with a list of bios that | |
2919 | * have bi_end_io, bi_sector, bi_bdev set, | |
2920 | * and bi_private set to the r10bio. | |
2921 | * For recovery, we may actually create several r10bios | |
2922 | * with 2 bios in each, that correspond to the bios in the main one. | |
2923 | * In this case, the subordinate r10bios link back through a | |
2924 | * borrowed master_bio pointer, and the counter in the master | |
2925 | * includes a ref from each subordinate. | |
2926 | */ | |
2927 | /* First, we decide what to do and set ->bi_end_io | |
2928 | * To end_sync_read if we want to read, and | |
2929 | * end_sync_write if we will want to write. | |
2930 | */ | |
2931 | ||
6cce3b23 | 2932 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
2933 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
2934 | /* recovery... the complicated one */ | |
e875ecea | 2935 | int j; |
1da177e4 LT |
2936 | r10_bio = NULL; |
2937 | ||
5cf00fcd | 2938 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 2939 | int still_degraded; |
9f2c9d12 | 2940 | struct r10bio *rb2; |
ab9d47e9 N |
2941 | sector_t sect; |
2942 | int must_sync; | |
e875ecea | 2943 | int any_working; |
dc280d98 | 2944 | struct raid10_info *mirror = &conf->mirrors[i]; |
24afd80d N |
2945 | |
2946 | if ((mirror->rdev == NULL || | |
2947 | test_bit(In_sync, &mirror->rdev->flags)) | |
2948 | && | |
2949 | (mirror->replacement == NULL || | |
2950 | test_bit(Faulty, | |
2951 | &mirror->replacement->flags))) | |
ab9d47e9 | 2952 | continue; |
1da177e4 | 2953 | |
ab9d47e9 N |
2954 | still_degraded = 0; |
2955 | /* want to reconstruct this device */ | |
2956 | rb2 = r10_bio; | |
2957 | sect = raid10_find_virt(conf, sector_nr, i); | |
fc448a18 N |
2958 | if (sect >= mddev->resync_max_sectors) { |
2959 | /* last stripe is not complete - don't | |
2960 | * try to recover this sector. | |
2961 | */ | |
2962 | continue; | |
2963 | } | |
24afd80d N |
2964 | /* Unless we are doing a full sync, or a replacement |
2965 | * we only need to recover the block if it is set in | |
2966 | * the bitmap | |
ab9d47e9 N |
2967 | */ |
2968 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
2969 | &sync_blocks, 1); | |
2970 | if (sync_blocks < max_sync) | |
2971 | max_sync = sync_blocks; | |
2972 | if (!must_sync && | |
24afd80d | 2973 | mirror->replacement == NULL && |
ab9d47e9 N |
2974 | !conf->fullsync) { |
2975 | /* yep, skip the sync_blocks here, but don't assume | |
2976 | * that there will never be anything to do here | |
2977 | */ | |
2978 | chunks_skipped = -1; | |
2979 | continue; | |
2980 | } | |
6cce3b23 | 2981 | |
ab9d47e9 | 2982 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 2983 | r10_bio->state = 0; |
ab9d47e9 N |
2984 | raise_barrier(conf, rb2 != NULL); |
2985 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 2986 | |
ab9d47e9 N |
2987 | r10_bio->master_bio = (struct bio*)rb2; |
2988 | if (rb2) | |
2989 | atomic_inc(&rb2->remaining); | |
2990 | r10_bio->mddev = mddev; | |
2991 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
2992 | r10_bio->sector = sect; | |
1da177e4 | 2993 | |
ab9d47e9 N |
2994 | raid10_find_phys(conf, r10_bio); |
2995 | ||
2996 | /* Need to check if the array will still be | |
2997 | * degraded | |
2998 | */ | |
5cf00fcd | 2999 | for (j = 0; j < conf->geo.raid_disks; j++) |
ab9d47e9 N |
3000 | if (conf->mirrors[j].rdev == NULL || |
3001 | test_bit(Faulty, &conf->mirrors[j].rdev->flags)) { | |
3002 | still_degraded = 1; | |
87fc767b | 3003 | break; |
1da177e4 | 3004 | } |
ab9d47e9 N |
3005 | |
3006 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3007 | &sync_blocks, still_degraded); | |
3008 | ||
e875ecea | 3009 | any_working = 0; |
ab9d47e9 | 3010 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3011 | int k; |
ab9d47e9 | 3012 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3013 | sector_t from_addr, to_addr; |
3cb03002 | 3014 | struct md_rdev *rdev; |
40c356ce N |
3015 | sector_t sector, first_bad; |
3016 | int bad_sectors; | |
ab9d47e9 N |
3017 | if (!conf->mirrors[d].rdev || |
3018 | !test_bit(In_sync, &conf->mirrors[d].rdev->flags)) | |
3019 | continue; | |
3020 | /* This is where we read from */ | |
e875ecea | 3021 | any_working = 1; |
40c356ce N |
3022 | rdev = conf->mirrors[d].rdev; |
3023 | sector = r10_bio->devs[j].addr; | |
3024 | ||
3025 | if (is_badblock(rdev, sector, max_sync, | |
3026 | &first_bad, &bad_sectors)) { | |
3027 | if (first_bad > sector) | |
3028 | max_sync = first_bad - sector; | |
3029 | else { | |
3030 | bad_sectors -= (sector | |
3031 | - first_bad); | |
3032 | if (max_sync > bad_sectors) | |
3033 | max_sync = bad_sectors; | |
3034 | continue; | |
3035 | } | |
3036 | } | |
ab9d47e9 | 3037 | bio = r10_bio->devs[0].bio; |
8be185f2 | 3038 | bio_reset(bio); |
ab9d47e9 N |
3039 | bio->bi_next = biolist; |
3040 | biolist = bio; | |
3041 | bio->bi_private = r10_bio; | |
3042 | bio->bi_end_io = end_sync_read; | |
3043 | bio->bi_rw = READ; | |
5e570289 | 3044 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3045 | bio->bi_iter.bi_sector = from_addr + |
3046 | rdev->data_offset; | |
24afd80d N |
3047 | bio->bi_bdev = rdev->bdev; |
3048 | atomic_inc(&rdev->nr_pending); | |
3049 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3050 | |
3051 | for (k=0; k<conf->copies; k++) | |
3052 | if (r10_bio->devs[k].devnum == i) | |
3053 | break; | |
3054 | BUG_ON(k == conf->copies); | |
5e570289 | 3055 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3056 | r10_bio->devs[0].devnum = d; |
5e570289 | 3057 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3058 | r10_bio->devs[1].devnum = i; |
5e570289 | 3059 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3060 | |
24afd80d N |
3061 | rdev = mirror->rdev; |
3062 | if (!test_bit(In_sync, &rdev->flags)) { | |
3063 | bio = r10_bio->devs[1].bio; | |
8be185f2 | 3064 | bio_reset(bio); |
24afd80d N |
3065 | bio->bi_next = biolist; |
3066 | biolist = bio; | |
3067 | bio->bi_private = r10_bio; | |
3068 | bio->bi_end_io = end_sync_write; | |
3069 | bio->bi_rw = WRITE; | |
4f024f37 | 3070 | bio->bi_iter.bi_sector = to_addr |
24afd80d N |
3071 | + rdev->data_offset; |
3072 | bio->bi_bdev = rdev->bdev; | |
3073 | atomic_inc(&r10_bio->remaining); | |
3074 | } else | |
3075 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3076 | ||
3077 | /* and maybe write to replacement */ | |
3078 | bio = r10_bio->devs[1].repl_bio; | |
3079 | if (bio) | |
3080 | bio->bi_end_io = NULL; | |
3081 | rdev = mirror->replacement; | |
3082 | /* Note: if rdev != NULL, then bio | |
3083 | * cannot be NULL as r10buf_pool_alloc will | |
3084 | * have allocated it. | |
3085 | * So the second test here is pointless. | |
3086 | * But it keeps semantic-checkers happy, and | |
3087 | * this comment keeps human reviewers | |
3088 | * happy. | |
3089 | */ | |
3090 | if (rdev == NULL || bio == NULL || | |
3091 | test_bit(Faulty, &rdev->flags)) | |
3092 | break; | |
8be185f2 | 3093 | bio_reset(bio); |
24afd80d N |
3094 | bio->bi_next = biolist; |
3095 | biolist = bio; | |
3096 | bio->bi_private = r10_bio; | |
3097 | bio->bi_end_io = end_sync_write; | |
3098 | bio->bi_rw = WRITE; | |
4f024f37 KO |
3099 | bio->bi_iter.bi_sector = to_addr + |
3100 | rdev->data_offset; | |
24afd80d N |
3101 | bio->bi_bdev = rdev->bdev; |
3102 | atomic_inc(&r10_bio->remaining); | |
ab9d47e9 N |
3103 | break; |
3104 | } | |
3105 | if (j == conf->copies) { | |
e875ecea N |
3106 | /* Cannot recover, so abort the recovery or |
3107 | * record a bad block */ | |
e875ecea N |
3108 | if (any_working) { |
3109 | /* problem is that there are bad blocks | |
3110 | * on other device(s) | |
3111 | */ | |
3112 | int k; | |
3113 | for (k = 0; k < conf->copies; k++) | |
3114 | if (r10_bio->devs[k].devnum == i) | |
3115 | break; | |
24afd80d N |
3116 | if (!test_bit(In_sync, |
3117 | &mirror->rdev->flags) | |
3118 | && !rdev_set_badblocks( | |
3119 | mirror->rdev, | |
3120 | r10_bio->devs[k].addr, | |
3121 | max_sync, 0)) | |
3122 | any_working = 0; | |
3123 | if (mirror->replacement && | |
3124 | !rdev_set_badblocks( | |
3125 | mirror->replacement, | |
e875ecea N |
3126 | r10_bio->devs[k].addr, |
3127 | max_sync, 0)) | |
3128 | any_working = 0; | |
3129 | } | |
3130 | if (!any_working) { | |
3131 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3132 | &mddev->recovery)) | |
3133 | printk(KERN_INFO "md/raid10:%s: insufficient " | |
3134 | "working devices for recovery.\n", | |
3135 | mdname(mddev)); | |
24afd80d | 3136 | mirror->recovery_disabled |
e875ecea N |
3137 | = mddev->recovery_disabled; |
3138 | } | |
e8b84915 N |
3139 | put_buf(r10_bio); |
3140 | if (rb2) | |
3141 | atomic_dec(&rb2->remaining); | |
3142 | r10_bio = rb2; | |
ab9d47e9 | 3143 | break; |
1da177e4 | 3144 | } |
ab9d47e9 | 3145 | } |
1da177e4 LT |
3146 | if (biolist == NULL) { |
3147 | while (r10_bio) { | |
9f2c9d12 N |
3148 | struct r10bio *rb2 = r10_bio; |
3149 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3150 | rb2->master_bio = NULL; |
3151 | put_buf(rb2); | |
3152 | } | |
3153 | goto giveup; | |
3154 | } | |
3155 | } else { | |
3156 | /* resync. Schedule a read for every block at this virt offset */ | |
3157 | int count = 0; | |
6cce3b23 | 3158 | |
c40f341f | 3159 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, 0); |
78200d45 | 3160 | |
6cce3b23 N |
3161 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
3162 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3163 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3164 | &mddev->recovery)) { | |
6cce3b23 N |
3165 | /* We can skip this block */ |
3166 | *skipped = 1; | |
3167 | return sync_blocks + sectors_skipped; | |
3168 | } | |
3169 | if (sync_blocks < max_sync) | |
3170 | max_sync = sync_blocks; | |
1da177e4 | 3171 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 3172 | r10_bio->state = 0; |
1da177e4 | 3173 | |
1da177e4 LT |
3174 | r10_bio->mddev = mddev; |
3175 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3176 | raise_barrier(conf, 0); |
3177 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3178 | |
3179 | r10_bio->master_bio = NULL; | |
3180 | r10_bio->sector = sector_nr; | |
3181 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3182 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3183 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3184 | |
5cf00fcd | 3185 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3186 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3187 | sector_t first_bad, sector; |
3188 | int bad_sectors; | |
3189 | ||
9ad1aefc N |
3190 | if (r10_bio->devs[i].repl_bio) |
3191 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3192 | ||
1da177e4 | 3193 | bio = r10_bio->devs[i].bio; |
8be185f2 | 3194 | bio_reset(bio); |
4246a0b6 | 3195 | bio->bi_error = -EIO; |
1da177e4 | 3196 | if (conf->mirrors[d].rdev == NULL || |
b2d444d7 | 3197 | test_bit(Faulty, &conf->mirrors[d].rdev->flags)) |
1da177e4 | 3198 | continue; |
40c356ce N |
3199 | sector = r10_bio->devs[i].addr; |
3200 | if (is_badblock(conf->mirrors[d].rdev, | |
3201 | sector, max_sync, | |
3202 | &first_bad, &bad_sectors)) { | |
3203 | if (first_bad > sector) | |
3204 | max_sync = first_bad - sector; | |
3205 | else { | |
3206 | bad_sectors -= (sector - first_bad); | |
3207 | if (max_sync > bad_sectors) | |
91502f09 | 3208 | max_sync = bad_sectors; |
40c356ce N |
3209 | continue; |
3210 | } | |
3211 | } | |
1da177e4 LT |
3212 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
3213 | atomic_inc(&r10_bio->remaining); | |
3214 | bio->bi_next = biolist; | |
3215 | biolist = bio; | |
3216 | bio->bi_private = r10_bio; | |
3217 | bio->bi_end_io = end_sync_read; | |
802ba064 | 3218 | bio->bi_rw = READ; |
4f024f37 | 3219 | bio->bi_iter.bi_sector = sector + |
1da177e4 LT |
3220 | conf->mirrors[d].rdev->data_offset; |
3221 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
3222 | count++; | |
9ad1aefc N |
3223 | |
3224 | if (conf->mirrors[d].replacement == NULL || | |
3225 | test_bit(Faulty, | |
3226 | &conf->mirrors[d].replacement->flags)) | |
3227 | continue; | |
3228 | ||
3229 | /* Need to set up for writing to the replacement */ | |
3230 | bio = r10_bio->devs[i].repl_bio; | |
8be185f2 | 3231 | bio_reset(bio); |
4246a0b6 | 3232 | bio->bi_error = -EIO; |
9ad1aefc N |
3233 | |
3234 | sector = r10_bio->devs[i].addr; | |
3235 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
3236 | bio->bi_next = biolist; | |
3237 | biolist = bio; | |
3238 | bio->bi_private = r10_bio; | |
3239 | bio->bi_end_io = end_sync_write; | |
3240 | bio->bi_rw = WRITE; | |
4f024f37 | 3241 | bio->bi_iter.bi_sector = sector + |
9ad1aefc N |
3242 | conf->mirrors[d].replacement->data_offset; |
3243 | bio->bi_bdev = conf->mirrors[d].replacement->bdev; | |
3244 | count++; | |
1da177e4 LT |
3245 | } |
3246 | ||
3247 | if (count < 2) { | |
3248 | for (i=0; i<conf->copies; i++) { | |
3249 | int d = r10_bio->devs[i].devnum; | |
3250 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3251 | rdev_dec_pending(conf->mirrors[d].rdev, |
3252 | mddev); | |
9ad1aefc N |
3253 | if (r10_bio->devs[i].repl_bio && |
3254 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3255 | rdev_dec_pending( | |
3256 | conf->mirrors[d].replacement, | |
3257 | mddev); | |
1da177e4 LT |
3258 | } |
3259 | put_buf(r10_bio); | |
3260 | biolist = NULL; | |
3261 | goto giveup; | |
3262 | } | |
3263 | } | |
3264 | ||
1da177e4 | 3265 | nr_sectors = 0; |
6cce3b23 N |
3266 | if (sector_nr + max_sync < max_sector) |
3267 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3268 | do { |
3269 | struct page *page; | |
3270 | int len = PAGE_SIZE; | |
1da177e4 LT |
3271 | if (sector_nr + (len>>9) > max_sector) |
3272 | len = (max_sector - sector_nr) << 9; | |
3273 | if (len == 0) | |
3274 | break; | |
3275 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
ab9d47e9 | 3276 | struct bio *bio2; |
1da177e4 | 3277 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
ab9d47e9 N |
3278 | if (bio_add_page(bio, page, len, 0)) |
3279 | continue; | |
3280 | ||
3281 | /* stop here */ | |
3282 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
3283 | for (bio2 = biolist; | |
3284 | bio2 && bio2 != bio; | |
3285 | bio2 = bio2->bi_next) { | |
3286 | /* remove last page from this bio */ | |
3287 | bio2->bi_vcnt--; | |
4f024f37 | 3288 | bio2->bi_iter.bi_size -= len; |
b7c44ed9 | 3289 | bio_clear_flag(bio2, BIO_SEG_VALID); |
1da177e4 | 3290 | } |
ab9d47e9 | 3291 | goto bio_full; |
1da177e4 LT |
3292 | } |
3293 | nr_sectors += len>>9; | |
3294 | sector_nr += len>>9; | |
3295 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
3296 | bio_full: | |
3297 | r10_bio->sectors = nr_sectors; | |
3298 | ||
3299 | while (biolist) { | |
3300 | bio = biolist; | |
3301 | biolist = biolist->bi_next; | |
3302 | ||
3303 | bio->bi_next = NULL; | |
3304 | r10_bio = bio->bi_private; | |
3305 | r10_bio->sectors = nr_sectors; | |
3306 | ||
3307 | if (bio->bi_end_io == end_sync_read) { | |
3308 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
4246a0b6 | 3309 | bio->bi_error = 0; |
1da177e4 LT |
3310 | generic_make_request(bio); |
3311 | } | |
3312 | } | |
3313 | ||
57afd89f N |
3314 | if (sectors_skipped) |
3315 | /* pretend they weren't skipped, it makes | |
3316 | * no important difference in this case | |
3317 | */ | |
3318 | md_done_sync(mddev, sectors_skipped, 1); | |
3319 | ||
1da177e4 LT |
3320 | return sectors_skipped + nr_sectors; |
3321 | giveup: | |
3322 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3323 | * drives must be failed or in resync, all drives |
3324 | * have a bad block, so try the next chunk... | |
1da177e4 | 3325 | */ |
09b4068a N |
3326 | if (sector_nr + max_sync < max_sector) |
3327 | max_sector = sector_nr + max_sync; | |
3328 | ||
3329 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3330 | chunks_skipped ++; |
3331 | sector_nr = max_sector; | |
1da177e4 | 3332 | goto skipped; |
1da177e4 LT |
3333 | } |
3334 | ||
80c3a6ce | 3335 | static sector_t |
fd01b88c | 3336 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3337 | { |
3338 | sector_t size; | |
e879a879 | 3339 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3340 | |
3341 | if (!raid_disks) | |
3ea7daa5 N |
3342 | raid_disks = min(conf->geo.raid_disks, |
3343 | conf->prev.raid_disks); | |
80c3a6ce | 3344 | if (!sectors) |
dab8b292 | 3345 | sectors = conf->dev_sectors; |
80c3a6ce | 3346 | |
5cf00fcd N |
3347 | size = sectors >> conf->geo.chunk_shift; |
3348 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3349 | size = size * raid_disks; |
5cf00fcd | 3350 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3351 | |
5cf00fcd | 3352 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3353 | } |
3354 | ||
6508fdbf N |
3355 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3356 | { | |
3357 | /* Calculate the number of sectors-per-device that will | |
3358 | * actually be used, and set conf->dev_sectors and | |
3359 | * conf->stride | |
3360 | */ | |
3361 | ||
5cf00fcd N |
3362 | size = size >> conf->geo.chunk_shift; |
3363 | sector_div(size, conf->geo.far_copies); | |
3364 | size = size * conf->geo.raid_disks; | |
3365 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3366 | /* 'size' is now the number of chunks in the array */ |
3367 | /* calculate "used chunks per device" */ | |
3368 | size = size * conf->copies; | |
3369 | ||
3370 | /* We need to round up when dividing by raid_disks to | |
3371 | * get the stride size. | |
3372 | */ | |
5cf00fcd | 3373 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3374 | |
5cf00fcd | 3375 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3376 | |
5cf00fcd N |
3377 | if (conf->geo.far_offset) |
3378 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3379 | else { |
5cf00fcd N |
3380 | sector_div(size, conf->geo.far_copies); |
3381 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3382 | } |
3383 | } | |
dab8b292 | 3384 | |
deb200d0 N |
3385 | enum geo_type {geo_new, geo_old, geo_start}; |
3386 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3387 | { | |
3388 | int nc, fc, fo; | |
3389 | int layout, chunk, disks; | |
3390 | switch (new) { | |
3391 | case geo_old: | |
3392 | layout = mddev->layout; | |
3393 | chunk = mddev->chunk_sectors; | |
3394 | disks = mddev->raid_disks - mddev->delta_disks; | |
3395 | break; | |
3396 | case geo_new: | |
3397 | layout = mddev->new_layout; | |
3398 | chunk = mddev->new_chunk_sectors; | |
3399 | disks = mddev->raid_disks; | |
3400 | break; | |
3401 | default: /* avoid 'may be unused' warnings */ | |
3402 | case geo_start: /* new when starting reshape - raid_disks not | |
3403 | * updated yet. */ | |
3404 | layout = mddev->new_layout; | |
3405 | chunk = mddev->new_chunk_sectors; | |
3406 | disks = mddev->raid_disks + mddev->delta_disks; | |
3407 | break; | |
3408 | } | |
8bce6d35 | 3409 | if (layout >> 19) |
deb200d0 N |
3410 | return -1; |
3411 | if (chunk < (PAGE_SIZE >> 9) || | |
3412 | !is_power_of_2(chunk)) | |
3413 | return -2; | |
3414 | nc = layout & 255; | |
3415 | fc = (layout >> 8) & 255; | |
3416 | fo = layout & (1<<16); | |
3417 | geo->raid_disks = disks; | |
3418 | geo->near_copies = nc; | |
3419 | geo->far_copies = fc; | |
3420 | geo->far_offset = fo; | |
8bce6d35 N |
3421 | switch (layout >> 17) { |
3422 | case 0: /* original layout. simple but not always optimal */ | |
3423 | geo->far_set_size = disks; | |
3424 | break; | |
3425 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3426 | * actually using this, but leave code here just in case.*/ | |
3427 | geo->far_set_size = disks/fc; | |
3428 | WARN(geo->far_set_size < fc, | |
3429 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3430 | break; | |
3431 | case 2: /* "improved" layout fixed to match documentation */ | |
3432 | geo->far_set_size = fc * nc; | |
3433 | break; | |
3434 | default: /* Not a valid layout */ | |
3435 | return -1; | |
3436 | } | |
deb200d0 N |
3437 | geo->chunk_mask = chunk - 1; |
3438 | geo->chunk_shift = ffz(~chunk); | |
3439 | return nc*fc; | |
3440 | } | |
3441 | ||
e879a879 | 3442 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3443 | { |
e879a879 | 3444 | struct r10conf *conf = NULL; |
dab8b292 | 3445 | int err = -EINVAL; |
deb200d0 N |
3446 | struct geom geo; |
3447 | int copies; | |
3448 | ||
3449 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3450 | |
deb200d0 | 3451 | if (copies == -2) { |
128595ed N |
3452 | printk(KERN_ERR "md/raid10:%s: chunk size must be " |
3453 | "at least PAGE_SIZE(%ld) and be a power of 2.\n", | |
3454 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3455 | goto out; |
1da177e4 | 3456 | } |
2604b703 | 3457 | |
deb200d0 | 3458 | if (copies < 2 || copies > mddev->raid_disks) { |
128595ed | 3459 | printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
f73ea873 | 3460 | mdname(mddev), mddev->new_layout); |
1da177e4 LT |
3461 | goto out; |
3462 | } | |
dab8b292 TM |
3463 | |
3464 | err = -ENOMEM; | |
e879a879 | 3465 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3466 | if (!conf) |
1da177e4 | 3467 | goto out; |
dab8b292 | 3468 | |
3ea7daa5 | 3469 | /* FIXME calc properly */ |
dc280d98 | 3470 | conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks + |
78eaa0d4 | 3471 | max(0,-mddev->delta_disks)), |
dab8b292 TM |
3472 | GFP_KERNEL); |
3473 | if (!conf->mirrors) | |
3474 | goto out; | |
4443ae10 N |
3475 | |
3476 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3477 | if (!conf->tmppage) | |
dab8b292 TM |
3478 | goto out; |
3479 | ||
deb200d0 N |
3480 | conf->geo = geo; |
3481 | conf->copies = copies; | |
dab8b292 TM |
3482 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, |
3483 | r10bio_pool_free, conf); | |
3484 | if (!conf->r10bio_pool) | |
3485 | goto out; | |
3486 | ||
6508fdbf | 3487 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3488 | if (mddev->reshape_position == MaxSector) { |
3489 | conf->prev = conf->geo; | |
3490 | conf->reshape_progress = MaxSector; | |
3491 | } else { | |
3492 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3493 | err = -EINVAL; | |
3494 | goto out; | |
3495 | } | |
3496 | conf->reshape_progress = mddev->reshape_position; | |
3497 | if (conf->prev.far_offset) | |
3498 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
3499 | else | |
3500 | /* far_copies must be 1 */ | |
3501 | conf->prev.stride = conf->dev_sectors; | |
3502 | } | |
299b0685 | 3503 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 3504 | spin_lock_init(&conf->device_lock); |
dab8b292 | 3505 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 3506 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 TM |
3507 | |
3508 | spin_lock_init(&conf->resync_lock); | |
3509 | init_waitqueue_head(&conf->wait_barrier); | |
3510 | ||
0232605d | 3511 | conf->thread = md_register_thread(raid10d, mddev, "raid10"); |
dab8b292 TM |
3512 | if (!conf->thread) |
3513 | goto out; | |
3514 | ||
dab8b292 TM |
3515 | conf->mddev = mddev; |
3516 | return conf; | |
3517 | ||
3518 | out: | |
3ea7daa5 N |
3519 | if (err == -ENOMEM) |
3520 | printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n", | |
3521 | mdname(mddev)); | |
dab8b292 | 3522 | if (conf) { |
644df1a8 | 3523 | mempool_destroy(conf->r10bio_pool); |
dab8b292 TM |
3524 | kfree(conf->mirrors); |
3525 | safe_put_page(conf->tmppage); | |
3526 | kfree(conf); | |
3527 | } | |
3528 | return ERR_PTR(err); | |
3529 | } | |
3530 | ||
849674e4 | 3531 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 3532 | { |
e879a879 | 3533 | struct r10conf *conf; |
dab8b292 | 3534 | int i, disk_idx, chunk_size; |
dc280d98 | 3535 | struct raid10_info *disk; |
3cb03002 | 3536 | struct md_rdev *rdev; |
dab8b292 | 3537 | sector_t size; |
3ea7daa5 N |
3538 | sector_t min_offset_diff = 0; |
3539 | int first = 1; | |
532a2a3f | 3540 | bool discard_supported = false; |
dab8b292 TM |
3541 | |
3542 | if (mddev->private == NULL) { | |
3543 | conf = setup_conf(mddev); | |
3544 | if (IS_ERR(conf)) | |
3545 | return PTR_ERR(conf); | |
3546 | mddev->private = conf; | |
3547 | } | |
3548 | conf = mddev->private; | |
3549 | if (!conf) | |
3550 | goto out; | |
3551 | ||
dab8b292 TM |
3552 | mddev->thread = conf->thread; |
3553 | conf->thread = NULL; | |
3554 | ||
8f6c2e4b | 3555 | chunk_size = mddev->chunk_sectors << 9; |
cc4d1efd | 3556 | if (mddev->queue) { |
532a2a3f SL |
3557 | blk_queue_max_discard_sectors(mddev->queue, |
3558 | mddev->chunk_sectors); | |
5026d7a9 | 3559 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
cc4d1efd JB |
3560 | blk_queue_io_min(mddev->queue, chunk_size); |
3561 | if (conf->geo.raid_disks % conf->geo.near_copies) | |
3562 | blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks); | |
3563 | else | |
3564 | blk_queue_io_opt(mddev->queue, chunk_size * | |
3565 | (conf->geo.raid_disks / conf->geo.near_copies)); | |
3566 | } | |
8f6c2e4b | 3567 | |
dafb20fa | 3568 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 3569 | long long diff; |
aba336bd | 3570 | struct request_queue *q; |
34b343cf | 3571 | |
1da177e4 | 3572 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
3573 | if (disk_idx < 0) |
3574 | continue; | |
3575 | if (disk_idx >= conf->geo.raid_disks && | |
3576 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
3577 | continue; |
3578 | disk = conf->mirrors + disk_idx; | |
3579 | ||
56a2559b N |
3580 | if (test_bit(Replacement, &rdev->flags)) { |
3581 | if (disk->replacement) | |
3582 | goto out_free_conf; | |
3583 | disk->replacement = rdev; | |
3584 | } else { | |
3585 | if (disk->rdev) | |
3586 | goto out_free_conf; | |
3587 | disk->rdev = rdev; | |
3588 | } | |
aba336bd | 3589 | q = bdev_get_queue(rdev->bdev); |
3ea7daa5 N |
3590 | diff = (rdev->new_data_offset - rdev->data_offset); |
3591 | if (!mddev->reshape_backwards) | |
3592 | diff = -diff; | |
3593 | if (diff < 0) | |
3594 | diff = 0; | |
3595 | if (first || diff < min_offset_diff) | |
3596 | min_offset_diff = diff; | |
56a2559b | 3597 | |
cc4d1efd JB |
3598 | if (mddev->gendisk) |
3599 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
3600 | rdev->data_offset << 9); | |
1da177e4 LT |
3601 | |
3602 | disk->head_position = 0; | |
532a2a3f SL |
3603 | |
3604 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) | |
3605 | discard_supported = true; | |
1da177e4 | 3606 | } |
3ea7daa5 | 3607 | |
ed30be07 JB |
3608 | if (mddev->queue) { |
3609 | if (discard_supported) | |
3610 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3611 | mddev->queue); | |
3612 | else | |
3613 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3614 | mddev->queue); | |
3615 | } | |
6d508242 | 3616 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3617 | if (!enough(conf, -1)) { |
128595ed | 3618 | printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3619 | mdname(mddev)); |
1da177e4 LT |
3620 | goto out_free_conf; |
3621 | } | |
3622 | ||
3ea7daa5 N |
3623 | if (conf->reshape_progress != MaxSector) { |
3624 | /* must ensure that shape change is supported */ | |
3625 | if (conf->geo.far_copies != 1 && | |
3626 | conf->geo.far_offset == 0) | |
3627 | goto out_free_conf; | |
3628 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 3629 | conf->prev.far_offset == 0) |
3ea7daa5 N |
3630 | goto out_free_conf; |
3631 | } | |
3632 | ||
1da177e4 | 3633 | mddev->degraded = 0; |
f8c9e74f N |
3634 | for (i = 0; |
3635 | i < conf->geo.raid_disks | |
3636 | || i < conf->prev.raid_disks; | |
3637 | i++) { | |
1da177e4 LT |
3638 | |
3639 | disk = conf->mirrors + i; | |
3640 | ||
56a2559b N |
3641 | if (!disk->rdev && disk->replacement) { |
3642 | /* The replacement is all we have - use it */ | |
3643 | disk->rdev = disk->replacement; | |
3644 | disk->replacement = NULL; | |
3645 | clear_bit(Replacement, &disk->rdev->flags); | |
3646 | } | |
3647 | ||
5fd6c1dc | 3648 | if (!disk->rdev || |
2e333e89 | 3649 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3650 | disk->head_position = 0; |
3651 | mddev->degraded++; | |
0b59bb64 N |
3652 | if (disk->rdev && |
3653 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 3654 | conf->fullsync = 1; |
1da177e4 | 3655 | } |
d890fa2b | 3656 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3657 | } |
3658 | ||
8c6ac868 | 3659 | if (mddev->recovery_cp != MaxSector) |
128595ed | 3660 | printk(KERN_NOTICE "md/raid10:%s: not clean" |
8c6ac868 AN |
3661 | " -- starting background reconstruction\n", |
3662 | mdname(mddev)); | |
1da177e4 | 3663 | printk(KERN_INFO |
128595ed | 3664 | "md/raid10:%s: active with %d out of %d devices\n", |
5cf00fcd N |
3665 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
3666 | conf->geo.raid_disks); | |
1da177e4 LT |
3667 | /* |
3668 | * Ok, everything is just fine now | |
3669 | */ | |
dab8b292 TM |
3670 | mddev->dev_sectors = conf->dev_sectors; |
3671 | size = raid10_size(mddev, 0, 0); | |
3672 | md_set_array_sectors(mddev, size); | |
3673 | mddev->resync_max_sectors = size; | |
1da177e4 | 3674 | |
cc4d1efd | 3675 | if (mddev->queue) { |
5cf00fcd | 3676 | int stripe = conf->geo.raid_disks * |
9d8f0363 | 3677 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
cc4d1efd JB |
3678 | |
3679 | /* Calculate max read-ahead size. | |
3680 | * We need to readahead at least twice a whole stripe.... | |
3681 | * maybe... | |
3682 | */ | |
5cf00fcd | 3683 | stripe /= conf->geo.near_copies; |
3ea7daa5 N |
3684 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
3685 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
1da177e4 LT |
3686 | } |
3687 | ||
a91a2785 MP |
3688 | if (md_integrity_register(mddev)) |
3689 | goto out_free_conf; | |
3690 | ||
3ea7daa5 N |
3691 | if (conf->reshape_progress != MaxSector) { |
3692 | unsigned long before_length, after_length; | |
3693 | ||
3694 | before_length = ((1 << conf->prev.chunk_shift) * | |
3695 | conf->prev.far_copies); | |
3696 | after_length = ((1 << conf->geo.chunk_shift) * | |
3697 | conf->geo.far_copies); | |
3698 | ||
3699 | if (max(before_length, after_length) > min_offset_diff) { | |
3700 | /* This cannot work */ | |
3701 | printk("md/raid10: offset difference not enough to continue reshape\n"); | |
3702 | goto out_free_conf; | |
3703 | } | |
3704 | conf->offset_diff = min_offset_diff; | |
3705 | ||
3ea7daa5 N |
3706 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
3707 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3708 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3709 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3710 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3711 | "reshape"); | |
3712 | } | |
3713 | ||
1da177e4 LT |
3714 | return 0; |
3715 | ||
3716 | out_free_conf: | |
01f96c0a | 3717 | md_unregister_thread(&mddev->thread); |
644df1a8 | 3718 | mempool_destroy(conf->r10bio_pool); |
1345b1d8 | 3719 | safe_put_page(conf->tmppage); |
990a8baf | 3720 | kfree(conf->mirrors); |
1da177e4 LT |
3721 | kfree(conf); |
3722 | mddev->private = NULL; | |
3723 | out: | |
3724 | return -EIO; | |
3725 | } | |
3726 | ||
afa0f557 | 3727 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 3728 | { |
afa0f557 | 3729 | struct r10conf *conf = priv; |
1da177e4 | 3730 | |
644df1a8 | 3731 | mempool_destroy(conf->r10bio_pool); |
0fea7ed8 | 3732 | safe_put_page(conf->tmppage); |
990a8baf | 3733 | kfree(conf->mirrors); |
c4796e21 N |
3734 | kfree(conf->mirrors_old); |
3735 | kfree(conf->mirrors_new); | |
1da177e4 | 3736 | kfree(conf); |
1da177e4 LT |
3737 | } |
3738 | ||
fd01b88c | 3739 | static void raid10_quiesce(struct mddev *mddev, int state) |
6cce3b23 | 3740 | { |
e879a879 | 3741 | struct r10conf *conf = mddev->private; |
6cce3b23 N |
3742 | |
3743 | switch(state) { | |
3744 | case 1: | |
3745 | raise_barrier(conf, 0); | |
3746 | break; | |
3747 | case 0: | |
3748 | lower_barrier(conf); | |
3749 | break; | |
3750 | } | |
6cce3b23 | 3751 | } |
1da177e4 | 3752 | |
006a09a0 N |
3753 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3754 | { | |
3755 | /* Resize of 'far' arrays is not supported. | |
3756 | * For 'near' and 'offset' arrays we can set the | |
3757 | * number of sectors used to be an appropriate multiple | |
3758 | * of the chunk size. | |
3759 | * For 'offset', this is far_copies*chunksize. | |
3760 | * For 'near' the multiplier is the LCM of | |
3761 | * near_copies and raid_disks. | |
3762 | * So if far_copies > 1 && !far_offset, fail. | |
3763 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
3764 | * multiply by chunk_size. Then round to this number. | |
3765 | * This is mostly done by raid10_size() | |
3766 | */ | |
3767 | struct r10conf *conf = mddev->private; | |
3768 | sector_t oldsize, size; | |
3769 | ||
f8c9e74f N |
3770 | if (mddev->reshape_position != MaxSector) |
3771 | return -EBUSY; | |
3772 | ||
5cf00fcd | 3773 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
3774 | return -EINVAL; |
3775 | ||
3776 | oldsize = raid10_size(mddev, 0, 0); | |
3777 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
3778 | if (mddev->external_size && |
3779 | mddev->array_sectors > size) | |
006a09a0 | 3780 | return -EINVAL; |
a4a6125a N |
3781 | if (mddev->bitmap) { |
3782 | int ret = bitmap_resize(mddev->bitmap, size, 0, 0); | |
3783 | if (ret) | |
3784 | return ret; | |
3785 | } | |
3786 | md_set_array_sectors(mddev, size); | |
859644f0 HM |
3787 | if (mddev->queue) { |
3788 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
3789 | revalidate_disk(mddev->gendisk); | |
3790 | } | |
006a09a0 N |
3791 | if (sectors > mddev->dev_sectors && |
3792 | mddev->recovery_cp > oldsize) { | |
3793 | mddev->recovery_cp = oldsize; | |
3794 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3795 | } | |
6508fdbf N |
3796 | calc_sectors(conf, sectors); |
3797 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
3798 | mddev->resync_max_sectors = size; |
3799 | return 0; | |
3800 | } | |
3801 | ||
53a6ab4d | 3802 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 3803 | { |
3cb03002 | 3804 | struct md_rdev *rdev; |
e879a879 | 3805 | struct r10conf *conf; |
dab8b292 TM |
3806 | |
3807 | if (mddev->degraded > 0) { | |
128595ed N |
3808 | printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n", |
3809 | mdname(mddev)); | |
dab8b292 TM |
3810 | return ERR_PTR(-EINVAL); |
3811 | } | |
53a6ab4d | 3812 | sector_div(size, devs); |
dab8b292 | 3813 | |
dab8b292 TM |
3814 | /* Set new parameters */ |
3815 | mddev->new_level = 10; | |
3816 | /* new layout: far_copies = 1, near_copies = 2 */ | |
3817 | mddev->new_layout = (1<<8) + 2; | |
3818 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
3819 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
3820 | mddev->raid_disks *= 2; |
3821 | /* make sure it will be not marked as dirty */ | |
3822 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 3823 | mddev->dev_sectors = size; |
dab8b292 TM |
3824 | |
3825 | conf = setup_conf(mddev); | |
02214dc5 | 3826 | if (!IS_ERR(conf)) { |
dafb20fa | 3827 | rdev_for_each(rdev, mddev) |
53a6ab4d | 3828 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 3829 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
3830 | rdev->sectors = size; |
3831 | } | |
02214dc5 KW |
3832 | conf->barrier = 1; |
3833 | } | |
3834 | ||
dab8b292 TM |
3835 | return conf; |
3836 | } | |
3837 | ||
fd01b88c | 3838 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 3839 | { |
e373ab10 | 3840 | struct r0conf *raid0_conf; |
dab8b292 TM |
3841 | |
3842 | /* raid10 can take over: | |
3843 | * raid0 - providing it has only two drives | |
3844 | */ | |
3845 | if (mddev->level == 0) { | |
3846 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
3847 | raid0_conf = mddev->private; |
3848 | if (raid0_conf->nr_strip_zones > 1) { | |
128595ed N |
3849 | printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0" |
3850 | " with more than one zone.\n", | |
3851 | mdname(mddev)); | |
dab8b292 TM |
3852 | return ERR_PTR(-EINVAL); |
3853 | } | |
53a6ab4d N |
3854 | return raid10_takeover_raid0(mddev, |
3855 | raid0_conf->strip_zone->zone_end, | |
3856 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
3857 | } |
3858 | return ERR_PTR(-EINVAL); | |
3859 | } | |
3860 | ||
3ea7daa5 N |
3861 | static int raid10_check_reshape(struct mddev *mddev) |
3862 | { | |
3863 | /* Called when there is a request to change | |
3864 | * - layout (to ->new_layout) | |
3865 | * - chunk size (to ->new_chunk_sectors) | |
3866 | * - raid_disks (by delta_disks) | |
3867 | * or when trying to restart a reshape that was ongoing. | |
3868 | * | |
3869 | * We need to validate the request and possibly allocate | |
3870 | * space if that might be an issue later. | |
3871 | * | |
3872 | * Currently we reject any reshape of a 'far' mode array, | |
3873 | * allow chunk size to change if new is generally acceptable, | |
3874 | * allow raid_disks to increase, and allow | |
3875 | * a switch between 'near' mode and 'offset' mode. | |
3876 | */ | |
3877 | struct r10conf *conf = mddev->private; | |
3878 | struct geom geo; | |
3879 | ||
3880 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
3881 | return -EINVAL; | |
3882 | ||
3883 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
3884 | /* mustn't change number of copies */ | |
3885 | return -EINVAL; | |
3886 | if (geo.far_copies > 1 && !geo.far_offset) | |
3887 | /* Cannot switch to 'far' mode */ | |
3888 | return -EINVAL; | |
3889 | ||
3890 | if (mddev->array_sectors & geo.chunk_mask) | |
3891 | /* not factor of array size */ | |
3892 | return -EINVAL; | |
3893 | ||
3ea7daa5 N |
3894 | if (!enough(conf, -1)) |
3895 | return -EINVAL; | |
3896 | ||
3897 | kfree(conf->mirrors_new); | |
3898 | conf->mirrors_new = NULL; | |
3899 | if (mddev->delta_disks > 0) { | |
3900 | /* allocate new 'mirrors' list */ | |
3901 | conf->mirrors_new = kzalloc( | |
dc280d98 | 3902 | sizeof(struct raid10_info) |
3ea7daa5 N |
3903 | *(mddev->raid_disks + |
3904 | mddev->delta_disks), | |
3905 | GFP_KERNEL); | |
3906 | if (!conf->mirrors_new) | |
3907 | return -ENOMEM; | |
3908 | } | |
3909 | return 0; | |
3910 | } | |
3911 | ||
3912 | /* | |
3913 | * Need to check if array has failed when deciding whether to: | |
3914 | * - start an array | |
3915 | * - remove non-faulty devices | |
3916 | * - add a spare | |
3917 | * - allow a reshape | |
3918 | * This determination is simple when no reshape is happening. | |
3919 | * However if there is a reshape, we need to carefully check | |
3920 | * both the before and after sections. | |
3921 | * This is because some failed devices may only affect one | |
3922 | * of the two sections, and some non-in_sync devices may | |
3923 | * be insync in the section most affected by failed devices. | |
3924 | */ | |
3925 | static int calc_degraded(struct r10conf *conf) | |
3926 | { | |
3927 | int degraded, degraded2; | |
3928 | int i; | |
3929 | ||
3930 | rcu_read_lock(); | |
3931 | degraded = 0; | |
3932 | /* 'prev' section first */ | |
3933 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
3934 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
3935 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
3936 | degraded++; | |
3937 | else if (!test_bit(In_sync, &rdev->flags)) | |
3938 | /* When we can reduce the number of devices in | |
3939 | * an array, this might not contribute to | |
3940 | * 'degraded'. It does now. | |
3941 | */ | |
3942 | degraded++; | |
3943 | } | |
3944 | rcu_read_unlock(); | |
3945 | if (conf->geo.raid_disks == conf->prev.raid_disks) | |
3946 | return degraded; | |
3947 | rcu_read_lock(); | |
3948 | degraded2 = 0; | |
3949 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3950 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
3951 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
3952 | degraded2++; | |
3953 | else if (!test_bit(In_sync, &rdev->flags)) { | |
3954 | /* If reshape is increasing the number of devices, | |
3955 | * this section has already been recovered, so | |
3956 | * it doesn't contribute to degraded. | |
3957 | * else it does. | |
3958 | */ | |
3959 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
3960 | degraded2++; | |
3961 | } | |
3962 | } | |
3963 | rcu_read_unlock(); | |
3964 | if (degraded2 > degraded) | |
3965 | return degraded2; | |
3966 | return degraded; | |
3967 | } | |
3968 | ||
3969 | static int raid10_start_reshape(struct mddev *mddev) | |
3970 | { | |
3971 | /* A 'reshape' has been requested. This commits | |
3972 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
3973 | * This also checks if there are enough spares and adds them | |
3974 | * to the array. | |
3975 | * We currently require enough spares to make the final | |
3976 | * array non-degraded. We also require that the difference | |
3977 | * between old and new data_offset - on each device - is | |
3978 | * enough that we never risk over-writing. | |
3979 | */ | |
3980 | ||
3981 | unsigned long before_length, after_length; | |
3982 | sector_t min_offset_diff = 0; | |
3983 | int first = 1; | |
3984 | struct geom new; | |
3985 | struct r10conf *conf = mddev->private; | |
3986 | struct md_rdev *rdev; | |
3987 | int spares = 0; | |
bb63a701 | 3988 | int ret; |
3ea7daa5 N |
3989 | |
3990 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
3991 | return -EBUSY; | |
3992 | ||
3993 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
3994 | return -EINVAL; | |
3995 | ||
3996 | before_length = ((1 << conf->prev.chunk_shift) * | |
3997 | conf->prev.far_copies); | |
3998 | after_length = ((1 << conf->geo.chunk_shift) * | |
3999 | conf->geo.far_copies); | |
4000 | ||
4001 | rdev_for_each(rdev, mddev) { | |
4002 | if (!test_bit(In_sync, &rdev->flags) | |
4003 | && !test_bit(Faulty, &rdev->flags)) | |
4004 | spares++; | |
4005 | if (rdev->raid_disk >= 0) { | |
4006 | long long diff = (rdev->new_data_offset | |
4007 | - rdev->data_offset); | |
4008 | if (!mddev->reshape_backwards) | |
4009 | diff = -diff; | |
4010 | if (diff < 0) | |
4011 | diff = 0; | |
4012 | if (first || diff < min_offset_diff) | |
4013 | min_offset_diff = diff; | |
4014 | } | |
4015 | } | |
4016 | ||
4017 | if (max(before_length, after_length) > min_offset_diff) | |
4018 | return -EINVAL; | |
4019 | ||
4020 | if (spares < mddev->delta_disks) | |
4021 | return -EINVAL; | |
4022 | ||
4023 | conf->offset_diff = min_offset_diff; | |
4024 | spin_lock_irq(&conf->device_lock); | |
4025 | if (conf->mirrors_new) { | |
4026 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4027 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4028 | smp_mb(); |
c4796e21 | 4029 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4030 | conf->mirrors_old = conf->mirrors; |
4031 | conf->mirrors = conf->mirrors_new; | |
4032 | conf->mirrors_new = NULL; | |
4033 | } | |
4034 | setup_geo(&conf->geo, mddev, geo_start); | |
4035 | smp_mb(); | |
4036 | if (mddev->reshape_backwards) { | |
4037 | sector_t size = raid10_size(mddev, 0, 0); | |
4038 | if (size < mddev->array_sectors) { | |
4039 | spin_unlock_irq(&conf->device_lock); | |
4040 | printk(KERN_ERR "md/raid10:%s: array size must be reduce before number of disks\n", | |
4041 | mdname(mddev)); | |
4042 | return -EINVAL; | |
4043 | } | |
4044 | mddev->resync_max_sectors = size; | |
4045 | conf->reshape_progress = size; | |
4046 | } else | |
4047 | conf->reshape_progress = 0; | |
299b0685 | 4048 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4049 | spin_unlock_irq(&conf->device_lock); |
4050 | ||
bb63a701 N |
4051 | if (mddev->delta_disks && mddev->bitmap) { |
4052 | ret = bitmap_resize(mddev->bitmap, | |
4053 | raid10_size(mddev, 0, | |
4054 | conf->geo.raid_disks), | |
4055 | 0, 0); | |
4056 | if (ret) | |
4057 | goto abort; | |
4058 | } | |
3ea7daa5 N |
4059 | if (mddev->delta_disks > 0) { |
4060 | rdev_for_each(rdev, mddev) | |
4061 | if (rdev->raid_disk < 0 && | |
4062 | !test_bit(Faulty, &rdev->flags)) { | |
4063 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4064 | if (rdev->raid_disk >= | |
4065 | conf->prev.raid_disks) | |
4066 | set_bit(In_sync, &rdev->flags); | |
4067 | else | |
4068 | rdev->recovery_offset = 0; | |
4069 | ||
4070 | if (sysfs_link_rdev(mddev, rdev)) | |
4071 | /* Failure here is OK */; | |
4072 | } | |
4073 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4074 | && !test_bit(Faulty, &rdev->flags)) { | |
4075 | /* This is a spare that was manually added */ | |
4076 | set_bit(In_sync, &rdev->flags); | |
4077 | } | |
4078 | } | |
4079 | /* When a reshape changes the number of devices, | |
4080 | * ->degraded is measured against the larger of the | |
4081 | * pre and post numbers. | |
4082 | */ | |
4083 | spin_lock_irq(&conf->device_lock); | |
4084 | mddev->degraded = calc_degraded(conf); | |
4085 | spin_unlock_irq(&conf->device_lock); | |
4086 | mddev->raid_disks = conf->geo.raid_disks; | |
4087 | mddev->reshape_position = conf->reshape_progress; | |
4088 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
4089 | ||
4090 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4091 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4092 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 N |
4093 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
4094 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4095 | ||
4096 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4097 | "reshape"); | |
4098 | if (!mddev->sync_thread) { | |
bb63a701 N |
4099 | ret = -EAGAIN; |
4100 | goto abort; | |
3ea7daa5 N |
4101 | } |
4102 | conf->reshape_checkpoint = jiffies; | |
4103 | md_wakeup_thread(mddev->sync_thread); | |
4104 | md_new_event(mddev); | |
4105 | return 0; | |
bb63a701 N |
4106 | |
4107 | abort: | |
4108 | mddev->recovery = 0; | |
4109 | spin_lock_irq(&conf->device_lock); | |
4110 | conf->geo = conf->prev; | |
4111 | mddev->raid_disks = conf->geo.raid_disks; | |
4112 | rdev_for_each(rdev, mddev) | |
4113 | rdev->new_data_offset = rdev->data_offset; | |
4114 | smp_wmb(); | |
4115 | conf->reshape_progress = MaxSector; | |
299b0685 | 4116 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4117 | mddev->reshape_position = MaxSector; |
4118 | spin_unlock_irq(&conf->device_lock); | |
4119 | return ret; | |
3ea7daa5 N |
4120 | } |
4121 | ||
4122 | /* Calculate the last device-address that could contain | |
4123 | * any block from the chunk that includes the array-address 's' | |
4124 | * and report the next address. | |
4125 | * i.e. the address returned will be chunk-aligned and after | |
4126 | * any data that is in the chunk containing 's'. | |
4127 | */ | |
4128 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4129 | { | |
4130 | s = (s | geo->chunk_mask) + 1; | |
4131 | s >>= geo->chunk_shift; | |
4132 | s *= geo->near_copies; | |
4133 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4134 | s *= geo->far_copies; | |
4135 | s <<= geo->chunk_shift; | |
4136 | return s; | |
4137 | } | |
4138 | ||
4139 | /* Calculate the first device-address that could contain | |
4140 | * any block from the chunk that includes the array-address 's'. | |
4141 | * This too will be the start of a chunk | |
4142 | */ | |
4143 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4144 | { | |
4145 | s >>= geo->chunk_shift; | |
4146 | s *= geo->near_copies; | |
4147 | sector_div(s, geo->raid_disks); | |
4148 | s *= geo->far_copies; | |
4149 | s <<= geo->chunk_shift; | |
4150 | return s; | |
4151 | } | |
4152 | ||
4153 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4154 | int *skipped) | |
4155 | { | |
4156 | /* We simply copy at most one chunk (smallest of old and new) | |
4157 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4158 | * or we hit a bad block or something. | |
4159 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4160 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4161 | * can record any location. |
4162 | * | |
4163 | * If we will want to write to a location that isn't | |
4164 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4165 | * we need to flush all reshape requests and update the metadata. | |
4166 | * | |
4167 | * When reshaping forwards (e.g. to more devices), we interpret | |
4168 | * 'safe' as the earliest block which might not have been copied | |
4169 | * down yet. We divide this by previous stripe size and multiply | |
4170 | * by previous stripe length to get lowest device offset that we | |
4171 | * cannot write to yet. | |
4172 | * We interpret 'sector_nr' as an address that we want to write to. | |
4173 | * From this we use last_device_address() to find where we might | |
4174 | * write to, and first_device_address on the 'safe' position. | |
4175 | * If this 'next' write position is after the 'safe' position, | |
4176 | * we must update the metadata to increase the 'safe' position. | |
4177 | * | |
4178 | * When reshaping backwards, we round in the opposite direction | |
4179 | * and perform the reverse test: next write position must not be | |
4180 | * less than current safe position. | |
4181 | * | |
4182 | * In all this the minimum difference in data offsets | |
4183 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4184 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4185 | * |
4186 | * We need to prepare all the bios here before we start any IO | |
4187 | * to ensure the size we choose is acceptable to all devices. | |
4188 | * The means one for each copy for write-out and an extra one for | |
4189 | * read-in. | |
4190 | * We store the read-in bio in ->master_bio and the others in | |
4191 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4192 | */ | |
4193 | struct r10conf *conf = mddev->private; | |
4194 | struct r10bio *r10_bio; | |
4195 | sector_t next, safe, last; | |
4196 | int max_sectors; | |
4197 | int nr_sectors; | |
4198 | int s; | |
4199 | struct md_rdev *rdev; | |
4200 | int need_flush = 0; | |
4201 | struct bio *blist; | |
4202 | struct bio *bio, *read_bio; | |
4203 | int sectors_done = 0; | |
4204 | ||
4205 | if (sector_nr == 0) { | |
4206 | /* If restarting in the middle, skip the initial sectors */ | |
4207 | if (mddev->reshape_backwards && | |
4208 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4209 | sector_nr = (raid10_size(mddev, 0, 0) | |
4210 | - conf->reshape_progress); | |
4211 | } else if (!mddev->reshape_backwards && | |
4212 | conf->reshape_progress > 0) | |
4213 | sector_nr = conf->reshape_progress; | |
4214 | if (sector_nr) { | |
4215 | mddev->curr_resync_completed = sector_nr; | |
4216 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
4217 | *skipped = 1; | |
4218 | return sector_nr; | |
4219 | } | |
4220 | } | |
4221 | ||
4222 | /* We don't use sector_nr to track where we are up to | |
4223 | * as that doesn't work well for ->reshape_backwards. | |
4224 | * So just use ->reshape_progress. | |
4225 | */ | |
4226 | if (mddev->reshape_backwards) { | |
4227 | /* 'next' is the earliest device address that we might | |
4228 | * write to for this chunk in the new layout | |
4229 | */ | |
4230 | next = first_dev_address(conf->reshape_progress - 1, | |
4231 | &conf->geo); | |
4232 | ||
4233 | /* 'safe' is the last device address that we might read from | |
4234 | * in the old layout after a restart | |
4235 | */ | |
4236 | safe = last_dev_address(conf->reshape_safe - 1, | |
4237 | &conf->prev); | |
4238 | ||
4239 | if (next + conf->offset_diff < safe) | |
4240 | need_flush = 1; | |
4241 | ||
4242 | last = conf->reshape_progress - 1; | |
4243 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4244 | & conf->prev.chunk_mask); | |
4245 | if (sector_nr + RESYNC_BLOCK_SIZE/512 < last) | |
4246 | sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512; | |
4247 | } else { | |
4248 | /* 'next' is after the last device address that we | |
4249 | * might write to for this chunk in the new layout | |
4250 | */ | |
4251 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4252 | ||
4253 | /* 'safe' is the earliest device address that we might | |
4254 | * read from in the old layout after a restart | |
4255 | */ | |
4256 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4257 | ||
4258 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4259 | * as that would possibly corrupt data | |
4260 | */ | |
4261 | if (next > safe + conf->offset_diff) | |
4262 | need_flush = 1; | |
4263 | ||
4264 | sector_nr = conf->reshape_progress; | |
4265 | last = sector_nr | (conf->geo.chunk_mask | |
4266 | & conf->prev.chunk_mask); | |
4267 | ||
4268 | if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last) | |
4269 | last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1; | |
4270 | } | |
4271 | ||
4272 | if (need_flush || | |
4273 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4274 | /* Need to update reshape_position in metadata */ | |
4275 | wait_barrier(conf); | |
4276 | mddev->reshape_position = conf->reshape_progress; | |
4277 | if (mddev->reshape_backwards) | |
4278 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4279 | - conf->reshape_progress; | |
4280 | else | |
4281 | mddev->curr_resync_completed = conf->reshape_progress; | |
4282 | conf->reshape_checkpoint = jiffies; | |
4283 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
4284 | md_wakeup_thread(mddev->thread); | |
4285 | wait_event(mddev->sb_wait, mddev->flags == 0 || | |
c91abf5a N |
4286 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4287 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4288 | allow_barrier(conf); | |
4289 | return sectors_done; | |
4290 | } | |
3ea7daa5 N |
4291 | conf->reshape_safe = mddev->reshape_position; |
4292 | allow_barrier(conf); | |
4293 | } | |
4294 | ||
4295 | read_more: | |
4296 | /* Now schedule reads for blocks from sector_nr to last */ | |
4297 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); | |
cb8b12b5 | 4298 | r10_bio->state = 0; |
3ea7daa5 N |
4299 | raise_barrier(conf, sectors_done != 0); |
4300 | atomic_set(&r10_bio->remaining, 0); | |
4301 | r10_bio->mddev = mddev; | |
4302 | r10_bio->sector = sector_nr; | |
4303 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4304 | r10_bio->sectors = last - sector_nr + 1; | |
4305 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4306 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4307 | ||
4308 | if (!rdev) { | |
4309 | /* Cannot read from here, so need to record bad blocks | |
4310 | * on all the target devices. | |
4311 | */ | |
4312 | // FIXME | |
e337aead | 4313 | mempool_free(r10_bio, conf->r10buf_pool); |
3ea7daa5 N |
4314 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4315 | return sectors_done; | |
4316 | } | |
4317 | ||
4318 | read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev); | |
4319 | ||
4320 | read_bio->bi_bdev = rdev->bdev; | |
4f024f37 | 4321 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4322 | + rdev->data_offset); |
4323 | read_bio->bi_private = r10_bio; | |
4324 | read_bio->bi_end_io = end_sync_read; | |
4325 | read_bio->bi_rw = READ; | |
ce0b0a46 | 4326 | read_bio->bi_flags &= (~0UL << BIO_RESET_BITS); |
4246a0b6 | 4327 | read_bio->bi_error = 0; |
3ea7daa5 | 4328 | read_bio->bi_vcnt = 0; |
4f024f37 | 4329 | read_bio->bi_iter.bi_size = 0; |
3ea7daa5 N |
4330 | r10_bio->master_bio = read_bio; |
4331 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4332 | ||
4333 | /* Now find the locations in the new layout */ | |
4334 | __raid10_find_phys(&conf->geo, r10_bio); | |
4335 | ||
4336 | blist = read_bio; | |
4337 | read_bio->bi_next = NULL; | |
4338 | ||
4339 | for (s = 0; s < conf->copies*2; s++) { | |
4340 | struct bio *b; | |
4341 | int d = r10_bio->devs[s/2].devnum; | |
4342 | struct md_rdev *rdev2; | |
4343 | if (s&1) { | |
4344 | rdev2 = conf->mirrors[d].replacement; | |
4345 | b = r10_bio->devs[s/2].repl_bio; | |
4346 | } else { | |
4347 | rdev2 = conf->mirrors[d].rdev; | |
4348 | b = r10_bio->devs[s/2].bio; | |
4349 | } | |
4350 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4351 | continue; | |
8be185f2 KO |
4352 | |
4353 | bio_reset(b); | |
3ea7daa5 | 4354 | b->bi_bdev = rdev2->bdev; |
4f024f37 KO |
4355 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4356 | rdev2->new_data_offset; | |
3ea7daa5 N |
4357 | b->bi_private = r10_bio; |
4358 | b->bi_end_io = end_reshape_write; | |
4359 | b->bi_rw = WRITE; | |
3ea7daa5 | 4360 | b->bi_next = blist; |
3ea7daa5 N |
4361 | blist = b; |
4362 | } | |
4363 | ||
4364 | /* Now add as many pages as possible to all of these bios. */ | |
4365 | ||
4366 | nr_sectors = 0; | |
4367 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { | |
4368 | struct page *page = r10_bio->devs[0].bio->bi_io_vec[s/(PAGE_SIZE>>9)].bv_page; | |
4369 | int len = (max_sectors - s) << 9; | |
4370 | if (len > PAGE_SIZE) | |
4371 | len = PAGE_SIZE; | |
4372 | for (bio = blist; bio ; bio = bio->bi_next) { | |
4373 | struct bio *bio2; | |
4374 | if (bio_add_page(bio, page, len, 0)) | |
4375 | continue; | |
4376 | ||
4377 | /* Didn't fit, must stop */ | |
4378 | for (bio2 = blist; | |
4379 | bio2 && bio2 != bio; | |
4380 | bio2 = bio2->bi_next) { | |
4381 | /* Remove last page from this bio */ | |
4382 | bio2->bi_vcnt--; | |
4f024f37 | 4383 | bio2->bi_iter.bi_size -= len; |
b7c44ed9 | 4384 | bio_clear_flag(bio2, BIO_SEG_VALID); |
3ea7daa5 N |
4385 | } |
4386 | goto bio_full; | |
4387 | } | |
4388 | sector_nr += len >> 9; | |
4389 | nr_sectors += len >> 9; | |
4390 | } | |
4391 | bio_full: | |
4392 | r10_bio->sectors = nr_sectors; | |
4393 | ||
4394 | /* Now submit the read */ | |
4395 | md_sync_acct(read_bio->bi_bdev, r10_bio->sectors); | |
4396 | atomic_inc(&r10_bio->remaining); | |
4397 | read_bio->bi_next = NULL; | |
4398 | generic_make_request(read_bio); | |
4399 | sector_nr += nr_sectors; | |
4400 | sectors_done += nr_sectors; | |
4401 | if (sector_nr <= last) | |
4402 | goto read_more; | |
4403 | ||
4404 | /* Now that we have done the whole section we can | |
4405 | * update reshape_progress | |
4406 | */ | |
4407 | if (mddev->reshape_backwards) | |
4408 | conf->reshape_progress -= sectors_done; | |
4409 | else | |
4410 | conf->reshape_progress += sectors_done; | |
4411 | ||
4412 | return sectors_done; | |
4413 | } | |
4414 | ||
4415 | static void end_reshape_request(struct r10bio *r10_bio); | |
4416 | static int handle_reshape_read_error(struct mddev *mddev, | |
4417 | struct r10bio *r10_bio); | |
4418 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4419 | { | |
4420 | /* Reshape read completed. Hopefully we have a block | |
4421 | * to write out. | |
4422 | * If we got a read error then we do sync 1-page reads from | |
4423 | * elsewhere until we find the data - or give up. | |
4424 | */ | |
4425 | struct r10conf *conf = mddev->private; | |
4426 | int s; | |
4427 | ||
4428 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4429 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4430 | /* Reshape has been aborted */ | |
4431 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4432 | return; | |
4433 | } | |
4434 | ||
4435 | /* We definitely have the data in the pages, schedule the | |
4436 | * writes. | |
4437 | */ | |
4438 | atomic_set(&r10_bio->remaining, 1); | |
4439 | for (s = 0; s < conf->copies*2; s++) { | |
4440 | struct bio *b; | |
4441 | int d = r10_bio->devs[s/2].devnum; | |
4442 | struct md_rdev *rdev; | |
4443 | if (s&1) { | |
4444 | rdev = conf->mirrors[d].replacement; | |
4445 | b = r10_bio->devs[s/2].repl_bio; | |
4446 | } else { | |
4447 | rdev = conf->mirrors[d].rdev; | |
4448 | b = r10_bio->devs[s/2].bio; | |
4449 | } | |
4450 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4451 | continue; | |
4452 | atomic_inc(&rdev->nr_pending); | |
4453 | md_sync_acct(b->bi_bdev, r10_bio->sectors); | |
4454 | atomic_inc(&r10_bio->remaining); | |
4455 | b->bi_next = NULL; | |
4456 | generic_make_request(b); | |
4457 | } | |
4458 | end_reshape_request(r10_bio); | |
4459 | } | |
4460 | ||
4461 | static void end_reshape(struct r10conf *conf) | |
4462 | { | |
4463 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
4464 | return; | |
4465 | ||
4466 | spin_lock_irq(&conf->device_lock); | |
4467 | conf->prev = conf->geo; | |
4468 | md_finish_reshape(conf->mddev); | |
4469 | smp_wmb(); | |
4470 | conf->reshape_progress = MaxSector; | |
299b0685 | 4471 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
4472 | spin_unlock_irq(&conf->device_lock); |
4473 | ||
4474 | /* read-ahead size must cover two whole stripes, which is | |
4475 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4476 | */ | |
4477 | if (conf->mddev->queue) { | |
4478 | int stripe = conf->geo.raid_disks * | |
4479 | ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE); | |
4480 | stripe /= conf->geo.near_copies; | |
4481 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
4482 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
4483 | } | |
4484 | conf->fullsync = 0; | |
4485 | } | |
4486 | ||
3ea7daa5 N |
4487 | static int handle_reshape_read_error(struct mddev *mddev, |
4488 | struct r10bio *r10_bio) | |
4489 | { | |
4490 | /* Use sync reads to get the blocks from somewhere else */ | |
4491 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 4492 | struct r10conf *conf = mddev->private; |
e0ee7785 N |
4493 | struct { |
4494 | struct r10bio r10_bio; | |
4495 | struct r10dev devs[conf->copies]; | |
4496 | } on_stack; | |
4497 | struct r10bio *r10b = &on_stack.r10_bio; | |
3ea7daa5 N |
4498 | int slot = 0; |
4499 | int idx = 0; | |
4500 | struct bio_vec *bvec = r10_bio->master_bio->bi_io_vec; | |
4501 | ||
e0ee7785 N |
4502 | r10b->sector = r10_bio->sector; |
4503 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
4504 | |
4505 | while (sectors) { | |
4506 | int s = sectors; | |
4507 | int success = 0; | |
4508 | int first_slot = slot; | |
4509 | ||
4510 | if (s > (PAGE_SIZE >> 9)) | |
4511 | s = PAGE_SIZE >> 9; | |
4512 | ||
4513 | while (!success) { | |
e0ee7785 | 4514 | int d = r10b->devs[slot].devnum; |
3ea7daa5 N |
4515 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
4516 | sector_t addr; | |
4517 | if (rdev == NULL || | |
4518 | test_bit(Faulty, &rdev->flags) || | |
4519 | !test_bit(In_sync, &rdev->flags)) | |
4520 | goto failed; | |
4521 | ||
e0ee7785 | 4522 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
3ea7daa5 N |
4523 | success = sync_page_io(rdev, |
4524 | addr, | |
4525 | s << 9, | |
4526 | bvec[idx].bv_page, | |
4527 | READ, false); | |
4528 | if (success) | |
4529 | break; | |
4530 | failed: | |
4531 | slot++; | |
4532 | if (slot >= conf->copies) | |
4533 | slot = 0; | |
4534 | if (slot == first_slot) | |
4535 | break; | |
4536 | } | |
4537 | if (!success) { | |
4538 | /* couldn't read this block, must give up */ | |
4539 | set_bit(MD_RECOVERY_INTR, | |
4540 | &mddev->recovery); | |
4541 | return -EIO; | |
4542 | } | |
4543 | sectors -= s; | |
4544 | idx++; | |
4545 | } | |
4546 | return 0; | |
4547 | } | |
4548 | ||
4246a0b6 | 4549 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 4550 | { |
3ea7daa5 N |
4551 | struct r10bio *r10_bio = bio->bi_private; |
4552 | struct mddev *mddev = r10_bio->mddev; | |
4553 | struct r10conf *conf = mddev->private; | |
4554 | int d; | |
4555 | int slot; | |
4556 | int repl; | |
4557 | struct md_rdev *rdev = NULL; | |
4558 | ||
4559 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
4560 | if (repl) | |
4561 | rdev = conf->mirrors[d].replacement; | |
4562 | if (!rdev) { | |
4563 | smp_mb(); | |
4564 | rdev = conf->mirrors[d].rdev; | |
4565 | } | |
4566 | ||
4246a0b6 | 4567 | if (bio->bi_error) { |
3ea7daa5 N |
4568 | /* FIXME should record badblock */ |
4569 | md_error(mddev, rdev); | |
4570 | } | |
4571 | ||
4572 | rdev_dec_pending(rdev, mddev); | |
4573 | end_reshape_request(r10_bio); | |
4574 | } | |
4575 | ||
4576 | static void end_reshape_request(struct r10bio *r10_bio) | |
4577 | { | |
4578 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
4579 | return; | |
4580 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
4581 | bio_put(r10_bio->master_bio); | |
4582 | put_buf(r10_bio); | |
4583 | } | |
4584 | ||
4585 | static void raid10_finish_reshape(struct mddev *mddev) | |
4586 | { | |
4587 | struct r10conf *conf = mddev->private; | |
4588 | ||
4589 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
4590 | return; | |
4591 | ||
4592 | if (mddev->delta_disks > 0) { | |
4593 | sector_t size = raid10_size(mddev, 0, 0); | |
4594 | md_set_array_sectors(mddev, size); | |
4595 | if (mddev->recovery_cp > mddev->resync_max_sectors) { | |
4596 | mddev->recovery_cp = mddev->resync_max_sectors; | |
4597 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4598 | } | |
4599 | mddev->resync_max_sectors = size; | |
859644f0 HM |
4600 | if (mddev->queue) { |
4601 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4602 | revalidate_disk(mddev->gendisk); | |
4603 | } | |
63aced61 N |
4604 | } else { |
4605 | int d; | |
4606 | for (d = conf->geo.raid_disks ; | |
4607 | d < conf->geo.raid_disks - mddev->delta_disks; | |
4608 | d++) { | |
4609 | struct md_rdev *rdev = conf->mirrors[d].rdev; | |
4610 | if (rdev) | |
4611 | clear_bit(In_sync, &rdev->flags); | |
4612 | rdev = conf->mirrors[d].replacement; | |
4613 | if (rdev) | |
4614 | clear_bit(In_sync, &rdev->flags); | |
4615 | } | |
3ea7daa5 N |
4616 | } |
4617 | mddev->layout = mddev->new_layout; | |
4618 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
4619 | mddev->reshape_position = MaxSector; | |
4620 | mddev->delta_disks = 0; | |
4621 | mddev->reshape_backwards = 0; | |
4622 | } | |
4623 | ||
84fc4b56 | 4624 | static struct md_personality raid10_personality = |
1da177e4 LT |
4625 | { |
4626 | .name = "raid10", | |
2604b703 | 4627 | .level = 10, |
1da177e4 | 4628 | .owner = THIS_MODULE, |
849674e4 SL |
4629 | .make_request = raid10_make_request, |
4630 | .run = raid10_run, | |
afa0f557 | 4631 | .free = raid10_free, |
849674e4 SL |
4632 | .status = raid10_status, |
4633 | .error_handler = raid10_error, | |
1da177e4 LT |
4634 | .hot_add_disk = raid10_add_disk, |
4635 | .hot_remove_disk= raid10_remove_disk, | |
4636 | .spare_active = raid10_spare_active, | |
849674e4 | 4637 | .sync_request = raid10_sync_request, |
6cce3b23 | 4638 | .quiesce = raid10_quiesce, |
80c3a6ce | 4639 | .size = raid10_size, |
006a09a0 | 4640 | .resize = raid10_resize, |
dab8b292 | 4641 | .takeover = raid10_takeover, |
3ea7daa5 N |
4642 | .check_reshape = raid10_check_reshape, |
4643 | .start_reshape = raid10_start_reshape, | |
4644 | .finish_reshape = raid10_finish_reshape, | |
5c675f83 | 4645 | .congested = raid10_congested, |
1da177e4 LT |
4646 | }; |
4647 | ||
4648 | static int __init raid_init(void) | |
4649 | { | |
2604b703 | 4650 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
4651 | } |
4652 | ||
4653 | static void raid_exit(void) | |
4654 | { | |
2604b703 | 4655 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
4656 | } |
4657 | ||
4658 | module_init(raid_init); | |
4659 | module_exit(raid_exit); | |
4660 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 4661 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 4662 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 4663 | MODULE_ALIAS("md-raid10"); |
2604b703 | 4664 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
4665 | |
4666 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |