]>
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> |
43b2e5d8 | 27 | #include "md.h" |
ef740c37 | 28 | #include "raid10.h" |
dab8b292 | 29 | #include "raid0.h" |
ef740c37 | 30 | #include "bitmap.h" |
1da177e4 LT |
31 | |
32 | /* | |
33 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
34 | * The layout of data is defined by | |
35 | * chunk_size | |
36 | * raid_disks | |
37 | * near_copies (stored in low byte of layout) | |
38 | * far_copies (stored in second byte of layout) | |
c93983bf | 39 | * far_offset (stored in bit 16 of layout ) |
1da177e4 LT |
40 | * |
41 | * The data to be stored is divided into chunks using chunksize. | |
42 | * Each device is divided into far_copies sections. | |
43 | * In each section, chunks are laid out in a style similar to raid0, but | |
44 | * near_copies copies of each chunk is stored (each on a different drive). | |
45 | * The starting device for each section is offset near_copies from the starting | |
46 | * device of the previous section. | |
c93983bf | 47 | * Thus they are (near_copies*far_copies) of each chunk, and each is on a different |
1da177e4 LT |
48 | * drive. |
49 | * near_copies and far_copies must be at least one, and their product is at most | |
50 | * raid_disks. | |
c93983bf N |
51 | * |
52 | * If far_offset is true, then the far_copies are handled a bit differently. | |
53 | * The copies are still in different stripes, but instead of be very far apart | |
54 | * on disk, there are adjacent stripes. | |
1da177e4 LT |
55 | */ |
56 | ||
57 | /* | |
58 | * Number of guaranteed r10bios in case of extreme VM load: | |
59 | */ | |
60 | #define NR_RAID10_BIOS 256 | |
61 | ||
34db0cd6 N |
62 | /* When there are this many requests queue to be written by |
63 | * the raid10 thread, we become 'congested' to provide back-pressure | |
64 | * for writeback. | |
65 | */ | |
66 | static int max_queued_requests = 1024; | |
67 | ||
e879a879 N |
68 | static void allow_barrier(struct r10conf *conf); |
69 | static void lower_barrier(struct r10conf *conf); | |
fae8cc5e | 70 | static int enough(struct r10conf *conf, int ignore); |
0a27ec96 | 71 | |
dd0fc66f | 72 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 73 | { |
e879a879 | 74 | struct r10conf *conf = data; |
9f2c9d12 | 75 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 76 | |
69335ef3 N |
77 | /* allocate a r10bio with room for raid_disks entries in the |
78 | * bios array */ | |
7eaceacc | 79 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
80 | } |
81 | ||
82 | static void r10bio_pool_free(void *r10_bio, void *data) | |
83 | { | |
84 | kfree(r10_bio); | |
85 | } | |
86 | ||
0310fa21 | 87 | /* Maximum size of each resync request */ |
1da177e4 | 88 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 89 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
90 | /* amount of memory to reserve for resync requests */ |
91 | #define RESYNC_WINDOW (1024*1024) | |
92 | /* maximum number of concurrent requests, memory permitting */ | |
93 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
94 | |
95 | /* | |
96 | * When performing a resync, we need to read and compare, so | |
97 | * we need as many pages are there are copies. | |
98 | * When performing a recovery, we need 2 bios, one for read, | |
99 | * one for write (we recover only one drive per r10buf) | |
100 | * | |
101 | */ | |
dd0fc66f | 102 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 103 | { |
e879a879 | 104 | struct r10conf *conf = data; |
1da177e4 | 105 | struct page *page; |
9f2c9d12 | 106 | struct r10bio *r10_bio; |
1da177e4 LT |
107 | struct bio *bio; |
108 | int i, j; | |
109 | int nalloc; | |
110 | ||
111 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 112 | if (!r10_bio) |
1da177e4 | 113 | return NULL; |
1da177e4 LT |
114 | |
115 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | |
116 | nalloc = conf->copies; /* resync */ | |
117 | else | |
118 | nalloc = 2; /* recovery */ | |
119 | ||
120 | /* | |
121 | * Allocate bios. | |
122 | */ | |
123 | for (j = nalloc ; j-- ; ) { | |
6746557f | 124 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
125 | if (!bio) |
126 | goto out_free_bio; | |
127 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
128 | if (!conf->have_replacement) |
129 | continue; | |
130 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
131 | if (!bio) | |
132 | goto out_free_bio; | |
133 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
134 | } |
135 | /* | |
136 | * Allocate RESYNC_PAGES data pages and attach them | |
137 | * where needed. | |
138 | */ | |
139 | for (j = 0 ; j < nalloc; j++) { | |
69335ef3 | 140 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
1da177e4 LT |
141 | bio = r10_bio->devs[j].bio; |
142 | for (i = 0; i < RESYNC_PAGES; i++) { | |
c65060ad NK |
143 | if (j == 1 && !test_bit(MD_RECOVERY_SYNC, |
144 | &conf->mddev->recovery)) { | |
145 | /* we can share bv_page's during recovery */ | |
146 | struct bio *rbio = r10_bio->devs[0].bio; | |
147 | page = rbio->bi_io_vec[i].bv_page; | |
148 | get_page(page); | |
149 | } else | |
150 | page = alloc_page(gfp_flags); | |
1da177e4 LT |
151 | if (unlikely(!page)) |
152 | goto out_free_pages; | |
153 | ||
154 | bio->bi_io_vec[i].bv_page = page; | |
69335ef3 N |
155 | if (rbio) |
156 | rbio->bi_io_vec[i].bv_page = page; | |
1da177e4 LT |
157 | } |
158 | } | |
159 | ||
160 | return r10_bio; | |
161 | ||
162 | out_free_pages: | |
163 | for ( ; i > 0 ; i--) | |
1345b1d8 | 164 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
165 | while (j--) |
166 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 167 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
168 | j = -1; |
169 | out_free_bio: | |
69335ef3 | 170 | while (++j < nalloc) { |
1da177e4 | 171 | bio_put(r10_bio->devs[j].bio); |
69335ef3 N |
172 | if (r10_bio->devs[j].repl_bio) |
173 | bio_put(r10_bio->devs[j].repl_bio); | |
174 | } | |
1da177e4 LT |
175 | r10bio_pool_free(r10_bio, conf); |
176 | return NULL; | |
177 | } | |
178 | ||
179 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
180 | { | |
181 | int i; | |
e879a879 | 182 | struct r10conf *conf = data; |
9f2c9d12 | 183 | struct r10bio *r10bio = __r10_bio; |
1da177e4 LT |
184 | int j; |
185 | ||
186 | for (j=0; j < conf->copies; j++) { | |
187 | struct bio *bio = r10bio->devs[j].bio; | |
188 | if (bio) { | |
189 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 190 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
191 | bio->bi_io_vec[i].bv_page = NULL; |
192 | } | |
193 | bio_put(bio); | |
194 | } | |
69335ef3 N |
195 | bio = r10bio->devs[j].repl_bio; |
196 | if (bio) | |
197 | bio_put(bio); | |
1da177e4 LT |
198 | } |
199 | r10bio_pool_free(r10bio, conf); | |
200 | } | |
201 | ||
e879a879 | 202 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
203 | { |
204 | int i; | |
205 | ||
206 | for (i = 0; i < conf->copies; i++) { | |
207 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 208 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
209 | bio_put(*bio); |
210 | *bio = NULL; | |
69335ef3 N |
211 | bio = &r10_bio->devs[i].repl_bio; |
212 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
213 | bio_put(*bio); | |
214 | *bio = NULL; | |
1da177e4 LT |
215 | } |
216 | } | |
217 | ||
9f2c9d12 | 218 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 219 | { |
e879a879 | 220 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 221 | |
1da177e4 LT |
222 | put_all_bios(conf, r10_bio); |
223 | mempool_free(r10_bio, conf->r10bio_pool); | |
224 | } | |
225 | ||
9f2c9d12 | 226 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 227 | { |
e879a879 | 228 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
229 | |
230 | mempool_free(r10_bio, conf->r10buf_pool); | |
231 | ||
0a27ec96 | 232 | lower_barrier(conf); |
1da177e4 LT |
233 | } |
234 | ||
9f2c9d12 | 235 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
236 | { |
237 | unsigned long flags; | |
fd01b88c | 238 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 239 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
240 | |
241 | spin_lock_irqsave(&conf->device_lock, flags); | |
242 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 243 | conf->nr_queued ++; |
1da177e4 LT |
244 | spin_unlock_irqrestore(&conf->device_lock, flags); |
245 | ||
388667be AJ |
246 | /* wake up frozen array... */ |
247 | wake_up(&conf->wait_barrier); | |
248 | ||
1da177e4 LT |
249 | md_wakeup_thread(mddev->thread); |
250 | } | |
251 | ||
252 | /* | |
253 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
254 | * operation and are ready to return a success/failure code to the buffer | |
255 | * cache layer. | |
256 | */ | |
9f2c9d12 | 257 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
258 | { |
259 | struct bio *bio = r10_bio->master_bio; | |
856e08e2 | 260 | int done; |
e879a879 | 261 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 262 | |
856e08e2 N |
263 | if (bio->bi_phys_segments) { |
264 | unsigned long flags; | |
265 | spin_lock_irqsave(&conf->device_lock, flags); | |
266 | bio->bi_phys_segments--; | |
267 | done = (bio->bi_phys_segments == 0); | |
268 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
269 | } else | |
270 | done = 1; | |
271 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
272 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
273 | if (done) { | |
274 | bio_endio(bio, 0); | |
275 | /* | |
276 | * Wake up any possible resync thread that waits for the device | |
277 | * to go idle. | |
278 | */ | |
279 | allow_barrier(conf); | |
280 | } | |
1da177e4 LT |
281 | free_r10bio(r10_bio); |
282 | } | |
283 | ||
284 | /* | |
285 | * Update disk head position estimator based on IRQ completion info. | |
286 | */ | |
9f2c9d12 | 287 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 288 | { |
e879a879 | 289 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
290 | |
291 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
292 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
293 | } | |
294 | ||
778ca018 NK |
295 | /* |
296 | * Find the disk number which triggered given bio | |
297 | */ | |
e879a879 | 298 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 299 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
300 | { |
301 | int slot; | |
69335ef3 | 302 | int repl = 0; |
778ca018 | 303 | |
69335ef3 | 304 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
305 | if (r10_bio->devs[slot].bio == bio) |
306 | break; | |
69335ef3 N |
307 | if (r10_bio->devs[slot].repl_bio == bio) { |
308 | repl = 1; | |
309 | break; | |
310 | } | |
311 | } | |
778ca018 NK |
312 | |
313 | BUG_ON(slot == conf->copies); | |
314 | update_head_pos(slot, r10_bio); | |
315 | ||
749c55e9 N |
316 | if (slotp) |
317 | *slotp = slot; | |
69335ef3 N |
318 | if (replp) |
319 | *replp = repl; | |
778ca018 NK |
320 | return r10_bio->devs[slot].devnum; |
321 | } | |
322 | ||
6712ecf8 | 323 | static void raid10_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
324 | { |
325 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 326 | struct r10bio *r10_bio = bio->bi_private; |
1da177e4 | 327 | int slot, dev; |
abbf098e | 328 | struct md_rdev *rdev; |
e879a879 | 329 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 330 | |
1da177e4 LT |
331 | |
332 | slot = r10_bio->read_slot; | |
333 | dev = r10_bio->devs[slot].devnum; | |
abbf098e | 334 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
335 | /* |
336 | * this branch is our 'one mirror IO has finished' event handler: | |
337 | */ | |
4443ae10 N |
338 | update_head_pos(slot, r10_bio); |
339 | ||
340 | if (uptodate) { | |
1da177e4 LT |
341 | /* |
342 | * Set R10BIO_Uptodate in our master bio, so that | |
343 | * we will return a good error code to the higher | |
344 | * levels even if IO on some other mirrored buffer fails. | |
345 | * | |
346 | * The 'master' represents the composite IO operation to | |
347 | * user-side. So if something waits for IO, then it will | |
348 | * wait for the 'master' bio. | |
349 | */ | |
350 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
351 | } else { |
352 | /* If all other devices that store this block have | |
353 | * failed, we want to return the error upwards rather | |
354 | * than fail the last device. Here we redefine | |
355 | * "uptodate" to mean "Don't want to retry" | |
356 | */ | |
357 | unsigned long flags; | |
358 | spin_lock_irqsave(&conf->device_lock, flags); | |
359 | if (!enough(conf, rdev->raid_disk)) | |
360 | uptodate = 1; | |
361 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
362 | } | |
363 | if (uptodate) { | |
1da177e4 | 364 | raid_end_bio_io(r10_bio); |
abbf098e | 365 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 366 | } else { |
1da177e4 | 367 | /* |
7c4e06ff | 368 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
369 | */ |
370 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
371 | printk_ratelimited(KERN_ERR |
372 | "md/raid10:%s: %s: rescheduling sector %llu\n", | |
373 | mdname(conf->mddev), | |
abbf098e | 374 | bdevname(rdev->bdev, b), |
8bda470e | 375 | (unsigned long long)r10_bio->sector); |
856e08e2 | 376 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
377 | reschedule_retry(r10_bio); |
378 | } | |
1da177e4 LT |
379 | } |
380 | ||
9f2c9d12 | 381 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
382 | { |
383 | /* clear the bitmap if all writes complete successfully */ | |
384 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
385 | r10_bio->sectors, | |
386 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
387 | 0); | |
388 | md_write_end(r10_bio->mddev); | |
389 | } | |
390 | ||
9f2c9d12 | 391 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
392 | { |
393 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
394 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
395 | reschedule_retry(r10_bio); | |
396 | else { | |
397 | close_write(r10_bio); | |
398 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
399 | reschedule_retry(r10_bio); | |
400 | else | |
401 | raid_end_bio_io(r10_bio); | |
402 | } | |
403 | } | |
404 | } | |
405 | ||
6712ecf8 | 406 | static void raid10_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
407 | { |
408 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 409 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 410 | int dev; |
749c55e9 | 411 | int dec_rdev = 1; |
e879a879 | 412 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 413 | int slot, repl; |
4ca40c2c | 414 | struct md_rdev *rdev = NULL; |
1da177e4 | 415 | |
475b0321 | 416 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 417 | |
475b0321 N |
418 | if (repl) |
419 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
420 | if (!rdev) { |
421 | smp_rmb(); | |
422 | repl = 0; | |
475b0321 | 423 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 424 | } |
1da177e4 LT |
425 | /* |
426 | * this branch is our 'one mirror IO has finished' event handler: | |
427 | */ | |
6cce3b23 | 428 | if (!uptodate) { |
475b0321 N |
429 | if (repl) |
430 | /* Never record new bad blocks to replacement, | |
431 | * just fail it. | |
432 | */ | |
433 | md_error(rdev->mddev, rdev); | |
434 | else { | |
435 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
436 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
437 | set_bit(MD_RECOVERY_NEEDED, | |
438 | &rdev->mddev->recovery); | |
475b0321 N |
439 | set_bit(R10BIO_WriteError, &r10_bio->state); |
440 | dec_rdev = 0; | |
441 | } | |
749c55e9 | 442 | } else { |
1da177e4 LT |
443 | /* |
444 | * Set R10BIO_Uptodate in our master bio, so that | |
445 | * we will return a good error code for to the higher | |
446 | * levels even if IO on some other mirrored buffer fails. | |
447 | * | |
448 | * The 'master' represents the composite IO operation to | |
449 | * user-side. So if something waits for IO, then it will | |
450 | * wait for the 'master' bio. | |
451 | */ | |
749c55e9 N |
452 | sector_t first_bad; |
453 | int bad_sectors; | |
454 | ||
1da177e4 LT |
455 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
456 | ||
749c55e9 | 457 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 458 | if (is_badblock(rdev, |
749c55e9 N |
459 | r10_bio->devs[slot].addr, |
460 | r10_bio->sectors, | |
461 | &first_bad, &bad_sectors)) { | |
462 | bio_put(bio); | |
475b0321 N |
463 | if (repl) |
464 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
465 | else | |
466 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
467 | dec_rdev = 0; |
468 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
469 | } | |
470 | } | |
471 | ||
1da177e4 LT |
472 | /* |
473 | * | |
474 | * Let's see if all mirrored write operations have finished | |
475 | * already. | |
476 | */ | |
19d5f834 | 477 | one_write_done(r10_bio); |
749c55e9 N |
478 | if (dec_rdev) |
479 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); | |
1da177e4 LT |
480 | } |
481 | ||
1da177e4 LT |
482 | /* |
483 | * RAID10 layout manager | |
25985edc | 484 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
485 | * parameters: near_copies and far_copies. |
486 | * near_copies * far_copies must be <= raid_disks. | |
487 | * Normally one of these will be 1. | |
488 | * If both are 1, we get raid0. | |
489 | * If near_copies == raid_disks, we get raid1. | |
490 | * | |
25985edc | 491 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
492 | * first chunk, followed by near_copies copies of the next chunk and |
493 | * so on. | |
494 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
495 | * as described above, we start again with a device offset of near_copies. | |
496 | * So we effectively have another copy of the whole array further down all | |
497 | * the drives, but with blocks on different drives. | |
498 | * With this layout, and block is never stored twice on the one device. | |
499 | * | |
500 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 501 | * on each device that it is on. |
1da177e4 LT |
502 | * |
503 | * raid10_find_virt does the reverse mapping, from a device and a | |
504 | * sector offset to a virtual address | |
505 | */ | |
506 | ||
f8c9e74f | 507 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
508 | { |
509 | int n,f; | |
510 | sector_t sector; | |
511 | sector_t chunk; | |
512 | sector_t stripe; | |
513 | int dev; | |
1da177e4 LT |
514 | int slot = 0; |
515 | ||
516 | /* now calculate first sector/dev */ | |
5cf00fcd N |
517 | chunk = r10bio->sector >> geo->chunk_shift; |
518 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 519 | |
5cf00fcd | 520 | chunk *= geo->near_copies; |
1da177e4 | 521 | stripe = chunk; |
5cf00fcd N |
522 | dev = sector_div(stripe, geo->raid_disks); |
523 | if (geo->far_offset) | |
524 | stripe *= geo->far_copies; | |
1da177e4 | 525 | |
5cf00fcd | 526 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
527 | |
528 | /* and calculate all the others */ | |
5cf00fcd | 529 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 LT |
530 | int d = dev; |
531 | sector_t s = sector; | |
532 | r10bio->devs[slot].addr = sector; | |
533 | r10bio->devs[slot].devnum = d; | |
534 | slot++; | |
535 | ||
5cf00fcd N |
536 | for (f = 1; f < geo->far_copies; f++) { |
537 | d += geo->near_copies; | |
538 | if (d >= geo->raid_disks) | |
539 | d -= geo->raid_disks; | |
540 | s += geo->stride; | |
1da177e4 LT |
541 | r10bio->devs[slot].devnum = d; |
542 | r10bio->devs[slot].addr = s; | |
543 | slot++; | |
544 | } | |
545 | dev++; | |
5cf00fcd | 546 | if (dev >= geo->raid_disks) { |
1da177e4 | 547 | dev = 0; |
5cf00fcd | 548 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
549 | } |
550 | } | |
f8c9e74f N |
551 | } |
552 | ||
553 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
554 | { | |
555 | struct geom *geo = &conf->geo; | |
556 | ||
557 | if (conf->reshape_progress != MaxSector && | |
558 | ((r10bio->sector >= conf->reshape_progress) != | |
559 | conf->mddev->reshape_backwards)) { | |
560 | set_bit(R10BIO_Previous, &r10bio->state); | |
561 | geo = &conf->prev; | |
562 | } else | |
563 | clear_bit(R10BIO_Previous, &r10bio->state); | |
564 | ||
565 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
566 | } |
567 | ||
e879a879 | 568 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
569 | { |
570 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
571 | /* Never use conf->prev as this is only called during resync |
572 | * or recovery, so reshape isn't happening | |
573 | */ | |
5cf00fcd | 574 | struct geom *geo = &conf->geo; |
1da177e4 | 575 | |
5cf00fcd N |
576 | offset = sector & geo->chunk_mask; |
577 | if (geo->far_offset) { | |
c93983bf | 578 | int fc; |
5cf00fcd N |
579 | chunk = sector >> geo->chunk_shift; |
580 | fc = sector_div(chunk, geo->far_copies); | |
581 | dev -= fc * geo->near_copies; | |
c93983bf | 582 | if (dev < 0) |
5cf00fcd | 583 | dev += geo->raid_disks; |
c93983bf | 584 | } else { |
5cf00fcd N |
585 | while (sector >= geo->stride) { |
586 | sector -= geo->stride; | |
587 | if (dev < geo->near_copies) | |
588 | dev += geo->raid_disks - geo->near_copies; | |
c93983bf | 589 | else |
5cf00fcd | 590 | dev -= geo->near_copies; |
c93983bf | 591 | } |
5cf00fcd | 592 | chunk = sector >> geo->chunk_shift; |
c93983bf | 593 | } |
5cf00fcd N |
594 | vchunk = chunk * geo->raid_disks + dev; |
595 | sector_div(vchunk, geo->near_copies); | |
596 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
597 | } |
598 | ||
599 | /** | |
600 | * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged | |
601 | * @q: request queue | |
cc371e66 | 602 | * @bvm: properties of new bio |
1da177e4 LT |
603 | * @biovec: the request that could be merged to it. |
604 | * | |
605 | * Return amount of bytes we can accept at this offset | |
050b6615 N |
606 | * This requires checking for end-of-chunk if near_copies != raid_disks, |
607 | * and for subordinate merge_bvec_fns if merge_check_needed. | |
1da177e4 | 608 | */ |
cc371e66 AK |
609 | static int raid10_mergeable_bvec(struct request_queue *q, |
610 | struct bvec_merge_data *bvm, | |
611 | struct bio_vec *biovec) | |
1da177e4 | 612 | { |
fd01b88c | 613 | struct mddev *mddev = q->queuedata; |
050b6615 | 614 | struct r10conf *conf = mddev->private; |
cc371e66 | 615 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
1da177e4 | 616 | int max; |
9d8f0363 | 617 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 618 | unsigned int bio_sectors = bvm->bi_size >> 9; |
5cf00fcd | 619 | struct geom *geo = &conf->geo; |
1da177e4 | 620 | |
f8c9e74f N |
621 | if (conf->reshape_progress != MaxSector && |
622 | ((sector >= conf->reshape_progress) != | |
623 | conf->mddev->reshape_backwards)) | |
624 | geo = &conf->prev; | |
625 | ||
5cf00fcd | 626 | if (geo->near_copies < geo->raid_disks) { |
050b6615 N |
627 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) |
628 | + bio_sectors)) << 9; | |
629 | if (max < 0) | |
630 | /* bio_add cannot handle a negative return */ | |
631 | max = 0; | |
632 | if (max <= biovec->bv_len && bio_sectors == 0) | |
633 | return biovec->bv_len; | |
634 | } else | |
635 | max = biovec->bv_len; | |
636 | ||
637 | if (mddev->merge_check_needed) { | |
638 | struct r10bio r10_bio; | |
639 | int s; | |
f8c9e74f N |
640 | if (conf->reshape_progress != MaxSector) { |
641 | /* Cannot give any guidance during reshape */ | |
642 | if (max <= biovec->bv_len && bio_sectors == 0) | |
643 | return biovec->bv_len; | |
644 | return 0; | |
645 | } | |
050b6615 N |
646 | r10_bio.sector = sector; |
647 | raid10_find_phys(conf, &r10_bio); | |
648 | rcu_read_lock(); | |
649 | for (s = 0; s < conf->copies; s++) { | |
650 | int disk = r10_bio.devs[s].devnum; | |
651 | struct md_rdev *rdev = rcu_dereference( | |
652 | conf->mirrors[disk].rdev); | |
653 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
654 | struct request_queue *q = | |
655 | bdev_get_queue(rdev->bdev); | |
656 | if (q->merge_bvec_fn) { | |
657 | bvm->bi_sector = r10_bio.devs[s].addr | |
658 | + rdev->data_offset; | |
659 | bvm->bi_bdev = rdev->bdev; | |
660 | max = min(max, q->merge_bvec_fn( | |
661 | q, bvm, biovec)); | |
662 | } | |
663 | } | |
664 | rdev = rcu_dereference(conf->mirrors[disk].replacement); | |
665 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
666 | struct request_queue *q = | |
667 | bdev_get_queue(rdev->bdev); | |
668 | if (q->merge_bvec_fn) { | |
669 | bvm->bi_sector = r10_bio.devs[s].addr | |
670 | + rdev->data_offset; | |
671 | bvm->bi_bdev = rdev->bdev; | |
672 | max = min(max, q->merge_bvec_fn( | |
673 | q, bvm, biovec)); | |
674 | } | |
675 | } | |
676 | } | |
677 | rcu_read_unlock(); | |
678 | } | |
679 | return max; | |
1da177e4 LT |
680 | } |
681 | ||
682 | /* | |
683 | * This routine returns the disk from which the requested read should | |
684 | * be done. There is a per-array 'next expected sequential IO' sector | |
685 | * number - if this matches on the next IO then we use the last disk. | |
686 | * There is also a per-disk 'last know head position' sector that is | |
687 | * maintained from IRQ contexts, both the normal and the resync IO | |
688 | * completion handlers update this position correctly. If there is no | |
689 | * perfect sequential match then we pick the disk whose head is closest. | |
690 | * | |
691 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
692 | * because position is mirror, not device based. | |
693 | * | |
694 | * The rdev for the device selected will have nr_pending incremented. | |
695 | */ | |
696 | ||
697 | /* | |
698 | * FIXME: possibly should rethink readbalancing and do it differently | |
699 | * depending on near_copies / far_copies geometry. | |
700 | */ | |
96c3fd1f N |
701 | static struct md_rdev *read_balance(struct r10conf *conf, |
702 | struct r10bio *r10_bio, | |
703 | int *max_sectors) | |
1da177e4 | 704 | { |
af3a2cd6 | 705 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 706 | int disk, slot; |
856e08e2 N |
707 | int sectors = r10_bio->sectors; |
708 | int best_good_sectors; | |
56d99121 | 709 | sector_t new_distance, best_dist; |
abbf098e | 710 | struct md_rdev *rdev, *best_rdev; |
56d99121 N |
711 | int do_balance; |
712 | int best_slot; | |
5cf00fcd | 713 | struct geom *geo = &conf->geo; |
1da177e4 LT |
714 | |
715 | raid10_find_phys(conf, r10_bio); | |
716 | rcu_read_lock(); | |
56d99121 | 717 | retry: |
856e08e2 | 718 | sectors = r10_bio->sectors; |
56d99121 | 719 | best_slot = -1; |
abbf098e | 720 | best_rdev = NULL; |
56d99121 | 721 | best_dist = MaxSector; |
856e08e2 | 722 | best_good_sectors = 0; |
56d99121 | 723 | do_balance = 1; |
1da177e4 LT |
724 | /* |
725 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
726 | * device if no resync is going on (recovery is ok), or below |
727 | * the resync window. We take the first readable disk when | |
728 | * above the resync window. | |
1da177e4 LT |
729 | */ |
730 | if (conf->mddev->recovery_cp < MaxSector | |
56d99121 N |
731 | && (this_sector + sectors >= conf->next_resync)) |
732 | do_balance = 0; | |
1da177e4 | 733 | |
56d99121 | 734 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
735 | sector_t first_bad; |
736 | int bad_sectors; | |
737 | sector_t dev_sector; | |
738 | ||
56d99121 N |
739 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
740 | continue; | |
1da177e4 | 741 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
742 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
743 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
050b6615 | 744 | test_bit(Unmerged, &rdev->flags) || |
abbf098e N |
745 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) |
746 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 N |
747 | if (rdev == NULL || |
748 | test_bit(Faulty, &rdev->flags) || | |
749 | test_bit(Unmerged, &rdev->flags)) | |
abbf098e N |
750 | continue; |
751 | if (!test_bit(In_sync, &rdev->flags) && | |
752 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
753 | continue; |
754 | ||
856e08e2 N |
755 | dev_sector = r10_bio->devs[slot].addr; |
756 | if (is_badblock(rdev, dev_sector, sectors, | |
757 | &first_bad, &bad_sectors)) { | |
758 | if (best_dist < MaxSector) | |
759 | /* Already have a better slot */ | |
760 | continue; | |
761 | if (first_bad <= dev_sector) { | |
762 | /* Cannot read here. If this is the | |
763 | * 'primary' device, then we must not read | |
764 | * beyond 'bad_sectors' from another device. | |
765 | */ | |
766 | bad_sectors -= (dev_sector - first_bad); | |
767 | if (!do_balance && sectors > bad_sectors) | |
768 | sectors = bad_sectors; | |
769 | if (best_good_sectors > sectors) | |
770 | best_good_sectors = sectors; | |
771 | } else { | |
772 | sector_t good_sectors = | |
773 | first_bad - dev_sector; | |
774 | if (good_sectors > best_good_sectors) { | |
775 | best_good_sectors = good_sectors; | |
776 | best_slot = slot; | |
abbf098e | 777 | best_rdev = rdev; |
856e08e2 N |
778 | } |
779 | if (!do_balance) | |
780 | /* Must read from here */ | |
781 | break; | |
782 | } | |
783 | continue; | |
784 | } else | |
785 | best_good_sectors = sectors; | |
786 | ||
56d99121 N |
787 | if (!do_balance) |
788 | break; | |
1da177e4 | 789 | |
22dfdf52 N |
790 | /* This optimisation is debatable, and completely destroys |
791 | * sequential read speed for 'far copies' arrays. So only | |
792 | * keep it for 'near' arrays, and review those later. | |
793 | */ | |
5cf00fcd | 794 | if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
1da177e4 | 795 | break; |
8ed3a195 KS |
796 | |
797 | /* for far > 1 always use the lowest address */ | |
5cf00fcd | 798 | if (geo->far_copies > 1) |
56d99121 | 799 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 800 | else |
56d99121 N |
801 | new_distance = abs(r10_bio->devs[slot].addr - |
802 | conf->mirrors[disk].head_position); | |
803 | if (new_distance < best_dist) { | |
804 | best_dist = new_distance; | |
805 | best_slot = slot; | |
abbf098e | 806 | best_rdev = rdev; |
1da177e4 LT |
807 | } |
808 | } | |
abbf098e | 809 | if (slot >= conf->copies) { |
56d99121 | 810 | slot = best_slot; |
abbf098e N |
811 | rdev = best_rdev; |
812 | } | |
1da177e4 | 813 | |
56d99121 | 814 | if (slot >= 0) { |
56d99121 N |
815 | atomic_inc(&rdev->nr_pending); |
816 | if (test_bit(Faulty, &rdev->flags)) { | |
817 | /* Cannot risk returning a device that failed | |
818 | * before we inc'ed nr_pending | |
819 | */ | |
820 | rdev_dec_pending(rdev, conf->mddev); | |
821 | goto retry; | |
822 | } | |
823 | r10_bio->read_slot = slot; | |
824 | } else | |
96c3fd1f | 825 | rdev = NULL; |
1da177e4 | 826 | rcu_read_unlock(); |
856e08e2 | 827 | *max_sectors = best_good_sectors; |
1da177e4 | 828 | |
96c3fd1f | 829 | return rdev; |
1da177e4 LT |
830 | } |
831 | ||
0d129228 N |
832 | static int raid10_congested(void *data, int bits) |
833 | { | |
fd01b88c | 834 | struct mddev *mddev = data; |
e879a879 | 835 | struct r10conf *conf = mddev->private; |
0d129228 N |
836 | int i, ret = 0; |
837 | ||
34db0cd6 N |
838 | if ((bits & (1 << BDI_async_congested)) && |
839 | conf->pending_count >= max_queued_requests) | |
840 | return 1; | |
841 | ||
3fa841d7 N |
842 | if (mddev_congested(mddev, bits)) |
843 | return 1; | |
0d129228 | 844 | rcu_read_lock(); |
f8c9e74f N |
845 | for (i = 0; |
846 | (i < conf->geo.raid_disks || i < conf->prev.raid_disks) | |
847 | && ret == 0; | |
848 | i++) { | |
3cb03002 | 849 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 850 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 851 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
852 | |
853 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
854 | } | |
855 | } | |
856 | rcu_read_unlock(); | |
857 | return ret; | |
858 | } | |
859 | ||
e879a879 | 860 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
861 | { |
862 | /* Any writes that have been queued but are awaiting | |
863 | * bitmap updates get flushed here. | |
a35e63ef | 864 | */ |
a35e63ef N |
865 | spin_lock_irq(&conf->device_lock); |
866 | ||
867 | if (conf->pending_bio_list.head) { | |
868 | struct bio *bio; | |
869 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 870 | conf->pending_count = 0; |
a35e63ef N |
871 | spin_unlock_irq(&conf->device_lock); |
872 | /* flush any pending bitmap writes to disk | |
873 | * before proceeding w/ I/O */ | |
874 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 875 | wake_up(&conf->wait_barrier); |
a35e63ef N |
876 | |
877 | while (bio) { /* submit pending writes */ | |
878 | struct bio *next = bio->bi_next; | |
879 | bio->bi_next = NULL; | |
880 | generic_make_request(bio); | |
881 | bio = next; | |
882 | } | |
a35e63ef N |
883 | } else |
884 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 885 | } |
7eaceacc | 886 | |
0a27ec96 N |
887 | /* Barriers.... |
888 | * Sometimes we need to suspend IO while we do something else, | |
889 | * either some resync/recovery, or reconfigure the array. | |
890 | * To do this we raise a 'barrier'. | |
891 | * The 'barrier' is a counter that can be raised multiple times | |
892 | * to count how many activities are happening which preclude | |
893 | * normal IO. | |
894 | * We can only raise the barrier if there is no pending IO. | |
895 | * i.e. if nr_pending == 0. | |
896 | * We choose only to raise the barrier if no-one is waiting for the | |
897 | * barrier to go down. This means that as soon as an IO request | |
898 | * is ready, no other operations which require a barrier will start | |
899 | * until the IO request has had a chance. | |
900 | * | |
901 | * So: regular IO calls 'wait_barrier'. When that returns there | |
902 | * is no backgroup IO happening, It must arrange to call | |
903 | * allow_barrier when it has finished its IO. | |
904 | * backgroup IO calls must call raise_barrier. Once that returns | |
905 | * there is no normal IO happeing. It must arrange to call | |
906 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 907 | */ |
1da177e4 | 908 | |
e879a879 | 909 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 910 | { |
6cce3b23 | 911 | BUG_ON(force && !conf->barrier); |
1da177e4 | 912 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 913 | |
6cce3b23 N |
914 | /* Wait until no block IO is waiting (unless 'force') */ |
915 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
c3b328ac | 916 | conf->resync_lock, ); |
0a27ec96 N |
917 | |
918 | /* block any new IO from starting */ | |
919 | conf->barrier++; | |
920 | ||
c3b328ac | 921 | /* Now wait for all pending IO to complete */ |
0a27ec96 N |
922 | wait_event_lock_irq(conf->wait_barrier, |
923 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
c3b328ac | 924 | conf->resync_lock, ); |
0a27ec96 N |
925 | |
926 | spin_unlock_irq(&conf->resync_lock); | |
927 | } | |
928 | ||
e879a879 | 929 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
930 | { |
931 | unsigned long flags; | |
932 | spin_lock_irqsave(&conf->resync_lock, flags); | |
933 | conf->barrier--; | |
934 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
935 | wake_up(&conf->wait_barrier); | |
936 | } | |
937 | ||
e879a879 | 938 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
939 | { |
940 | spin_lock_irq(&conf->resync_lock); | |
941 | if (conf->barrier) { | |
942 | conf->nr_waiting++; | |
d6b42dcb N |
943 | /* Wait for the barrier to drop. |
944 | * However if there are already pending | |
945 | * requests (preventing the barrier from | |
946 | * rising completely), and the | |
947 | * pre-process bio queue isn't empty, | |
948 | * then don't wait, as we need to empty | |
949 | * that queue to get the nr_pending | |
950 | * count down. | |
951 | */ | |
952 | wait_event_lock_irq(conf->wait_barrier, | |
953 | !conf->barrier || | |
954 | (conf->nr_pending && | |
955 | current->bio_list && | |
956 | !bio_list_empty(current->bio_list)), | |
0a27ec96 | 957 | conf->resync_lock, |
d6b42dcb | 958 | ); |
0a27ec96 | 959 | conf->nr_waiting--; |
1da177e4 | 960 | } |
0a27ec96 | 961 | conf->nr_pending++; |
1da177e4 LT |
962 | spin_unlock_irq(&conf->resync_lock); |
963 | } | |
964 | ||
e879a879 | 965 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 N |
966 | { |
967 | unsigned long flags; | |
968 | spin_lock_irqsave(&conf->resync_lock, flags); | |
969 | conf->nr_pending--; | |
970 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
971 | wake_up(&conf->wait_barrier); | |
972 | } | |
973 | ||
e879a879 | 974 | static void freeze_array(struct r10conf *conf) |
4443ae10 N |
975 | { |
976 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 977 | * go quiet. |
4443ae10 | 978 | * We increment barrier and nr_waiting, and then |
1c830532 N |
979 | * wait until nr_pending match nr_queued+1 |
980 | * This is called in the context of one normal IO request | |
981 | * that has failed. Thus any sync request that might be pending | |
982 | * will be blocked by nr_pending, and we need to wait for | |
983 | * pending IO requests to complete or be queued for re-try. | |
984 | * Thus the number queued (nr_queued) plus this request (1) | |
985 | * must match the number of pending IOs (nr_pending) before | |
986 | * we continue. | |
4443ae10 N |
987 | */ |
988 | spin_lock_irq(&conf->resync_lock); | |
989 | conf->barrier++; | |
990 | conf->nr_waiting++; | |
991 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 992 | conf->nr_pending == conf->nr_queued+1, |
4443ae10 | 993 | conf->resync_lock, |
c3b328ac N |
994 | flush_pending_writes(conf)); |
995 | ||
4443ae10 N |
996 | spin_unlock_irq(&conf->resync_lock); |
997 | } | |
998 | ||
e879a879 | 999 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1000 | { |
1001 | /* reverse the effect of the freeze */ | |
1002 | spin_lock_irq(&conf->resync_lock); | |
1003 | conf->barrier--; | |
1004 | conf->nr_waiting--; | |
1005 | wake_up(&conf->wait_barrier); | |
1006 | spin_unlock_irq(&conf->resync_lock); | |
1007 | } | |
1008 | ||
f8c9e74f N |
1009 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1010 | struct md_rdev *rdev) | |
1011 | { | |
1012 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1013 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1014 | return rdev->data_offset; | |
1015 | else | |
1016 | return rdev->new_data_offset; | |
1017 | } | |
1018 | ||
b4fdcb02 | 1019 | static void make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 1020 | { |
e879a879 | 1021 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 1022 | struct r10bio *r10_bio; |
1da177e4 LT |
1023 | struct bio *read_bio; |
1024 | int i; | |
f8c9e74f | 1025 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); |
5cf00fcd | 1026 | int chunk_sects = chunk_mask + 1; |
a362357b | 1027 | const int rw = bio_data_dir(bio); |
2c7d46ec | 1028 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 1029 | const unsigned long do_fua = (bio->bi_rw & REQ_FUA); |
6cce3b23 | 1030 | unsigned long flags; |
3cb03002 | 1031 | struct md_rdev *blocked_rdev; |
c3b328ac | 1032 | int plugged; |
d4432c23 N |
1033 | int sectors_handled; |
1034 | int max_sectors; | |
1da177e4 | 1035 | |
e9c7469b TH |
1036 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
1037 | md_flush_request(mddev, bio); | |
5a7bbad2 | 1038 | return; |
e5dcdd80 N |
1039 | } |
1040 | ||
1da177e4 LT |
1041 | /* If this request crosses a chunk boundary, we need to |
1042 | * split it. This will only happen for 1 PAGE (or less) requests. | |
1043 | */ | |
5cf00fcd N |
1044 | if (unlikely((bio->bi_sector & chunk_mask) + (bio->bi_size >> 9) |
1045 | > chunk_sects | |
f8c9e74f N |
1046 | && (conf->geo.near_copies < conf->geo.raid_disks |
1047 | || conf->prev.near_copies < conf->prev.raid_disks))) { | |
1da177e4 LT |
1048 | struct bio_pair *bp; |
1049 | /* Sanity check -- queue functions should prevent this happening */ | |
1050 | if (bio->bi_vcnt != 1 || | |
1051 | bio->bi_idx != 0) | |
1052 | goto bad_map; | |
1053 | /* This is a one page bio that upper layers | |
1054 | * refuse to split for us, so we need to split it. | |
1055 | */ | |
6feef531 | 1056 | bp = bio_split(bio, |
1da177e4 | 1057 | chunk_sects - (bio->bi_sector & (chunk_sects - 1)) ); |
51e9ac77 N |
1058 | |
1059 | /* Each of these 'make_request' calls will call 'wait_barrier'. | |
1060 | * If the first succeeds but the second blocks due to the resync | |
1061 | * thread raising the barrier, we will deadlock because the | |
1062 | * IO to the underlying device will be queued in generic_make_request | |
1063 | * and will never complete, so will never reduce nr_pending. | |
1064 | * So increment nr_waiting here so no new raise_barriers will | |
1065 | * succeed, and so the second wait_barrier cannot block. | |
1066 | */ | |
1067 | spin_lock_irq(&conf->resync_lock); | |
1068 | conf->nr_waiting++; | |
1069 | spin_unlock_irq(&conf->resync_lock); | |
1070 | ||
5a7bbad2 CH |
1071 | make_request(mddev, &bp->bio1); |
1072 | make_request(mddev, &bp->bio2); | |
1da177e4 | 1073 | |
51e9ac77 N |
1074 | spin_lock_irq(&conf->resync_lock); |
1075 | conf->nr_waiting--; | |
1076 | wake_up(&conf->wait_barrier); | |
1077 | spin_unlock_irq(&conf->resync_lock); | |
1078 | ||
1da177e4 | 1079 | bio_pair_release(bp); |
5a7bbad2 | 1080 | return; |
1da177e4 | 1081 | bad_map: |
128595ed N |
1082 | printk("md/raid10:%s: make_request bug: can't convert block across chunks" |
1083 | " or bigger than %dk %llu %d\n", mdname(mddev), chunk_sects/2, | |
1da177e4 LT |
1084 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); |
1085 | ||
6712ecf8 | 1086 | bio_io_error(bio); |
5a7bbad2 | 1087 | return; |
1da177e4 LT |
1088 | } |
1089 | ||
3d310eb7 | 1090 | md_write_start(mddev, bio); |
06d91a5f | 1091 | |
1da177e4 LT |
1092 | /* |
1093 | * Register the new request and wait if the reconstruction | |
1094 | * thread has put up a bar for new requests. | |
1095 | * Continue immediately if no resync is active currently. | |
1096 | */ | |
0a27ec96 | 1097 | wait_barrier(conf); |
1da177e4 | 1098 | |
1da177e4 LT |
1099 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); |
1100 | ||
1101 | r10_bio->master_bio = bio; | |
1102 | r10_bio->sectors = bio->bi_size >> 9; | |
1103 | ||
1104 | r10_bio->mddev = mddev; | |
1105 | r10_bio->sector = bio->bi_sector; | |
6cce3b23 | 1106 | r10_bio->state = 0; |
1da177e4 | 1107 | |
856e08e2 N |
1108 | /* We might need to issue multiple reads to different |
1109 | * devices if there are bad blocks around, so we keep | |
1110 | * track of the number of reads in bio->bi_phys_segments. | |
1111 | * If this is 0, there is only one r10_bio and no locking | |
1112 | * will be needed when the request completes. If it is | |
1113 | * non-zero, then it is the number of not-completed requests. | |
1114 | */ | |
1115 | bio->bi_phys_segments = 0; | |
1116 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
1117 | ||
a362357b | 1118 | if (rw == READ) { |
1da177e4 LT |
1119 | /* |
1120 | * read balancing logic: | |
1121 | */ | |
96c3fd1f | 1122 | struct md_rdev *rdev; |
856e08e2 N |
1123 | int slot; |
1124 | ||
1125 | read_again: | |
96c3fd1f N |
1126 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1127 | if (!rdev) { | |
1da177e4 | 1128 | raid_end_bio_io(r10_bio); |
5a7bbad2 | 1129 | return; |
1da177e4 | 1130 | } |
96c3fd1f | 1131 | slot = r10_bio->read_slot; |
1da177e4 | 1132 | |
a167f663 | 1133 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
856e08e2 N |
1134 | md_trim_bio(read_bio, r10_bio->sector - bio->bi_sector, |
1135 | max_sectors); | |
1da177e4 LT |
1136 | |
1137 | r10_bio->devs[slot].bio = read_bio; | |
abbf098e | 1138 | r10_bio->devs[slot].rdev = rdev; |
1da177e4 LT |
1139 | |
1140 | read_bio->bi_sector = r10_bio->devs[slot].addr + | |
f8c9e74f | 1141 | choose_data_offset(r10_bio, rdev); |
96c3fd1f | 1142 | read_bio->bi_bdev = rdev->bdev; |
1da177e4 | 1143 | read_bio->bi_end_io = raid10_end_read_request; |
7b6d91da | 1144 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1145 | read_bio->bi_private = r10_bio; |
1146 | ||
856e08e2 N |
1147 | if (max_sectors < r10_bio->sectors) { |
1148 | /* Could not read all from this device, so we will | |
1149 | * need another r10_bio. | |
1150 | */ | |
856e08e2 N |
1151 | sectors_handled = (r10_bio->sectors + max_sectors |
1152 | - bio->bi_sector); | |
1153 | r10_bio->sectors = max_sectors; | |
1154 | spin_lock_irq(&conf->device_lock); | |
1155 | if (bio->bi_phys_segments == 0) | |
1156 | bio->bi_phys_segments = 2; | |
1157 | else | |
1158 | bio->bi_phys_segments++; | |
1159 | spin_unlock(&conf->device_lock); | |
1160 | /* Cannot call generic_make_request directly | |
1161 | * as that will be queued in __generic_make_request | |
1162 | * and subsequent mempool_alloc might block | |
1163 | * waiting for it. so hand bio over to raid10d. | |
1164 | */ | |
1165 | reschedule_retry(r10_bio); | |
1166 | ||
1167 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1168 | ||
1169 | r10_bio->master_bio = bio; | |
1170 | r10_bio->sectors = ((bio->bi_size >> 9) | |
1171 | - sectors_handled); | |
1172 | r10_bio->state = 0; | |
1173 | r10_bio->mddev = mddev; | |
1174 | r10_bio->sector = bio->bi_sector + sectors_handled; | |
1175 | goto read_again; | |
1176 | } else | |
1177 | generic_make_request(read_bio); | |
5a7bbad2 | 1178 | return; |
1da177e4 LT |
1179 | } |
1180 | ||
1181 | /* | |
1182 | * WRITE: | |
1183 | */ | |
34db0cd6 N |
1184 | if (conf->pending_count >= max_queued_requests) { |
1185 | md_wakeup_thread(mddev->thread); | |
1186 | wait_event(conf->wait_barrier, | |
1187 | conf->pending_count < max_queued_requests); | |
1188 | } | |
6bfe0b49 | 1189 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1190 | * inc refcount on their rdev. Record them by setting |
1191 | * bios[x] to bio | |
d4432c23 N |
1192 | * If there are known/acknowledged bad blocks on any device |
1193 | * on which we have seen a write error, we want to avoid | |
1194 | * writing to those blocks. This potentially requires several | |
1195 | * writes to write around the bad blocks. Each set of writes | |
1196 | * gets its own r10_bio with a set of bios attached. The number | |
1197 | * of r10_bios is recored in bio->bi_phys_segments just as with | |
1198 | * the read case. | |
1da177e4 | 1199 | */ |
c3b328ac N |
1200 | plugged = mddev_check_plugged(mddev); |
1201 | ||
69335ef3 | 1202 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1203 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1204 | retry_write: |
cb6969e8 | 1205 | blocked_rdev = NULL; |
1da177e4 | 1206 | rcu_read_lock(); |
d4432c23 N |
1207 | max_sectors = r10_bio->sectors; |
1208 | ||
1da177e4 LT |
1209 | for (i = 0; i < conf->copies; i++) { |
1210 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1211 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1212 | struct md_rdev *rrdev = rcu_dereference( |
1213 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1214 | if (rdev == rrdev) |
1215 | rrdev = NULL; | |
6bfe0b49 DW |
1216 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1217 | atomic_inc(&rdev->nr_pending); | |
1218 | blocked_rdev = rdev; | |
1219 | break; | |
1220 | } | |
475b0321 N |
1221 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { |
1222 | atomic_inc(&rrdev->nr_pending); | |
1223 | blocked_rdev = rrdev; | |
1224 | break; | |
1225 | } | |
050b6615 N |
1226 | if (rrdev && (test_bit(Faulty, &rrdev->flags) |
1227 | || test_bit(Unmerged, &rrdev->flags))) | |
475b0321 N |
1228 | rrdev = NULL; |
1229 | ||
d4432c23 | 1230 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1231 | r10_bio->devs[i].repl_bio = NULL; |
050b6615 N |
1232 | if (!rdev || test_bit(Faulty, &rdev->flags) || |
1233 | test_bit(Unmerged, &rdev->flags)) { | |
6cce3b23 | 1234 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1235 | continue; |
1236 | } | |
1237 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1238 | sector_t first_bad; | |
1239 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1240 | int bad_sectors; | |
1241 | int is_bad; | |
1242 | ||
1243 | is_bad = is_badblock(rdev, dev_sector, | |
1244 | max_sectors, | |
1245 | &first_bad, &bad_sectors); | |
1246 | if (is_bad < 0) { | |
1247 | /* Mustn't write here until the bad block | |
1248 | * is acknowledged | |
1249 | */ | |
1250 | atomic_inc(&rdev->nr_pending); | |
1251 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1252 | blocked_rdev = rdev; | |
1253 | break; | |
1254 | } | |
1255 | if (is_bad && first_bad <= dev_sector) { | |
1256 | /* Cannot write here at all */ | |
1257 | bad_sectors -= (dev_sector - first_bad); | |
1258 | if (bad_sectors < max_sectors) | |
1259 | /* Mustn't write more than bad_sectors | |
1260 | * to other devices yet | |
1261 | */ | |
1262 | max_sectors = bad_sectors; | |
1263 | /* We don't set R10BIO_Degraded as that | |
1264 | * only applies if the disk is missing, | |
1265 | * so it might be re-added, and we want to | |
1266 | * know to recover this chunk. | |
1267 | * In this case the device is here, and the | |
1268 | * fact that this chunk is not in-sync is | |
1269 | * recorded in the bad block log. | |
1270 | */ | |
1271 | continue; | |
1272 | } | |
1273 | if (is_bad) { | |
1274 | int good_sectors = first_bad - dev_sector; | |
1275 | if (good_sectors < max_sectors) | |
1276 | max_sectors = good_sectors; | |
1277 | } | |
6cce3b23 | 1278 | } |
d4432c23 N |
1279 | r10_bio->devs[i].bio = bio; |
1280 | atomic_inc(&rdev->nr_pending); | |
475b0321 N |
1281 | if (rrdev) { |
1282 | r10_bio->devs[i].repl_bio = bio; | |
1283 | atomic_inc(&rrdev->nr_pending); | |
1284 | } | |
1da177e4 LT |
1285 | } |
1286 | rcu_read_unlock(); | |
1287 | ||
6bfe0b49 DW |
1288 | if (unlikely(blocked_rdev)) { |
1289 | /* Have to wait for this device to get unblocked, then retry */ | |
1290 | int j; | |
1291 | int d; | |
1292 | ||
475b0321 | 1293 | for (j = 0; j < i; j++) { |
6bfe0b49 DW |
1294 | if (r10_bio->devs[j].bio) { |
1295 | d = r10_bio->devs[j].devnum; | |
1296 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1297 | } | |
475b0321 | 1298 | if (r10_bio->devs[j].repl_bio) { |
4ca40c2c | 1299 | struct md_rdev *rdev; |
475b0321 | 1300 | d = r10_bio->devs[j].devnum; |
4ca40c2c N |
1301 | rdev = conf->mirrors[d].replacement; |
1302 | if (!rdev) { | |
1303 | /* Race with remove_disk */ | |
1304 | smp_mb(); | |
1305 | rdev = conf->mirrors[d].rdev; | |
1306 | } | |
1307 | rdev_dec_pending(rdev, mddev); | |
475b0321 N |
1308 | } |
1309 | } | |
6bfe0b49 DW |
1310 | allow_barrier(conf); |
1311 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1312 | wait_barrier(conf); | |
1313 | goto retry_write; | |
1314 | } | |
1315 | ||
d4432c23 N |
1316 | if (max_sectors < r10_bio->sectors) { |
1317 | /* We are splitting this into multiple parts, so | |
1318 | * we need to prepare for allocating another r10_bio. | |
1319 | */ | |
1320 | r10_bio->sectors = max_sectors; | |
1321 | spin_lock_irq(&conf->device_lock); | |
1322 | if (bio->bi_phys_segments == 0) | |
1323 | bio->bi_phys_segments = 2; | |
1324 | else | |
1325 | bio->bi_phys_segments++; | |
1326 | spin_unlock_irq(&conf->device_lock); | |
1327 | } | |
1328 | sectors_handled = r10_bio->sector + max_sectors - bio->bi_sector; | |
1329 | ||
4e78064f | 1330 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1331 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1332 | |
1da177e4 LT |
1333 | for (i = 0; i < conf->copies; i++) { |
1334 | struct bio *mbio; | |
1335 | int d = r10_bio->devs[i].devnum; | |
1336 | if (!r10_bio->devs[i].bio) | |
1337 | continue; | |
1338 | ||
a167f663 | 1339 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
d4432c23 N |
1340 | md_trim_bio(mbio, r10_bio->sector - bio->bi_sector, |
1341 | max_sectors); | |
1da177e4 LT |
1342 | r10_bio->devs[i].bio = mbio; |
1343 | ||
d4432c23 | 1344 | mbio->bi_sector = (r10_bio->devs[i].addr+ |
f8c9e74f N |
1345 | choose_data_offset(r10_bio, |
1346 | conf->mirrors[d].rdev)); | |
1da177e4 LT |
1347 | mbio->bi_bdev = conf->mirrors[d].rdev->bdev; |
1348 | mbio->bi_end_io = raid10_end_write_request; | |
e9c7469b | 1349 | mbio->bi_rw = WRITE | do_sync | do_fua; |
1da177e4 LT |
1350 | mbio->bi_private = r10_bio; |
1351 | ||
1352 | atomic_inc(&r10_bio->remaining); | |
4e78064f N |
1353 | spin_lock_irqsave(&conf->device_lock, flags); |
1354 | bio_list_add(&conf->pending_bio_list, mbio); | |
34db0cd6 | 1355 | conf->pending_count++; |
4e78064f | 1356 | spin_unlock_irqrestore(&conf->device_lock, flags); |
475b0321 N |
1357 | |
1358 | if (!r10_bio->devs[i].repl_bio) | |
1359 | continue; | |
1360 | ||
1361 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
1362 | md_trim_bio(mbio, r10_bio->sector - bio->bi_sector, | |
1363 | max_sectors); | |
1364 | r10_bio->devs[i].repl_bio = mbio; | |
1365 | ||
4ca40c2c N |
1366 | /* We are actively writing to the original device |
1367 | * so it cannot disappear, so the replacement cannot | |
1368 | * become NULL here | |
1369 | */ | |
f8c9e74f N |
1370 | mbio->bi_sector = (r10_bio->devs[i].addr + |
1371 | choose_data_offset( | |
1372 | r10_bio, | |
1373 | conf->mirrors[d].replacement)); | |
475b0321 N |
1374 | mbio->bi_bdev = conf->mirrors[d].replacement->bdev; |
1375 | mbio->bi_end_io = raid10_end_write_request; | |
1376 | mbio->bi_rw = WRITE | do_sync | do_fua; | |
1377 | mbio->bi_private = r10_bio; | |
1378 | ||
1379 | atomic_inc(&r10_bio->remaining); | |
1380 | spin_lock_irqsave(&conf->device_lock, flags); | |
1381 | bio_list_add(&conf->pending_bio_list, mbio); | |
1382 | conf->pending_count++; | |
1383 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1384 | } |
1385 | ||
079fa166 N |
1386 | /* Don't remove the bias on 'remaining' (one_write_done) until |
1387 | * after checking if we need to go around again. | |
1388 | */ | |
a35e63ef | 1389 | |
d4432c23 | 1390 | if (sectors_handled < (bio->bi_size >> 9)) { |
079fa166 | 1391 | one_write_done(r10_bio); |
5e570289 | 1392 | /* We need another r10_bio. It has already been counted |
d4432c23 N |
1393 | * in bio->bi_phys_segments. |
1394 | */ | |
1395 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1396 | ||
1397 | r10_bio->master_bio = bio; | |
1398 | r10_bio->sectors = (bio->bi_size >> 9) - sectors_handled; | |
1399 | ||
1400 | r10_bio->mddev = mddev; | |
1401 | r10_bio->sector = bio->bi_sector + sectors_handled; | |
1402 | r10_bio->state = 0; | |
1403 | goto retry_write; | |
1404 | } | |
079fa166 N |
1405 | one_write_done(r10_bio); |
1406 | ||
1407 | /* In case raid10d snuck in to freeze_array */ | |
1408 | wake_up(&conf->wait_barrier); | |
d4432c23 | 1409 | |
c3b328ac | 1410 | if (do_sync || !mddev->bitmap || !plugged) |
e3881a68 | 1411 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
1412 | } |
1413 | ||
fd01b88c | 1414 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1415 | { |
e879a879 | 1416 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1417 | int i; |
1418 | ||
5cf00fcd | 1419 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1420 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1421 | if (conf->geo.near_copies > 1) |
1422 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1423 | if (conf->geo.far_copies > 1) { | |
1424 | if (conf->geo.far_offset) | |
1425 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1426 | else |
5cf00fcd | 1427 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
c93983bf | 1428 | } |
5cf00fcd N |
1429 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1430 | conf->geo.raid_disks - mddev->degraded); | |
1431 | for (i = 0; i < conf->geo.raid_disks; i++) | |
1da177e4 LT |
1432 | seq_printf(seq, "%s", |
1433 | conf->mirrors[i].rdev && | |
b2d444d7 | 1434 | test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
1435 | seq_printf(seq, "]"); |
1436 | } | |
1437 | ||
700c7213 N |
1438 | /* check if there are enough drives for |
1439 | * every block to appear on atleast one. | |
1440 | * Don't consider the device numbered 'ignore' | |
1441 | * as we might be about to remove it. | |
1442 | */ | |
f8c9e74f | 1443 | static int _enough(struct r10conf *conf, struct geom *geo, int ignore) |
700c7213 N |
1444 | { |
1445 | int first = 0; | |
1446 | ||
1447 | do { | |
1448 | int n = conf->copies; | |
1449 | int cnt = 0; | |
1450 | while (n--) { | |
1451 | if (conf->mirrors[first].rdev && | |
1452 | first != ignore) | |
1453 | cnt++; | |
f8c9e74f | 1454 | first = (first+1) % geo->raid_disks; |
700c7213 N |
1455 | } |
1456 | if (cnt == 0) | |
1457 | return 0; | |
1458 | } while (first != 0); | |
1459 | return 1; | |
1460 | } | |
1461 | ||
f8c9e74f N |
1462 | static int enough(struct r10conf *conf, int ignore) |
1463 | { | |
1464 | return _enough(conf, &conf->geo, ignore) && | |
1465 | _enough(conf, &conf->prev, ignore); | |
1466 | } | |
1467 | ||
fd01b88c | 1468 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1469 | { |
1470 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1471 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1472 | |
1473 | /* | |
1474 | * If it is not operational, then we have already marked it as dead | |
1475 | * else if it is the last working disks, ignore the error, let the | |
1476 | * next level up know. | |
1477 | * else mark the drive as failed | |
1478 | */ | |
b2d444d7 | 1479 | if (test_bit(In_sync, &rdev->flags) |
700c7213 | 1480 | && !enough(conf, rdev->raid_disk)) |
1da177e4 LT |
1481 | /* |
1482 | * Don't fail the drive, just return an IO error. | |
1da177e4 LT |
1483 | */ |
1484 | return; | |
c04be0aa N |
1485 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1486 | unsigned long flags; | |
1487 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1488 | mddev->degraded++; |
c04be0aa | 1489 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1490 | /* |
1491 | * if recovery is running, make sure it aborts. | |
1492 | */ | |
dfc70645 | 1493 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1494 | } |
de393cde | 1495 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1496 | set_bit(Faulty, &rdev->flags); |
850b2b42 | 1497 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
1498 | printk(KERN_ALERT |
1499 | "md/raid10:%s: Disk failure on %s, disabling device.\n" | |
1500 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
128595ed | 1501 | mdname(mddev), bdevname(rdev->bdev, b), |
5cf00fcd | 1502 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); |
1da177e4 LT |
1503 | } |
1504 | ||
e879a879 | 1505 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1506 | { |
1507 | int i; | |
0f6d02d5 | 1508 | struct mirror_info *tmp; |
1da177e4 | 1509 | |
128595ed | 1510 | printk(KERN_DEBUG "RAID10 conf printout:\n"); |
1da177e4 | 1511 | if (!conf) { |
128595ed | 1512 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1513 | return; |
1514 | } | |
5cf00fcd N |
1515 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1516 | conf->geo.raid_disks); | |
1da177e4 | 1517 | |
5cf00fcd | 1518 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 LT |
1519 | char b[BDEVNAME_SIZE]; |
1520 | tmp = conf->mirrors + i; | |
1521 | if (tmp->rdev) | |
128595ed | 1522 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
b2d444d7 N |
1523 | i, !test_bit(In_sync, &tmp->rdev->flags), |
1524 | !test_bit(Faulty, &tmp->rdev->flags), | |
1da177e4 LT |
1525 | bdevname(tmp->rdev->bdev,b)); |
1526 | } | |
1527 | } | |
1528 | ||
e879a879 | 1529 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1530 | { |
0a27ec96 N |
1531 | wait_barrier(conf); |
1532 | allow_barrier(conf); | |
1da177e4 LT |
1533 | |
1534 | mempool_destroy(conf->r10buf_pool); | |
1535 | conf->r10buf_pool = NULL; | |
1536 | } | |
1537 | ||
fd01b88c | 1538 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1539 | { |
1540 | int i; | |
e879a879 | 1541 | struct r10conf *conf = mddev->private; |
0f6d02d5 | 1542 | struct mirror_info *tmp; |
6b965620 N |
1543 | int count = 0; |
1544 | unsigned long flags; | |
1da177e4 LT |
1545 | |
1546 | /* | |
1547 | * Find all non-in_sync disks within the RAID10 configuration | |
1548 | * and mark them in_sync | |
1549 | */ | |
5cf00fcd | 1550 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1551 | tmp = conf->mirrors + i; |
4ca40c2c N |
1552 | if (tmp->replacement |
1553 | && tmp->replacement->recovery_offset == MaxSector | |
1554 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1555 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1556 | /* Replacement has just become active */ | |
1557 | if (!tmp->rdev | |
1558 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1559 | count++; | |
1560 | if (tmp->rdev) { | |
1561 | /* Replaced device not technically faulty, | |
1562 | * but we need to be sure it gets removed | |
1563 | * and never re-added. | |
1564 | */ | |
1565 | set_bit(Faulty, &tmp->rdev->flags); | |
1566 | sysfs_notify_dirent_safe( | |
1567 | tmp->rdev->sysfs_state); | |
1568 | } | |
1569 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1570 | } else if (tmp->rdev | |
1571 | && !test_bit(Faulty, &tmp->rdev->flags) | |
1572 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1573 | count++; |
e6ffbcb6 | 1574 | sysfs_notify_dirent(tmp->rdev->sysfs_state); |
1da177e4 LT |
1575 | } |
1576 | } | |
6b965620 N |
1577 | spin_lock_irqsave(&conf->device_lock, flags); |
1578 | mddev->degraded -= count; | |
1579 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1580 | |
1581 | print_conf(conf); | |
6b965620 | 1582 | return count; |
1da177e4 LT |
1583 | } |
1584 | ||
1585 | ||
fd01b88c | 1586 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1587 | { |
e879a879 | 1588 | struct r10conf *conf = mddev->private; |
199050ea | 1589 | int err = -EEXIST; |
1da177e4 | 1590 | int mirror; |
6c2fce2e | 1591 | int first = 0; |
5cf00fcd | 1592 | int last = conf->geo.raid_disks - 1; |
050b6615 | 1593 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
1594 | |
1595 | if (mddev->recovery_cp < MaxSector) | |
1596 | /* only hot-add to in-sync arrays, as recovery is | |
1597 | * very different from resync | |
1598 | */ | |
199050ea | 1599 | return -EBUSY; |
f8c9e74f | 1600 | if (rdev->saved_raid_disk < 0 && !_enough(conf, &conf->prev, -1)) |
199050ea | 1601 | return -EINVAL; |
1da177e4 | 1602 | |
a53a6c85 | 1603 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1604 | first = last = rdev->raid_disk; |
1da177e4 | 1605 | |
050b6615 N |
1606 | if (q->merge_bvec_fn) { |
1607 | set_bit(Unmerged, &rdev->flags); | |
1608 | mddev->merge_check_needed = 1; | |
1609 | } | |
1610 | ||
2c4193df | 1611 | if (rdev->saved_raid_disk >= first && |
6cce3b23 N |
1612 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1613 | mirror = rdev->saved_raid_disk; | |
1614 | else | |
6c2fce2e | 1615 | mirror = first; |
2bb77736 | 1616 | for ( ; mirror <= last ; mirror++) { |
0f6d02d5 | 1617 | struct mirror_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1618 | if (p->recovery_disabled == mddev->recovery_disabled) |
1619 | continue; | |
b7044d41 N |
1620 | if (p->rdev) { |
1621 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1622 | p->replacement != NULL) | |
1623 | continue; | |
1624 | clear_bit(In_sync, &rdev->flags); | |
1625 | set_bit(Replacement, &rdev->flags); | |
1626 | rdev->raid_disk = mirror; | |
1627 | err = 0; | |
1628 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1629 | rdev->data_offset << 9); | |
b7044d41 N |
1630 | conf->fullsync = 1; |
1631 | rcu_assign_pointer(p->replacement, rdev); | |
1632 | break; | |
1633 | } | |
1da177e4 | 1634 | |
2bb77736 N |
1635 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1636 | rdev->data_offset << 9); | |
1da177e4 | 1637 | |
2bb77736 | 1638 | p->head_position = 0; |
d890fa2b | 1639 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1640 | rdev->raid_disk = mirror; |
1641 | err = 0; | |
1642 | if (rdev->saved_raid_disk != mirror) | |
1643 | conf->fullsync = 1; | |
1644 | rcu_assign_pointer(p->rdev, rdev); | |
1645 | break; | |
1646 | } | |
050b6615 N |
1647 | if (err == 0 && test_bit(Unmerged, &rdev->flags)) { |
1648 | /* Some requests might not have seen this new | |
1649 | * merge_bvec_fn. We must wait for them to complete | |
1650 | * before merging the device fully. | |
1651 | * First we make sure any code which has tested | |
1652 | * our function has submitted the request, then | |
1653 | * we wait for all outstanding requests to complete. | |
1654 | */ | |
1655 | synchronize_sched(); | |
1656 | raise_barrier(conf, 0); | |
1657 | lower_barrier(conf); | |
1658 | clear_bit(Unmerged, &rdev->flags); | |
1659 | } | |
ac5e7113 | 1660 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1661 | print_conf(conf); |
199050ea | 1662 | return err; |
1da177e4 LT |
1663 | } |
1664 | ||
b8321b68 | 1665 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1666 | { |
e879a879 | 1667 | struct r10conf *conf = mddev->private; |
1da177e4 | 1668 | int err = 0; |
b8321b68 | 1669 | int number = rdev->raid_disk; |
c8ab903e N |
1670 | struct md_rdev **rdevp; |
1671 | struct mirror_info *p = conf->mirrors + number; | |
1da177e4 LT |
1672 | |
1673 | print_conf(conf); | |
c8ab903e N |
1674 | if (rdev == p->rdev) |
1675 | rdevp = &p->rdev; | |
1676 | else if (rdev == p->replacement) | |
1677 | rdevp = &p->replacement; | |
1678 | else | |
1679 | return 0; | |
1680 | ||
1681 | if (test_bit(In_sync, &rdev->flags) || | |
1682 | atomic_read(&rdev->nr_pending)) { | |
1683 | err = -EBUSY; | |
1684 | goto abort; | |
1685 | } | |
1686 | /* Only remove faulty devices if recovery | |
1687 | * is not possible. | |
1688 | */ | |
1689 | if (!test_bit(Faulty, &rdev->flags) && | |
1690 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1691 | (!p->replacement || p->replacement == rdev) && |
c8ab903e N |
1692 | enough(conf, -1)) { |
1693 | err = -EBUSY; | |
1694 | goto abort; | |
1da177e4 | 1695 | } |
c8ab903e N |
1696 | *rdevp = NULL; |
1697 | synchronize_rcu(); | |
1698 | if (atomic_read(&rdev->nr_pending)) { | |
1699 | /* lost the race, try later */ | |
1700 | err = -EBUSY; | |
1701 | *rdevp = rdev; | |
1702 | goto abort; | |
4ca40c2c N |
1703 | } else if (p->replacement) { |
1704 | /* We must have just cleared 'rdev' */ | |
1705 | p->rdev = p->replacement; | |
1706 | clear_bit(Replacement, &p->replacement->flags); | |
1707 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1708 | * but will never see neither -- if they are careful. | |
1709 | */ | |
1710 | p->replacement = NULL; | |
1711 | clear_bit(WantReplacement, &rdev->flags); | |
1712 | } else | |
1713 | /* We might have just remove the Replacement as faulty | |
1714 | * Clear the flag just in case | |
1715 | */ | |
1716 | clear_bit(WantReplacement, &rdev->flags); | |
1717 | ||
c8ab903e N |
1718 | err = md_integrity_register(mddev); |
1719 | ||
1da177e4 LT |
1720 | abort: |
1721 | ||
1722 | print_conf(conf); | |
1723 | return err; | |
1724 | } | |
1725 | ||
1726 | ||
6712ecf8 | 1727 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1728 | { |
9f2c9d12 | 1729 | struct r10bio *r10_bio = bio->bi_private; |
e879a879 | 1730 | struct r10conf *conf = r10_bio->mddev->private; |
778ca018 | 1731 | int d; |
1da177e4 | 1732 | |
69335ef3 | 1733 | d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); |
0eb3ff12 N |
1734 | |
1735 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1736 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
e684e41d N |
1737 | else |
1738 | /* The write handler will notice the lack of | |
1739 | * R10BIO_Uptodate and record any errors etc | |
1740 | */ | |
4dbcdc75 N |
1741 | atomic_add(r10_bio->sectors, |
1742 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1743 | |
1744 | /* for reconstruct, we always reschedule after a read. | |
1745 | * for resync, only after all reads | |
1746 | */ | |
73d5c38a | 1747 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1748 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1749 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1750 | /* we have read all the blocks, | |
1751 | * do the comparison in process context in raid10d | |
1752 | */ | |
1753 | reschedule_retry(r10_bio); | |
1754 | } | |
1da177e4 LT |
1755 | } |
1756 | ||
9f2c9d12 | 1757 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1758 | { |
fd01b88c | 1759 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1760 | |
1da177e4 LT |
1761 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1762 | if (r10_bio->master_bio == NULL) { | |
1763 | /* the primary of several recovery bios */ | |
73d5c38a | 1764 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1765 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1766 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1767 | reschedule_retry(r10_bio); |
1768 | else | |
1769 | put_buf(r10_bio); | |
73d5c38a | 1770 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1771 | break; |
1772 | } else { | |
9f2c9d12 | 1773 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1774 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1775 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1776 | reschedule_retry(r10_bio); |
1777 | else | |
1778 | put_buf(r10_bio); | |
1da177e4 LT |
1779 | r10_bio = r10_bio2; |
1780 | } | |
1781 | } | |
1da177e4 LT |
1782 | } |
1783 | ||
5e570289 N |
1784 | static void end_sync_write(struct bio *bio, int error) |
1785 | { | |
1786 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 1787 | struct r10bio *r10_bio = bio->bi_private; |
fd01b88c | 1788 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1789 | struct r10conf *conf = mddev->private; |
5e570289 N |
1790 | int d; |
1791 | sector_t first_bad; | |
1792 | int bad_sectors; | |
1793 | int slot; | |
9ad1aefc | 1794 | int repl; |
4ca40c2c | 1795 | struct md_rdev *rdev = NULL; |
5e570289 | 1796 | |
9ad1aefc N |
1797 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1798 | if (repl) | |
1799 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 1800 | else |
9ad1aefc | 1801 | rdev = conf->mirrors[d].rdev; |
5e570289 N |
1802 | |
1803 | if (!uptodate) { | |
9ad1aefc N |
1804 | if (repl) |
1805 | md_error(mddev, rdev); | |
1806 | else { | |
1807 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
1808 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
1809 | set_bit(MD_RECOVERY_NEEDED, | |
1810 | &rdev->mddev->recovery); | |
9ad1aefc N |
1811 | set_bit(R10BIO_WriteError, &r10_bio->state); |
1812 | } | |
1813 | } else if (is_badblock(rdev, | |
5e570289 N |
1814 | r10_bio->devs[slot].addr, |
1815 | r10_bio->sectors, | |
1816 | &first_bad, &bad_sectors)) | |
1817 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
1818 | ||
9ad1aefc | 1819 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
1820 | |
1821 | end_sync_request(r10_bio); | |
1822 | } | |
1823 | ||
1da177e4 LT |
1824 | /* |
1825 | * Note: sync and recover and handled very differently for raid10 | |
1826 | * This code is for resync. | |
1827 | * For resync, we read through virtual addresses and read all blocks. | |
1828 | * If there is any error, we schedule a write. The lowest numbered | |
1829 | * drive is authoritative. | |
1830 | * However requests come for physical address, so we need to map. | |
1831 | * For every physical address there are raid_disks/copies virtual addresses, | |
1832 | * which is always are least one, but is not necessarly an integer. | |
1833 | * This means that a physical address can span multiple chunks, so we may | |
1834 | * have to submit multiple io requests for a single sync request. | |
1835 | */ | |
1836 | /* | |
1837 | * We check if all blocks are in-sync and only write to blocks that | |
1838 | * aren't in sync | |
1839 | */ | |
9f2c9d12 | 1840 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1841 | { |
e879a879 | 1842 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1843 | int i, first; |
1844 | struct bio *tbio, *fbio; | |
f4380a91 | 1845 | int vcnt; |
1da177e4 LT |
1846 | |
1847 | atomic_set(&r10_bio->remaining, 1); | |
1848 | ||
1849 | /* find the first device with a block */ | |
1850 | for (i=0; i<conf->copies; i++) | |
1851 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) | |
1852 | break; | |
1853 | ||
1854 | if (i == conf->copies) | |
1855 | goto done; | |
1856 | ||
1857 | first = i; | |
1858 | fbio = r10_bio->devs[i].bio; | |
1859 | ||
f4380a91 | 1860 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 1861 | /* now find blocks with errors */ |
0eb3ff12 N |
1862 | for (i=0 ; i < conf->copies ; i++) { |
1863 | int j, d; | |
1da177e4 | 1864 | |
1da177e4 | 1865 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
1866 | |
1867 | if (tbio->bi_end_io != end_sync_read) | |
1868 | continue; | |
1869 | if (i == first) | |
1da177e4 | 1870 | continue; |
0eb3ff12 N |
1871 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) { |
1872 | /* We know that the bi_io_vec layout is the same for | |
1873 | * both 'first' and 'i', so we just compare them. | |
1874 | * All vec entries are PAGE_SIZE; | |
1875 | */ | |
1876 | for (j = 0; j < vcnt; j++) | |
1877 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), | |
1878 | page_address(tbio->bi_io_vec[j].bv_page), | |
5020ad7d | 1879 | fbio->bi_io_vec[j].bv_len)) |
0eb3ff12 N |
1880 | break; |
1881 | if (j == vcnt) | |
1882 | continue; | |
1883 | mddev->resync_mismatches += r10_bio->sectors; | |
f84ee364 N |
1884 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
1885 | /* Don't fix anything. */ | |
1886 | continue; | |
0eb3ff12 | 1887 | } |
f84ee364 N |
1888 | /* Ok, we need to write this bio, either to correct an |
1889 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
1890 | * First we need to fixup bv_offset, bv_len and |
1891 | * bi_vecs, as the read request might have corrupted these | |
1892 | */ | |
1893 | tbio->bi_vcnt = vcnt; | |
1894 | tbio->bi_size = r10_bio->sectors << 9; | |
1895 | tbio->bi_idx = 0; | |
1896 | tbio->bi_phys_segments = 0; | |
1da177e4 LT |
1897 | tbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
1898 | tbio->bi_flags |= 1 << BIO_UPTODATE; | |
1899 | tbio->bi_next = NULL; | |
1900 | tbio->bi_rw = WRITE; | |
1901 | tbio->bi_private = r10_bio; | |
1902 | tbio->bi_sector = r10_bio->devs[i].addr; | |
1903 | ||
1904 | for (j=0; j < vcnt ; j++) { | |
1905 | tbio->bi_io_vec[j].bv_offset = 0; | |
1906 | tbio->bi_io_vec[j].bv_len = PAGE_SIZE; | |
1907 | ||
1908 | memcpy(page_address(tbio->bi_io_vec[j].bv_page), | |
1909 | page_address(fbio->bi_io_vec[j].bv_page), | |
1910 | PAGE_SIZE); | |
1911 | } | |
1912 | tbio->bi_end_io = end_sync_write; | |
1913 | ||
1914 | d = r10_bio->devs[i].devnum; | |
1915 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1916 | atomic_inc(&r10_bio->remaining); | |
1917 | md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9); | |
1918 | ||
1919 | tbio->bi_sector += conf->mirrors[d].rdev->data_offset; | |
1920 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
1921 | generic_make_request(tbio); | |
1922 | } | |
1923 | ||
9ad1aefc N |
1924 | /* Now write out to any replacement devices |
1925 | * that are active | |
1926 | */ | |
1927 | for (i = 0; i < conf->copies; i++) { | |
1928 | int j, d; | |
9ad1aefc N |
1929 | |
1930 | tbio = r10_bio->devs[i].repl_bio; | |
1931 | if (!tbio || !tbio->bi_end_io) | |
1932 | continue; | |
1933 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
1934 | && r10_bio->devs[i].bio != fbio) | |
1935 | for (j = 0; j < vcnt; j++) | |
1936 | memcpy(page_address(tbio->bi_io_vec[j].bv_page), | |
1937 | page_address(fbio->bi_io_vec[j].bv_page), | |
1938 | PAGE_SIZE); | |
1939 | d = r10_bio->devs[i].devnum; | |
1940 | atomic_inc(&r10_bio->remaining); | |
1941 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
1942 | tbio->bi_size >> 9); | |
1943 | generic_make_request(tbio); | |
1944 | } | |
1945 | ||
1da177e4 LT |
1946 | done: |
1947 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
1948 | md_done_sync(mddev, r10_bio->sectors, 1); | |
1949 | put_buf(r10_bio); | |
1950 | } | |
1951 | } | |
1952 | ||
1953 | /* | |
1954 | * Now for the recovery code. | |
1955 | * Recovery happens across physical sectors. | |
1956 | * We recover all non-is_sync drives by finding the virtual address of | |
1957 | * each, and then choose a working drive that also has that virt address. | |
1958 | * There is a separate r10_bio for each non-in_sync drive. | |
1959 | * Only the first two slots are in use. The first for reading, | |
1960 | * The second for writing. | |
1961 | * | |
1962 | */ | |
9f2c9d12 | 1963 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
1964 | { |
1965 | /* We got a read error during recovery. | |
1966 | * We repeat the read in smaller page-sized sections. | |
1967 | * If a read succeeds, write it to the new device or record | |
1968 | * a bad block if we cannot. | |
1969 | * If a read fails, record a bad block on both old and | |
1970 | * new devices. | |
1971 | */ | |
fd01b88c | 1972 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1973 | struct r10conf *conf = mddev->private; |
5e570289 N |
1974 | struct bio *bio = r10_bio->devs[0].bio; |
1975 | sector_t sect = 0; | |
1976 | int sectors = r10_bio->sectors; | |
1977 | int idx = 0; | |
1978 | int dr = r10_bio->devs[0].devnum; | |
1979 | int dw = r10_bio->devs[1].devnum; | |
1980 | ||
1981 | while (sectors) { | |
1982 | int s = sectors; | |
3cb03002 | 1983 | struct md_rdev *rdev; |
5e570289 N |
1984 | sector_t addr; |
1985 | int ok; | |
1986 | ||
1987 | if (s > (PAGE_SIZE>>9)) | |
1988 | s = PAGE_SIZE >> 9; | |
1989 | ||
1990 | rdev = conf->mirrors[dr].rdev; | |
1991 | addr = r10_bio->devs[0].addr + sect, | |
1992 | ok = sync_page_io(rdev, | |
1993 | addr, | |
1994 | s << 9, | |
1995 | bio->bi_io_vec[idx].bv_page, | |
1996 | READ, false); | |
1997 | if (ok) { | |
1998 | rdev = conf->mirrors[dw].rdev; | |
1999 | addr = r10_bio->devs[1].addr + sect; | |
2000 | ok = sync_page_io(rdev, | |
2001 | addr, | |
2002 | s << 9, | |
2003 | bio->bi_io_vec[idx].bv_page, | |
2004 | WRITE, false); | |
b7044d41 | 2005 | if (!ok) { |
5e570289 | 2006 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2007 | if (!test_and_set_bit(WantReplacement, |
2008 | &rdev->flags)) | |
2009 | set_bit(MD_RECOVERY_NEEDED, | |
2010 | &rdev->mddev->recovery); | |
2011 | } | |
5e570289 N |
2012 | } |
2013 | if (!ok) { | |
2014 | /* We don't worry if we cannot set a bad block - | |
2015 | * it really is bad so there is no loss in not | |
2016 | * recording it yet | |
2017 | */ | |
2018 | rdev_set_badblocks(rdev, addr, s, 0); | |
2019 | ||
2020 | if (rdev != conf->mirrors[dw].rdev) { | |
2021 | /* need bad block on destination too */ | |
3cb03002 | 2022 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2023 | addr = r10_bio->devs[1].addr + sect; |
2024 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2025 | if (!ok) { | |
2026 | /* just abort the recovery */ | |
2027 | printk(KERN_NOTICE | |
2028 | "md/raid10:%s: recovery aborted" | |
2029 | " due to read error\n", | |
2030 | mdname(mddev)); | |
2031 | ||
2032 | conf->mirrors[dw].recovery_disabled | |
2033 | = mddev->recovery_disabled; | |
2034 | set_bit(MD_RECOVERY_INTR, | |
2035 | &mddev->recovery); | |
2036 | break; | |
2037 | } | |
2038 | } | |
2039 | } | |
2040 | ||
2041 | sectors -= s; | |
2042 | sect += s; | |
2043 | idx++; | |
2044 | } | |
2045 | } | |
1da177e4 | 2046 | |
9f2c9d12 | 2047 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2048 | { |
e879a879 | 2049 | struct r10conf *conf = mddev->private; |
c65060ad | 2050 | int d; |
24afd80d | 2051 | struct bio *wbio, *wbio2; |
1da177e4 | 2052 | |
5e570289 N |
2053 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2054 | fix_recovery_read_error(r10_bio); | |
2055 | end_sync_request(r10_bio); | |
2056 | return; | |
2057 | } | |
2058 | ||
c65060ad NK |
2059 | /* |
2060 | * share the pages with the first bio | |
1da177e4 LT |
2061 | * and submit the write request |
2062 | */ | |
1da177e4 | 2063 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2064 | wbio = r10_bio->devs[1].bio; |
2065 | wbio2 = r10_bio->devs[1].repl_bio; | |
2066 | if (wbio->bi_end_io) { | |
2067 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
2068 | md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9); | |
2069 | generic_make_request(wbio); | |
2070 | } | |
2071 | if (wbio2 && wbio2->bi_end_io) { | |
2072 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); | |
2073 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
2074 | wbio2->bi_size >> 9); | |
2075 | generic_make_request(wbio2); | |
2076 | } | |
1da177e4 LT |
2077 | } |
2078 | ||
2079 | ||
1e50915f RB |
2080 | /* |
2081 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2082 | * We halve the read error count for every hour that has elapsed | |
2083 | * since the last recorded read error. | |
2084 | * | |
2085 | */ | |
fd01b88c | 2086 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f RB |
2087 | { |
2088 | struct timespec cur_time_mon; | |
2089 | unsigned long hours_since_last; | |
2090 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2091 | ||
2092 | ktime_get_ts(&cur_time_mon); | |
2093 | ||
2094 | if (rdev->last_read_error.tv_sec == 0 && | |
2095 | rdev->last_read_error.tv_nsec == 0) { | |
2096 | /* first time we've seen a read error */ | |
2097 | rdev->last_read_error = cur_time_mon; | |
2098 | return; | |
2099 | } | |
2100 | ||
2101 | hours_since_last = (cur_time_mon.tv_sec - | |
2102 | rdev->last_read_error.tv_sec) / 3600; | |
2103 | ||
2104 | rdev->last_read_error = cur_time_mon; | |
2105 | ||
2106 | /* | |
2107 | * if hours_since_last is > the number of bits in read_errors | |
2108 | * just set read errors to 0. We do this to avoid | |
2109 | * overflowing the shift of read_errors by hours_since_last. | |
2110 | */ | |
2111 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2112 | atomic_set(&rdev->read_errors, 0); | |
2113 | else | |
2114 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2115 | } | |
2116 | ||
3cb03002 | 2117 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2118 | int sectors, struct page *page, int rw) |
2119 | { | |
2120 | sector_t first_bad; | |
2121 | int bad_sectors; | |
2122 | ||
2123 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2124 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2125 | return -1; | |
2126 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
2127 | /* success */ | |
2128 | return 1; | |
b7044d41 | 2129 | if (rw == WRITE) { |
58c54fcc | 2130 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2131 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2132 | set_bit(MD_RECOVERY_NEEDED, | |
2133 | &rdev->mddev->recovery); | |
2134 | } | |
58c54fcc N |
2135 | /* need to record an error - either for the block or the device */ |
2136 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2137 | md_error(rdev->mddev, rdev); | |
2138 | return 0; | |
2139 | } | |
2140 | ||
1da177e4 LT |
2141 | /* |
2142 | * This is a kernel thread which: | |
2143 | * | |
2144 | * 1. Retries failed read operations on working mirrors. | |
2145 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2146 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2147 | */ |
2148 | ||
e879a879 | 2149 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2150 | { |
2151 | int sect = 0; /* Offset from r10_bio->sector */ | |
2152 | int sectors = r10_bio->sectors; | |
3cb03002 | 2153 | struct md_rdev*rdev; |
1e50915f | 2154 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2155 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2156 | |
7c4e06ff N |
2157 | /* still own a reference to this rdev, so it cannot |
2158 | * have been cleared recently. | |
2159 | */ | |
2160 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2161 | |
7c4e06ff N |
2162 | if (test_bit(Faulty, &rdev->flags)) |
2163 | /* drive has already been failed, just ignore any | |
2164 | more fix_read_error() attempts */ | |
2165 | return; | |
1e50915f | 2166 | |
7c4e06ff N |
2167 | check_decay_read_errors(mddev, rdev); |
2168 | atomic_inc(&rdev->read_errors); | |
2169 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2170 | char b[BDEVNAME_SIZE]; | |
2171 | bdevname(rdev->bdev, b); | |
1e50915f | 2172 | |
7c4e06ff N |
2173 | printk(KERN_NOTICE |
2174 | "md/raid10:%s: %s: Raid device exceeded " | |
2175 | "read_error threshold [cur %d:max %d]\n", | |
2176 | mdname(mddev), b, | |
2177 | atomic_read(&rdev->read_errors), max_read_errors); | |
2178 | printk(KERN_NOTICE | |
2179 | "md/raid10:%s: %s: Failing raid device\n", | |
2180 | mdname(mddev), b); | |
2181 | md_error(mddev, conf->mirrors[d].rdev); | |
fae8cc5e | 2182 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2183 | return; |
1e50915f | 2184 | } |
1e50915f | 2185 | |
6814d536 N |
2186 | while(sectors) { |
2187 | int s = sectors; | |
2188 | int sl = r10_bio->read_slot; | |
2189 | int success = 0; | |
2190 | int start; | |
2191 | ||
2192 | if (s > (PAGE_SIZE>>9)) | |
2193 | s = PAGE_SIZE >> 9; | |
2194 | ||
2195 | rcu_read_lock(); | |
2196 | do { | |
8dbed5ce N |
2197 | sector_t first_bad; |
2198 | int bad_sectors; | |
2199 | ||
0544a21d | 2200 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2201 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2202 | if (rdev && | |
050b6615 | 2203 | !test_bit(Unmerged, &rdev->flags) && |
8dbed5ce N |
2204 | test_bit(In_sync, &rdev->flags) && |
2205 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, | |
2206 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2207 | atomic_inc(&rdev->nr_pending); |
2208 | rcu_read_unlock(); | |
2b193363 | 2209 | success = sync_page_io(rdev, |
6814d536 | 2210 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2211 | sect, |
6814d536 | 2212 | s<<9, |
ccebd4c4 | 2213 | conf->tmppage, READ, false); |
6814d536 N |
2214 | rdev_dec_pending(rdev, mddev); |
2215 | rcu_read_lock(); | |
2216 | if (success) | |
2217 | break; | |
2218 | } | |
2219 | sl++; | |
2220 | if (sl == conf->copies) | |
2221 | sl = 0; | |
2222 | } while (!success && sl != r10_bio->read_slot); | |
2223 | rcu_read_unlock(); | |
2224 | ||
2225 | if (!success) { | |
58c54fcc N |
2226 | /* Cannot read from anywhere, just mark the block |
2227 | * as bad on the first device to discourage future | |
2228 | * reads. | |
2229 | */ | |
6814d536 | 2230 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2231 | rdev = conf->mirrors[dn].rdev; |
2232 | ||
2233 | if (!rdev_set_badblocks( | |
2234 | rdev, | |
2235 | r10_bio->devs[r10_bio->read_slot].addr | |
2236 | + sect, | |
fae8cc5e | 2237 | s, 0)) { |
58c54fcc | 2238 | md_error(mddev, rdev); |
fae8cc5e N |
2239 | r10_bio->devs[r10_bio->read_slot].bio |
2240 | = IO_BLOCKED; | |
2241 | } | |
6814d536 N |
2242 | break; |
2243 | } | |
2244 | ||
2245 | start = sl; | |
2246 | /* write it back and re-read */ | |
2247 | rcu_read_lock(); | |
2248 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2249 | char b[BDEVNAME_SIZE]; |
0544a21d | 2250 | |
6814d536 N |
2251 | if (sl==0) |
2252 | sl = conf->copies; | |
2253 | sl--; | |
2254 | d = r10_bio->devs[sl].devnum; | |
2255 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2256 | if (!rdev || |
050b6615 | 2257 | test_bit(Unmerged, &rdev->flags) || |
1294b9c9 N |
2258 | !test_bit(In_sync, &rdev->flags)) |
2259 | continue; | |
2260 | ||
2261 | atomic_inc(&rdev->nr_pending); | |
2262 | rcu_read_unlock(); | |
58c54fcc N |
2263 | if (r10_sync_page_io(rdev, |
2264 | r10_bio->devs[sl].addr + | |
2265 | sect, | |
2266 | s<<9, conf->tmppage, WRITE) | |
1294b9c9 N |
2267 | == 0) { |
2268 | /* Well, this device is dead */ | |
2269 | printk(KERN_NOTICE | |
2270 | "md/raid10:%s: read correction " | |
2271 | "write failed" | |
2272 | " (%d sectors at %llu on %s)\n", | |
2273 | mdname(mddev), s, | |
2274 | (unsigned long long)( | |
f8c9e74f N |
2275 | sect + |
2276 | choose_data_offset(r10_bio, | |
2277 | rdev)), | |
1294b9c9 N |
2278 | bdevname(rdev->bdev, b)); |
2279 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
2280 | "drive\n", | |
2281 | mdname(mddev), | |
2282 | bdevname(rdev->bdev, b)); | |
6814d536 | 2283 | } |
1294b9c9 N |
2284 | rdev_dec_pending(rdev, mddev); |
2285 | rcu_read_lock(); | |
6814d536 N |
2286 | } |
2287 | sl = start; | |
2288 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2289 | char b[BDEVNAME_SIZE]; |
0544a21d | 2290 | |
6814d536 N |
2291 | if (sl==0) |
2292 | sl = conf->copies; | |
2293 | sl--; | |
2294 | d = r10_bio->devs[sl].devnum; | |
2295 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 N |
2296 | if (!rdev || |
2297 | !test_bit(In_sync, &rdev->flags)) | |
2298 | continue; | |
6814d536 | 2299 | |
1294b9c9 N |
2300 | atomic_inc(&rdev->nr_pending); |
2301 | rcu_read_unlock(); | |
58c54fcc N |
2302 | switch (r10_sync_page_io(rdev, |
2303 | r10_bio->devs[sl].addr + | |
2304 | sect, | |
2305 | s<<9, conf->tmppage, | |
2306 | READ)) { | |
2307 | case 0: | |
1294b9c9 N |
2308 | /* Well, this device is dead */ |
2309 | printk(KERN_NOTICE | |
2310 | "md/raid10:%s: unable to read back " | |
2311 | "corrected sectors" | |
2312 | " (%d sectors at %llu on %s)\n", | |
2313 | mdname(mddev), s, | |
2314 | (unsigned long long)( | |
f8c9e74f N |
2315 | sect + |
2316 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2317 | bdevname(rdev->bdev, b)); |
2318 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
2319 | "drive\n", | |
2320 | mdname(mddev), | |
2321 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2322 | break; |
2323 | case 1: | |
1294b9c9 N |
2324 | printk(KERN_INFO |
2325 | "md/raid10:%s: read error corrected" | |
2326 | " (%d sectors at %llu on %s)\n", | |
2327 | mdname(mddev), s, | |
2328 | (unsigned long long)( | |
f8c9e74f N |
2329 | sect + |
2330 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2331 | bdevname(rdev->bdev, b)); |
2332 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2333 | } |
1294b9c9 N |
2334 | |
2335 | rdev_dec_pending(rdev, mddev); | |
2336 | rcu_read_lock(); | |
6814d536 N |
2337 | } |
2338 | rcu_read_unlock(); | |
2339 | ||
2340 | sectors -= s; | |
2341 | sect += s; | |
2342 | } | |
2343 | } | |
2344 | ||
bd870a16 N |
2345 | static void bi_complete(struct bio *bio, int error) |
2346 | { | |
2347 | complete((struct completion *)bio->bi_private); | |
2348 | } | |
2349 | ||
2350 | static int submit_bio_wait(int rw, struct bio *bio) | |
2351 | { | |
2352 | struct completion event; | |
2353 | rw |= REQ_SYNC; | |
2354 | ||
2355 | init_completion(&event); | |
2356 | bio->bi_private = &event; | |
2357 | bio->bi_end_io = bi_complete; | |
2358 | submit_bio(rw, bio); | |
2359 | wait_for_completion(&event); | |
2360 | ||
2361 | return test_bit(BIO_UPTODATE, &bio->bi_flags); | |
2362 | } | |
2363 | ||
9f2c9d12 | 2364 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2365 | { |
2366 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2367 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2368 | struct r10conf *conf = mddev->private; |
3cb03002 | 2369 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2370 | /* bio has the data to be written to slot 'i' where |
2371 | * we just recently had a write error. | |
2372 | * We repeatedly clone the bio and trim down to one block, | |
2373 | * then try the write. Where the write fails we record | |
2374 | * a bad block. | |
2375 | * It is conceivable that the bio doesn't exactly align with | |
2376 | * blocks. We must handle this. | |
2377 | * | |
2378 | * We currently own a reference to the rdev. | |
2379 | */ | |
2380 | ||
2381 | int block_sectors; | |
2382 | sector_t sector; | |
2383 | int sectors; | |
2384 | int sect_to_write = r10_bio->sectors; | |
2385 | int ok = 1; | |
2386 | ||
2387 | if (rdev->badblocks.shift < 0) | |
2388 | return 0; | |
2389 | ||
2390 | block_sectors = 1 << rdev->badblocks.shift; | |
2391 | sector = r10_bio->sector; | |
2392 | sectors = ((r10_bio->sector + block_sectors) | |
2393 | & ~(sector_t)(block_sectors - 1)) | |
2394 | - sector; | |
2395 | ||
2396 | while (sect_to_write) { | |
2397 | struct bio *wbio; | |
2398 | if (sectors > sect_to_write) | |
2399 | sectors = sect_to_write; | |
2400 | /* Write at 'sector' for 'sectors' */ | |
2401 | wbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
2402 | md_trim_bio(wbio, sector - bio->bi_sector, sectors); | |
2403 | wbio->bi_sector = (r10_bio->devs[i].addr+ | |
f8c9e74f | 2404 | choose_data_offset(r10_bio, rdev) + |
bd870a16 N |
2405 | (sector - r10_bio->sector)); |
2406 | wbio->bi_bdev = rdev->bdev; | |
2407 | if (submit_bio_wait(WRITE, wbio) == 0) | |
2408 | /* Failure! */ | |
2409 | ok = rdev_set_badblocks(rdev, sector, | |
2410 | sectors, 0) | |
2411 | && ok; | |
2412 | ||
2413 | bio_put(wbio); | |
2414 | sect_to_write -= sectors; | |
2415 | sector += sectors; | |
2416 | sectors = block_sectors; | |
2417 | } | |
2418 | return ok; | |
2419 | } | |
2420 | ||
9f2c9d12 | 2421 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2422 | { |
2423 | int slot = r10_bio->read_slot; | |
560f8e55 | 2424 | struct bio *bio; |
e879a879 | 2425 | struct r10conf *conf = mddev->private; |
abbf098e | 2426 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2427 | char b[BDEVNAME_SIZE]; |
2428 | unsigned long do_sync; | |
856e08e2 | 2429 | int max_sectors; |
560f8e55 N |
2430 | |
2431 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2432 | * the block and we can fix it. | |
2433 | * We freeze all other IO, and try reading the block from | |
2434 | * other devices. When we find one, we re-write | |
2435 | * and check it that fixes the read error. | |
2436 | * This is all done synchronously while the array is | |
2437 | * frozen. | |
2438 | */ | |
fae8cc5e N |
2439 | bio = r10_bio->devs[slot].bio; |
2440 | bdevname(bio->bi_bdev, b); | |
2441 | bio_put(bio); | |
2442 | r10_bio->devs[slot].bio = NULL; | |
2443 | ||
560f8e55 N |
2444 | if (mddev->ro == 0) { |
2445 | freeze_array(conf); | |
2446 | fix_read_error(conf, mddev, r10_bio); | |
2447 | unfreeze_array(conf); | |
fae8cc5e N |
2448 | } else |
2449 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2450 | ||
abbf098e | 2451 | rdev_dec_pending(rdev, mddev); |
560f8e55 | 2452 | |
7399c31b | 2453 | read_more: |
96c3fd1f N |
2454 | rdev = read_balance(conf, r10_bio, &max_sectors); |
2455 | if (rdev == NULL) { | |
560f8e55 N |
2456 | printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O" |
2457 | " read error for block %llu\n", | |
7399c31b | 2458 | mdname(mddev), b, |
560f8e55 N |
2459 | (unsigned long long)r10_bio->sector); |
2460 | raid_end_bio_io(r10_bio); | |
560f8e55 N |
2461 | return; |
2462 | } | |
2463 | ||
2464 | do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC); | |
560f8e55 | 2465 | slot = r10_bio->read_slot; |
560f8e55 N |
2466 | printk_ratelimited( |
2467 | KERN_ERR | |
2468 | "md/raid10:%s: %s: redirecting" | |
2469 | "sector %llu to another mirror\n", | |
2470 | mdname(mddev), | |
2471 | bdevname(rdev->bdev, b), | |
2472 | (unsigned long long)r10_bio->sector); | |
2473 | bio = bio_clone_mddev(r10_bio->master_bio, | |
2474 | GFP_NOIO, mddev); | |
7399c31b N |
2475 | md_trim_bio(bio, |
2476 | r10_bio->sector - bio->bi_sector, | |
2477 | max_sectors); | |
560f8e55 | 2478 | r10_bio->devs[slot].bio = bio; |
abbf098e | 2479 | r10_bio->devs[slot].rdev = rdev; |
560f8e55 | 2480 | bio->bi_sector = r10_bio->devs[slot].addr |
f8c9e74f | 2481 | + choose_data_offset(r10_bio, rdev); |
560f8e55 N |
2482 | bio->bi_bdev = rdev->bdev; |
2483 | bio->bi_rw = READ | do_sync; | |
2484 | bio->bi_private = r10_bio; | |
2485 | bio->bi_end_io = raid10_end_read_request; | |
7399c31b N |
2486 | if (max_sectors < r10_bio->sectors) { |
2487 | /* Drat - have to split this up more */ | |
2488 | struct bio *mbio = r10_bio->master_bio; | |
2489 | int sectors_handled = | |
2490 | r10_bio->sector + max_sectors | |
2491 | - mbio->bi_sector; | |
2492 | r10_bio->sectors = max_sectors; | |
2493 | spin_lock_irq(&conf->device_lock); | |
2494 | if (mbio->bi_phys_segments == 0) | |
2495 | mbio->bi_phys_segments = 2; | |
2496 | else | |
2497 | mbio->bi_phys_segments++; | |
2498 | spin_unlock_irq(&conf->device_lock); | |
2499 | generic_make_request(bio); | |
7399c31b N |
2500 | |
2501 | r10_bio = mempool_alloc(conf->r10bio_pool, | |
2502 | GFP_NOIO); | |
2503 | r10_bio->master_bio = mbio; | |
2504 | r10_bio->sectors = (mbio->bi_size >> 9) | |
2505 | - sectors_handled; | |
2506 | r10_bio->state = 0; | |
2507 | set_bit(R10BIO_ReadError, | |
2508 | &r10_bio->state); | |
2509 | r10_bio->mddev = mddev; | |
2510 | r10_bio->sector = mbio->bi_sector | |
2511 | + sectors_handled; | |
2512 | ||
2513 | goto read_more; | |
2514 | } else | |
2515 | generic_make_request(bio); | |
560f8e55 N |
2516 | } |
2517 | ||
e879a879 | 2518 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2519 | { |
2520 | /* Some sort of write request has finished and it | |
2521 | * succeeded in writing where we thought there was a | |
2522 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2523 | * Or possibly if failed and we need to record |
2524 | * a bad block. | |
749c55e9 N |
2525 | */ |
2526 | int m; | |
3cb03002 | 2527 | struct md_rdev *rdev; |
749c55e9 N |
2528 | |
2529 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2530 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2531 | for (m = 0; m < conf->copies; m++) { |
2532 | int dev = r10_bio->devs[m].devnum; | |
2533 | rdev = conf->mirrors[dev].rdev; | |
2534 | if (r10_bio->devs[m].bio == NULL) | |
2535 | continue; | |
2536 | if (test_bit(BIO_UPTODATE, | |
749c55e9 | 2537 | &r10_bio->devs[m].bio->bi_flags)) { |
749c55e9 N |
2538 | rdev_clear_badblocks( |
2539 | rdev, | |
2540 | r10_bio->devs[m].addr, | |
c6563a8c | 2541 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2542 | } else { |
2543 | if (!rdev_set_badblocks( | |
2544 | rdev, | |
2545 | r10_bio->devs[m].addr, | |
2546 | r10_bio->sectors, 0)) | |
2547 | md_error(conf->mddev, rdev); | |
749c55e9 | 2548 | } |
9ad1aefc N |
2549 | rdev = conf->mirrors[dev].replacement; |
2550 | if (r10_bio->devs[m].repl_bio == NULL) | |
2551 | continue; | |
2552 | if (test_bit(BIO_UPTODATE, | |
2553 | &r10_bio->devs[m].repl_bio->bi_flags)) { | |
2554 | rdev_clear_badblocks( | |
2555 | rdev, | |
2556 | r10_bio->devs[m].addr, | |
c6563a8c | 2557 | r10_bio->sectors, 0); |
9ad1aefc N |
2558 | } else { |
2559 | if (!rdev_set_badblocks( | |
2560 | rdev, | |
2561 | r10_bio->devs[m].addr, | |
2562 | r10_bio->sectors, 0)) | |
2563 | md_error(conf->mddev, rdev); | |
2564 | } | |
1a0b7cd8 | 2565 | } |
749c55e9 N |
2566 | put_buf(r10_bio); |
2567 | } else { | |
bd870a16 N |
2568 | for (m = 0; m < conf->copies; m++) { |
2569 | int dev = r10_bio->devs[m].devnum; | |
2570 | struct bio *bio = r10_bio->devs[m].bio; | |
2571 | rdev = conf->mirrors[dev].rdev; | |
2572 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2573 | rdev_clear_badblocks( |
2574 | rdev, | |
2575 | r10_bio->devs[m].addr, | |
c6563a8c | 2576 | r10_bio->sectors, 0); |
749c55e9 | 2577 | rdev_dec_pending(rdev, conf->mddev); |
bd870a16 N |
2578 | } else if (bio != NULL && |
2579 | !test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
2580 | if (!narrow_write_error(r10_bio, m)) { | |
2581 | md_error(conf->mddev, rdev); | |
2582 | set_bit(R10BIO_Degraded, | |
2583 | &r10_bio->state); | |
2584 | } | |
2585 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2586 | } |
475b0321 N |
2587 | bio = r10_bio->devs[m].repl_bio; |
2588 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2589 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2590 | rdev_clear_badblocks( |
2591 | rdev, | |
2592 | r10_bio->devs[m].addr, | |
c6563a8c | 2593 | r10_bio->sectors, 0); |
475b0321 N |
2594 | rdev_dec_pending(rdev, conf->mddev); |
2595 | } | |
bd870a16 N |
2596 | } |
2597 | if (test_bit(R10BIO_WriteError, | |
2598 | &r10_bio->state)) | |
2599 | close_write(r10_bio); | |
749c55e9 N |
2600 | raid_end_bio_io(r10_bio); |
2601 | } | |
2602 | } | |
2603 | ||
fd01b88c | 2604 | static void raid10d(struct mddev *mddev) |
1da177e4 | 2605 | { |
9f2c9d12 | 2606 | struct r10bio *r10_bio; |
1da177e4 | 2607 | unsigned long flags; |
e879a879 | 2608 | struct r10conf *conf = mddev->private; |
1da177e4 | 2609 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2610 | struct blk_plug plug; |
1da177e4 LT |
2611 | |
2612 | md_check_recovery(mddev); | |
1da177e4 | 2613 | |
e1dfa0a2 | 2614 | blk_start_plug(&plug); |
1da177e4 | 2615 | for (;;) { |
6cce3b23 | 2616 | |
7eaceacc | 2617 | flush_pending_writes(conf); |
6cce3b23 | 2618 | |
a35e63ef N |
2619 | spin_lock_irqsave(&conf->device_lock, flags); |
2620 | if (list_empty(head)) { | |
2621 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2622 | break; |
a35e63ef | 2623 | } |
9f2c9d12 | 2624 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2625 | list_del(head->prev); |
4443ae10 | 2626 | conf->nr_queued--; |
1da177e4 LT |
2627 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2628 | ||
2629 | mddev = r10_bio->mddev; | |
070ec55d | 2630 | conf = mddev->private; |
bd870a16 N |
2631 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2632 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
2633 | handle_write_completed(conf, r10_bio); |
2634 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) | |
1da177e4 | 2635 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2636 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2637 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2638 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2639 | handle_read_error(mddev, r10_bio); |
856e08e2 N |
2640 | else { |
2641 | /* just a partial read to be scheduled from a | |
2642 | * separate context | |
2643 | */ | |
2644 | int slot = r10_bio->read_slot; | |
2645 | generic_make_request(r10_bio->devs[slot].bio); | |
2646 | } | |
560f8e55 | 2647 | |
1d9d5241 | 2648 | cond_resched(); |
de393cde N |
2649 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2650 | md_check_recovery(mddev); | |
1da177e4 | 2651 | } |
e1dfa0a2 | 2652 | blk_finish_plug(&plug); |
1da177e4 LT |
2653 | } |
2654 | ||
2655 | ||
e879a879 | 2656 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2657 | { |
2658 | int buffs; | |
69335ef3 | 2659 | int i; |
1da177e4 LT |
2660 | |
2661 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2662 | BUG_ON(conf->r10buf_pool); |
69335ef3 | 2663 | conf->have_replacement = 0; |
5cf00fcd | 2664 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
2665 | if (conf->mirrors[i].replacement) |
2666 | conf->have_replacement = 1; | |
1da177e4 LT |
2667 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2668 | if (!conf->r10buf_pool) | |
2669 | return -ENOMEM; | |
2670 | conf->next_resync = 0; | |
2671 | return 0; | |
2672 | } | |
2673 | ||
2674 | /* | |
2675 | * perform a "sync" on one "block" | |
2676 | * | |
2677 | * We need to make sure that no normal I/O request - particularly write | |
2678 | * requests - conflict with active sync requests. | |
2679 | * | |
2680 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2681 | * that can be installed to exclude normal IO requests. | |
2682 | * | |
2683 | * Resync and recovery are handled very differently. | |
2684 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2685 | * | |
2686 | * For resync, we iterate over virtual addresses, read all copies, | |
2687 | * and update if there are differences. If only one copy is live, | |
2688 | * skip it. | |
2689 | * For recovery, we iterate over physical addresses, read a good | |
2690 | * value for each non-in_sync drive, and over-write. | |
2691 | * | |
2692 | * So, for recovery we may have several outstanding complex requests for a | |
2693 | * given address, one for each out-of-sync device. We model this by allocating | |
2694 | * a number of r10_bio structures, one for each out-of-sync device. | |
2695 | * As we setup these structures, we collect all bio's together into a list | |
2696 | * which we then process collectively to add pages, and then process again | |
2697 | * to pass to generic_make_request. | |
2698 | * | |
2699 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2700 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2701 | * has its remaining count decremented to 0, the whole complex operation | |
2702 | * is complete. | |
2703 | * | |
2704 | */ | |
2705 | ||
fd01b88c | 2706 | static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, |
ab9d47e9 | 2707 | int *skipped, int go_faster) |
1da177e4 | 2708 | { |
e879a879 | 2709 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2710 | struct r10bio *r10_bio; |
1da177e4 LT |
2711 | struct bio *biolist = NULL, *bio; |
2712 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2713 | int i; |
6cce3b23 | 2714 | int max_sync; |
57dab0bd | 2715 | sector_t sync_blocks; |
1da177e4 LT |
2716 | sector_t sectors_skipped = 0; |
2717 | int chunks_skipped = 0; | |
5cf00fcd | 2718 | sector_t chunk_mask = conf->geo.chunk_mask; |
1da177e4 LT |
2719 | |
2720 | if (!conf->r10buf_pool) | |
2721 | if (init_resync(conf)) | |
57afd89f | 2722 | return 0; |
1da177e4 LT |
2723 | |
2724 | skipped: | |
58c0fed4 | 2725 | max_sector = mddev->dev_sectors; |
1da177e4 LT |
2726 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
2727 | max_sector = mddev->resync_max_sectors; | |
2728 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
2729 | /* If we aborted, we need to abort the |
2730 | * sync on the 'current' bitmap chucks (there can | |
2731 | * be several when recovering multiple devices). | |
2732 | * as we may have started syncing it but not finished. | |
2733 | * We can find the current address in | |
2734 | * mddev->curr_resync, but for recovery, | |
2735 | * we need to convert that to several | |
2736 | * virtual addresses. | |
2737 | */ | |
2738 | if (mddev->curr_resync < max_sector) { /* aborted */ | |
2739 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2740 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
2741 | &sync_blocks, 1); | |
5cf00fcd | 2742 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
2743 | sector_t sect = |
2744 | raid10_find_virt(conf, mddev->curr_resync, i); | |
2745 | bitmap_end_sync(mddev->bitmap, sect, | |
2746 | &sync_blocks, 1); | |
2747 | } | |
9ad1aefc N |
2748 | } else { |
2749 | /* completed sync */ | |
2750 | if ((!mddev->bitmap || conf->fullsync) | |
2751 | && conf->have_replacement | |
2752 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
2753 | /* Completed a full sync so the replacements | |
2754 | * are now fully recovered. | |
2755 | */ | |
5cf00fcd | 2756 | for (i = 0; i < conf->geo.raid_disks; i++) |
9ad1aefc N |
2757 | if (conf->mirrors[i].replacement) |
2758 | conf->mirrors[i].replacement | |
2759 | ->recovery_offset | |
2760 | = MaxSector; | |
2761 | } | |
6cce3b23 | 2762 | conf->fullsync = 0; |
9ad1aefc | 2763 | } |
6cce3b23 | 2764 | bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2765 | close_sync(conf); |
57afd89f | 2766 | *skipped = 1; |
1da177e4 LT |
2767 | return sectors_skipped; |
2768 | } | |
5cf00fcd | 2769 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
2770 | /* if there has been nothing to do on any drive, |
2771 | * then there is nothing to do at all.. | |
2772 | */ | |
57afd89f N |
2773 | *skipped = 1; |
2774 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
2775 | } |
2776 | ||
c6207277 N |
2777 | if (max_sector > mddev->resync_max) |
2778 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
2779 | ||
1da177e4 LT |
2780 | /* make sure whole request will fit in a chunk - if chunks |
2781 | * are meaningful | |
2782 | */ | |
5cf00fcd N |
2783 | if (conf->geo.near_copies < conf->geo.raid_disks && |
2784 | max_sector > (sector_nr | chunk_mask)) | |
2785 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 LT |
2786 | /* |
2787 | * If there is non-resync activity waiting for us then | |
2788 | * put in a delay to throttle resync. | |
2789 | */ | |
0a27ec96 | 2790 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 2791 | msleep_interruptible(1000); |
1da177e4 LT |
2792 | |
2793 | /* Again, very different code for resync and recovery. | |
2794 | * Both must result in an r10bio with a list of bios that | |
2795 | * have bi_end_io, bi_sector, bi_bdev set, | |
2796 | * and bi_private set to the r10bio. | |
2797 | * For recovery, we may actually create several r10bios | |
2798 | * with 2 bios in each, that correspond to the bios in the main one. | |
2799 | * In this case, the subordinate r10bios link back through a | |
2800 | * borrowed master_bio pointer, and the counter in the master | |
2801 | * includes a ref from each subordinate. | |
2802 | */ | |
2803 | /* First, we decide what to do and set ->bi_end_io | |
2804 | * To end_sync_read if we want to read, and | |
2805 | * end_sync_write if we will want to write. | |
2806 | */ | |
2807 | ||
6cce3b23 | 2808 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
2809 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
2810 | /* recovery... the complicated one */ | |
e875ecea | 2811 | int j; |
1da177e4 LT |
2812 | r10_bio = NULL; |
2813 | ||
5cf00fcd | 2814 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 2815 | int still_degraded; |
9f2c9d12 | 2816 | struct r10bio *rb2; |
ab9d47e9 N |
2817 | sector_t sect; |
2818 | int must_sync; | |
e875ecea | 2819 | int any_working; |
24afd80d N |
2820 | struct mirror_info *mirror = &conf->mirrors[i]; |
2821 | ||
2822 | if ((mirror->rdev == NULL || | |
2823 | test_bit(In_sync, &mirror->rdev->flags)) | |
2824 | && | |
2825 | (mirror->replacement == NULL || | |
2826 | test_bit(Faulty, | |
2827 | &mirror->replacement->flags))) | |
ab9d47e9 | 2828 | continue; |
1da177e4 | 2829 | |
ab9d47e9 N |
2830 | still_degraded = 0; |
2831 | /* want to reconstruct this device */ | |
2832 | rb2 = r10_bio; | |
2833 | sect = raid10_find_virt(conf, sector_nr, i); | |
24afd80d N |
2834 | /* Unless we are doing a full sync, or a replacement |
2835 | * we only need to recover the block if it is set in | |
2836 | * the bitmap | |
ab9d47e9 N |
2837 | */ |
2838 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
2839 | &sync_blocks, 1); | |
2840 | if (sync_blocks < max_sync) | |
2841 | max_sync = sync_blocks; | |
2842 | if (!must_sync && | |
24afd80d | 2843 | mirror->replacement == NULL && |
ab9d47e9 N |
2844 | !conf->fullsync) { |
2845 | /* yep, skip the sync_blocks here, but don't assume | |
2846 | * that there will never be anything to do here | |
2847 | */ | |
2848 | chunks_skipped = -1; | |
2849 | continue; | |
2850 | } | |
6cce3b23 | 2851 | |
ab9d47e9 N |
2852 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
2853 | raise_barrier(conf, rb2 != NULL); | |
2854 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 2855 | |
ab9d47e9 N |
2856 | r10_bio->master_bio = (struct bio*)rb2; |
2857 | if (rb2) | |
2858 | atomic_inc(&rb2->remaining); | |
2859 | r10_bio->mddev = mddev; | |
2860 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
2861 | r10_bio->sector = sect; | |
1da177e4 | 2862 | |
ab9d47e9 N |
2863 | raid10_find_phys(conf, r10_bio); |
2864 | ||
2865 | /* Need to check if the array will still be | |
2866 | * degraded | |
2867 | */ | |
5cf00fcd | 2868 | for (j = 0; j < conf->geo.raid_disks; j++) |
ab9d47e9 N |
2869 | if (conf->mirrors[j].rdev == NULL || |
2870 | test_bit(Faulty, &conf->mirrors[j].rdev->flags)) { | |
2871 | still_degraded = 1; | |
87fc767b | 2872 | break; |
1da177e4 | 2873 | } |
ab9d47e9 N |
2874 | |
2875 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
2876 | &sync_blocks, still_degraded); | |
2877 | ||
e875ecea | 2878 | any_working = 0; |
ab9d47e9 | 2879 | for (j=0; j<conf->copies;j++) { |
e875ecea | 2880 | int k; |
ab9d47e9 | 2881 | int d = r10_bio->devs[j].devnum; |
5e570289 | 2882 | sector_t from_addr, to_addr; |
3cb03002 | 2883 | struct md_rdev *rdev; |
40c356ce N |
2884 | sector_t sector, first_bad; |
2885 | int bad_sectors; | |
ab9d47e9 N |
2886 | if (!conf->mirrors[d].rdev || |
2887 | !test_bit(In_sync, &conf->mirrors[d].rdev->flags)) | |
2888 | continue; | |
2889 | /* This is where we read from */ | |
e875ecea | 2890 | any_working = 1; |
40c356ce N |
2891 | rdev = conf->mirrors[d].rdev; |
2892 | sector = r10_bio->devs[j].addr; | |
2893 | ||
2894 | if (is_badblock(rdev, sector, max_sync, | |
2895 | &first_bad, &bad_sectors)) { | |
2896 | if (first_bad > sector) | |
2897 | max_sync = first_bad - sector; | |
2898 | else { | |
2899 | bad_sectors -= (sector | |
2900 | - first_bad); | |
2901 | if (max_sync > bad_sectors) | |
2902 | max_sync = bad_sectors; | |
2903 | continue; | |
2904 | } | |
2905 | } | |
ab9d47e9 N |
2906 | bio = r10_bio->devs[0].bio; |
2907 | bio->bi_next = biolist; | |
2908 | biolist = bio; | |
2909 | bio->bi_private = r10_bio; | |
2910 | bio->bi_end_io = end_sync_read; | |
2911 | bio->bi_rw = READ; | |
5e570289 | 2912 | from_addr = r10_bio->devs[j].addr; |
24afd80d N |
2913 | bio->bi_sector = from_addr + rdev->data_offset; |
2914 | bio->bi_bdev = rdev->bdev; | |
2915 | atomic_inc(&rdev->nr_pending); | |
2916 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
2917 | |
2918 | for (k=0; k<conf->copies; k++) | |
2919 | if (r10_bio->devs[k].devnum == i) | |
2920 | break; | |
2921 | BUG_ON(k == conf->copies); | |
5e570289 | 2922 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 2923 | r10_bio->devs[0].devnum = d; |
5e570289 | 2924 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 2925 | r10_bio->devs[1].devnum = i; |
5e570289 | 2926 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 2927 | |
24afd80d N |
2928 | rdev = mirror->rdev; |
2929 | if (!test_bit(In_sync, &rdev->flags)) { | |
2930 | bio = r10_bio->devs[1].bio; | |
2931 | bio->bi_next = biolist; | |
2932 | biolist = bio; | |
2933 | bio->bi_private = r10_bio; | |
2934 | bio->bi_end_io = end_sync_write; | |
2935 | bio->bi_rw = WRITE; | |
2936 | bio->bi_sector = to_addr | |
2937 | + rdev->data_offset; | |
2938 | bio->bi_bdev = rdev->bdev; | |
2939 | atomic_inc(&r10_bio->remaining); | |
2940 | } else | |
2941 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
2942 | ||
2943 | /* and maybe write to replacement */ | |
2944 | bio = r10_bio->devs[1].repl_bio; | |
2945 | if (bio) | |
2946 | bio->bi_end_io = NULL; | |
2947 | rdev = mirror->replacement; | |
2948 | /* Note: if rdev != NULL, then bio | |
2949 | * cannot be NULL as r10buf_pool_alloc will | |
2950 | * have allocated it. | |
2951 | * So the second test here is pointless. | |
2952 | * But it keeps semantic-checkers happy, and | |
2953 | * this comment keeps human reviewers | |
2954 | * happy. | |
2955 | */ | |
2956 | if (rdev == NULL || bio == NULL || | |
2957 | test_bit(Faulty, &rdev->flags)) | |
2958 | break; | |
2959 | bio->bi_next = biolist; | |
2960 | biolist = bio; | |
2961 | bio->bi_private = r10_bio; | |
2962 | bio->bi_end_io = end_sync_write; | |
2963 | bio->bi_rw = WRITE; | |
2964 | bio->bi_sector = to_addr + rdev->data_offset; | |
2965 | bio->bi_bdev = rdev->bdev; | |
2966 | atomic_inc(&r10_bio->remaining); | |
ab9d47e9 N |
2967 | break; |
2968 | } | |
2969 | if (j == conf->copies) { | |
e875ecea N |
2970 | /* Cannot recover, so abort the recovery or |
2971 | * record a bad block */ | |
ab9d47e9 N |
2972 | put_buf(r10_bio); |
2973 | if (rb2) | |
2974 | atomic_dec(&rb2->remaining); | |
2975 | r10_bio = rb2; | |
e875ecea N |
2976 | if (any_working) { |
2977 | /* problem is that there are bad blocks | |
2978 | * on other device(s) | |
2979 | */ | |
2980 | int k; | |
2981 | for (k = 0; k < conf->copies; k++) | |
2982 | if (r10_bio->devs[k].devnum == i) | |
2983 | break; | |
24afd80d N |
2984 | if (!test_bit(In_sync, |
2985 | &mirror->rdev->flags) | |
2986 | && !rdev_set_badblocks( | |
2987 | mirror->rdev, | |
2988 | r10_bio->devs[k].addr, | |
2989 | max_sync, 0)) | |
2990 | any_working = 0; | |
2991 | if (mirror->replacement && | |
2992 | !rdev_set_badblocks( | |
2993 | mirror->replacement, | |
e875ecea N |
2994 | r10_bio->devs[k].addr, |
2995 | max_sync, 0)) | |
2996 | any_working = 0; | |
2997 | } | |
2998 | if (!any_working) { | |
2999 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3000 | &mddev->recovery)) | |
3001 | printk(KERN_INFO "md/raid10:%s: insufficient " | |
3002 | "working devices for recovery.\n", | |
3003 | mdname(mddev)); | |
24afd80d | 3004 | mirror->recovery_disabled |
e875ecea N |
3005 | = mddev->recovery_disabled; |
3006 | } | |
ab9d47e9 | 3007 | break; |
1da177e4 | 3008 | } |
ab9d47e9 | 3009 | } |
1da177e4 LT |
3010 | if (biolist == NULL) { |
3011 | while (r10_bio) { | |
9f2c9d12 N |
3012 | struct r10bio *rb2 = r10_bio; |
3013 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3014 | rb2->master_bio = NULL; |
3015 | put_buf(rb2); | |
3016 | } | |
3017 | goto giveup; | |
3018 | } | |
3019 | } else { | |
3020 | /* resync. Schedule a read for every block at this virt offset */ | |
3021 | int count = 0; | |
6cce3b23 | 3022 | |
78200d45 N |
3023 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
3024 | ||
6cce3b23 N |
3025 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
3026 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3027 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3028 | &mddev->recovery)) { | |
6cce3b23 N |
3029 | /* We can skip this block */ |
3030 | *skipped = 1; | |
3031 | return sync_blocks + sectors_skipped; | |
3032 | } | |
3033 | if (sync_blocks < max_sync) | |
3034 | max_sync = sync_blocks; | |
1da177e4 LT |
3035 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
3036 | ||
1da177e4 LT |
3037 | r10_bio->mddev = mddev; |
3038 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3039 | raise_barrier(conf, 0); |
3040 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3041 | |
3042 | r10_bio->master_bio = NULL; | |
3043 | r10_bio->sector = sector_nr; | |
3044 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3045 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3046 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3047 | |
5cf00fcd | 3048 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3049 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3050 | sector_t first_bad, sector; |
3051 | int bad_sectors; | |
3052 | ||
9ad1aefc N |
3053 | if (r10_bio->devs[i].repl_bio) |
3054 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3055 | ||
1da177e4 LT |
3056 | bio = r10_bio->devs[i].bio; |
3057 | bio->bi_end_io = NULL; | |
af03b8e4 | 3058 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
1da177e4 | 3059 | if (conf->mirrors[d].rdev == NULL || |
b2d444d7 | 3060 | test_bit(Faulty, &conf->mirrors[d].rdev->flags)) |
1da177e4 | 3061 | continue; |
40c356ce N |
3062 | sector = r10_bio->devs[i].addr; |
3063 | if (is_badblock(conf->mirrors[d].rdev, | |
3064 | sector, max_sync, | |
3065 | &first_bad, &bad_sectors)) { | |
3066 | if (first_bad > sector) | |
3067 | max_sync = first_bad - sector; | |
3068 | else { | |
3069 | bad_sectors -= (sector - first_bad); | |
3070 | if (max_sync > bad_sectors) | |
3071 | max_sync = max_sync; | |
3072 | continue; | |
3073 | } | |
3074 | } | |
1da177e4 LT |
3075 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
3076 | atomic_inc(&r10_bio->remaining); | |
3077 | bio->bi_next = biolist; | |
3078 | biolist = bio; | |
3079 | bio->bi_private = r10_bio; | |
3080 | bio->bi_end_io = end_sync_read; | |
802ba064 | 3081 | bio->bi_rw = READ; |
40c356ce | 3082 | bio->bi_sector = sector + |
1da177e4 LT |
3083 | conf->mirrors[d].rdev->data_offset; |
3084 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
3085 | count++; | |
9ad1aefc N |
3086 | |
3087 | if (conf->mirrors[d].replacement == NULL || | |
3088 | test_bit(Faulty, | |
3089 | &conf->mirrors[d].replacement->flags)) | |
3090 | continue; | |
3091 | ||
3092 | /* Need to set up for writing to the replacement */ | |
3093 | bio = r10_bio->devs[i].repl_bio; | |
3094 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
3095 | ||
3096 | sector = r10_bio->devs[i].addr; | |
3097 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
3098 | bio->bi_next = biolist; | |
3099 | biolist = bio; | |
3100 | bio->bi_private = r10_bio; | |
3101 | bio->bi_end_io = end_sync_write; | |
3102 | bio->bi_rw = WRITE; | |
3103 | bio->bi_sector = sector + | |
3104 | conf->mirrors[d].replacement->data_offset; | |
3105 | bio->bi_bdev = conf->mirrors[d].replacement->bdev; | |
3106 | count++; | |
1da177e4 LT |
3107 | } |
3108 | ||
3109 | if (count < 2) { | |
3110 | for (i=0; i<conf->copies; i++) { | |
3111 | int d = r10_bio->devs[i].devnum; | |
3112 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3113 | rdev_dec_pending(conf->mirrors[d].rdev, |
3114 | mddev); | |
9ad1aefc N |
3115 | if (r10_bio->devs[i].repl_bio && |
3116 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3117 | rdev_dec_pending( | |
3118 | conf->mirrors[d].replacement, | |
3119 | mddev); | |
1da177e4 LT |
3120 | } |
3121 | put_buf(r10_bio); | |
3122 | biolist = NULL; | |
3123 | goto giveup; | |
3124 | } | |
3125 | } | |
3126 | ||
3127 | for (bio = biolist; bio ; bio=bio->bi_next) { | |
3128 | ||
3129 | bio->bi_flags &= ~(BIO_POOL_MASK - 1); | |
3130 | if (bio->bi_end_io) | |
3131 | bio->bi_flags |= 1 << BIO_UPTODATE; | |
3132 | bio->bi_vcnt = 0; | |
3133 | bio->bi_idx = 0; | |
3134 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
3135 | bio->bi_size = 0; |
3136 | } | |
3137 | ||
3138 | nr_sectors = 0; | |
6cce3b23 N |
3139 | if (sector_nr + max_sync < max_sector) |
3140 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3141 | do { |
3142 | struct page *page; | |
3143 | int len = PAGE_SIZE; | |
1da177e4 LT |
3144 | if (sector_nr + (len>>9) > max_sector) |
3145 | len = (max_sector - sector_nr) << 9; | |
3146 | if (len == 0) | |
3147 | break; | |
3148 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
ab9d47e9 | 3149 | struct bio *bio2; |
1da177e4 | 3150 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
ab9d47e9 N |
3151 | if (bio_add_page(bio, page, len, 0)) |
3152 | continue; | |
3153 | ||
3154 | /* stop here */ | |
3155 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
3156 | for (bio2 = biolist; | |
3157 | bio2 && bio2 != bio; | |
3158 | bio2 = bio2->bi_next) { | |
3159 | /* remove last page from this bio */ | |
3160 | bio2->bi_vcnt--; | |
3161 | bio2->bi_size -= len; | |
3162 | bio2->bi_flags &= ~(1<< BIO_SEG_VALID); | |
1da177e4 | 3163 | } |
ab9d47e9 | 3164 | goto bio_full; |
1da177e4 LT |
3165 | } |
3166 | nr_sectors += len>>9; | |
3167 | sector_nr += len>>9; | |
3168 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
3169 | bio_full: | |
3170 | r10_bio->sectors = nr_sectors; | |
3171 | ||
3172 | while (biolist) { | |
3173 | bio = biolist; | |
3174 | biolist = biolist->bi_next; | |
3175 | ||
3176 | bio->bi_next = NULL; | |
3177 | r10_bio = bio->bi_private; | |
3178 | r10_bio->sectors = nr_sectors; | |
3179 | ||
3180 | if (bio->bi_end_io == end_sync_read) { | |
3181 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
3182 | generic_make_request(bio); | |
3183 | } | |
3184 | } | |
3185 | ||
57afd89f N |
3186 | if (sectors_skipped) |
3187 | /* pretend they weren't skipped, it makes | |
3188 | * no important difference in this case | |
3189 | */ | |
3190 | md_done_sync(mddev, sectors_skipped, 1); | |
3191 | ||
1da177e4 LT |
3192 | return sectors_skipped + nr_sectors; |
3193 | giveup: | |
3194 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3195 | * drives must be failed or in resync, all drives |
3196 | * have a bad block, so try the next chunk... | |
1da177e4 | 3197 | */ |
09b4068a N |
3198 | if (sector_nr + max_sync < max_sector) |
3199 | max_sector = sector_nr + max_sync; | |
3200 | ||
3201 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3202 | chunks_skipped ++; |
3203 | sector_nr = max_sector; | |
1da177e4 | 3204 | goto skipped; |
1da177e4 LT |
3205 | } |
3206 | ||
80c3a6ce | 3207 | static sector_t |
fd01b88c | 3208 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3209 | { |
3210 | sector_t size; | |
e879a879 | 3211 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3212 | |
3213 | if (!raid_disks) | |
5cf00fcd | 3214 | raid_disks = conf->geo.raid_disks; |
80c3a6ce | 3215 | if (!sectors) |
dab8b292 | 3216 | sectors = conf->dev_sectors; |
80c3a6ce | 3217 | |
5cf00fcd N |
3218 | size = sectors >> conf->geo.chunk_shift; |
3219 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3220 | size = size * raid_disks; |
5cf00fcd | 3221 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3222 | |
5cf00fcd | 3223 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3224 | } |
3225 | ||
6508fdbf N |
3226 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3227 | { | |
3228 | /* Calculate the number of sectors-per-device that will | |
3229 | * actually be used, and set conf->dev_sectors and | |
3230 | * conf->stride | |
3231 | */ | |
3232 | ||
5cf00fcd N |
3233 | size = size >> conf->geo.chunk_shift; |
3234 | sector_div(size, conf->geo.far_copies); | |
3235 | size = size * conf->geo.raid_disks; | |
3236 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3237 | /* 'size' is now the number of chunks in the array */ |
3238 | /* calculate "used chunks per device" */ | |
3239 | size = size * conf->copies; | |
3240 | ||
3241 | /* We need to round up when dividing by raid_disks to | |
3242 | * get the stride size. | |
3243 | */ | |
5cf00fcd | 3244 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3245 | |
5cf00fcd | 3246 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3247 | |
5cf00fcd N |
3248 | if (conf->geo.far_offset) |
3249 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3250 | else { |
5cf00fcd N |
3251 | sector_div(size, conf->geo.far_copies); |
3252 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3253 | } |
3254 | } | |
dab8b292 | 3255 | |
e879a879 | 3256 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3257 | { |
e879a879 | 3258 | struct r10conf *conf = NULL; |
c93983bf | 3259 | int nc, fc, fo; |
dab8b292 | 3260 | int err = -EINVAL; |
1da177e4 | 3261 | |
f73ea873 MT |
3262 | if (mddev->new_chunk_sectors < (PAGE_SIZE >> 9) || |
3263 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
128595ed N |
3264 | printk(KERN_ERR "md/raid10:%s: chunk size must be " |
3265 | "at least PAGE_SIZE(%ld) and be a power of 2.\n", | |
3266 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3267 | goto out; |
1da177e4 | 3268 | } |
2604b703 | 3269 | |
f73ea873 MT |
3270 | nc = mddev->new_layout & 255; |
3271 | fc = (mddev->new_layout >> 8) & 255; | |
3272 | fo = mddev->new_layout & (1<<16); | |
dab8b292 | 3273 | |
1da177e4 | 3274 | if ((nc*fc) <2 || (nc*fc) > mddev->raid_disks || |
f73ea873 | 3275 | (mddev->new_layout >> 17)) { |
128595ed | 3276 | printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
f73ea873 | 3277 | mdname(mddev), mddev->new_layout); |
1da177e4 LT |
3278 | goto out; |
3279 | } | |
dab8b292 TM |
3280 | |
3281 | err = -ENOMEM; | |
e879a879 | 3282 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3283 | if (!conf) |
1da177e4 | 3284 | goto out; |
dab8b292 | 3285 | |
4443ae10 | 3286 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
dab8b292 TM |
3287 | GFP_KERNEL); |
3288 | if (!conf->mirrors) | |
3289 | goto out; | |
4443ae10 N |
3290 | |
3291 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3292 | if (!conf->tmppage) | |
dab8b292 TM |
3293 | goto out; |
3294 | ||
1da177e4 | 3295 | |
5cf00fcd N |
3296 | conf->geo.raid_disks = mddev->raid_disks; |
3297 | conf->geo.near_copies = nc; | |
3298 | conf->geo.far_copies = fc; | |
1da177e4 | 3299 | conf->copies = nc*fc; |
5cf00fcd N |
3300 | conf->geo.far_offset = fo; |
3301 | conf->geo.chunk_mask = mddev->new_chunk_sectors - 1; | |
3302 | conf->geo.chunk_shift = ffz(~mddev->new_chunk_sectors); | |
dab8b292 TM |
3303 | |
3304 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, | |
3305 | r10bio_pool_free, conf); | |
3306 | if (!conf->r10bio_pool) | |
3307 | goto out; | |
3308 | ||
6508fdbf | 3309 | calc_sectors(conf, mddev->dev_sectors); |
f8c9e74f N |
3310 | conf->prev = conf->geo; |
3311 | conf->reshape_progress = MaxSector; | |
1da177e4 | 3312 | |
e7e72bf6 | 3313 | spin_lock_init(&conf->device_lock); |
dab8b292 TM |
3314 | INIT_LIST_HEAD(&conf->retry_list); |
3315 | ||
3316 | spin_lock_init(&conf->resync_lock); | |
3317 | init_waitqueue_head(&conf->wait_barrier); | |
3318 | ||
3319 | conf->thread = md_register_thread(raid10d, mddev, NULL); | |
3320 | if (!conf->thread) | |
3321 | goto out; | |
3322 | ||
dab8b292 TM |
3323 | conf->mddev = mddev; |
3324 | return conf; | |
3325 | ||
3326 | out: | |
128595ed | 3327 | printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n", |
dab8b292 TM |
3328 | mdname(mddev)); |
3329 | if (conf) { | |
3330 | if (conf->r10bio_pool) | |
3331 | mempool_destroy(conf->r10bio_pool); | |
3332 | kfree(conf->mirrors); | |
3333 | safe_put_page(conf->tmppage); | |
3334 | kfree(conf); | |
3335 | } | |
3336 | return ERR_PTR(err); | |
3337 | } | |
3338 | ||
fd01b88c | 3339 | static int run(struct mddev *mddev) |
dab8b292 | 3340 | { |
e879a879 | 3341 | struct r10conf *conf; |
dab8b292 | 3342 | int i, disk_idx, chunk_size; |
0f6d02d5 | 3343 | struct mirror_info *disk; |
3cb03002 | 3344 | struct md_rdev *rdev; |
dab8b292 TM |
3345 | sector_t size; |
3346 | ||
3347 | /* | |
3348 | * copy the already verified devices into our private RAID10 | |
3349 | * bookkeeping area. [whatever we allocate in run(), | |
3350 | * should be freed in stop()] | |
3351 | */ | |
3352 | ||
3353 | if (mddev->private == NULL) { | |
3354 | conf = setup_conf(mddev); | |
3355 | if (IS_ERR(conf)) | |
3356 | return PTR_ERR(conf); | |
3357 | mddev->private = conf; | |
3358 | } | |
3359 | conf = mddev->private; | |
3360 | if (!conf) | |
3361 | goto out; | |
3362 | ||
dab8b292 TM |
3363 | mddev->thread = conf->thread; |
3364 | conf->thread = NULL; | |
3365 | ||
8f6c2e4b MP |
3366 | chunk_size = mddev->chunk_sectors << 9; |
3367 | blk_queue_io_min(mddev->queue, chunk_size); | |
5cf00fcd N |
3368 | if (conf->geo.raid_disks % conf->geo.near_copies) |
3369 | blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks); | |
8f6c2e4b MP |
3370 | else |
3371 | blk_queue_io_opt(mddev->queue, chunk_size * | |
5cf00fcd | 3372 | (conf->geo.raid_disks / conf->geo.near_copies)); |
8f6c2e4b | 3373 | |
dafb20fa | 3374 | rdev_for_each(rdev, mddev) { |
34b343cf | 3375 | |
1da177e4 | 3376 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
3377 | if (disk_idx < 0) |
3378 | continue; | |
3379 | if (disk_idx >= conf->geo.raid_disks && | |
3380 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
3381 | continue; |
3382 | disk = conf->mirrors + disk_idx; | |
3383 | ||
56a2559b N |
3384 | if (test_bit(Replacement, &rdev->flags)) { |
3385 | if (disk->replacement) | |
3386 | goto out_free_conf; | |
3387 | disk->replacement = rdev; | |
3388 | } else { | |
3389 | if (disk->rdev) | |
3390 | goto out_free_conf; | |
3391 | disk->rdev = rdev; | |
3392 | } | |
3393 | ||
8f6c2e4b MP |
3394 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3395 | rdev->data_offset << 9); | |
1da177e4 LT |
3396 | |
3397 | disk->head_position = 0; | |
1da177e4 | 3398 | } |
6d508242 | 3399 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3400 | if (!enough(conf, -1)) { |
128595ed | 3401 | printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3402 | mdname(mddev)); |
1da177e4 LT |
3403 | goto out_free_conf; |
3404 | } | |
3405 | ||
3406 | mddev->degraded = 0; | |
f8c9e74f N |
3407 | for (i = 0; |
3408 | i < conf->geo.raid_disks | |
3409 | || i < conf->prev.raid_disks; | |
3410 | i++) { | |
1da177e4 LT |
3411 | |
3412 | disk = conf->mirrors + i; | |
3413 | ||
56a2559b N |
3414 | if (!disk->rdev && disk->replacement) { |
3415 | /* The replacement is all we have - use it */ | |
3416 | disk->rdev = disk->replacement; | |
3417 | disk->replacement = NULL; | |
3418 | clear_bit(Replacement, &disk->rdev->flags); | |
3419 | } | |
3420 | ||
5fd6c1dc | 3421 | if (!disk->rdev || |
2e333e89 | 3422 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3423 | disk->head_position = 0; |
3424 | mddev->degraded++; | |
8c2e870a NB |
3425 | if (disk->rdev) |
3426 | conf->fullsync = 1; | |
1da177e4 | 3427 | } |
d890fa2b | 3428 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3429 | } |
3430 | ||
8c6ac868 | 3431 | if (mddev->recovery_cp != MaxSector) |
128595ed | 3432 | printk(KERN_NOTICE "md/raid10:%s: not clean" |
8c6ac868 AN |
3433 | " -- starting background reconstruction\n", |
3434 | mdname(mddev)); | |
1da177e4 | 3435 | printk(KERN_INFO |
128595ed | 3436 | "md/raid10:%s: active with %d out of %d devices\n", |
5cf00fcd N |
3437 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
3438 | conf->geo.raid_disks); | |
1da177e4 LT |
3439 | /* |
3440 | * Ok, everything is just fine now | |
3441 | */ | |
dab8b292 TM |
3442 | mddev->dev_sectors = conf->dev_sectors; |
3443 | size = raid10_size(mddev, 0, 0); | |
3444 | md_set_array_sectors(mddev, size); | |
3445 | mddev->resync_max_sectors = size; | |
1da177e4 | 3446 | |
0d129228 N |
3447 | mddev->queue->backing_dev_info.congested_fn = raid10_congested; |
3448 | mddev->queue->backing_dev_info.congested_data = mddev; | |
7a5febe9 | 3449 | |
1da177e4 LT |
3450 | /* Calculate max read-ahead size. |
3451 | * We need to readahead at least twice a whole stripe.... | |
3452 | * maybe... | |
3453 | */ | |
3454 | { | |
5cf00fcd | 3455 | int stripe = conf->geo.raid_disks * |
9d8f0363 | 3456 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
5cf00fcd | 3457 | stripe /= conf->geo.near_copies; |
1da177e4 LT |
3458 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) |
3459 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
3460 | } | |
3461 | ||
050b6615 | 3462 | blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec); |
a91a2785 MP |
3463 | |
3464 | if (md_integrity_register(mddev)) | |
3465 | goto out_free_conf; | |
3466 | ||
1da177e4 LT |
3467 | return 0; |
3468 | ||
3469 | out_free_conf: | |
01f96c0a | 3470 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
3471 | if (conf->r10bio_pool) |
3472 | mempool_destroy(conf->r10bio_pool); | |
1345b1d8 | 3473 | safe_put_page(conf->tmppage); |
990a8baf | 3474 | kfree(conf->mirrors); |
1da177e4 LT |
3475 | kfree(conf); |
3476 | mddev->private = NULL; | |
3477 | out: | |
3478 | return -EIO; | |
3479 | } | |
3480 | ||
fd01b88c | 3481 | static int stop(struct mddev *mddev) |
1da177e4 | 3482 | { |
e879a879 | 3483 | struct r10conf *conf = mddev->private; |
1da177e4 | 3484 | |
409c57f3 N |
3485 | raise_barrier(conf, 0); |
3486 | lower_barrier(conf); | |
3487 | ||
01f96c0a | 3488 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
3489 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
3490 | if (conf->r10bio_pool) | |
3491 | mempool_destroy(conf->r10bio_pool); | |
990a8baf | 3492 | kfree(conf->mirrors); |
1da177e4 LT |
3493 | kfree(conf); |
3494 | mddev->private = NULL; | |
3495 | return 0; | |
3496 | } | |
3497 | ||
fd01b88c | 3498 | static void raid10_quiesce(struct mddev *mddev, int state) |
6cce3b23 | 3499 | { |
e879a879 | 3500 | struct r10conf *conf = mddev->private; |
6cce3b23 N |
3501 | |
3502 | switch(state) { | |
3503 | case 1: | |
3504 | raise_barrier(conf, 0); | |
3505 | break; | |
3506 | case 0: | |
3507 | lower_barrier(conf); | |
3508 | break; | |
3509 | } | |
6cce3b23 | 3510 | } |
1da177e4 | 3511 | |
006a09a0 N |
3512 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3513 | { | |
3514 | /* Resize of 'far' arrays is not supported. | |
3515 | * For 'near' and 'offset' arrays we can set the | |
3516 | * number of sectors used to be an appropriate multiple | |
3517 | * of the chunk size. | |
3518 | * For 'offset', this is far_copies*chunksize. | |
3519 | * For 'near' the multiplier is the LCM of | |
3520 | * near_copies and raid_disks. | |
3521 | * So if far_copies > 1 && !far_offset, fail. | |
3522 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
3523 | * multiply by chunk_size. Then round to this number. | |
3524 | * This is mostly done by raid10_size() | |
3525 | */ | |
3526 | struct r10conf *conf = mddev->private; | |
3527 | sector_t oldsize, size; | |
3528 | ||
f8c9e74f N |
3529 | if (mddev->reshape_position != MaxSector) |
3530 | return -EBUSY; | |
3531 | ||
5cf00fcd | 3532 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
3533 | return -EINVAL; |
3534 | ||
3535 | oldsize = raid10_size(mddev, 0, 0); | |
3536 | size = raid10_size(mddev, sectors, 0); | |
3537 | md_set_array_sectors(mddev, size); | |
3538 | if (mddev->array_sectors > size) | |
3539 | return -EINVAL; | |
3540 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
3541 | revalidate_disk(mddev->gendisk); | |
3542 | if (sectors > mddev->dev_sectors && | |
3543 | mddev->recovery_cp > oldsize) { | |
3544 | mddev->recovery_cp = oldsize; | |
3545 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3546 | } | |
6508fdbf N |
3547 | calc_sectors(conf, sectors); |
3548 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
3549 | mddev->resync_max_sectors = size; |
3550 | return 0; | |
3551 | } | |
3552 | ||
fd01b88c | 3553 | static void *raid10_takeover_raid0(struct mddev *mddev) |
dab8b292 | 3554 | { |
3cb03002 | 3555 | struct md_rdev *rdev; |
e879a879 | 3556 | struct r10conf *conf; |
dab8b292 TM |
3557 | |
3558 | if (mddev->degraded > 0) { | |
128595ed N |
3559 | printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n", |
3560 | mdname(mddev)); | |
dab8b292 TM |
3561 | return ERR_PTR(-EINVAL); |
3562 | } | |
3563 | ||
dab8b292 TM |
3564 | /* Set new parameters */ |
3565 | mddev->new_level = 10; | |
3566 | /* new layout: far_copies = 1, near_copies = 2 */ | |
3567 | mddev->new_layout = (1<<8) + 2; | |
3568 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
3569 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
3570 | mddev->raid_disks *= 2; |
3571 | /* make sure it will be not marked as dirty */ | |
3572 | mddev->recovery_cp = MaxSector; | |
3573 | ||
3574 | conf = setup_conf(mddev); | |
02214dc5 | 3575 | if (!IS_ERR(conf)) { |
dafb20fa | 3576 | rdev_for_each(rdev, mddev) |
e93f68a1 N |
3577 | if (rdev->raid_disk >= 0) |
3578 | rdev->new_raid_disk = rdev->raid_disk * 2; | |
02214dc5 KW |
3579 | conf->barrier = 1; |
3580 | } | |
3581 | ||
dab8b292 TM |
3582 | return conf; |
3583 | } | |
3584 | ||
fd01b88c | 3585 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 3586 | { |
e373ab10 | 3587 | struct r0conf *raid0_conf; |
dab8b292 TM |
3588 | |
3589 | /* raid10 can take over: | |
3590 | * raid0 - providing it has only two drives | |
3591 | */ | |
3592 | if (mddev->level == 0) { | |
3593 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
3594 | raid0_conf = mddev->private; |
3595 | if (raid0_conf->nr_strip_zones > 1) { | |
128595ed N |
3596 | printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0" |
3597 | " with more than one zone.\n", | |
3598 | mdname(mddev)); | |
dab8b292 TM |
3599 | return ERR_PTR(-EINVAL); |
3600 | } | |
3601 | return raid10_takeover_raid0(mddev); | |
3602 | } | |
3603 | return ERR_PTR(-EINVAL); | |
3604 | } | |
3605 | ||
84fc4b56 | 3606 | static struct md_personality raid10_personality = |
1da177e4 LT |
3607 | { |
3608 | .name = "raid10", | |
2604b703 | 3609 | .level = 10, |
1da177e4 LT |
3610 | .owner = THIS_MODULE, |
3611 | .make_request = make_request, | |
3612 | .run = run, | |
3613 | .stop = stop, | |
3614 | .status = status, | |
3615 | .error_handler = error, | |
3616 | .hot_add_disk = raid10_add_disk, | |
3617 | .hot_remove_disk= raid10_remove_disk, | |
3618 | .spare_active = raid10_spare_active, | |
3619 | .sync_request = sync_request, | |
6cce3b23 | 3620 | .quiesce = raid10_quiesce, |
80c3a6ce | 3621 | .size = raid10_size, |
006a09a0 | 3622 | .resize = raid10_resize, |
dab8b292 | 3623 | .takeover = raid10_takeover, |
1da177e4 LT |
3624 | }; |
3625 | ||
3626 | static int __init raid_init(void) | |
3627 | { | |
2604b703 | 3628 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
3629 | } |
3630 | ||
3631 | static void raid_exit(void) | |
3632 | { | |
2604b703 | 3633 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
3634 | } |
3635 | ||
3636 | module_init(raid_init); | |
3637 | module_exit(raid_exit); | |
3638 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3639 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 3640 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 3641 | MODULE_ALIAS("md-raid10"); |
2604b703 | 3642 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
3643 | |
3644 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |