]>
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); | |
0a27ec96 | 70 | |
dd0fc66f | 71 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 72 | { |
e879a879 | 73 | struct r10conf *conf = data; |
9f2c9d12 | 74 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 75 | |
69335ef3 N |
76 | /* allocate a r10bio with room for raid_disks entries in the |
77 | * bios array */ | |
7eaceacc | 78 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
79 | } |
80 | ||
81 | static void r10bio_pool_free(void *r10_bio, void *data) | |
82 | { | |
83 | kfree(r10_bio); | |
84 | } | |
85 | ||
0310fa21 | 86 | /* Maximum size of each resync request */ |
1da177e4 | 87 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 88 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
89 | /* amount of memory to reserve for resync requests */ |
90 | #define RESYNC_WINDOW (1024*1024) | |
91 | /* maximum number of concurrent requests, memory permitting */ | |
92 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
93 | |
94 | /* | |
95 | * When performing a resync, we need to read and compare, so | |
96 | * we need as many pages are there are copies. | |
97 | * When performing a recovery, we need 2 bios, one for read, | |
98 | * one for write (we recover only one drive per r10buf) | |
99 | * | |
100 | */ | |
dd0fc66f | 101 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 102 | { |
e879a879 | 103 | struct r10conf *conf = data; |
1da177e4 | 104 | struct page *page; |
9f2c9d12 | 105 | struct r10bio *r10_bio; |
1da177e4 LT |
106 | struct bio *bio; |
107 | int i, j; | |
108 | int nalloc; | |
109 | ||
110 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 111 | if (!r10_bio) |
1da177e4 | 112 | return NULL; |
1da177e4 LT |
113 | |
114 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) | |
115 | nalloc = conf->copies; /* resync */ | |
116 | else | |
117 | nalloc = 2; /* recovery */ | |
118 | ||
119 | /* | |
120 | * Allocate bios. | |
121 | */ | |
122 | for (j = nalloc ; j-- ; ) { | |
6746557f | 123 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
124 | if (!bio) |
125 | goto out_free_bio; | |
126 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
127 | if (!conf->have_replacement) |
128 | continue; | |
129 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
130 | if (!bio) | |
131 | goto out_free_bio; | |
132 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
133 | } |
134 | /* | |
135 | * Allocate RESYNC_PAGES data pages and attach them | |
136 | * where needed. | |
137 | */ | |
138 | for (j = 0 ; j < nalloc; j++) { | |
69335ef3 | 139 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
1da177e4 LT |
140 | bio = r10_bio->devs[j].bio; |
141 | for (i = 0; i < RESYNC_PAGES; i++) { | |
c65060ad NK |
142 | if (j == 1 && !test_bit(MD_RECOVERY_SYNC, |
143 | &conf->mddev->recovery)) { | |
144 | /* we can share bv_page's during recovery */ | |
145 | struct bio *rbio = r10_bio->devs[0].bio; | |
146 | page = rbio->bi_io_vec[i].bv_page; | |
147 | get_page(page); | |
148 | } else | |
149 | page = alloc_page(gfp_flags); | |
1da177e4 LT |
150 | if (unlikely(!page)) |
151 | goto out_free_pages; | |
152 | ||
153 | bio->bi_io_vec[i].bv_page = page; | |
69335ef3 N |
154 | if (rbio) |
155 | rbio->bi_io_vec[i].bv_page = page; | |
1da177e4 LT |
156 | } |
157 | } | |
158 | ||
159 | return r10_bio; | |
160 | ||
161 | out_free_pages: | |
162 | for ( ; i > 0 ; i--) | |
1345b1d8 | 163 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
164 | while (j--) |
165 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 166 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
167 | j = -1; |
168 | out_free_bio: | |
69335ef3 | 169 | while (++j < nalloc) { |
1da177e4 | 170 | bio_put(r10_bio->devs[j].bio); |
69335ef3 N |
171 | if (r10_bio->devs[j].repl_bio) |
172 | bio_put(r10_bio->devs[j].repl_bio); | |
173 | } | |
1da177e4 LT |
174 | r10bio_pool_free(r10_bio, conf); |
175 | return NULL; | |
176 | } | |
177 | ||
178 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
179 | { | |
180 | int i; | |
e879a879 | 181 | struct r10conf *conf = data; |
9f2c9d12 | 182 | struct r10bio *r10bio = __r10_bio; |
1da177e4 LT |
183 | int j; |
184 | ||
185 | for (j=0; j < conf->copies; j++) { | |
186 | struct bio *bio = r10bio->devs[j].bio; | |
187 | if (bio) { | |
188 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 189 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
190 | bio->bi_io_vec[i].bv_page = NULL; |
191 | } | |
192 | bio_put(bio); | |
193 | } | |
69335ef3 N |
194 | bio = r10bio->devs[j].repl_bio; |
195 | if (bio) | |
196 | bio_put(bio); | |
1da177e4 LT |
197 | } |
198 | r10bio_pool_free(r10bio, conf); | |
199 | } | |
200 | ||
e879a879 | 201 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
202 | { |
203 | int i; | |
204 | ||
205 | for (i = 0; i < conf->copies; i++) { | |
206 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 207 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
208 | bio_put(*bio); |
209 | *bio = NULL; | |
69335ef3 N |
210 | bio = &r10_bio->devs[i].repl_bio; |
211 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
212 | bio_put(*bio); | |
213 | *bio = NULL; | |
1da177e4 LT |
214 | } |
215 | } | |
216 | ||
9f2c9d12 | 217 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 218 | { |
e879a879 | 219 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 220 | |
1da177e4 LT |
221 | put_all_bios(conf, r10_bio); |
222 | mempool_free(r10_bio, conf->r10bio_pool); | |
223 | } | |
224 | ||
9f2c9d12 | 225 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 226 | { |
e879a879 | 227 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
228 | |
229 | mempool_free(r10_bio, conf->r10buf_pool); | |
230 | ||
0a27ec96 | 231 | lower_barrier(conf); |
1da177e4 LT |
232 | } |
233 | ||
9f2c9d12 | 234 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
235 | { |
236 | unsigned long flags; | |
fd01b88c | 237 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 238 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
239 | |
240 | spin_lock_irqsave(&conf->device_lock, flags); | |
241 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 242 | conf->nr_queued ++; |
1da177e4 LT |
243 | spin_unlock_irqrestore(&conf->device_lock, flags); |
244 | ||
388667be AJ |
245 | /* wake up frozen array... */ |
246 | wake_up(&conf->wait_barrier); | |
247 | ||
1da177e4 LT |
248 | md_wakeup_thread(mddev->thread); |
249 | } | |
250 | ||
251 | /* | |
252 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
253 | * operation and are ready to return a success/failure code to the buffer | |
254 | * cache layer. | |
255 | */ | |
9f2c9d12 | 256 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
257 | { |
258 | struct bio *bio = r10_bio->master_bio; | |
856e08e2 | 259 | int done; |
e879a879 | 260 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 261 | |
856e08e2 N |
262 | if (bio->bi_phys_segments) { |
263 | unsigned long flags; | |
264 | spin_lock_irqsave(&conf->device_lock, flags); | |
265 | bio->bi_phys_segments--; | |
266 | done = (bio->bi_phys_segments == 0); | |
267 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
268 | } else | |
269 | done = 1; | |
270 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
271 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
272 | if (done) { | |
273 | bio_endio(bio, 0); | |
274 | /* | |
275 | * Wake up any possible resync thread that waits for the device | |
276 | * to go idle. | |
277 | */ | |
278 | allow_barrier(conf); | |
279 | } | |
1da177e4 LT |
280 | free_r10bio(r10_bio); |
281 | } | |
282 | ||
283 | /* | |
284 | * Update disk head position estimator based on IRQ completion info. | |
285 | */ | |
9f2c9d12 | 286 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 287 | { |
e879a879 | 288 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
289 | |
290 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
291 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
292 | } | |
293 | ||
778ca018 NK |
294 | /* |
295 | * Find the disk number which triggered given bio | |
296 | */ | |
e879a879 | 297 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 298 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
299 | { |
300 | int slot; | |
69335ef3 | 301 | int repl = 0; |
778ca018 | 302 | |
69335ef3 | 303 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
304 | if (r10_bio->devs[slot].bio == bio) |
305 | break; | |
69335ef3 N |
306 | if (r10_bio->devs[slot].repl_bio == bio) { |
307 | repl = 1; | |
308 | break; | |
309 | } | |
310 | } | |
778ca018 NK |
311 | |
312 | BUG_ON(slot == conf->copies); | |
313 | update_head_pos(slot, r10_bio); | |
314 | ||
749c55e9 N |
315 | if (slotp) |
316 | *slotp = slot; | |
69335ef3 N |
317 | if (replp) |
318 | *replp = repl; | |
778ca018 NK |
319 | return r10_bio->devs[slot].devnum; |
320 | } | |
321 | ||
6712ecf8 | 322 | static void raid10_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
323 | { |
324 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 325 | struct r10bio *r10_bio = bio->bi_private; |
1da177e4 | 326 | int slot, dev; |
e879a879 | 327 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 328 | |
1da177e4 LT |
329 | |
330 | slot = r10_bio->read_slot; | |
331 | dev = r10_bio->devs[slot].devnum; | |
332 | /* | |
333 | * this branch is our 'one mirror IO has finished' event handler: | |
334 | */ | |
4443ae10 N |
335 | update_head_pos(slot, r10_bio); |
336 | ||
337 | if (uptodate) { | |
1da177e4 LT |
338 | /* |
339 | * Set R10BIO_Uptodate in our master bio, so that | |
340 | * we will return a good error code to the higher | |
341 | * levels even if IO on some other mirrored buffer fails. | |
342 | * | |
343 | * The 'master' represents the composite IO operation to | |
344 | * user-side. So if something waits for IO, then it will | |
345 | * wait for the 'master' bio. | |
346 | */ | |
347 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 348 | raid_end_bio_io(r10_bio); |
7c4e06ff | 349 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); |
4443ae10 | 350 | } else { |
1da177e4 | 351 | /* |
7c4e06ff | 352 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
353 | */ |
354 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
355 | printk_ratelimited(KERN_ERR |
356 | "md/raid10:%s: %s: rescheduling sector %llu\n", | |
357 | mdname(conf->mddev), | |
358 | bdevname(conf->mirrors[dev].rdev->bdev, b), | |
359 | (unsigned long long)r10_bio->sector); | |
856e08e2 | 360 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
361 | reschedule_retry(r10_bio); |
362 | } | |
1da177e4 LT |
363 | } |
364 | ||
9f2c9d12 | 365 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
366 | { |
367 | /* clear the bitmap if all writes complete successfully */ | |
368 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
369 | r10_bio->sectors, | |
370 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
371 | 0); | |
372 | md_write_end(r10_bio->mddev); | |
373 | } | |
374 | ||
9f2c9d12 | 375 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
376 | { |
377 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
378 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
379 | reschedule_retry(r10_bio); | |
380 | else { | |
381 | close_write(r10_bio); | |
382 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
383 | reschedule_retry(r10_bio); | |
384 | else | |
385 | raid_end_bio_io(r10_bio); | |
386 | } | |
387 | } | |
388 | } | |
389 | ||
6712ecf8 | 390 | static void raid10_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
391 | { |
392 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 393 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 394 | int dev; |
749c55e9 | 395 | int dec_rdev = 1; |
e879a879 | 396 | struct r10conf *conf = r10_bio->mddev->private; |
749c55e9 | 397 | int slot; |
1da177e4 | 398 | |
69335ef3 | 399 | dev = find_bio_disk(conf, r10_bio, bio, &slot, NULL); |
1da177e4 LT |
400 | |
401 | /* | |
402 | * this branch is our 'one mirror IO has finished' event handler: | |
403 | */ | |
6cce3b23 | 404 | if (!uptodate) { |
bd870a16 N |
405 | set_bit(WriteErrorSeen, &conf->mirrors[dev].rdev->flags); |
406 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
407 | dec_rdev = 0; | |
749c55e9 | 408 | } else { |
1da177e4 LT |
409 | /* |
410 | * Set R10BIO_Uptodate in our master bio, so that | |
411 | * we will return a good error code for to the higher | |
412 | * levels even if IO on some other mirrored buffer fails. | |
413 | * | |
414 | * The 'master' represents the composite IO operation to | |
415 | * user-side. So if something waits for IO, then it will | |
416 | * wait for the 'master' bio. | |
417 | */ | |
749c55e9 N |
418 | sector_t first_bad; |
419 | int bad_sectors; | |
420 | ||
1da177e4 LT |
421 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
422 | ||
749c55e9 N |
423 | /* Maybe we can clear some bad blocks. */ |
424 | if (is_badblock(conf->mirrors[dev].rdev, | |
425 | r10_bio->devs[slot].addr, | |
426 | r10_bio->sectors, | |
427 | &first_bad, &bad_sectors)) { | |
428 | bio_put(bio); | |
429 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
430 | dec_rdev = 0; | |
431 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
432 | } | |
433 | } | |
434 | ||
1da177e4 LT |
435 | /* |
436 | * | |
437 | * Let's see if all mirrored write operations have finished | |
438 | * already. | |
439 | */ | |
19d5f834 | 440 | one_write_done(r10_bio); |
749c55e9 N |
441 | if (dec_rdev) |
442 | rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); | |
1da177e4 LT |
443 | } |
444 | ||
445 | ||
446 | /* | |
447 | * RAID10 layout manager | |
25985edc | 448 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
449 | * parameters: near_copies and far_copies. |
450 | * near_copies * far_copies must be <= raid_disks. | |
451 | * Normally one of these will be 1. | |
452 | * If both are 1, we get raid0. | |
453 | * If near_copies == raid_disks, we get raid1. | |
454 | * | |
25985edc | 455 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
456 | * first chunk, followed by near_copies copies of the next chunk and |
457 | * so on. | |
458 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
459 | * as described above, we start again with a device offset of near_copies. | |
460 | * So we effectively have another copy of the whole array further down all | |
461 | * the drives, but with blocks on different drives. | |
462 | * With this layout, and block is never stored twice on the one device. | |
463 | * | |
464 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 465 | * on each device that it is on. |
1da177e4 LT |
466 | * |
467 | * raid10_find_virt does the reverse mapping, from a device and a | |
468 | * sector offset to a virtual address | |
469 | */ | |
470 | ||
e879a879 | 471 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) |
1da177e4 LT |
472 | { |
473 | int n,f; | |
474 | sector_t sector; | |
475 | sector_t chunk; | |
476 | sector_t stripe; | |
477 | int dev; | |
478 | ||
479 | int slot = 0; | |
480 | ||
481 | /* now calculate first sector/dev */ | |
482 | chunk = r10bio->sector >> conf->chunk_shift; | |
483 | sector = r10bio->sector & conf->chunk_mask; | |
484 | ||
485 | chunk *= conf->near_copies; | |
486 | stripe = chunk; | |
487 | dev = sector_div(stripe, conf->raid_disks); | |
c93983bf N |
488 | if (conf->far_offset) |
489 | stripe *= conf->far_copies; | |
1da177e4 LT |
490 | |
491 | sector += stripe << conf->chunk_shift; | |
492 | ||
493 | /* and calculate all the others */ | |
494 | for (n=0; n < conf->near_copies; n++) { | |
495 | int d = dev; | |
496 | sector_t s = sector; | |
497 | r10bio->devs[slot].addr = sector; | |
498 | r10bio->devs[slot].devnum = d; | |
499 | slot++; | |
500 | ||
501 | for (f = 1; f < conf->far_copies; f++) { | |
502 | d += conf->near_copies; | |
503 | if (d >= conf->raid_disks) | |
504 | d -= conf->raid_disks; | |
505 | s += conf->stride; | |
506 | r10bio->devs[slot].devnum = d; | |
507 | r10bio->devs[slot].addr = s; | |
508 | slot++; | |
509 | } | |
510 | dev++; | |
511 | if (dev >= conf->raid_disks) { | |
512 | dev = 0; | |
513 | sector += (conf->chunk_mask + 1); | |
514 | } | |
515 | } | |
516 | BUG_ON(slot != conf->copies); | |
517 | } | |
518 | ||
e879a879 | 519 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
520 | { |
521 | sector_t offset, chunk, vchunk; | |
522 | ||
1da177e4 | 523 | offset = sector & conf->chunk_mask; |
c93983bf N |
524 | if (conf->far_offset) { |
525 | int fc; | |
526 | chunk = sector >> conf->chunk_shift; | |
527 | fc = sector_div(chunk, conf->far_copies); | |
528 | dev -= fc * conf->near_copies; | |
529 | if (dev < 0) | |
530 | dev += conf->raid_disks; | |
531 | } else { | |
64a742bc | 532 | while (sector >= conf->stride) { |
c93983bf N |
533 | sector -= conf->stride; |
534 | if (dev < conf->near_copies) | |
535 | dev += conf->raid_disks - conf->near_copies; | |
536 | else | |
537 | dev -= conf->near_copies; | |
538 | } | |
539 | chunk = sector >> conf->chunk_shift; | |
540 | } | |
1da177e4 LT |
541 | vchunk = chunk * conf->raid_disks + dev; |
542 | sector_div(vchunk, conf->near_copies); | |
543 | return (vchunk << conf->chunk_shift) + offset; | |
544 | } | |
545 | ||
546 | /** | |
547 | * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged | |
548 | * @q: request queue | |
cc371e66 | 549 | * @bvm: properties of new bio |
1da177e4 LT |
550 | * @biovec: the request that could be merged to it. |
551 | * | |
552 | * Return amount of bytes we can accept at this offset | |
553 | * If near_copies == raid_disk, there are no striping issues, | |
554 | * but in that case, the function isn't called at all. | |
555 | */ | |
cc371e66 AK |
556 | static int raid10_mergeable_bvec(struct request_queue *q, |
557 | struct bvec_merge_data *bvm, | |
558 | struct bio_vec *biovec) | |
1da177e4 | 559 | { |
fd01b88c | 560 | struct mddev *mddev = q->queuedata; |
cc371e66 | 561 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
1da177e4 | 562 | int max; |
9d8f0363 | 563 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 564 | unsigned int bio_sectors = bvm->bi_size >> 9; |
1da177e4 LT |
565 | |
566 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
567 | if (max < 0) max = 0; /* bio_add cannot handle a negative return */ | |
cc371e66 AK |
568 | if (max <= biovec->bv_len && bio_sectors == 0) |
569 | return biovec->bv_len; | |
1da177e4 LT |
570 | else |
571 | return max; | |
572 | } | |
573 | ||
574 | /* | |
575 | * This routine returns the disk from which the requested read should | |
576 | * be done. There is a per-array 'next expected sequential IO' sector | |
577 | * number - if this matches on the next IO then we use the last disk. | |
578 | * There is also a per-disk 'last know head position' sector that is | |
579 | * maintained from IRQ contexts, both the normal and the resync IO | |
580 | * completion handlers update this position correctly. If there is no | |
581 | * perfect sequential match then we pick the disk whose head is closest. | |
582 | * | |
583 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
584 | * because position is mirror, not device based. | |
585 | * | |
586 | * The rdev for the device selected will have nr_pending incremented. | |
587 | */ | |
588 | ||
589 | /* | |
590 | * FIXME: possibly should rethink readbalancing and do it differently | |
591 | * depending on near_copies / far_copies geometry. | |
592 | */ | |
96c3fd1f N |
593 | static struct md_rdev *read_balance(struct r10conf *conf, |
594 | struct r10bio *r10_bio, | |
595 | int *max_sectors) | |
1da177e4 | 596 | { |
af3a2cd6 | 597 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 598 | int disk, slot; |
856e08e2 N |
599 | int sectors = r10_bio->sectors; |
600 | int best_good_sectors; | |
56d99121 | 601 | sector_t new_distance, best_dist; |
3cb03002 | 602 | struct md_rdev *rdev; |
56d99121 N |
603 | int do_balance; |
604 | int best_slot; | |
1da177e4 LT |
605 | |
606 | raid10_find_phys(conf, r10_bio); | |
607 | rcu_read_lock(); | |
56d99121 | 608 | retry: |
856e08e2 | 609 | sectors = r10_bio->sectors; |
56d99121 N |
610 | best_slot = -1; |
611 | best_dist = MaxSector; | |
856e08e2 | 612 | best_good_sectors = 0; |
56d99121 | 613 | do_balance = 1; |
1da177e4 LT |
614 | /* |
615 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
616 | * device if no resync is going on (recovery is ok), or below |
617 | * the resync window. We take the first readable disk when | |
618 | * above the resync window. | |
1da177e4 LT |
619 | */ |
620 | if (conf->mddev->recovery_cp < MaxSector | |
56d99121 N |
621 | && (this_sector + sectors >= conf->next_resync)) |
622 | do_balance = 0; | |
1da177e4 | 623 | |
56d99121 | 624 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
625 | sector_t first_bad; |
626 | int bad_sectors; | |
627 | sector_t dev_sector; | |
628 | ||
56d99121 N |
629 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
630 | continue; | |
1da177e4 | 631 | disk = r10_bio->devs[slot].devnum; |
56d99121 N |
632 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
633 | if (rdev == NULL) | |
1da177e4 | 634 | continue; |
56d99121 N |
635 | if (!test_bit(In_sync, &rdev->flags)) |
636 | continue; | |
637 | ||
856e08e2 N |
638 | dev_sector = r10_bio->devs[slot].addr; |
639 | if (is_badblock(rdev, dev_sector, sectors, | |
640 | &first_bad, &bad_sectors)) { | |
641 | if (best_dist < MaxSector) | |
642 | /* Already have a better slot */ | |
643 | continue; | |
644 | if (first_bad <= dev_sector) { | |
645 | /* Cannot read here. If this is the | |
646 | * 'primary' device, then we must not read | |
647 | * beyond 'bad_sectors' from another device. | |
648 | */ | |
649 | bad_sectors -= (dev_sector - first_bad); | |
650 | if (!do_balance && sectors > bad_sectors) | |
651 | sectors = bad_sectors; | |
652 | if (best_good_sectors > sectors) | |
653 | best_good_sectors = sectors; | |
654 | } else { | |
655 | sector_t good_sectors = | |
656 | first_bad - dev_sector; | |
657 | if (good_sectors > best_good_sectors) { | |
658 | best_good_sectors = good_sectors; | |
659 | best_slot = slot; | |
660 | } | |
661 | if (!do_balance) | |
662 | /* Must read from here */ | |
663 | break; | |
664 | } | |
665 | continue; | |
666 | } else | |
667 | best_good_sectors = sectors; | |
668 | ||
56d99121 N |
669 | if (!do_balance) |
670 | break; | |
1da177e4 | 671 | |
22dfdf52 N |
672 | /* This optimisation is debatable, and completely destroys |
673 | * sequential read speed for 'far copies' arrays. So only | |
674 | * keep it for 'near' arrays, and review those later. | |
675 | */ | |
56d99121 | 676 | if (conf->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
1da177e4 | 677 | break; |
8ed3a195 KS |
678 | |
679 | /* for far > 1 always use the lowest address */ | |
680 | if (conf->far_copies > 1) | |
56d99121 | 681 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 682 | else |
56d99121 N |
683 | new_distance = abs(r10_bio->devs[slot].addr - |
684 | conf->mirrors[disk].head_position); | |
685 | if (new_distance < best_dist) { | |
686 | best_dist = new_distance; | |
687 | best_slot = slot; | |
1da177e4 LT |
688 | } |
689 | } | |
56d99121 N |
690 | if (slot == conf->copies) |
691 | slot = best_slot; | |
1da177e4 | 692 | |
56d99121 N |
693 | if (slot >= 0) { |
694 | disk = r10_bio->devs[slot].devnum; | |
695 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
696 | if (!rdev) | |
697 | goto retry; | |
698 | atomic_inc(&rdev->nr_pending); | |
699 | if (test_bit(Faulty, &rdev->flags)) { | |
700 | /* Cannot risk returning a device that failed | |
701 | * before we inc'ed nr_pending | |
702 | */ | |
703 | rdev_dec_pending(rdev, conf->mddev); | |
704 | goto retry; | |
705 | } | |
706 | r10_bio->read_slot = slot; | |
707 | } else | |
96c3fd1f | 708 | rdev = NULL; |
1da177e4 | 709 | rcu_read_unlock(); |
856e08e2 | 710 | *max_sectors = best_good_sectors; |
1da177e4 | 711 | |
96c3fd1f | 712 | return rdev; |
1da177e4 LT |
713 | } |
714 | ||
0d129228 N |
715 | static int raid10_congested(void *data, int bits) |
716 | { | |
fd01b88c | 717 | struct mddev *mddev = data; |
e879a879 | 718 | struct r10conf *conf = mddev->private; |
0d129228 N |
719 | int i, ret = 0; |
720 | ||
34db0cd6 N |
721 | if ((bits & (1 << BDI_async_congested)) && |
722 | conf->pending_count >= max_queued_requests) | |
723 | return 1; | |
724 | ||
3fa841d7 N |
725 | if (mddev_congested(mddev, bits)) |
726 | return 1; | |
0d129228 | 727 | rcu_read_lock(); |
84707f38 | 728 | for (i = 0; i < conf->raid_disks && ret == 0; i++) { |
3cb03002 | 729 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 730 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 731 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 N |
732 | |
733 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
734 | } | |
735 | } | |
736 | rcu_read_unlock(); | |
737 | return ret; | |
738 | } | |
739 | ||
e879a879 | 740 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
741 | { |
742 | /* Any writes that have been queued but are awaiting | |
743 | * bitmap updates get flushed here. | |
a35e63ef | 744 | */ |
a35e63ef N |
745 | spin_lock_irq(&conf->device_lock); |
746 | ||
747 | if (conf->pending_bio_list.head) { | |
748 | struct bio *bio; | |
749 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 750 | conf->pending_count = 0; |
a35e63ef N |
751 | spin_unlock_irq(&conf->device_lock); |
752 | /* flush any pending bitmap writes to disk | |
753 | * before proceeding w/ I/O */ | |
754 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 755 | wake_up(&conf->wait_barrier); |
a35e63ef N |
756 | |
757 | while (bio) { /* submit pending writes */ | |
758 | struct bio *next = bio->bi_next; | |
759 | bio->bi_next = NULL; | |
760 | generic_make_request(bio); | |
761 | bio = next; | |
762 | } | |
a35e63ef N |
763 | } else |
764 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 765 | } |
7eaceacc | 766 | |
0a27ec96 N |
767 | /* Barriers.... |
768 | * Sometimes we need to suspend IO while we do something else, | |
769 | * either some resync/recovery, or reconfigure the array. | |
770 | * To do this we raise a 'barrier'. | |
771 | * The 'barrier' is a counter that can be raised multiple times | |
772 | * to count how many activities are happening which preclude | |
773 | * normal IO. | |
774 | * We can only raise the barrier if there is no pending IO. | |
775 | * i.e. if nr_pending == 0. | |
776 | * We choose only to raise the barrier if no-one is waiting for the | |
777 | * barrier to go down. This means that as soon as an IO request | |
778 | * is ready, no other operations which require a barrier will start | |
779 | * until the IO request has had a chance. | |
780 | * | |
781 | * So: regular IO calls 'wait_barrier'. When that returns there | |
782 | * is no backgroup IO happening, It must arrange to call | |
783 | * allow_barrier when it has finished its IO. | |
784 | * backgroup IO calls must call raise_barrier. Once that returns | |
785 | * there is no normal IO happeing. It must arrange to call | |
786 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 787 | */ |
1da177e4 | 788 | |
e879a879 | 789 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 790 | { |
6cce3b23 | 791 | BUG_ON(force && !conf->barrier); |
1da177e4 | 792 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 793 | |
6cce3b23 N |
794 | /* Wait until no block IO is waiting (unless 'force') */ |
795 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
c3b328ac | 796 | conf->resync_lock, ); |
0a27ec96 N |
797 | |
798 | /* block any new IO from starting */ | |
799 | conf->barrier++; | |
800 | ||
c3b328ac | 801 | /* Now wait for all pending IO to complete */ |
0a27ec96 N |
802 | wait_event_lock_irq(conf->wait_barrier, |
803 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
c3b328ac | 804 | conf->resync_lock, ); |
0a27ec96 N |
805 | |
806 | spin_unlock_irq(&conf->resync_lock); | |
807 | } | |
808 | ||
e879a879 | 809 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
810 | { |
811 | unsigned long flags; | |
812 | spin_lock_irqsave(&conf->resync_lock, flags); | |
813 | conf->barrier--; | |
814 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
815 | wake_up(&conf->wait_barrier); | |
816 | } | |
817 | ||
e879a879 | 818 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
819 | { |
820 | spin_lock_irq(&conf->resync_lock); | |
821 | if (conf->barrier) { | |
822 | conf->nr_waiting++; | |
823 | wait_event_lock_irq(conf->wait_barrier, !conf->barrier, | |
824 | conf->resync_lock, | |
c3b328ac | 825 | ); |
0a27ec96 | 826 | conf->nr_waiting--; |
1da177e4 | 827 | } |
0a27ec96 | 828 | conf->nr_pending++; |
1da177e4 LT |
829 | spin_unlock_irq(&conf->resync_lock); |
830 | } | |
831 | ||
e879a879 | 832 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 N |
833 | { |
834 | unsigned long flags; | |
835 | spin_lock_irqsave(&conf->resync_lock, flags); | |
836 | conf->nr_pending--; | |
837 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
838 | wake_up(&conf->wait_barrier); | |
839 | } | |
840 | ||
e879a879 | 841 | static void freeze_array(struct r10conf *conf) |
4443ae10 N |
842 | { |
843 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 844 | * go quiet. |
4443ae10 | 845 | * We increment barrier and nr_waiting, and then |
1c830532 N |
846 | * wait until nr_pending match nr_queued+1 |
847 | * This is called in the context of one normal IO request | |
848 | * that has failed. Thus any sync request that might be pending | |
849 | * will be blocked by nr_pending, and we need to wait for | |
850 | * pending IO requests to complete or be queued for re-try. | |
851 | * Thus the number queued (nr_queued) plus this request (1) | |
852 | * must match the number of pending IOs (nr_pending) before | |
853 | * we continue. | |
4443ae10 N |
854 | */ |
855 | spin_lock_irq(&conf->resync_lock); | |
856 | conf->barrier++; | |
857 | conf->nr_waiting++; | |
858 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 859 | conf->nr_pending == conf->nr_queued+1, |
4443ae10 | 860 | conf->resync_lock, |
c3b328ac N |
861 | flush_pending_writes(conf)); |
862 | ||
4443ae10 N |
863 | spin_unlock_irq(&conf->resync_lock); |
864 | } | |
865 | ||
e879a879 | 866 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
867 | { |
868 | /* reverse the effect of the freeze */ | |
869 | spin_lock_irq(&conf->resync_lock); | |
870 | conf->barrier--; | |
871 | conf->nr_waiting--; | |
872 | wake_up(&conf->wait_barrier); | |
873 | spin_unlock_irq(&conf->resync_lock); | |
874 | } | |
875 | ||
b4fdcb02 | 876 | static void make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 877 | { |
e879a879 | 878 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 879 | struct r10bio *r10_bio; |
1da177e4 LT |
880 | struct bio *read_bio; |
881 | int i; | |
882 | int chunk_sects = conf->chunk_mask + 1; | |
a362357b | 883 | const int rw = bio_data_dir(bio); |
2c7d46ec | 884 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 885 | const unsigned long do_fua = (bio->bi_rw & REQ_FUA); |
6cce3b23 | 886 | unsigned long flags; |
3cb03002 | 887 | struct md_rdev *blocked_rdev; |
c3b328ac | 888 | int plugged; |
d4432c23 N |
889 | int sectors_handled; |
890 | int max_sectors; | |
1da177e4 | 891 | |
e9c7469b TH |
892 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
893 | md_flush_request(mddev, bio); | |
5a7bbad2 | 894 | return; |
e5dcdd80 N |
895 | } |
896 | ||
1da177e4 LT |
897 | /* If this request crosses a chunk boundary, we need to |
898 | * split it. This will only happen for 1 PAGE (or less) requests. | |
899 | */ | |
900 | if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9) | |
901 | > chunk_sects && | |
902 | conf->near_copies < conf->raid_disks)) { | |
903 | struct bio_pair *bp; | |
904 | /* Sanity check -- queue functions should prevent this happening */ | |
905 | if (bio->bi_vcnt != 1 || | |
906 | bio->bi_idx != 0) | |
907 | goto bad_map; | |
908 | /* This is a one page bio that upper layers | |
909 | * refuse to split for us, so we need to split it. | |
910 | */ | |
6feef531 | 911 | bp = bio_split(bio, |
1da177e4 | 912 | chunk_sects - (bio->bi_sector & (chunk_sects - 1)) ); |
51e9ac77 N |
913 | |
914 | /* Each of these 'make_request' calls will call 'wait_barrier'. | |
915 | * If the first succeeds but the second blocks due to the resync | |
916 | * thread raising the barrier, we will deadlock because the | |
917 | * IO to the underlying device will be queued in generic_make_request | |
918 | * and will never complete, so will never reduce nr_pending. | |
919 | * So increment nr_waiting here so no new raise_barriers will | |
920 | * succeed, and so the second wait_barrier cannot block. | |
921 | */ | |
922 | spin_lock_irq(&conf->resync_lock); | |
923 | conf->nr_waiting++; | |
924 | spin_unlock_irq(&conf->resync_lock); | |
925 | ||
5a7bbad2 CH |
926 | make_request(mddev, &bp->bio1); |
927 | make_request(mddev, &bp->bio2); | |
1da177e4 | 928 | |
51e9ac77 N |
929 | spin_lock_irq(&conf->resync_lock); |
930 | conf->nr_waiting--; | |
931 | wake_up(&conf->wait_barrier); | |
932 | spin_unlock_irq(&conf->resync_lock); | |
933 | ||
1da177e4 | 934 | bio_pair_release(bp); |
5a7bbad2 | 935 | return; |
1da177e4 | 936 | bad_map: |
128595ed N |
937 | printk("md/raid10:%s: make_request bug: can't convert block across chunks" |
938 | " or bigger than %dk %llu %d\n", mdname(mddev), chunk_sects/2, | |
1da177e4 LT |
939 | (unsigned long long)bio->bi_sector, bio->bi_size >> 10); |
940 | ||
6712ecf8 | 941 | bio_io_error(bio); |
5a7bbad2 | 942 | return; |
1da177e4 LT |
943 | } |
944 | ||
3d310eb7 | 945 | md_write_start(mddev, bio); |
06d91a5f | 946 | |
1da177e4 LT |
947 | /* |
948 | * Register the new request and wait if the reconstruction | |
949 | * thread has put up a bar for new requests. | |
950 | * Continue immediately if no resync is active currently. | |
951 | */ | |
0a27ec96 | 952 | wait_barrier(conf); |
1da177e4 | 953 | |
1da177e4 LT |
954 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); |
955 | ||
956 | r10_bio->master_bio = bio; | |
957 | r10_bio->sectors = bio->bi_size >> 9; | |
958 | ||
959 | r10_bio->mddev = mddev; | |
960 | r10_bio->sector = bio->bi_sector; | |
6cce3b23 | 961 | r10_bio->state = 0; |
1da177e4 | 962 | |
856e08e2 N |
963 | /* We might need to issue multiple reads to different |
964 | * devices if there are bad blocks around, so we keep | |
965 | * track of the number of reads in bio->bi_phys_segments. | |
966 | * If this is 0, there is only one r10_bio and no locking | |
967 | * will be needed when the request completes. If it is | |
968 | * non-zero, then it is the number of not-completed requests. | |
969 | */ | |
970 | bio->bi_phys_segments = 0; | |
971 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
972 | ||
a362357b | 973 | if (rw == READ) { |
1da177e4 LT |
974 | /* |
975 | * read balancing logic: | |
976 | */ | |
96c3fd1f | 977 | struct md_rdev *rdev; |
856e08e2 N |
978 | int slot; |
979 | ||
980 | read_again: | |
96c3fd1f N |
981 | rdev = read_balance(conf, r10_bio, &max_sectors); |
982 | if (!rdev) { | |
1da177e4 | 983 | raid_end_bio_io(r10_bio); |
5a7bbad2 | 984 | return; |
1da177e4 | 985 | } |
96c3fd1f | 986 | slot = r10_bio->read_slot; |
1da177e4 | 987 | |
a167f663 | 988 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
856e08e2 N |
989 | md_trim_bio(read_bio, r10_bio->sector - bio->bi_sector, |
990 | max_sectors); | |
1da177e4 LT |
991 | |
992 | r10_bio->devs[slot].bio = read_bio; | |
993 | ||
994 | read_bio->bi_sector = r10_bio->devs[slot].addr + | |
96c3fd1f N |
995 | rdev->data_offset; |
996 | read_bio->bi_bdev = rdev->bdev; | |
1da177e4 | 997 | read_bio->bi_end_io = raid10_end_read_request; |
7b6d91da | 998 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
999 | read_bio->bi_private = r10_bio; |
1000 | ||
856e08e2 N |
1001 | if (max_sectors < r10_bio->sectors) { |
1002 | /* Could not read all from this device, so we will | |
1003 | * need another r10_bio. | |
1004 | */ | |
856e08e2 N |
1005 | sectors_handled = (r10_bio->sectors + max_sectors |
1006 | - bio->bi_sector); | |
1007 | r10_bio->sectors = max_sectors; | |
1008 | spin_lock_irq(&conf->device_lock); | |
1009 | if (bio->bi_phys_segments == 0) | |
1010 | bio->bi_phys_segments = 2; | |
1011 | else | |
1012 | bio->bi_phys_segments++; | |
1013 | spin_unlock(&conf->device_lock); | |
1014 | /* Cannot call generic_make_request directly | |
1015 | * as that will be queued in __generic_make_request | |
1016 | * and subsequent mempool_alloc might block | |
1017 | * waiting for it. so hand bio over to raid10d. | |
1018 | */ | |
1019 | reschedule_retry(r10_bio); | |
1020 | ||
1021 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1022 | ||
1023 | r10_bio->master_bio = bio; | |
1024 | r10_bio->sectors = ((bio->bi_size >> 9) | |
1025 | - sectors_handled); | |
1026 | r10_bio->state = 0; | |
1027 | r10_bio->mddev = mddev; | |
1028 | r10_bio->sector = bio->bi_sector + sectors_handled; | |
1029 | goto read_again; | |
1030 | } else | |
1031 | generic_make_request(read_bio); | |
5a7bbad2 | 1032 | return; |
1da177e4 LT |
1033 | } |
1034 | ||
1035 | /* | |
1036 | * WRITE: | |
1037 | */ | |
34db0cd6 N |
1038 | if (conf->pending_count >= max_queued_requests) { |
1039 | md_wakeup_thread(mddev->thread); | |
1040 | wait_event(conf->wait_barrier, | |
1041 | conf->pending_count < max_queued_requests); | |
1042 | } | |
6bfe0b49 | 1043 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1044 | * inc refcount on their rdev. Record them by setting |
1045 | * bios[x] to bio | |
d4432c23 N |
1046 | * If there are known/acknowledged bad blocks on any device |
1047 | * on which we have seen a write error, we want to avoid | |
1048 | * writing to those blocks. This potentially requires several | |
1049 | * writes to write around the bad blocks. Each set of writes | |
1050 | * gets its own r10_bio with a set of bios attached. The number | |
1051 | * of r10_bios is recored in bio->bi_phys_segments just as with | |
1052 | * the read case. | |
1da177e4 | 1053 | */ |
c3b328ac N |
1054 | plugged = mddev_check_plugged(mddev); |
1055 | ||
69335ef3 | 1056 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1057 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1058 | retry_write: |
cb6969e8 | 1059 | blocked_rdev = NULL; |
1da177e4 | 1060 | rcu_read_lock(); |
d4432c23 N |
1061 | max_sectors = r10_bio->sectors; |
1062 | ||
1da177e4 LT |
1063 | for (i = 0; i < conf->copies; i++) { |
1064 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1065 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
6bfe0b49 DW |
1066 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1067 | atomic_inc(&rdev->nr_pending); | |
1068 | blocked_rdev = rdev; | |
1069 | break; | |
1070 | } | |
d4432c23 N |
1071 | r10_bio->devs[i].bio = NULL; |
1072 | if (!rdev || test_bit(Faulty, &rdev->flags)) { | |
6cce3b23 | 1073 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1074 | continue; |
1075 | } | |
1076 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1077 | sector_t first_bad; | |
1078 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1079 | int bad_sectors; | |
1080 | int is_bad; | |
1081 | ||
1082 | is_bad = is_badblock(rdev, dev_sector, | |
1083 | max_sectors, | |
1084 | &first_bad, &bad_sectors); | |
1085 | if (is_bad < 0) { | |
1086 | /* Mustn't write here until the bad block | |
1087 | * is acknowledged | |
1088 | */ | |
1089 | atomic_inc(&rdev->nr_pending); | |
1090 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1091 | blocked_rdev = rdev; | |
1092 | break; | |
1093 | } | |
1094 | if (is_bad && first_bad <= dev_sector) { | |
1095 | /* Cannot write here at all */ | |
1096 | bad_sectors -= (dev_sector - first_bad); | |
1097 | if (bad_sectors < max_sectors) | |
1098 | /* Mustn't write more than bad_sectors | |
1099 | * to other devices yet | |
1100 | */ | |
1101 | max_sectors = bad_sectors; | |
1102 | /* We don't set R10BIO_Degraded as that | |
1103 | * only applies if the disk is missing, | |
1104 | * so it might be re-added, and we want to | |
1105 | * know to recover this chunk. | |
1106 | * In this case the device is here, and the | |
1107 | * fact that this chunk is not in-sync is | |
1108 | * recorded in the bad block log. | |
1109 | */ | |
1110 | continue; | |
1111 | } | |
1112 | if (is_bad) { | |
1113 | int good_sectors = first_bad - dev_sector; | |
1114 | if (good_sectors < max_sectors) | |
1115 | max_sectors = good_sectors; | |
1116 | } | |
6cce3b23 | 1117 | } |
d4432c23 N |
1118 | r10_bio->devs[i].bio = bio; |
1119 | atomic_inc(&rdev->nr_pending); | |
1da177e4 LT |
1120 | } |
1121 | rcu_read_unlock(); | |
1122 | ||
6bfe0b49 DW |
1123 | if (unlikely(blocked_rdev)) { |
1124 | /* Have to wait for this device to get unblocked, then retry */ | |
1125 | int j; | |
1126 | int d; | |
1127 | ||
1128 | for (j = 0; j < i; j++) | |
1129 | if (r10_bio->devs[j].bio) { | |
1130 | d = r10_bio->devs[j].devnum; | |
1131 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1132 | } | |
1133 | allow_barrier(conf); | |
1134 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1135 | wait_barrier(conf); | |
1136 | goto retry_write; | |
1137 | } | |
1138 | ||
d4432c23 N |
1139 | if (max_sectors < r10_bio->sectors) { |
1140 | /* We are splitting this into multiple parts, so | |
1141 | * we need to prepare for allocating another r10_bio. | |
1142 | */ | |
1143 | r10_bio->sectors = max_sectors; | |
1144 | spin_lock_irq(&conf->device_lock); | |
1145 | if (bio->bi_phys_segments == 0) | |
1146 | bio->bi_phys_segments = 2; | |
1147 | else | |
1148 | bio->bi_phys_segments++; | |
1149 | spin_unlock_irq(&conf->device_lock); | |
1150 | } | |
1151 | sectors_handled = r10_bio->sector + max_sectors - bio->bi_sector; | |
1152 | ||
4e78064f | 1153 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1154 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1155 | |
1da177e4 LT |
1156 | for (i = 0; i < conf->copies; i++) { |
1157 | struct bio *mbio; | |
1158 | int d = r10_bio->devs[i].devnum; | |
1159 | if (!r10_bio->devs[i].bio) | |
1160 | continue; | |
1161 | ||
a167f663 | 1162 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
d4432c23 N |
1163 | md_trim_bio(mbio, r10_bio->sector - bio->bi_sector, |
1164 | max_sectors); | |
1da177e4 LT |
1165 | r10_bio->devs[i].bio = mbio; |
1166 | ||
d4432c23 N |
1167 | mbio->bi_sector = (r10_bio->devs[i].addr+ |
1168 | conf->mirrors[d].rdev->data_offset); | |
1da177e4 LT |
1169 | mbio->bi_bdev = conf->mirrors[d].rdev->bdev; |
1170 | mbio->bi_end_io = raid10_end_write_request; | |
e9c7469b | 1171 | mbio->bi_rw = WRITE | do_sync | do_fua; |
1da177e4 LT |
1172 | mbio->bi_private = r10_bio; |
1173 | ||
1174 | atomic_inc(&r10_bio->remaining); | |
4e78064f N |
1175 | spin_lock_irqsave(&conf->device_lock, flags); |
1176 | bio_list_add(&conf->pending_bio_list, mbio); | |
34db0cd6 | 1177 | conf->pending_count++; |
4e78064f | 1178 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1179 | } |
1180 | ||
079fa166 N |
1181 | /* Don't remove the bias on 'remaining' (one_write_done) until |
1182 | * after checking if we need to go around again. | |
1183 | */ | |
a35e63ef | 1184 | |
d4432c23 | 1185 | if (sectors_handled < (bio->bi_size >> 9)) { |
079fa166 | 1186 | one_write_done(r10_bio); |
5e570289 | 1187 | /* We need another r10_bio. It has already been counted |
d4432c23 N |
1188 | * in bio->bi_phys_segments. |
1189 | */ | |
1190 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1191 | ||
1192 | r10_bio->master_bio = bio; | |
1193 | r10_bio->sectors = (bio->bi_size >> 9) - sectors_handled; | |
1194 | ||
1195 | r10_bio->mddev = mddev; | |
1196 | r10_bio->sector = bio->bi_sector + sectors_handled; | |
1197 | r10_bio->state = 0; | |
1198 | goto retry_write; | |
1199 | } | |
079fa166 N |
1200 | one_write_done(r10_bio); |
1201 | ||
1202 | /* In case raid10d snuck in to freeze_array */ | |
1203 | wake_up(&conf->wait_barrier); | |
d4432c23 | 1204 | |
c3b328ac | 1205 | if (do_sync || !mddev->bitmap || !plugged) |
e3881a68 | 1206 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
1207 | } |
1208 | ||
fd01b88c | 1209 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1210 | { |
e879a879 | 1211 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1212 | int i; |
1213 | ||
1214 | if (conf->near_copies < conf->raid_disks) | |
9d8f0363 | 1215 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
1da177e4 LT |
1216 | if (conf->near_copies > 1) |
1217 | seq_printf(seq, " %d near-copies", conf->near_copies); | |
c93983bf N |
1218 | if (conf->far_copies > 1) { |
1219 | if (conf->far_offset) | |
1220 | seq_printf(seq, " %d offset-copies", conf->far_copies); | |
1221 | else | |
1222 | seq_printf(seq, " %d far-copies", conf->far_copies); | |
1223 | } | |
1da177e4 | 1224 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
76186dd8 | 1225 | conf->raid_disks - mddev->degraded); |
1da177e4 LT |
1226 | for (i = 0; i < conf->raid_disks; i++) |
1227 | seq_printf(seq, "%s", | |
1228 | conf->mirrors[i].rdev && | |
b2d444d7 | 1229 | test_bit(In_sync, &conf->mirrors[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
1230 | seq_printf(seq, "]"); |
1231 | } | |
1232 | ||
700c7213 N |
1233 | /* check if there are enough drives for |
1234 | * every block to appear on atleast one. | |
1235 | * Don't consider the device numbered 'ignore' | |
1236 | * as we might be about to remove it. | |
1237 | */ | |
e879a879 | 1238 | static int enough(struct r10conf *conf, int ignore) |
700c7213 N |
1239 | { |
1240 | int first = 0; | |
1241 | ||
1242 | do { | |
1243 | int n = conf->copies; | |
1244 | int cnt = 0; | |
1245 | while (n--) { | |
1246 | if (conf->mirrors[first].rdev && | |
1247 | first != ignore) | |
1248 | cnt++; | |
1249 | first = (first+1) % conf->raid_disks; | |
1250 | } | |
1251 | if (cnt == 0) | |
1252 | return 0; | |
1253 | } while (first != 0); | |
1254 | return 1; | |
1255 | } | |
1256 | ||
fd01b88c | 1257 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1258 | { |
1259 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1260 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1261 | |
1262 | /* | |
1263 | * If it is not operational, then we have already marked it as dead | |
1264 | * else if it is the last working disks, ignore the error, let the | |
1265 | * next level up know. | |
1266 | * else mark the drive as failed | |
1267 | */ | |
b2d444d7 | 1268 | if (test_bit(In_sync, &rdev->flags) |
700c7213 | 1269 | && !enough(conf, rdev->raid_disk)) |
1da177e4 LT |
1270 | /* |
1271 | * Don't fail the drive, just return an IO error. | |
1da177e4 LT |
1272 | */ |
1273 | return; | |
c04be0aa N |
1274 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1275 | unsigned long flags; | |
1276 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1277 | mddev->degraded++; |
c04be0aa | 1278 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1279 | /* |
1280 | * if recovery is running, make sure it aborts. | |
1281 | */ | |
dfc70645 | 1282 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1da177e4 | 1283 | } |
de393cde | 1284 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1285 | set_bit(Faulty, &rdev->flags); |
850b2b42 | 1286 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
1287 | printk(KERN_ALERT |
1288 | "md/raid10:%s: Disk failure on %s, disabling device.\n" | |
1289 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
128595ed N |
1290 | mdname(mddev), bdevname(rdev->bdev, b), |
1291 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1292 | } |
1293 | ||
e879a879 | 1294 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1295 | { |
1296 | int i; | |
0f6d02d5 | 1297 | struct mirror_info *tmp; |
1da177e4 | 1298 | |
128595ed | 1299 | printk(KERN_DEBUG "RAID10 conf printout:\n"); |
1da177e4 | 1300 | if (!conf) { |
128595ed | 1301 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1302 | return; |
1303 | } | |
128595ed | 1304 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1305 | conf->raid_disks); |
1306 | ||
1307 | for (i = 0; i < conf->raid_disks; i++) { | |
1308 | char b[BDEVNAME_SIZE]; | |
1309 | tmp = conf->mirrors + i; | |
1310 | if (tmp->rdev) | |
128595ed | 1311 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
b2d444d7 N |
1312 | i, !test_bit(In_sync, &tmp->rdev->flags), |
1313 | !test_bit(Faulty, &tmp->rdev->flags), | |
1da177e4 LT |
1314 | bdevname(tmp->rdev->bdev,b)); |
1315 | } | |
1316 | } | |
1317 | ||
e879a879 | 1318 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1319 | { |
0a27ec96 N |
1320 | wait_barrier(conf); |
1321 | allow_barrier(conf); | |
1da177e4 LT |
1322 | |
1323 | mempool_destroy(conf->r10buf_pool); | |
1324 | conf->r10buf_pool = NULL; | |
1325 | } | |
1326 | ||
fd01b88c | 1327 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1328 | { |
1329 | int i; | |
e879a879 | 1330 | struct r10conf *conf = mddev->private; |
0f6d02d5 | 1331 | struct mirror_info *tmp; |
6b965620 N |
1332 | int count = 0; |
1333 | unsigned long flags; | |
1da177e4 LT |
1334 | |
1335 | /* | |
1336 | * Find all non-in_sync disks within the RAID10 configuration | |
1337 | * and mark them in_sync | |
1338 | */ | |
1339 | for (i = 0; i < conf->raid_disks; i++) { | |
1340 | tmp = conf->mirrors + i; | |
1341 | if (tmp->rdev | |
b2d444d7 | 1342 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 1343 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 1344 | count++; |
e6ffbcb6 | 1345 | sysfs_notify_dirent(tmp->rdev->sysfs_state); |
1da177e4 LT |
1346 | } |
1347 | } | |
6b965620 N |
1348 | spin_lock_irqsave(&conf->device_lock, flags); |
1349 | mddev->degraded -= count; | |
1350 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1351 | |
1352 | print_conf(conf); | |
6b965620 | 1353 | return count; |
1da177e4 LT |
1354 | } |
1355 | ||
1356 | ||
fd01b88c | 1357 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1358 | { |
e879a879 | 1359 | struct r10conf *conf = mddev->private; |
199050ea | 1360 | int err = -EEXIST; |
1da177e4 | 1361 | int mirror; |
6c2fce2e | 1362 | int first = 0; |
84707f38 | 1363 | int last = conf->raid_disks - 1; |
1da177e4 LT |
1364 | |
1365 | if (mddev->recovery_cp < MaxSector) | |
1366 | /* only hot-add to in-sync arrays, as recovery is | |
1367 | * very different from resync | |
1368 | */ | |
199050ea | 1369 | return -EBUSY; |
700c7213 | 1370 | if (!enough(conf, -1)) |
199050ea | 1371 | return -EINVAL; |
1da177e4 | 1372 | |
a53a6c85 | 1373 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1374 | first = last = rdev->raid_disk; |
1da177e4 | 1375 | |
2c4193df | 1376 | if (rdev->saved_raid_disk >= first && |
6cce3b23 N |
1377 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1378 | mirror = rdev->saved_raid_disk; | |
1379 | else | |
6c2fce2e | 1380 | mirror = first; |
2bb77736 | 1381 | for ( ; mirror <= last ; mirror++) { |
0f6d02d5 | 1382 | struct mirror_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1383 | if (p->recovery_disabled == mddev->recovery_disabled) |
1384 | continue; | |
7fcc7c8a | 1385 | if (p->rdev) |
2bb77736 | 1386 | continue; |
1da177e4 | 1387 | |
2bb77736 N |
1388 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1389 | rdev->data_offset << 9); | |
1390 | /* as we don't honour merge_bvec_fn, we must | |
1391 | * never risk violating it, so limit | |
1392 | * ->max_segments to one lying with a single | |
1393 | * page, as a one page request is never in | |
1394 | * violation. | |
1395 | */ | |
1396 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { | |
1397 | blk_queue_max_segments(mddev->queue, 1); | |
1398 | blk_queue_segment_boundary(mddev->queue, | |
1399 | PAGE_CACHE_SIZE - 1); | |
1da177e4 LT |
1400 | } |
1401 | ||
2bb77736 | 1402 | p->head_position = 0; |
d890fa2b | 1403 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1404 | rdev->raid_disk = mirror; |
1405 | err = 0; | |
1406 | if (rdev->saved_raid_disk != mirror) | |
1407 | conf->fullsync = 1; | |
1408 | rcu_assign_pointer(p->rdev, rdev); | |
1409 | break; | |
1410 | } | |
1411 | ||
ac5e7113 | 1412 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1413 | print_conf(conf); |
199050ea | 1414 | return err; |
1da177e4 LT |
1415 | } |
1416 | ||
b8321b68 | 1417 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1418 | { |
e879a879 | 1419 | struct r10conf *conf = mddev->private; |
1da177e4 | 1420 | int err = 0; |
b8321b68 | 1421 | int number = rdev->raid_disk; |
0f6d02d5 | 1422 | struct mirror_info *p = conf->mirrors+ number; |
1da177e4 LT |
1423 | |
1424 | print_conf(conf); | |
b8321b68 | 1425 | if (rdev == p->rdev) { |
b2d444d7 | 1426 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1427 | atomic_read(&rdev->nr_pending)) { |
1428 | err = -EBUSY; | |
1429 | goto abort; | |
1430 | } | |
dfc70645 N |
1431 | /* Only remove faulty devices in recovery |
1432 | * is not possible. | |
1433 | */ | |
1434 | if (!test_bit(Faulty, &rdev->flags) && | |
2bb77736 | 1435 | mddev->recovery_disabled != p->recovery_disabled && |
700c7213 | 1436 | enough(conf, -1)) { |
dfc70645 N |
1437 | err = -EBUSY; |
1438 | goto abort; | |
1439 | } | |
1da177e4 | 1440 | p->rdev = NULL; |
fbd568a3 | 1441 | synchronize_rcu(); |
1da177e4 LT |
1442 | if (atomic_read(&rdev->nr_pending)) { |
1443 | /* lost the race, try later */ | |
1444 | err = -EBUSY; | |
1445 | p->rdev = rdev; | |
ac5e7113 | 1446 | goto abort; |
1da177e4 | 1447 | } |
a91a2785 | 1448 | err = md_integrity_register(mddev); |
1da177e4 LT |
1449 | } |
1450 | abort: | |
1451 | ||
1452 | print_conf(conf); | |
1453 | return err; | |
1454 | } | |
1455 | ||
1456 | ||
6712ecf8 | 1457 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1458 | { |
9f2c9d12 | 1459 | struct r10bio *r10_bio = bio->bi_private; |
e879a879 | 1460 | struct r10conf *conf = r10_bio->mddev->private; |
778ca018 | 1461 | int d; |
1da177e4 | 1462 | |
69335ef3 | 1463 | d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); |
0eb3ff12 N |
1464 | |
1465 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1466 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
e684e41d N |
1467 | else |
1468 | /* The write handler will notice the lack of | |
1469 | * R10BIO_Uptodate and record any errors etc | |
1470 | */ | |
4dbcdc75 N |
1471 | atomic_add(r10_bio->sectors, |
1472 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1473 | |
1474 | /* for reconstruct, we always reschedule after a read. | |
1475 | * for resync, only after all reads | |
1476 | */ | |
73d5c38a | 1477 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1478 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1479 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1480 | /* we have read all the blocks, | |
1481 | * do the comparison in process context in raid10d | |
1482 | */ | |
1483 | reschedule_retry(r10_bio); | |
1484 | } | |
1da177e4 LT |
1485 | } |
1486 | ||
9f2c9d12 | 1487 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1488 | { |
fd01b88c | 1489 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1490 | |
1da177e4 LT |
1491 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1492 | if (r10_bio->master_bio == NULL) { | |
1493 | /* the primary of several recovery bios */ | |
73d5c38a | 1494 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1495 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1496 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1497 | reschedule_retry(r10_bio); |
1498 | else | |
1499 | put_buf(r10_bio); | |
73d5c38a | 1500 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1501 | break; |
1502 | } else { | |
9f2c9d12 | 1503 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1504 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1505 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1506 | reschedule_retry(r10_bio); |
1507 | else | |
1508 | put_buf(r10_bio); | |
1da177e4 LT |
1509 | r10_bio = r10_bio2; |
1510 | } | |
1511 | } | |
1da177e4 LT |
1512 | } |
1513 | ||
5e570289 N |
1514 | static void end_sync_write(struct bio *bio, int error) |
1515 | { | |
1516 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 1517 | struct r10bio *r10_bio = bio->bi_private; |
fd01b88c | 1518 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1519 | struct r10conf *conf = mddev->private; |
5e570289 N |
1520 | int d; |
1521 | sector_t first_bad; | |
1522 | int bad_sectors; | |
1523 | int slot; | |
1524 | ||
69335ef3 | 1525 | d = find_bio_disk(conf, r10_bio, bio, &slot, NULL); |
5e570289 N |
1526 | |
1527 | if (!uptodate) { | |
1528 | set_bit(WriteErrorSeen, &conf->mirrors[d].rdev->flags); | |
1529 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
1530 | } else if (is_badblock(conf->mirrors[d].rdev, | |
1531 | r10_bio->devs[slot].addr, | |
1532 | r10_bio->sectors, | |
1533 | &first_bad, &bad_sectors)) | |
1534 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
1535 | ||
1536 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1537 | ||
1538 | end_sync_request(r10_bio); | |
1539 | } | |
1540 | ||
1da177e4 LT |
1541 | /* |
1542 | * Note: sync and recover and handled very differently for raid10 | |
1543 | * This code is for resync. | |
1544 | * For resync, we read through virtual addresses and read all blocks. | |
1545 | * If there is any error, we schedule a write. The lowest numbered | |
1546 | * drive is authoritative. | |
1547 | * However requests come for physical address, so we need to map. | |
1548 | * For every physical address there are raid_disks/copies virtual addresses, | |
1549 | * which is always are least one, but is not necessarly an integer. | |
1550 | * This means that a physical address can span multiple chunks, so we may | |
1551 | * have to submit multiple io requests for a single sync request. | |
1552 | */ | |
1553 | /* | |
1554 | * We check if all blocks are in-sync and only write to blocks that | |
1555 | * aren't in sync | |
1556 | */ | |
9f2c9d12 | 1557 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1558 | { |
e879a879 | 1559 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1560 | int i, first; |
1561 | struct bio *tbio, *fbio; | |
1562 | ||
1563 | atomic_set(&r10_bio->remaining, 1); | |
1564 | ||
1565 | /* find the first device with a block */ | |
1566 | for (i=0; i<conf->copies; i++) | |
1567 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) | |
1568 | break; | |
1569 | ||
1570 | if (i == conf->copies) | |
1571 | goto done; | |
1572 | ||
1573 | first = i; | |
1574 | fbio = r10_bio->devs[i].bio; | |
1575 | ||
1576 | /* now find blocks with errors */ | |
0eb3ff12 N |
1577 | for (i=0 ; i < conf->copies ; i++) { |
1578 | int j, d; | |
1579 | int vcnt = r10_bio->sectors >> (PAGE_SHIFT-9); | |
1da177e4 | 1580 | |
1da177e4 | 1581 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
1582 | |
1583 | if (tbio->bi_end_io != end_sync_read) | |
1584 | continue; | |
1585 | if (i == first) | |
1da177e4 | 1586 | continue; |
0eb3ff12 N |
1587 | if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) { |
1588 | /* We know that the bi_io_vec layout is the same for | |
1589 | * both 'first' and 'i', so we just compare them. | |
1590 | * All vec entries are PAGE_SIZE; | |
1591 | */ | |
1592 | for (j = 0; j < vcnt; j++) | |
1593 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), | |
1594 | page_address(tbio->bi_io_vec[j].bv_page), | |
1595 | PAGE_SIZE)) | |
1596 | break; | |
1597 | if (j == vcnt) | |
1598 | continue; | |
1599 | mddev->resync_mismatches += r10_bio->sectors; | |
f84ee364 N |
1600 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
1601 | /* Don't fix anything. */ | |
1602 | continue; | |
0eb3ff12 | 1603 | } |
f84ee364 N |
1604 | /* Ok, we need to write this bio, either to correct an |
1605 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
1606 | * First we need to fixup bv_offset, bv_len and |
1607 | * bi_vecs, as the read request might have corrupted these | |
1608 | */ | |
1609 | tbio->bi_vcnt = vcnt; | |
1610 | tbio->bi_size = r10_bio->sectors << 9; | |
1611 | tbio->bi_idx = 0; | |
1612 | tbio->bi_phys_segments = 0; | |
1da177e4 LT |
1613 | tbio->bi_flags &= ~(BIO_POOL_MASK - 1); |
1614 | tbio->bi_flags |= 1 << BIO_UPTODATE; | |
1615 | tbio->bi_next = NULL; | |
1616 | tbio->bi_rw = WRITE; | |
1617 | tbio->bi_private = r10_bio; | |
1618 | tbio->bi_sector = r10_bio->devs[i].addr; | |
1619 | ||
1620 | for (j=0; j < vcnt ; j++) { | |
1621 | tbio->bi_io_vec[j].bv_offset = 0; | |
1622 | tbio->bi_io_vec[j].bv_len = PAGE_SIZE; | |
1623 | ||
1624 | memcpy(page_address(tbio->bi_io_vec[j].bv_page), | |
1625 | page_address(fbio->bi_io_vec[j].bv_page), | |
1626 | PAGE_SIZE); | |
1627 | } | |
1628 | tbio->bi_end_io = end_sync_write; | |
1629 | ||
1630 | d = r10_bio->devs[i].devnum; | |
1631 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1632 | atomic_inc(&r10_bio->remaining); | |
1633 | md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9); | |
1634 | ||
1635 | tbio->bi_sector += conf->mirrors[d].rdev->data_offset; | |
1636 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
1637 | generic_make_request(tbio); | |
1638 | } | |
1639 | ||
1640 | done: | |
1641 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
1642 | md_done_sync(mddev, r10_bio->sectors, 1); | |
1643 | put_buf(r10_bio); | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | /* | |
1648 | * Now for the recovery code. | |
1649 | * Recovery happens across physical sectors. | |
1650 | * We recover all non-is_sync drives by finding the virtual address of | |
1651 | * each, and then choose a working drive that also has that virt address. | |
1652 | * There is a separate r10_bio for each non-in_sync drive. | |
1653 | * Only the first two slots are in use. The first for reading, | |
1654 | * The second for writing. | |
1655 | * | |
1656 | */ | |
9f2c9d12 | 1657 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
1658 | { |
1659 | /* We got a read error during recovery. | |
1660 | * We repeat the read in smaller page-sized sections. | |
1661 | * If a read succeeds, write it to the new device or record | |
1662 | * a bad block if we cannot. | |
1663 | * If a read fails, record a bad block on both old and | |
1664 | * new devices. | |
1665 | */ | |
fd01b88c | 1666 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1667 | struct r10conf *conf = mddev->private; |
5e570289 N |
1668 | struct bio *bio = r10_bio->devs[0].bio; |
1669 | sector_t sect = 0; | |
1670 | int sectors = r10_bio->sectors; | |
1671 | int idx = 0; | |
1672 | int dr = r10_bio->devs[0].devnum; | |
1673 | int dw = r10_bio->devs[1].devnum; | |
1674 | ||
1675 | while (sectors) { | |
1676 | int s = sectors; | |
3cb03002 | 1677 | struct md_rdev *rdev; |
5e570289 N |
1678 | sector_t addr; |
1679 | int ok; | |
1680 | ||
1681 | if (s > (PAGE_SIZE>>9)) | |
1682 | s = PAGE_SIZE >> 9; | |
1683 | ||
1684 | rdev = conf->mirrors[dr].rdev; | |
1685 | addr = r10_bio->devs[0].addr + sect, | |
1686 | ok = sync_page_io(rdev, | |
1687 | addr, | |
1688 | s << 9, | |
1689 | bio->bi_io_vec[idx].bv_page, | |
1690 | READ, false); | |
1691 | if (ok) { | |
1692 | rdev = conf->mirrors[dw].rdev; | |
1693 | addr = r10_bio->devs[1].addr + sect; | |
1694 | ok = sync_page_io(rdev, | |
1695 | addr, | |
1696 | s << 9, | |
1697 | bio->bi_io_vec[idx].bv_page, | |
1698 | WRITE, false); | |
1699 | if (!ok) | |
1700 | set_bit(WriteErrorSeen, &rdev->flags); | |
1701 | } | |
1702 | if (!ok) { | |
1703 | /* We don't worry if we cannot set a bad block - | |
1704 | * it really is bad so there is no loss in not | |
1705 | * recording it yet | |
1706 | */ | |
1707 | rdev_set_badblocks(rdev, addr, s, 0); | |
1708 | ||
1709 | if (rdev != conf->mirrors[dw].rdev) { | |
1710 | /* need bad block on destination too */ | |
3cb03002 | 1711 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
1712 | addr = r10_bio->devs[1].addr + sect; |
1713 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
1714 | if (!ok) { | |
1715 | /* just abort the recovery */ | |
1716 | printk(KERN_NOTICE | |
1717 | "md/raid10:%s: recovery aborted" | |
1718 | " due to read error\n", | |
1719 | mdname(mddev)); | |
1720 | ||
1721 | conf->mirrors[dw].recovery_disabled | |
1722 | = mddev->recovery_disabled; | |
1723 | set_bit(MD_RECOVERY_INTR, | |
1724 | &mddev->recovery); | |
1725 | break; | |
1726 | } | |
1727 | } | |
1728 | } | |
1729 | ||
1730 | sectors -= s; | |
1731 | sect += s; | |
1732 | idx++; | |
1733 | } | |
1734 | } | |
1da177e4 | 1735 | |
9f2c9d12 | 1736 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 1737 | { |
e879a879 | 1738 | struct r10conf *conf = mddev->private; |
c65060ad NK |
1739 | int d; |
1740 | struct bio *wbio; | |
1da177e4 | 1741 | |
5e570289 N |
1742 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
1743 | fix_recovery_read_error(r10_bio); | |
1744 | end_sync_request(r10_bio); | |
1745 | return; | |
1746 | } | |
1747 | ||
c65060ad NK |
1748 | /* |
1749 | * share the pages with the first bio | |
1da177e4 LT |
1750 | * and submit the write request |
1751 | */ | |
1da177e4 | 1752 | wbio = r10_bio->devs[1].bio; |
1da177e4 LT |
1753 | d = r10_bio->devs[1].devnum; |
1754 | ||
1755 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
1756 | md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9); | |
5e570289 | 1757 | generic_make_request(wbio); |
1da177e4 LT |
1758 | } |
1759 | ||
1760 | ||
1e50915f RB |
1761 | /* |
1762 | * Used by fix_read_error() to decay the per rdev read_errors. | |
1763 | * We halve the read error count for every hour that has elapsed | |
1764 | * since the last recorded read error. | |
1765 | * | |
1766 | */ | |
fd01b88c | 1767 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f RB |
1768 | { |
1769 | struct timespec cur_time_mon; | |
1770 | unsigned long hours_since_last; | |
1771 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
1772 | ||
1773 | ktime_get_ts(&cur_time_mon); | |
1774 | ||
1775 | if (rdev->last_read_error.tv_sec == 0 && | |
1776 | rdev->last_read_error.tv_nsec == 0) { | |
1777 | /* first time we've seen a read error */ | |
1778 | rdev->last_read_error = cur_time_mon; | |
1779 | return; | |
1780 | } | |
1781 | ||
1782 | hours_since_last = (cur_time_mon.tv_sec - | |
1783 | rdev->last_read_error.tv_sec) / 3600; | |
1784 | ||
1785 | rdev->last_read_error = cur_time_mon; | |
1786 | ||
1787 | /* | |
1788 | * if hours_since_last is > the number of bits in read_errors | |
1789 | * just set read errors to 0. We do this to avoid | |
1790 | * overflowing the shift of read_errors by hours_since_last. | |
1791 | */ | |
1792 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
1793 | atomic_set(&rdev->read_errors, 0); | |
1794 | else | |
1795 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
1796 | } | |
1797 | ||
3cb03002 | 1798 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
1799 | int sectors, struct page *page, int rw) |
1800 | { | |
1801 | sector_t first_bad; | |
1802 | int bad_sectors; | |
1803 | ||
1804 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
1805 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
1806 | return -1; | |
1807 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
1808 | /* success */ | |
1809 | return 1; | |
1810 | if (rw == WRITE) | |
1811 | set_bit(WriteErrorSeen, &rdev->flags); | |
1812 | /* need to record an error - either for the block or the device */ | |
1813 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1814 | md_error(rdev->mddev, rdev); | |
1815 | return 0; | |
1816 | } | |
1817 | ||
1da177e4 LT |
1818 | /* |
1819 | * This is a kernel thread which: | |
1820 | * | |
1821 | * 1. Retries failed read operations on working mirrors. | |
1822 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 1823 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
1824 | */ |
1825 | ||
e879a879 | 1826 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
1827 | { |
1828 | int sect = 0; /* Offset from r10_bio->sector */ | |
1829 | int sectors = r10_bio->sectors; | |
3cb03002 | 1830 | struct md_rdev*rdev; |
1e50915f | 1831 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 1832 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 1833 | |
7c4e06ff N |
1834 | /* still own a reference to this rdev, so it cannot |
1835 | * have been cleared recently. | |
1836 | */ | |
1837 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 1838 | |
7c4e06ff N |
1839 | if (test_bit(Faulty, &rdev->flags)) |
1840 | /* drive has already been failed, just ignore any | |
1841 | more fix_read_error() attempts */ | |
1842 | return; | |
1e50915f | 1843 | |
7c4e06ff N |
1844 | check_decay_read_errors(mddev, rdev); |
1845 | atomic_inc(&rdev->read_errors); | |
1846 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
1847 | char b[BDEVNAME_SIZE]; | |
1848 | bdevname(rdev->bdev, b); | |
1e50915f | 1849 | |
7c4e06ff N |
1850 | printk(KERN_NOTICE |
1851 | "md/raid10:%s: %s: Raid device exceeded " | |
1852 | "read_error threshold [cur %d:max %d]\n", | |
1853 | mdname(mddev), b, | |
1854 | atomic_read(&rdev->read_errors), max_read_errors); | |
1855 | printk(KERN_NOTICE | |
1856 | "md/raid10:%s: %s: Failing raid device\n", | |
1857 | mdname(mddev), b); | |
1858 | md_error(mddev, conf->mirrors[d].rdev); | |
1859 | return; | |
1e50915f | 1860 | } |
1e50915f | 1861 | |
6814d536 N |
1862 | while(sectors) { |
1863 | int s = sectors; | |
1864 | int sl = r10_bio->read_slot; | |
1865 | int success = 0; | |
1866 | int start; | |
1867 | ||
1868 | if (s > (PAGE_SIZE>>9)) | |
1869 | s = PAGE_SIZE >> 9; | |
1870 | ||
1871 | rcu_read_lock(); | |
1872 | do { | |
8dbed5ce N |
1873 | sector_t first_bad; |
1874 | int bad_sectors; | |
1875 | ||
0544a21d | 1876 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
1877 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
1878 | if (rdev && | |
8dbed5ce N |
1879 | test_bit(In_sync, &rdev->flags) && |
1880 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, | |
1881 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
1882 | atomic_inc(&rdev->nr_pending); |
1883 | rcu_read_unlock(); | |
2b193363 | 1884 | success = sync_page_io(rdev, |
6814d536 | 1885 | r10_bio->devs[sl].addr + |
ccebd4c4 | 1886 | sect, |
6814d536 | 1887 | s<<9, |
ccebd4c4 | 1888 | conf->tmppage, READ, false); |
6814d536 N |
1889 | rdev_dec_pending(rdev, mddev); |
1890 | rcu_read_lock(); | |
1891 | if (success) | |
1892 | break; | |
1893 | } | |
1894 | sl++; | |
1895 | if (sl == conf->copies) | |
1896 | sl = 0; | |
1897 | } while (!success && sl != r10_bio->read_slot); | |
1898 | rcu_read_unlock(); | |
1899 | ||
1900 | if (!success) { | |
58c54fcc N |
1901 | /* Cannot read from anywhere, just mark the block |
1902 | * as bad on the first device to discourage future | |
1903 | * reads. | |
1904 | */ | |
6814d536 | 1905 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
1906 | rdev = conf->mirrors[dn].rdev; |
1907 | ||
1908 | if (!rdev_set_badblocks( | |
1909 | rdev, | |
1910 | r10_bio->devs[r10_bio->read_slot].addr | |
1911 | + sect, | |
1912 | s, 0)) | |
1913 | md_error(mddev, rdev); | |
6814d536 N |
1914 | break; |
1915 | } | |
1916 | ||
1917 | start = sl; | |
1918 | /* write it back and re-read */ | |
1919 | rcu_read_lock(); | |
1920 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 1921 | char b[BDEVNAME_SIZE]; |
0544a21d | 1922 | |
6814d536 N |
1923 | if (sl==0) |
1924 | sl = conf->copies; | |
1925 | sl--; | |
1926 | d = r10_bio->devs[sl].devnum; | |
1927 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 N |
1928 | if (!rdev || |
1929 | !test_bit(In_sync, &rdev->flags)) | |
1930 | continue; | |
1931 | ||
1932 | atomic_inc(&rdev->nr_pending); | |
1933 | rcu_read_unlock(); | |
58c54fcc N |
1934 | if (r10_sync_page_io(rdev, |
1935 | r10_bio->devs[sl].addr + | |
1936 | sect, | |
1937 | s<<9, conf->tmppage, WRITE) | |
1294b9c9 N |
1938 | == 0) { |
1939 | /* Well, this device is dead */ | |
1940 | printk(KERN_NOTICE | |
1941 | "md/raid10:%s: read correction " | |
1942 | "write failed" | |
1943 | " (%d sectors at %llu on %s)\n", | |
1944 | mdname(mddev), s, | |
1945 | (unsigned long long)( | |
1946 | sect + rdev->data_offset), | |
1947 | bdevname(rdev->bdev, b)); | |
1948 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
1949 | "drive\n", | |
1950 | mdname(mddev), | |
1951 | bdevname(rdev->bdev, b)); | |
6814d536 | 1952 | } |
1294b9c9 N |
1953 | rdev_dec_pending(rdev, mddev); |
1954 | rcu_read_lock(); | |
6814d536 N |
1955 | } |
1956 | sl = start; | |
1957 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 1958 | char b[BDEVNAME_SIZE]; |
0544a21d | 1959 | |
6814d536 N |
1960 | if (sl==0) |
1961 | sl = conf->copies; | |
1962 | sl--; | |
1963 | d = r10_bio->devs[sl].devnum; | |
1964 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 N |
1965 | if (!rdev || |
1966 | !test_bit(In_sync, &rdev->flags)) | |
1967 | continue; | |
6814d536 | 1968 | |
1294b9c9 N |
1969 | atomic_inc(&rdev->nr_pending); |
1970 | rcu_read_unlock(); | |
58c54fcc N |
1971 | switch (r10_sync_page_io(rdev, |
1972 | r10_bio->devs[sl].addr + | |
1973 | sect, | |
1974 | s<<9, conf->tmppage, | |
1975 | READ)) { | |
1976 | case 0: | |
1294b9c9 N |
1977 | /* Well, this device is dead */ |
1978 | printk(KERN_NOTICE | |
1979 | "md/raid10:%s: unable to read back " | |
1980 | "corrected sectors" | |
1981 | " (%d sectors at %llu on %s)\n", | |
1982 | mdname(mddev), s, | |
1983 | (unsigned long long)( | |
1984 | sect + rdev->data_offset), | |
1985 | bdevname(rdev->bdev, b)); | |
1986 | printk(KERN_NOTICE "md/raid10:%s: %s: failing " | |
1987 | "drive\n", | |
1988 | mdname(mddev), | |
1989 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
1990 | break; |
1991 | case 1: | |
1294b9c9 N |
1992 | printk(KERN_INFO |
1993 | "md/raid10:%s: read error corrected" | |
1994 | " (%d sectors at %llu on %s)\n", | |
1995 | mdname(mddev), s, | |
1996 | (unsigned long long)( | |
1997 | sect + rdev->data_offset), | |
1998 | bdevname(rdev->bdev, b)); | |
1999 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2000 | } |
1294b9c9 N |
2001 | |
2002 | rdev_dec_pending(rdev, mddev); | |
2003 | rcu_read_lock(); | |
6814d536 N |
2004 | } |
2005 | rcu_read_unlock(); | |
2006 | ||
2007 | sectors -= s; | |
2008 | sect += s; | |
2009 | } | |
2010 | } | |
2011 | ||
bd870a16 N |
2012 | static void bi_complete(struct bio *bio, int error) |
2013 | { | |
2014 | complete((struct completion *)bio->bi_private); | |
2015 | } | |
2016 | ||
2017 | static int submit_bio_wait(int rw, struct bio *bio) | |
2018 | { | |
2019 | struct completion event; | |
2020 | rw |= REQ_SYNC; | |
2021 | ||
2022 | init_completion(&event); | |
2023 | bio->bi_private = &event; | |
2024 | bio->bi_end_io = bi_complete; | |
2025 | submit_bio(rw, bio); | |
2026 | wait_for_completion(&event); | |
2027 | ||
2028 | return test_bit(BIO_UPTODATE, &bio->bi_flags); | |
2029 | } | |
2030 | ||
9f2c9d12 | 2031 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2032 | { |
2033 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2034 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2035 | struct r10conf *conf = mddev->private; |
3cb03002 | 2036 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2037 | /* bio has the data to be written to slot 'i' where |
2038 | * we just recently had a write error. | |
2039 | * We repeatedly clone the bio and trim down to one block, | |
2040 | * then try the write. Where the write fails we record | |
2041 | * a bad block. | |
2042 | * It is conceivable that the bio doesn't exactly align with | |
2043 | * blocks. We must handle this. | |
2044 | * | |
2045 | * We currently own a reference to the rdev. | |
2046 | */ | |
2047 | ||
2048 | int block_sectors; | |
2049 | sector_t sector; | |
2050 | int sectors; | |
2051 | int sect_to_write = r10_bio->sectors; | |
2052 | int ok = 1; | |
2053 | ||
2054 | if (rdev->badblocks.shift < 0) | |
2055 | return 0; | |
2056 | ||
2057 | block_sectors = 1 << rdev->badblocks.shift; | |
2058 | sector = r10_bio->sector; | |
2059 | sectors = ((r10_bio->sector + block_sectors) | |
2060 | & ~(sector_t)(block_sectors - 1)) | |
2061 | - sector; | |
2062 | ||
2063 | while (sect_to_write) { | |
2064 | struct bio *wbio; | |
2065 | if (sectors > sect_to_write) | |
2066 | sectors = sect_to_write; | |
2067 | /* Write at 'sector' for 'sectors' */ | |
2068 | wbio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
2069 | md_trim_bio(wbio, sector - bio->bi_sector, sectors); | |
2070 | wbio->bi_sector = (r10_bio->devs[i].addr+ | |
2071 | rdev->data_offset+ | |
2072 | (sector - r10_bio->sector)); | |
2073 | wbio->bi_bdev = rdev->bdev; | |
2074 | if (submit_bio_wait(WRITE, wbio) == 0) | |
2075 | /* Failure! */ | |
2076 | ok = rdev_set_badblocks(rdev, sector, | |
2077 | sectors, 0) | |
2078 | && ok; | |
2079 | ||
2080 | bio_put(wbio); | |
2081 | sect_to_write -= sectors; | |
2082 | sector += sectors; | |
2083 | sectors = block_sectors; | |
2084 | } | |
2085 | return ok; | |
2086 | } | |
2087 | ||
9f2c9d12 | 2088 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2089 | { |
2090 | int slot = r10_bio->read_slot; | |
2091 | int mirror = r10_bio->devs[slot].devnum; | |
2092 | struct bio *bio; | |
e879a879 | 2093 | struct r10conf *conf = mddev->private; |
3cb03002 | 2094 | struct md_rdev *rdev; |
560f8e55 N |
2095 | char b[BDEVNAME_SIZE]; |
2096 | unsigned long do_sync; | |
856e08e2 | 2097 | int max_sectors; |
560f8e55 N |
2098 | |
2099 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2100 | * the block and we can fix it. | |
2101 | * We freeze all other IO, and try reading the block from | |
2102 | * other devices. When we find one, we re-write | |
2103 | * and check it that fixes the read error. | |
2104 | * This is all done synchronously while the array is | |
2105 | * frozen. | |
2106 | */ | |
2107 | if (mddev->ro == 0) { | |
2108 | freeze_array(conf); | |
2109 | fix_read_error(conf, mddev, r10_bio); | |
2110 | unfreeze_array(conf); | |
2111 | } | |
2112 | rdev_dec_pending(conf->mirrors[mirror].rdev, mddev); | |
2113 | ||
2114 | bio = r10_bio->devs[slot].bio; | |
7399c31b | 2115 | bdevname(bio->bi_bdev, b); |
560f8e55 N |
2116 | r10_bio->devs[slot].bio = |
2117 | mddev->ro ? IO_BLOCKED : NULL; | |
7399c31b | 2118 | read_more: |
96c3fd1f N |
2119 | rdev = read_balance(conf, r10_bio, &max_sectors); |
2120 | if (rdev == NULL) { | |
560f8e55 N |
2121 | printk(KERN_ALERT "md/raid10:%s: %s: unrecoverable I/O" |
2122 | " read error for block %llu\n", | |
7399c31b | 2123 | mdname(mddev), b, |
560f8e55 N |
2124 | (unsigned long long)r10_bio->sector); |
2125 | raid_end_bio_io(r10_bio); | |
2126 | bio_put(bio); | |
2127 | return; | |
2128 | } | |
2129 | ||
2130 | do_sync = (r10_bio->master_bio->bi_rw & REQ_SYNC); | |
7399c31b N |
2131 | if (bio) |
2132 | bio_put(bio); | |
560f8e55 | 2133 | slot = r10_bio->read_slot; |
560f8e55 N |
2134 | printk_ratelimited( |
2135 | KERN_ERR | |
2136 | "md/raid10:%s: %s: redirecting" | |
2137 | "sector %llu to another mirror\n", | |
2138 | mdname(mddev), | |
2139 | bdevname(rdev->bdev, b), | |
2140 | (unsigned long long)r10_bio->sector); | |
2141 | bio = bio_clone_mddev(r10_bio->master_bio, | |
2142 | GFP_NOIO, mddev); | |
7399c31b N |
2143 | md_trim_bio(bio, |
2144 | r10_bio->sector - bio->bi_sector, | |
2145 | max_sectors); | |
560f8e55 N |
2146 | r10_bio->devs[slot].bio = bio; |
2147 | bio->bi_sector = r10_bio->devs[slot].addr | |
2148 | + rdev->data_offset; | |
2149 | bio->bi_bdev = rdev->bdev; | |
2150 | bio->bi_rw = READ | do_sync; | |
2151 | bio->bi_private = r10_bio; | |
2152 | bio->bi_end_io = raid10_end_read_request; | |
7399c31b N |
2153 | if (max_sectors < r10_bio->sectors) { |
2154 | /* Drat - have to split this up more */ | |
2155 | struct bio *mbio = r10_bio->master_bio; | |
2156 | int sectors_handled = | |
2157 | r10_bio->sector + max_sectors | |
2158 | - mbio->bi_sector; | |
2159 | r10_bio->sectors = max_sectors; | |
2160 | spin_lock_irq(&conf->device_lock); | |
2161 | if (mbio->bi_phys_segments == 0) | |
2162 | mbio->bi_phys_segments = 2; | |
2163 | else | |
2164 | mbio->bi_phys_segments++; | |
2165 | spin_unlock_irq(&conf->device_lock); | |
2166 | generic_make_request(bio); | |
2167 | bio = NULL; | |
2168 | ||
2169 | r10_bio = mempool_alloc(conf->r10bio_pool, | |
2170 | GFP_NOIO); | |
2171 | r10_bio->master_bio = mbio; | |
2172 | r10_bio->sectors = (mbio->bi_size >> 9) | |
2173 | - sectors_handled; | |
2174 | r10_bio->state = 0; | |
2175 | set_bit(R10BIO_ReadError, | |
2176 | &r10_bio->state); | |
2177 | r10_bio->mddev = mddev; | |
2178 | r10_bio->sector = mbio->bi_sector | |
2179 | + sectors_handled; | |
2180 | ||
2181 | goto read_more; | |
2182 | } else | |
2183 | generic_make_request(bio); | |
560f8e55 N |
2184 | } |
2185 | ||
e879a879 | 2186 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2187 | { |
2188 | /* Some sort of write request has finished and it | |
2189 | * succeeded in writing where we thought there was a | |
2190 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2191 | * Or possibly if failed and we need to record |
2192 | * a bad block. | |
749c55e9 N |
2193 | */ |
2194 | int m; | |
3cb03002 | 2195 | struct md_rdev *rdev; |
749c55e9 N |
2196 | |
2197 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2198 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2199 | for (m = 0; m < conf->copies; m++) { |
2200 | int dev = r10_bio->devs[m].devnum; | |
2201 | rdev = conf->mirrors[dev].rdev; | |
2202 | if (r10_bio->devs[m].bio == NULL) | |
2203 | continue; | |
2204 | if (test_bit(BIO_UPTODATE, | |
749c55e9 | 2205 | &r10_bio->devs[m].bio->bi_flags)) { |
749c55e9 N |
2206 | rdev_clear_badblocks( |
2207 | rdev, | |
2208 | r10_bio->devs[m].addr, | |
2209 | r10_bio->sectors); | |
1a0b7cd8 N |
2210 | } else { |
2211 | if (!rdev_set_badblocks( | |
2212 | rdev, | |
2213 | r10_bio->devs[m].addr, | |
2214 | r10_bio->sectors, 0)) | |
2215 | md_error(conf->mddev, rdev); | |
749c55e9 | 2216 | } |
1a0b7cd8 | 2217 | } |
749c55e9 N |
2218 | put_buf(r10_bio); |
2219 | } else { | |
bd870a16 N |
2220 | for (m = 0; m < conf->copies; m++) { |
2221 | int dev = r10_bio->devs[m].devnum; | |
2222 | struct bio *bio = r10_bio->devs[m].bio; | |
2223 | rdev = conf->mirrors[dev].rdev; | |
2224 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2225 | rdev_clear_badblocks( |
2226 | rdev, | |
2227 | r10_bio->devs[m].addr, | |
2228 | r10_bio->sectors); | |
2229 | rdev_dec_pending(rdev, conf->mddev); | |
bd870a16 N |
2230 | } else if (bio != NULL && |
2231 | !test_bit(BIO_UPTODATE, &bio->bi_flags)) { | |
2232 | if (!narrow_write_error(r10_bio, m)) { | |
2233 | md_error(conf->mddev, rdev); | |
2234 | set_bit(R10BIO_Degraded, | |
2235 | &r10_bio->state); | |
2236 | } | |
2237 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2238 | } |
bd870a16 N |
2239 | } |
2240 | if (test_bit(R10BIO_WriteError, | |
2241 | &r10_bio->state)) | |
2242 | close_write(r10_bio); | |
749c55e9 N |
2243 | raid_end_bio_io(r10_bio); |
2244 | } | |
2245 | } | |
2246 | ||
fd01b88c | 2247 | static void raid10d(struct mddev *mddev) |
1da177e4 | 2248 | { |
9f2c9d12 | 2249 | struct r10bio *r10_bio; |
1da177e4 | 2250 | unsigned long flags; |
e879a879 | 2251 | struct r10conf *conf = mddev->private; |
1da177e4 | 2252 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2253 | struct blk_plug plug; |
1da177e4 LT |
2254 | |
2255 | md_check_recovery(mddev); | |
1da177e4 | 2256 | |
e1dfa0a2 | 2257 | blk_start_plug(&plug); |
1da177e4 | 2258 | for (;;) { |
6cce3b23 | 2259 | |
7eaceacc | 2260 | flush_pending_writes(conf); |
6cce3b23 | 2261 | |
a35e63ef N |
2262 | spin_lock_irqsave(&conf->device_lock, flags); |
2263 | if (list_empty(head)) { | |
2264 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2265 | break; |
a35e63ef | 2266 | } |
9f2c9d12 | 2267 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2268 | list_del(head->prev); |
4443ae10 | 2269 | conf->nr_queued--; |
1da177e4 LT |
2270 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2271 | ||
2272 | mddev = r10_bio->mddev; | |
070ec55d | 2273 | conf = mddev->private; |
bd870a16 N |
2274 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2275 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
2276 | handle_write_completed(conf, r10_bio); |
2277 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) | |
1da177e4 | 2278 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2279 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2280 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2281 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2282 | handle_read_error(mddev, r10_bio); |
856e08e2 N |
2283 | else { |
2284 | /* just a partial read to be scheduled from a | |
2285 | * separate context | |
2286 | */ | |
2287 | int slot = r10_bio->read_slot; | |
2288 | generic_make_request(r10_bio->devs[slot].bio); | |
2289 | } | |
560f8e55 | 2290 | |
1d9d5241 | 2291 | cond_resched(); |
de393cde N |
2292 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2293 | md_check_recovery(mddev); | |
1da177e4 | 2294 | } |
e1dfa0a2 | 2295 | blk_finish_plug(&plug); |
1da177e4 LT |
2296 | } |
2297 | ||
2298 | ||
e879a879 | 2299 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2300 | { |
2301 | int buffs; | |
69335ef3 | 2302 | int i; |
1da177e4 LT |
2303 | |
2304 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2305 | BUG_ON(conf->r10buf_pool); |
69335ef3 N |
2306 | conf->have_replacement = 0; |
2307 | for (i = 0; i < conf->raid_disks; i++) | |
2308 | if (conf->mirrors[i].replacement) | |
2309 | conf->have_replacement = 1; | |
1da177e4 LT |
2310 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2311 | if (!conf->r10buf_pool) | |
2312 | return -ENOMEM; | |
2313 | conf->next_resync = 0; | |
2314 | return 0; | |
2315 | } | |
2316 | ||
2317 | /* | |
2318 | * perform a "sync" on one "block" | |
2319 | * | |
2320 | * We need to make sure that no normal I/O request - particularly write | |
2321 | * requests - conflict with active sync requests. | |
2322 | * | |
2323 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2324 | * that can be installed to exclude normal IO requests. | |
2325 | * | |
2326 | * Resync and recovery are handled very differently. | |
2327 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2328 | * | |
2329 | * For resync, we iterate over virtual addresses, read all copies, | |
2330 | * and update if there are differences. If only one copy is live, | |
2331 | * skip it. | |
2332 | * For recovery, we iterate over physical addresses, read a good | |
2333 | * value for each non-in_sync drive, and over-write. | |
2334 | * | |
2335 | * So, for recovery we may have several outstanding complex requests for a | |
2336 | * given address, one for each out-of-sync device. We model this by allocating | |
2337 | * a number of r10_bio structures, one for each out-of-sync device. | |
2338 | * As we setup these structures, we collect all bio's together into a list | |
2339 | * which we then process collectively to add pages, and then process again | |
2340 | * to pass to generic_make_request. | |
2341 | * | |
2342 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2343 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2344 | * has its remaining count decremented to 0, the whole complex operation | |
2345 | * is complete. | |
2346 | * | |
2347 | */ | |
2348 | ||
fd01b88c | 2349 | static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, |
ab9d47e9 | 2350 | int *skipped, int go_faster) |
1da177e4 | 2351 | { |
e879a879 | 2352 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2353 | struct r10bio *r10_bio; |
1da177e4 LT |
2354 | struct bio *biolist = NULL, *bio; |
2355 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2356 | int i; |
6cce3b23 | 2357 | int max_sync; |
57dab0bd | 2358 | sector_t sync_blocks; |
1da177e4 LT |
2359 | sector_t sectors_skipped = 0; |
2360 | int chunks_skipped = 0; | |
2361 | ||
2362 | if (!conf->r10buf_pool) | |
2363 | if (init_resync(conf)) | |
57afd89f | 2364 | return 0; |
1da177e4 LT |
2365 | |
2366 | skipped: | |
58c0fed4 | 2367 | max_sector = mddev->dev_sectors; |
1da177e4 LT |
2368 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) |
2369 | max_sector = mddev->resync_max_sectors; | |
2370 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
2371 | /* If we aborted, we need to abort the |
2372 | * sync on the 'current' bitmap chucks (there can | |
2373 | * be several when recovering multiple devices). | |
2374 | * as we may have started syncing it but not finished. | |
2375 | * We can find the current address in | |
2376 | * mddev->curr_resync, but for recovery, | |
2377 | * we need to convert that to several | |
2378 | * virtual addresses. | |
2379 | */ | |
2380 | if (mddev->curr_resync < max_sector) { /* aborted */ | |
2381 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2382 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
2383 | &sync_blocks, 1); | |
2384 | else for (i=0; i<conf->raid_disks; i++) { | |
2385 | sector_t sect = | |
2386 | raid10_find_virt(conf, mddev->curr_resync, i); | |
2387 | bitmap_end_sync(mddev->bitmap, sect, | |
2388 | &sync_blocks, 1); | |
2389 | } | |
2390 | } else /* completed sync */ | |
2391 | conf->fullsync = 0; | |
2392 | ||
2393 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2394 | close_sync(conf); |
57afd89f | 2395 | *skipped = 1; |
1da177e4 LT |
2396 | return sectors_skipped; |
2397 | } | |
2398 | if (chunks_skipped >= conf->raid_disks) { | |
2399 | /* if there has been nothing to do on any drive, | |
2400 | * then there is nothing to do at all.. | |
2401 | */ | |
57afd89f N |
2402 | *skipped = 1; |
2403 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
2404 | } |
2405 | ||
c6207277 N |
2406 | if (max_sector > mddev->resync_max) |
2407 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
2408 | ||
1da177e4 LT |
2409 | /* make sure whole request will fit in a chunk - if chunks |
2410 | * are meaningful | |
2411 | */ | |
2412 | if (conf->near_copies < conf->raid_disks && | |
2413 | max_sector > (sector_nr | conf->chunk_mask)) | |
2414 | max_sector = (sector_nr | conf->chunk_mask) + 1; | |
2415 | /* | |
2416 | * If there is non-resync activity waiting for us then | |
2417 | * put in a delay to throttle resync. | |
2418 | */ | |
0a27ec96 | 2419 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 2420 | msleep_interruptible(1000); |
1da177e4 LT |
2421 | |
2422 | /* Again, very different code for resync and recovery. | |
2423 | * Both must result in an r10bio with a list of bios that | |
2424 | * have bi_end_io, bi_sector, bi_bdev set, | |
2425 | * and bi_private set to the r10bio. | |
2426 | * For recovery, we may actually create several r10bios | |
2427 | * with 2 bios in each, that correspond to the bios in the main one. | |
2428 | * In this case, the subordinate r10bios link back through a | |
2429 | * borrowed master_bio pointer, and the counter in the master | |
2430 | * includes a ref from each subordinate. | |
2431 | */ | |
2432 | /* First, we decide what to do and set ->bi_end_io | |
2433 | * To end_sync_read if we want to read, and | |
2434 | * end_sync_write if we will want to write. | |
2435 | */ | |
2436 | ||
6cce3b23 | 2437 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
2438 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
2439 | /* recovery... the complicated one */ | |
e875ecea | 2440 | int j; |
1da177e4 LT |
2441 | r10_bio = NULL; |
2442 | ||
ab9d47e9 N |
2443 | for (i=0 ; i<conf->raid_disks; i++) { |
2444 | int still_degraded; | |
9f2c9d12 | 2445 | struct r10bio *rb2; |
ab9d47e9 N |
2446 | sector_t sect; |
2447 | int must_sync; | |
e875ecea | 2448 | int any_working; |
1da177e4 | 2449 | |
ab9d47e9 N |
2450 | if (conf->mirrors[i].rdev == NULL || |
2451 | test_bit(In_sync, &conf->mirrors[i].rdev->flags)) | |
2452 | continue; | |
1da177e4 | 2453 | |
ab9d47e9 N |
2454 | still_degraded = 0; |
2455 | /* want to reconstruct this device */ | |
2456 | rb2 = r10_bio; | |
2457 | sect = raid10_find_virt(conf, sector_nr, i); | |
2458 | /* Unless we are doing a full sync, we only need | |
2459 | * to recover the block if it is set in the bitmap | |
2460 | */ | |
2461 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
2462 | &sync_blocks, 1); | |
2463 | if (sync_blocks < max_sync) | |
2464 | max_sync = sync_blocks; | |
2465 | if (!must_sync && | |
2466 | !conf->fullsync) { | |
2467 | /* yep, skip the sync_blocks here, but don't assume | |
2468 | * that there will never be anything to do here | |
2469 | */ | |
2470 | chunks_skipped = -1; | |
2471 | continue; | |
2472 | } | |
6cce3b23 | 2473 | |
ab9d47e9 N |
2474 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
2475 | raise_barrier(conf, rb2 != NULL); | |
2476 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 2477 | |
ab9d47e9 N |
2478 | r10_bio->master_bio = (struct bio*)rb2; |
2479 | if (rb2) | |
2480 | atomic_inc(&rb2->remaining); | |
2481 | r10_bio->mddev = mddev; | |
2482 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
2483 | r10_bio->sector = sect; | |
1da177e4 | 2484 | |
ab9d47e9 N |
2485 | raid10_find_phys(conf, r10_bio); |
2486 | ||
2487 | /* Need to check if the array will still be | |
2488 | * degraded | |
2489 | */ | |
2490 | for (j=0; j<conf->raid_disks; j++) | |
2491 | if (conf->mirrors[j].rdev == NULL || | |
2492 | test_bit(Faulty, &conf->mirrors[j].rdev->flags)) { | |
2493 | still_degraded = 1; | |
87fc767b | 2494 | break; |
1da177e4 | 2495 | } |
ab9d47e9 N |
2496 | |
2497 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
2498 | &sync_blocks, still_degraded); | |
2499 | ||
e875ecea | 2500 | any_working = 0; |
ab9d47e9 | 2501 | for (j=0; j<conf->copies;j++) { |
e875ecea | 2502 | int k; |
ab9d47e9 | 2503 | int d = r10_bio->devs[j].devnum; |
5e570289 | 2504 | sector_t from_addr, to_addr; |
3cb03002 | 2505 | struct md_rdev *rdev; |
40c356ce N |
2506 | sector_t sector, first_bad; |
2507 | int bad_sectors; | |
ab9d47e9 N |
2508 | if (!conf->mirrors[d].rdev || |
2509 | !test_bit(In_sync, &conf->mirrors[d].rdev->flags)) | |
2510 | continue; | |
2511 | /* This is where we read from */ | |
e875ecea | 2512 | any_working = 1; |
40c356ce N |
2513 | rdev = conf->mirrors[d].rdev; |
2514 | sector = r10_bio->devs[j].addr; | |
2515 | ||
2516 | if (is_badblock(rdev, sector, max_sync, | |
2517 | &first_bad, &bad_sectors)) { | |
2518 | if (first_bad > sector) | |
2519 | max_sync = first_bad - sector; | |
2520 | else { | |
2521 | bad_sectors -= (sector | |
2522 | - first_bad); | |
2523 | if (max_sync > bad_sectors) | |
2524 | max_sync = bad_sectors; | |
2525 | continue; | |
2526 | } | |
2527 | } | |
ab9d47e9 N |
2528 | bio = r10_bio->devs[0].bio; |
2529 | bio->bi_next = biolist; | |
2530 | biolist = bio; | |
2531 | bio->bi_private = r10_bio; | |
2532 | bio->bi_end_io = end_sync_read; | |
2533 | bio->bi_rw = READ; | |
5e570289 N |
2534 | from_addr = r10_bio->devs[j].addr; |
2535 | bio->bi_sector = from_addr + | |
ab9d47e9 N |
2536 | conf->mirrors[d].rdev->data_offset; |
2537 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
2538 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
2539 | atomic_inc(&r10_bio->remaining); | |
2540 | /* and we write to 'i' */ | |
2541 | ||
2542 | for (k=0; k<conf->copies; k++) | |
2543 | if (r10_bio->devs[k].devnum == i) | |
2544 | break; | |
2545 | BUG_ON(k == conf->copies); | |
2546 | bio = r10_bio->devs[1].bio; | |
2547 | bio->bi_next = biolist; | |
2548 | biolist = bio; | |
2549 | bio->bi_private = r10_bio; | |
2550 | bio->bi_end_io = end_sync_write; | |
2551 | bio->bi_rw = WRITE; | |
5e570289 N |
2552 | to_addr = r10_bio->devs[k].addr; |
2553 | bio->bi_sector = to_addr + | |
ab9d47e9 N |
2554 | conf->mirrors[i].rdev->data_offset; |
2555 | bio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
2556 | ||
2557 | r10_bio->devs[0].devnum = d; | |
5e570289 | 2558 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 2559 | r10_bio->devs[1].devnum = i; |
5e570289 | 2560 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 N |
2561 | |
2562 | break; | |
2563 | } | |
2564 | if (j == conf->copies) { | |
e875ecea N |
2565 | /* Cannot recover, so abort the recovery or |
2566 | * record a bad block */ | |
ab9d47e9 N |
2567 | put_buf(r10_bio); |
2568 | if (rb2) | |
2569 | atomic_dec(&rb2->remaining); | |
2570 | r10_bio = rb2; | |
e875ecea N |
2571 | if (any_working) { |
2572 | /* problem is that there are bad blocks | |
2573 | * on other device(s) | |
2574 | */ | |
2575 | int k; | |
2576 | for (k = 0; k < conf->copies; k++) | |
2577 | if (r10_bio->devs[k].devnum == i) | |
2578 | break; | |
2579 | if (!rdev_set_badblocks( | |
2580 | conf->mirrors[i].rdev, | |
2581 | r10_bio->devs[k].addr, | |
2582 | max_sync, 0)) | |
2583 | any_working = 0; | |
2584 | } | |
2585 | if (!any_working) { | |
2586 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
2587 | &mddev->recovery)) | |
2588 | printk(KERN_INFO "md/raid10:%s: insufficient " | |
2589 | "working devices for recovery.\n", | |
2590 | mdname(mddev)); | |
2591 | conf->mirrors[i].recovery_disabled | |
2592 | = mddev->recovery_disabled; | |
2593 | } | |
ab9d47e9 | 2594 | break; |
1da177e4 | 2595 | } |
ab9d47e9 | 2596 | } |
1da177e4 LT |
2597 | if (biolist == NULL) { |
2598 | while (r10_bio) { | |
9f2c9d12 N |
2599 | struct r10bio *rb2 = r10_bio; |
2600 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
2601 | rb2->master_bio = NULL; |
2602 | put_buf(rb2); | |
2603 | } | |
2604 | goto giveup; | |
2605 | } | |
2606 | } else { | |
2607 | /* resync. Schedule a read for every block at this virt offset */ | |
2608 | int count = 0; | |
6cce3b23 | 2609 | |
78200d45 N |
2610 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
2611 | ||
6cce3b23 N |
2612 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
2613 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
2614 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
2615 | &mddev->recovery)) { | |
6cce3b23 N |
2616 | /* We can skip this block */ |
2617 | *skipped = 1; | |
2618 | return sync_blocks + sectors_skipped; | |
2619 | } | |
2620 | if (sync_blocks < max_sync) | |
2621 | max_sync = sync_blocks; | |
1da177e4 LT |
2622 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
2623 | ||
1da177e4 LT |
2624 | r10_bio->mddev = mddev; |
2625 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
2626 | raise_barrier(conf, 0); |
2627 | conf->next_resync = sector_nr; | |
1da177e4 LT |
2628 | |
2629 | r10_bio->master_bio = NULL; | |
2630 | r10_bio->sector = sector_nr; | |
2631 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
2632 | raid10_find_phys(conf, r10_bio); | |
2633 | r10_bio->sectors = (sector_nr | conf->chunk_mask) - sector_nr +1; | |
2634 | ||
2635 | for (i=0; i<conf->copies; i++) { | |
2636 | int d = r10_bio->devs[i].devnum; | |
40c356ce N |
2637 | sector_t first_bad, sector; |
2638 | int bad_sectors; | |
2639 | ||
1da177e4 LT |
2640 | bio = r10_bio->devs[i].bio; |
2641 | bio->bi_end_io = NULL; | |
af03b8e4 | 2642 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
1da177e4 | 2643 | if (conf->mirrors[d].rdev == NULL || |
b2d444d7 | 2644 | test_bit(Faulty, &conf->mirrors[d].rdev->flags)) |
1da177e4 | 2645 | continue; |
40c356ce N |
2646 | sector = r10_bio->devs[i].addr; |
2647 | if (is_badblock(conf->mirrors[d].rdev, | |
2648 | sector, max_sync, | |
2649 | &first_bad, &bad_sectors)) { | |
2650 | if (first_bad > sector) | |
2651 | max_sync = first_bad - sector; | |
2652 | else { | |
2653 | bad_sectors -= (sector - first_bad); | |
2654 | if (max_sync > bad_sectors) | |
2655 | max_sync = max_sync; | |
2656 | continue; | |
2657 | } | |
2658 | } | |
1da177e4 LT |
2659 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2660 | atomic_inc(&r10_bio->remaining); | |
2661 | bio->bi_next = biolist; | |
2662 | biolist = bio; | |
2663 | bio->bi_private = r10_bio; | |
2664 | bio->bi_end_io = end_sync_read; | |
802ba064 | 2665 | bio->bi_rw = READ; |
40c356ce | 2666 | bio->bi_sector = sector + |
1da177e4 LT |
2667 | conf->mirrors[d].rdev->data_offset; |
2668 | bio->bi_bdev = conf->mirrors[d].rdev->bdev; | |
2669 | count++; | |
2670 | } | |
2671 | ||
2672 | if (count < 2) { | |
2673 | for (i=0; i<conf->copies; i++) { | |
2674 | int d = r10_bio->devs[i].devnum; | |
2675 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
2676 | rdev_dec_pending(conf->mirrors[d].rdev, |
2677 | mddev); | |
1da177e4 LT |
2678 | } |
2679 | put_buf(r10_bio); | |
2680 | biolist = NULL; | |
2681 | goto giveup; | |
2682 | } | |
2683 | } | |
2684 | ||
2685 | for (bio = biolist; bio ; bio=bio->bi_next) { | |
2686 | ||
2687 | bio->bi_flags &= ~(BIO_POOL_MASK - 1); | |
2688 | if (bio->bi_end_io) | |
2689 | bio->bi_flags |= 1 << BIO_UPTODATE; | |
2690 | bio->bi_vcnt = 0; | |
2691 | bio->bi_idx = 0; | |
2692 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
2693 | bio->bi_size = 0; |
2694 | } | |
2695 | ||
2696 | nr_sectors = 0; | |
6cce3b23 N |
2697 | if (sector_nr + max_sync < max_sector) |
2698 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
2699 | do { |
2700 | struct page *page; | |
2701 | int len = PAGE_SIZE; | |
1da177e4 LT |
2702 | if (sector_nr + (len>>9) > max_sector) |
2703 | len = (max_sector - sector_nr) << 9; | |
2704 | if (len == 0) | |
2705 | break; | |
2706 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
ab9d47e9 | 2707 | struct bio *bio2; |
1da177e4 | 2708 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
ab9d47e9 N |
2709 | if (bio_add_page(bio, page, len, 0)) |
2710 | continue; | |
2711 | ||
2712 | /* stop here */ | |
2713 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
2714 | for (bio2 = biolist; | |
2715 | bio2 && bio2 != bio; | |
2716 | bio2 = bio2->bi_next) { | |
2717 | /* remove last page from this bio */ | |
2718 | bio2->bi_vcnt--; | |
2719 | bio2->bi_size -= len; | |
2720 | bio2->bi_flags &= ~(1<< BIO_SEG_VALID); | |
1da177e4 | 2721 | } |
ab9d47e9 | 2722 | goto bio_full; |
1da177e4 LT |
2723 | } |
2724 | nr_sectors += len>>9; | |
2725 | sector_nr += len>>9; | |
2726 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
2727 | bio_full: | |
2728 | r10_bio->sectors = nr_sectors; | |
2729 | ||
2730 | while (biolist) { | |
2731 | bio = biolist; | |
2732 | biolist = biolist->bi_next; | |
2733 | ||
2734 | bio->bi_next = NULL; | |
2735 | r10_bio = bio->bi_private; | |
2736 | r10_bio->sectors = nr_sectors; | |
2737 | ||
2738 | if (bio->bi_end_io == end_sync_read) { | |
2739 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
2740 | generic_make_request(bio); | |
2741 | } | |
2742 | } | |
2743 | ||
57afd89f N |
2744 | if (sectors_skipped) |
2745 | /* pretend they weren't skipped, it makes | |
2746 | * no important difference in this case | |
2747 | */ | |
2748 | md_done_sync(mddev, sectors_skipped, 1); | |
2749 | ||
1da177e4 LT |
2750 | return sectors_skipped + nr_sectors; |
2751 | giveup: | |
2752 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
2753 | * drives must be failed or in resync, all drives |
2754 | * have a bad block, so try the next chunk... | |
1da177e4 | 2755 | */ |
09b4068a N |
2756 | if (sector_nr + max_sync < max_sector) |
2757 | max_sector = sector_nr + max_sync; | |
2758 | ||
2759 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
2760 | chunks_skipped ++; |
2761 | sector_nr = max_sector; | |
1da177e4 | 2762 | goto skipped; |
1da177e4 LT |
2763 | } |
2764 | ||
80c3a6ce | 2765 | static sector_t |
fd01b88c | 2766 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2767 | { |
2768 | sector_t size; | |
e879a879 | 2769 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
2770 | |
2771 | if (!raid_disks) | |
84707f38 | 2772 | raid_disks = conf->raid_disks; |
80c3a6ce | 2773 | if (!sectors) |
dab8b292 | 2774 | sectors = conf->dev_sectors; |
80c3a6ce DW |
2775 | |
2776 | size = sectors >> conf->chunk_shift; | |
2777 | sector_div(size, conf->far_copies); | |
2778 | size = size * raid_disks; | |
2779 | sector_div(size, conf->near_copies); | |
2780 | ||
2781 | return size << conf->chunk_shift; | |
2782 | } | |
2783 | ||
dab8b292 | 2784 | |
e879a879 | 2785 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2786 | { |
e879a879 | 2787 | struct r10conf *conf = NULL; |
c93983bf | 2788 | int nc, fc, fo; |
1da177e4 | 2789 | sector_t stride, size; |
dab8b292 | 2790 | int err = -EINVAL; |
1da177e4 | 2791 | |
f73ea873 MT |
2792 | if (mddev->new_chunk_sectors < (PAGE_SIZE >> 9) || |
2793 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
128595ed N |
2794 | printk(KERN_ERR "md/raid10:%s: chunk size must be " |
2795 | "at least PAGE_SIZE(%ld) and be a power of 2.\n", | |
2796 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 2797 | goto out; |
1da177e4 | 2798 | } |
2604b703 | 2799 | |
f73ea873 MT |
2800 | nc = mddev->new_layout & 255; |
2801 | fc = (mddev->new_layout >> 8) & 255; | |
2802 | fo = mddev->new_layout & (1<<16); | |
dab8b292 | 2803 | |
1da177e4 | 2804 | if ((nc*fc) <2 || (nc*fc) > mddev->raid_disks || |
f73ea873 | 2805 | (mddev->new_layout >> 17)) { |
128595ed | 2806 | printk(KERN_ERR "md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
f73ea873 | 2807 | mdname(mddev), mddev->new_layout); |
1da177e4 LT |
2808 | goto out; |
2809 | } | |
dab8b292 TM |
2810 | |
2811 | err = -ENOMEM; | |
e879a879 | 2812 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 2813 | if (!conf) |
1da177e4 | 2814 | goto out; |
dab8b292 | 2815 | |
4443ae10 | 2816 | conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks, |
dab8b292 TM |
2817 | GFP_KERNEL); |
2818 | if (!conf->mirrors) | |
2819 | goto out; | |
4443ae10 N |
2820 | |
2821 | conf->tmppage = alloc_page(GFP_KERNEL); | |
2822 | if (!conf->tmppage) | |
dab8b292 TM |
2823 | goto out; |
2824 | ||
1da177e4 | 2825 | |
64a742bc | 2826 | conf->raid_disks = mddev->raid_disks; |
1da177e4 LT |
2827 | conf->near_copies = nc; |
2828 | conf->far_copies = fc; | |
2829 | conf->copies = nc*fc; | |
c93983bf | 2830 | conf->far_offset = fo; |
dab8b292 TM |
2831 | conf->chunk_mask = mddev->new_chunk_sectors - 1; |
2832 | conf->chunk_shift = ffz(~mddev->new_chunk_sectors); | |
2833 | ||
2834 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, | |
2835 | r10bio_pool_free, conf); | |
2836 | if (!conf->r10bio_pool) | |
2837 | goto out; | |
2838 | ||
58c0fed4 | 2839 | size = mddev->dev_sectors >> conf->chunk_shift; |
64a742bc N |
2840 | sector_div(size, fc); |
2841 | size = size * conf->raid_disks; | |
2842 | sector_div(size, nc); | |
2843 | /* 'size' is now the number of chunks in the array */ | |
2844 | /* calculate "used chunks per device" in 'stride' */ | |
2845 | stride = size * conf->copies; | |
af03b8e4 N |
2846 | |
2847 | /* We need to round up when dividing by raid_disks to | |
2848 | * get the stride size. | |
2849 | */ | |
2850 | stride += conf->raid_disks - 1; | |
64a742bc | 2851 | sector_div(stride, conf->raid_disks); |
dab8b292 TM |
2852 | |
2853 | conf->dev_sectors = stride << conf->chunk_shift; | |
64a742bc | 2854 | |
c93983bf | 2855 | if (fo) |
64a742bc N |
2856 | stride = 1; |
2857 | else | |
c93983bf | 2858 | sector_div(stride, fc); |
64a742bc N |
2859 | conf->stride = stride << conf->chunk_shift; |
2860 | ||
1da177e4 | 2861 | |
e7e72bf6 | 2862 | spin_lock_init(&conf->device_lock); |
dab8b292 TM |
2863 | INIT_LIST_HEAD(&conf->retry_list); |
2864 | ||
2865 | spin_lock_init(&conf->resync_lock); | |
2866 | init_waitqueue_head(&conf->wait_barrier); | |
2867 | ||
2868 | conf->thread = md_register_thread(raid10d, mddev, NULL); | |
2869 | if (!conf->thread) | |
2870 | goto out; | |
2871 | ||
dab8b292 TM |
2872 | conf->mddev = mddev; |
2873 | return conf; | |
2874 | ||
2875 | out: | |
128595ed | 2876 | printk(KERN_ERR "md/raid10:%s: couldn't allocate memory.\n", |
dab8b292 TM |
2877 | mdname(mddev)); |
2878 | if (conf) { | |
2879 | if (conf->r10bio_pool) | |
2880 | mempool_destroy(conf->r10bio_pool); | |
2881 | kfree(conf->mirrors); | |
2882 | safe_put_page(conf->tmppage); | |
2883 | kfree(conf); | |
2884 | } | |
2885 | return ERR_PTR(err); | |
2886 | } | |
2887 | ||
fd01b88c | 2888 | static int run(struct mddev *mddev) |
dab8b292 | 2889 | { |
e879a879 | 2890 | struct r10conf *conf; |
dab8b292 | 2891 | int i, disk_idx, chunk_size; |
0f6d02d5 | 2892 | struct mirror_info *disk; |
3cb03002 | 2893 | struct md_rdev *rdev; |
dab8b292 TM |
2894 | sector_t size; |
2895 | ||
2896 | /* | |
2897 | * copy the already verified devices into our private RAID10 | |
2898 | * bookkeeping area. [whatever we allocate in run(), | |
2899 | * should be freed in stop()] | |
2900 | */ | |
2901 | ||
2902 | if (mddev->private == NULL) { | |
2903 | conf = setup_conf(mddev); | |
2904 | if (IS_ERR(conf)) | |
2905 | return PTR_ERR(conf); | |
2906 | mddev->private = conf; | |
2907 | } | |
2908 | conf = mddev->private; | |
2909 | if (!conf) | |
2910 | goto out; | |
2911 | ||
dab8b292 TM |
2912 | mddev->thread = conf->thread; |
2913 | conf->thread = NULL; | |
2914 | ||
8f6c2e4b MP |
2915 | chunk_size = mddev->chunk_sectors << 9; |
2916 | blk_queue_io_min(mddev->queue, chunk_size); | |
2917 | if (conf->raid_disks % conf->near_copies) | |
2918 | blk_queue_io_opt(mddev->queue, chunk_size * conf->raid_disks); | |
2919 | else | |
2920 | blk_queue_io_opt(mddev->queue, chunk_size * | |
2921 | (conf->raid_disks / conf->near_copies)); | |
2922 | ||
159ec1fc | 2923 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
34b343cf | 2924 | |
1da177e4 | 2925 | disk_idx = rdev->raid_disk; |
84707f38 | 2926 | if (disk_idx >= conf->raid_disks |
1da177e4 LT |
2927 | || disk_idx < 0) |
2928 | continue; | |
2929 | disk = conf->mirrors + disk_idx; | |
2930 | ||
2931 | disk->rdev = rdev; | |
8f6c2e4b MP |
2932 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2933 | rdev->data_offset << 9); | |
1da177e4 | 2934 | /* as we don't honour merge_bvec_fn, we must never risk |
627a2d3c N |
2935 | * violating it, so limit max_segments to 1 lying |
2936 | * within a single page. | |
1da177e4 | 2937 | */ |
627a2d3c N |
2938 | if (rdev->bdev->bd_disk->queue->merge_bvec_fn) { |
2939 | blk_queue_max_segments(mddev->queue, 1); | |
2940 | blk_queue_segment_boundary(mddev->queue, | |
2941 | PAGE_CACHE_SIZE - 1); | |
2942 | } | |
1da177e4 LT |
2943 | |
2944 | disk->head_position = 0; | |
1da177e4 | 2945 | } |
6d508242 | 2946 | /* need to check that every block has at least one working mirror */ |
700c7213 | 2947 | if (!enough(conf, -1)) { |
128595ed | 2948 | printk(KERN_ERR "md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 2949 | mdname(mddev)); |
1da177e4 LT |
2950 | goto out_free_conf; |
2951 | } | |
2952 | ||
2953 | mddev->degraded = 0; | |
2954 | for (i = 0; i < conf->raid_disks; i++) { | |
2955 | ||
2956 | disk = conf->mirrors + i; | |
2957 | ||
5fd6c1dc | 2958 | if (!disk->rdev || |
2e333e89 | 2959 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
2960 | disk->head_position = 0; |
2961 | mddev->degraded++; | |
8c2e870a NB |
2962 | if (disk->rdev) |
2963 | conf->fullsync = 1; | |
1da177e4 | 2964 | } |
d890fa2b | 2965 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
2966 | } |
2967 | ||
8c6ac868 | 2968 | if (mddev->recovery_cp != MaxSector) |
128595ed | 2969 | printk(KERN_NOTICE "md/raid10:%s: not clean" |
8c6ac868 AN |
2970 | " -- starting background reconstruction\n", |
2971 | mdname(mddev)); | |
1da177e4 | 2972 | printk(KERN_INFO |
128595ed | 2973 | "md/raid10:%s: active with %d out of %d devices\n", |
84707f38 N |
2974 | mdname(mddev), conf->raid_disks - mddev->degraded, |
2975 | conf->raid_disks); | |
1da177e4 LT |
2976 | /* |
2977 | * Ok, everything is just fine now | |
2978 | */ | |
dab8b292 TM |
2979 | mddev->dev_sectors = conf->dev_sectors; |
2980 | size = raid10_size(mddev, 0, 0); | |
2981 | md_set_array_sectors(mddev, size); | |
2982 | mddev->resync_max_sectors = size; | |
1da177e4 | 2983 | |
0d129228 N |
2984 | mddev->queue->backing_dev_info.congested_fn = raid10_congested; |
2985 | mddev->queue->backing_dev_info.congested_data = mddev; | |
7a5febe9 | 2986 | |
1da177e4 LT |
2987 | /* Calculate max read-ahead size. |
2988 | * We need to readahead at least twice a whole stripe.... | |
2989 | * maybe... | |
2990 | */ | |
2991 | { | |
9d8f0363 AN |
2992 | int stripe = conf->raid_disks * |
2993 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
1da177e4 LT |
2994 | stripe /= conf->near_copies; |
2995 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) | |
2996 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; | |
2997 | } | |
2998 | ||
84707f38 | 2999 | if (conf->near_copies < conf->raid_disks) |
1da177e4 | 3000 | blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec); |
a91a2785 MP |
3001 | |
3002 | if (md_integrity_register(mddev)) | |
3003 | goto out_free_conf; | |
3004 | ||
1da177e4 LT |
3005 | return 0; |
3006 | ||
3007 | out_free_conf: | |
01f96c0a | 3008 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
3009 | if (conf->r10bio_pool) |
3010 | mempool_destroy(conf->r10bio_pool); | |
1345b1d8 | 3011 | safe_put_page(conf->tmppage); |
990a8baf | 3012 | kfree(conf->mirrors); |
1da177e4 LT |
3013 | kfree(conf); |
3014 | mddev->private = NULL; | |
3015 | out: | |
3016 | return -EIO; | |
3017 | } | |
3018 | ||
fd01b88c | 3019 | static int stop(struct mddev *mddev) |
1da177e4 | 3020 | { |
e879a879 | 3021 | struct r10conf *conf = mddev->private; |
1da177e4 | 3022 | |
409c57f3 N |
3023 | raise_barrier(conf, 0); |
3024 | lower_barrier(conf); | |
3025 | ||
01f96c0a | 3026 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
3027 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
3028 | if (conf->r10bio_pool) | |
3029 | mempool_destroy(conf->r10bio_pool); | |
990a8baf | 3030 | kfree(conf->mirrors); |
1da177e4 LT |
3031 | kfree(conf); |
3032 | mddev->private = NULL; | |
3033 | return 0; | |
3034 | } | |
3035 | ||
fd01b88c | 3036 | static void raid10_quiesce(struct mddev *mddev, int state) |
6cce3b23 | 3037 | { |
e879a879 | 3038 | struct r10conf *conf = mddev->private; |
6cce3b23 N |
3039 | |
3040 | switch(state) { | |
3041 | case 1: | |
3042 | raise_barrier(conf, 0); | |
3043 | break; | |
3044 | case 0: | |
3045 | lower_barrier(conf); | |
3046 | break; | |
3047 | } | |
6cce3b23 | 3048 | } |
1da177e4 | 3049 | |
fd01b88c | 3050 | static void *raid10_takeover_raid0(struct mddev *mddev) |
dab8b292 | 3051 | { |
3cb03002 | 3052 | struct md_rdev *rdev; |
e879a879 | 3053 | struct r10conf *conf; |
dab8b292 TM |
3054 | |
3055 | if (mddev->degraded > 0) { | |
128595ed N |
3056 | printk(KERN_ERR "md/raid10:%s: Error: degraded raid0!\n", |
3057 | mdname(mddev)); | |
dab8b292 TM |
3058 | return ERR_PTR(-EINVAL); |
3059 | } | |
3060 | ||
dab8b292 TM |
3061 | /* Set new parameters */ |
3062 | mddev->new_level = 10; | |
3063 | /* new layout: far_copies = 1, near_copies = 2 */ | |
3064 | mddev->new_layout = (1<<8) + 2; | |
3065 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
3066 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
3067 | mddev->raid_disks *= 2; |
3068 | /* make sure it will be not marked as dirty */ | |
3069 | mddev->recovery_cp = MaxSector; | |
3070 | ||
3071 | conf = setup_conf(mddev); | |
02214dc5 | 3072 | if (!IS_ERR(conf)) { |
e93f68a1 N |
3073 | list_for_each_entry(rdev, &mddev->disks, same_set) |
3074 | if (rdev->raid_disk >= 0) | |
3075 | rdev->new_raid_disk = rdev->raid_disk * 2; | |
02214dc5 KW |
3076 | conf->barrier = 1; |
3077 | } | |
3078 | ||
dab8b292 TM |
3079 | return conf; |
3080 | } | |
3081 | ||
fd01b88c | 3082 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 3083 | { |
e373ab10 | 3084 | struct r0conf *raid0_conf; |
dab8b292 TM |
3085 | |
3086 | /* raid10 can take over: | |
3087 | * raid0 - providing it has only two drives | |
3088 | */ | |
3089 | if (mddev->level == 0) { | |
3090 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
3091 | raid0_conf = mddev->private; |
3092 | if (raid0_conf->nr_strip_zones > 1) { | |
128595ed N |
3093 | printk(KERN_ERR "md/raid10:%s: cannot takeover raid 0" |
3094 | " with more than one zone.\n", | |
3095 | mdname(mddev)); | |
dab8b292 TM |
3096 | return ERR_PTR(-EINVAL); |
3097 | } | |
3098 | return raid10_takeover_raid0(mddev); | |
3099 | } | |
3100 | return ERR_PTR(-EINVAL); | |
3101 | } | |
3102 | ||
84fc4b56 | 3103 | static struct md_personality raid10_personality = |
1da177e4 LT |
3104 | { |
3105 | .name = "raid10", | |
2604b703 | 3106 | .level = 10, |
1da177e4 LT |
3107 | .owner = THIS_MODULE, |
3108 | .make_request = make_request, | |
3109 | .run = run, | |
3110 | .stop = stop, | |
3111 | .status = status, | |
3112 | .error_handler = error, | |
3113 | .hot_add_disk = raid10_add_disk, | |
3114 | .hot_remove_disk= raid10_remove_disk, | |
3115 | .spare_active = raid10_spare_active, | |
3116 | .sync_request = sync_request, | |
6cce3b23 | 3117 | .quiesce = raid10_quiesce, |
80c3a6ce | 3118 | .size = raid10_size, |
dab8b292 | 3119 | .takeover = raid10_takeover, |
1da177e4 LT |
3120 | }; |
3121 | ||
3122 | static int __init raid_init(void) | |
3123 | { | |
2604b703 | 3124 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
3125 | } |
3126 | ||
3127 | static void raid_exit(void) | |
3128 | { | |
2604b703 | 3129 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
3130 | } |
3131 | ||
3132 | module_init(raid_init); | |
3133 | module_exit(raid_exit); | |
3134 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3135 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 3136 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 3137 | MODULE_ALIAS("md-raid10"); |
2604b703 | 3138 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
3139 | |
3140 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |