]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid1.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat | |
5 | * | |
6 | * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
7 | * | |
8 | * RAID-1 management functions. | |
9 | * | |
10 | * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 | |
11 | * | |
96de0e25 | 12 | * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> |
1da177e4 LT |
13 | * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> |
14 | * | |
191ea9b2 N |
15 | * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support |
16 | * bitmapped intelligence in resync: | |
17 | * | |
18 | * - bitmap marked during normal i/o | |
19 | * - bitmap used to skip nondirty blocks during sync | |
20 | * | |
21 | * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: | |
22 | * - persistent bitmap code | |
23 | * | |
1da177e4 LT |
24 | * This program is free software; you can redistribute it and/or modify |
25 | * it under the terms of the GNU General Public License as published by | |
26 | * the Free Software Foundation; either version 2, or (at your option) | |
27 | * any later version. | |
28 | * | |
29 | * You should have received a copy of the GNU General Public License | |
30 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
31 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
32 | */ | |
33 | ||
5a0e3ad6 | 34 | #include <linux/slab.h> |
25570727 | 35 | #include <linux/delay.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
056075c7 | 37 | #include <linux/module.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
8bda470e | 39 | #include <linux/ratelimit.h> |
43b2e5d8 | 40 | #include "md.h" |
ef740c37 CH |
41 | #include "raid1.h" |
42 | #include "bitmap.h" | |
191ea9b2 | 43 | |
1da177e4 LT |
44 | /* |
45 | * Number of guaranteed r1bios in case of extreme VM load: | |
46 | */ | |
47 | #define NR_RAID1_BIOS 256 | |
48 | ||
34db0cd6 N |
49 | /* When there are this many requests queue to be written by |
50 | * the raid1 thread, we become 'congested' to provide back-pressure | |
51 | * for writeback. | |
52 | */ | |
53 | static int max_queued_requests = 1024; | |
1da177e4 | 54 | |
e8096360 N |
55 | static void allow_barrier(struct r1conf *conf); |
56 | static void lower_barrier(struct r1conf *conf); | |
1da177e4 | 57 | |
dd0fc66f | 58 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
59 | { |
60 | struct pool_info *pi = data; | |
9f2c9d12 | 61 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
62 | |
63 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 64 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
65 | } |
66 | ||
67 | static void r1bio_pool_free(void *r1_bio, void *data) | |
68 | { | |
69 | kfree(r1_bio); | |
70 | } | |
71 | ||
72 | #define RESYNC_BLOCK_SIZE (64*1024) | |
73 | //#define RESYNC_BLOCK_SIZE PAGE_SIZE | |
74 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) | |
75 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
76 | #define RESYNC_WINDOW (2048*1024) | |
77 | ||
dd0fc66f | 78 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
79 | { |
80 | struct pool_info *pi = data; | |
81 | struct page *page; | |
9f2c9d12 | 82 | struct r1bio *r1_bio; |
1da177e4 LT |
83 | struct bio *bio; |
84 | int i, j; | |
85 | ||
86 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 87 | if (!r1_bio) |
1da177e4 | 88 | return NULL; |
1da177e4 LT |
89 | |
90 | /* | |
91 | * Allocate bios : 1 for reading, n-1 for writing | |
92 | */ | |
93 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 94 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
95 | if (!bio) |
96 | goto out_free_bio; | |
97 | r1_bio->bios[j] = bio; | |
98 | } | |
99 | /* | |
100 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
101 | * the first bio. |
102 | * If this is a user-requested check/repair, allocate | |
103 | * RESYNC_PAGES for each bio. | |
1da177e4 | 104 | */ |
d11c171e N |
105 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
106 | j = pi->raid_disks; | |
107 | else | |
108 | j = 1; | |
109 | while(j--) { | |
110 | bio = r1_bio->bios[j]; | |
111 | for (i = 0; i < RESYNC_PAGES; i++) { | |
112 | page = alloc_page(gfp_flags); | |
113 | if (unlikely(!page)) | |
114 | goto out_free_pages; | |
115 | ||
116 | bio->bi_io_vec[i].bv_page = page; | |
303a0e11 | 117 | bio->bi_vcnt = i+1; |
d11c171e N |
118 | } |
119 | } | |
120 | /* If not user-requests, copy the page pointers to all bios */ | |
121 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
122 | for (i=0; i<RESYNC_PAGES ; i++) | |
123 | for (j=1; j<pi->raid_disks; j++) | |
124 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
125 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
126 | } |
127 | ||
128 | r1_bio->master_bio = NULL; | |
129 | ||
130 | return r1_bio; | |
131 | ||
132 | out_free_pages: | |
303a0e11 N |
133 | for (j=0 ; j < pi->raid_disks; j++) |
134 | for (i=0; i < r1_bio->bios[j]->bi_vcnt ; i++) | |
135 | put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page); | |
d11c171e | 136 | j = -1; |
1da177e4 | 137 | out_free_bio: |
8f19ccb2 | 138 | while (++j < pi->raid_disks) |
1da177e4 LT |
139 | bio_put(r1_bio->bios[j]); |
140 | r1bio_pool_free(r1_bio, data); | |
141 | return NULL; | |
142 | } | |
143 | ||
144 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
145 | { | |
146 | struct pool_info *pi = data; | |
d11c171e | 147 | int i,j; |
9f2c9d12 | 148 | struct r1bio *r1bio = __r1_bio; |
1da177e4 | 149 | |
d11c171e N |
150 | for (i = 0; i < RESYNC_PAGES; i++) |
151 | for (j = pi->raid_disks; j-- ;) { | |
152 | if (j == 0 || | |
153 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
154 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 155 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 156 | } |
1da177e4 LT |
157 | for (i=0 ; i < pi->raid_disks; i++) |
158 | bio_put(r1bio->bios[i]); | |
159 | ||
160 | r1bio_pool_free(r1bio, data); | |
161 | } | |
162 | ||
e8096360 | 163 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
164 | { |
165 | int i; | |
166 | ||
8f19ccb2 | 167 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 168 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 169 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
170 | bio_put(*bio); |
171 | *bio = NULL; | |
172 | } | |
173 | } | |
174 | ||
9f2c9d12 | 175 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 176 | { |
e8096360 | 177 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 178 | |
1da177e4 LT |
179 | put_all_bios(conf, r1_bio); |
180 | mempool_free(r1_bio, conf->r1bio_pool); | |
181 | } | |
182 | ||
9f2c9d12 | 183 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 184 | { |
e8096360 | 185 | struct r1conf *conf = r1_bio->mddev->private; |
3e198f78 N |
186 | int i; |
187 | ||
8f19ccb2 | 188 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
189 | struct bio *bio = r1_bio->bios[i]; |
190 | if (bio->bi_end_io) | |
191 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
192 | } | |
1da177e4 LT |
193 | |
194 | mempool_free(r1_bio, conf->r1buf_pool); | |
195 | ||
17999be4 | 196 | lower_barrier(conf); |
1da177e4 LT |
197 | } |
198 | ||
9f2c9d12 | 199 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
200 | { |
201 | unsigned long flags; | |
fd01b88c | 202 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 203 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
204 | |
205 | spin_lock_irqsave(&conf->device_lock, flags); | |
206 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 207 | conf->nr_queued ++; |
1da177e4 LT |
208 | spin_unlock_irqrestore(&conf->device_lock, flags); |
209 | ||
17999be4 | 210 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
211 | md_wakeup_thread(mddev->thread); |
212 | } | |
213 | ||
214 | /* | |
215 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
216 | * operation and are ready to return a success/failure code to the buffer | |
217 | * cache layer. | |
218 | */ | |
9f2c9d12 | 219 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
220 | { |
221 | struct bio *bio = r1_bio->master_bio; | |
222 | int done; | |
e8096360 | 223 | struct r1conf *conf = r1_bio->mddev->private; |
d2eb35ac N |
224 | |
225 | if (bio->bi_phys_segments) { | |
226 | unsigned long flags; | |
227 | spin_lock_irqsave(&conf->device_lock, flags); | |
228 | bio->bi_phys_segments--; | |
229 | done = (bio->bi_phys_segments == 0); | |
230 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
231 | } else | |
232 | done = 1; | |
233 | ||
234 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
235 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
236 | if (done) { | |
237 | bio_endio(bio, 0); | |
238 | /* | |
239 | * Wake up any possible resync thread that waits for the device | |
240 | * to go idle. | |
241 | */ | |
242 | allow_barrier(conf); | |
243 | } | |
244 | } | |
245 | ||
9f2c9d12 | 246 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
247 | { |
248 | struct bio *bio = r1_bio->master_bio; | |
249 | ||
4b6d287f N |
250 | /* if nobody has done the final endio yet, do it now */ |
251 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
252 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
253 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
254 | (unsigned long long) bio->bi_sector, | |
255 | (unsigned long long) bio->bi_sector + | |
256 | (bio->bi_size >> 9) - 1); | |
4b6d287f | 257 | |
d2eb35ac | 258 | call_bio_endio(r1_bio); |
4b6d287f | 259 | } |
1da177e4 LT |
260 | free_r1bio(r1_bio); |
261 | } | |
262 | ||
263 | /* | |
264 | * Update disk head position estimator based on IRQ completion info. | |
265 | */ | |
9f2c9d12 | 266 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 267 | { |
e8096360 | 268 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
269 | |
270 | conf->mirrors[disk].head_position = | |
271 | r1_bio->sector + (r1_bio->sectors); | |
272 | } | |
273 | ||
ba3ae3be NK |
274 | /* |
275 | * Find the disk number which triggered given bio | |
276 | */ | |
9f2c9d12 | 277 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
278 | { |
279 | int mirror; | |
30194636 N |
280 | struct r1conf *conf = r1_bio->mddev->private; |
281 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 282 | |
8f19ccb2 | 283 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
284 | if (r1_bio->bios[mirror] == bio) |
285 | break; | |
286 | ||
8f19ccb2 | 287 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
288 | update_head_pos(mirror, r1_bio); |
289 | ||
290 | return mirror; | |
291 | } | |
292 | ||
6712ecf8 | 293 | static void raid1_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
294 | { |
295 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 296 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 297 | int mirror; |
e8096360 | 298 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 299 | |
1da177e4 LT |
300 | mirror = r1_bio->read_disk; |
301 | /* | |
302 | * this branch is our 'one mirror IO has finished' event handler: | |
303 | */ | |
ddaf22ab N |
304 | update_head_pos(mirror, r1_bio); |
305 | ||
dd00a99e N |
306 | if (uptodate) |
307 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
308 | else { | |
309 | /* If all other devices have failed, we want to return | |
310 | * the error upwards rather than fail the last device. | |
311 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 312 | */ |
dd00a99e N |
313 | unsigned long flags; |
314 | spin_lock_irqsave(&conf->device_lock, flags); | |
315 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
316 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
317 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) | |
318 | uptodate = 1; | |
319 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
320 | } | |
1da177e4 | 321 | |
dd00a99e | 322 | if (uptodate) |
1da177e4 | 323 | raid_end_bio_io(r1_bio); |
dd00a99e | 324 | else { |
1da177e4 LT |
325 | /* |
326 | * oops, read error: | |
327 | */ | |
328 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
329 | printk_ratelimited( |
330 | KERN_ERR "md/raid1:%s: %s: " | |
331 | "rescheduling sector %llu\n", | |
332 | mdname(conf->mddev), | |
333 | bdevname(conf->mirrors[mirror].rdev->bdev, | |
334 | b), | |
335 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 336 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 LT |
337 | reschedule_retry(r1_bio); |
338 | } | |
339 | ||
340 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); | |
1da177e4 LT |
341 | } |
342 | ||
9f2c9d12 | 343 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
344 | { |
345 | /* it really is the end of this request */ | |
346 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
347 | /* free extra copy of the data pages */ | |
348 | int i = r1_bio->behind_page_count; | |
349 | while (i--) | |
350 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
351 | kfree(r1_bio->behind_bvecs); | |
352 | r1_bio->behind_bvecs = NULL; | |
353 | } | |
354 | /* clear the bitmap if all writes complete successfully */ | |
355 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
356 | r1_bio->sectors, | |
357 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
358 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
359 | md_write_end(r1_bio->mddev); | |
360 | } | |
361 | ||
9f2c9d12 | 362 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 363 | { |
cd5ff9a1 N |
364 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
365 | return; | |
366 | ||
367 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
368 | reschedule_retry(r1_bio); | |
369 | else { | |
370 | close_write(r1_bio); | |
4367af55 N |
371 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
372 | reschedule_retry(r1_bio); | |
373 | else | |
374 | raid_end_bio_io(r1_bio); | |
4e78064f N |
375 | } |
376 | } | |
377 | ||
6712ecf8 | 378 | static void raid1_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
379 | { |
380 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 381 | struct r1bio *r1_bio = bio->bi_private; |
a9701a30 | 382 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 383 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 384 | struct bio *to_put = NULL; |
1da177e4 | 385 | |
ba3ae3be | 386 | mirror = find_bio_disk(r1_bio, bio); |
1da177e4 | 387 | |
e9c7469b TH |
388 | /* |
389 | * 'one mirror IO has finished' event handler: | |
390 | */ | |
e9c7469b | 391 | if (!uptodate) { |
cd5ff9a1 N |
392 | set_bit(WriteErrorSeen, |
393 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
394 | if (!test_and_set_bit(WantReplacement, |
395 | &conf->mirrors[mirror].rdev->flags)) | |
396 | set_bit(MD_RECOVERY_NEEDED, & | |
397 | conf->mddev->recovery); | |
398 | ||
cd5ff9a1 | 399 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 | 400 | } else { |
1da177e4 | 401 | /* |
e9c7469b TH |
402 | * Set R1BIO_Uptodate in our master bio, so that we |
403 | * will return a good error code for to the higher | |
404 | * levels even if IO on some other mirrored buffer | |
405 | * fails. | |
406 | * | |
407 | * The 'master' represents the composite IO operation | |
408 | * to user-side. So if something waits for IO, then it | |
409 | * will wait for the 'master' bio. | |
1da177e4 | 410 | */ |
4367af55 N |
411 | sector_t first_bad; |
412 | int bad_sectors; | |
413 | ||
cd5ff9a1 N |
414 | r1_bio->bios[mirror] = NULL; |
415 | to_put = bio; | |
e9c7469b TH |
416 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
417 | ||
4367af55 N |
418 | /* Maybe we can clear some bad blocks. */ |
419 | if (is_badblock(conf->mirrors[mirror].rdev, | |
420 | r1_bio->sector, r1_bio->sectors, | |
421 | &first_bad, &bad_sectors)) { | |
422 | r1_bio->bios[mirror] = IO_MADE_GOOD; | |
423 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
424 | } | |
425 | } | |
426 | ||
e9c7469b TH |
427 | if (behind) { |
428 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) | |
429 | atomic_dec(&r1_bio->behind_remaining); | |
430 | ||
431 | /* | |
432 | * In behind mode, we ACK the master bio once the I/O | |
433 | * has safely reached all non-writemostly | |
434 | * disks. Setting the Returned bit ensures that this | |
435 | * gets done only once -- we don't ever want to return | |
436 | * -EIO here, instead we'll wait | |
437 | */ | |
438 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
439 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
440 | /* Maybe we can return now */ | |
441 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
442 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
443 | pr_debug("raid1: behind end write sectors" |
444 | " %llu-%llu\n", | |
445 | (unsigned long long) mbio->bi_sector, | |
446 | (unsigned long long) mbio->bi_sector + | |
447 | (mbio->bi_size >> 9) - 1); | |
d2eb35ac | 448 | call_bio_endio(r1_bio); |
4b6d287f N |
449 | } |
450 | } | |
451 | } | |
4367af55 N |
452 | if (r1_bio->bios[mirror] == NULL) |
453 | rdev_dec_pending(conf->mirrors[mirror].rdev, | |
454 | conf->mddev); | |
e9c7469b | 455 | |
1da177e4 | 456 | /* |
1da177e4 LT |
457 | * Let's see if all mirrored write operations have finished |
458 | * already. | |
459 | */ | |
af6d7b76 | 460 | r1_bio_write_done(r1_bio); |
c70810b3 | 461 | |
04b857f7 N |
462 | if (to_put) |
463 | bio_put(to_put); | |
1da177e4 LT |
464 | } |
465 | ||
466 | ||
467 | /* | |
468 | * This routine returns the disk from which the requested read should | |
469 | * be done. There is a per-array 'next expected sequential IO' sector | |
470 | * number - if this matches on the next IO then we use the last disk. | |
471 | * There is also a per-disk 'last know head position' sector that is | |
472 | * maintained from IRQ contexts, both the normal and the resync IO | |
473 | * completion handlers update this position correctly. If there is no | |
474 | * perfect sequential match then we pick the disk whose head is closest. | |
475 | * | |
476 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
477 | * because position is mirror, not device based. | |
478 | * | |
479 | * The rdev for the device selected will have nr_pending incremented. | |
480 | */ | |
e8096360 | 481 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 482 | { |
af3a2cd6 | 483 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
484 | int sectors; |
485 | int best_good_sectors; | |
f3ac8bf7 | 486 | int start_disk; |
76073054 | 487 | int best_disk; |
f3ac8bf7 | 488 | int i; |
76073054 | 489 | sector_t best_dist; |
3cb03002 | 490 | struct md_rdev *rdev; |
f3ac8bf7 | 491 | int choose_first; |
1da177e4 LT |
492 | |
493 | rcu_read_lock(); | |
494 | /* | |
8ddf9efe | 495 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
496 | * device if no resync is going on, or below the resync window. |
497 | * We take the first readable disk when above the resync window. | |
498 | */ | |
499 | retry: | |
d2eb35ac | 500 | sectors = r1_bio->sectors; |
76073054 N |
501 | best_disk = -1; |
502 | best_dist = MaxSector; | |
d2eb35ac N |
503 | best_good_sectors = 0; |
504 | ||
1da177e4 LT |
505 | if (conf->mddev->recovery_cp < MaxSector && |
506 | (this_sector + sectors >= conf->next_resync)) { | |
f3ac8bf7 N |
507 | choose_first = 1; |
508 | start_disk = 0; | |
509 | } else { | |
510 | choose_first = 0; | |
511 | start_disk = conf->last_used; | |
1da177e4 LT |
512 | } |
513 | ||
8f19ccb2 | 514 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) { |
76073054 | 515 | sector_t dist; |
d2eb35ac N |
516 | sector_t first_bad; |
517 | int bad_sectors; | |
518 | ||
f3ac8bf7 | 519 | int disk = start_disk + i; |
32644afd N |
520 | if (disk >= conf->raid_disks * 2) |
521 | disk -= conf->raid_disks * 2; | |
f3ac8bf7 N |
522 | |
523 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
524 | if (r1_bio->bios[disk] == IO_BLOCKED | |
525 | || rdev == NULL | |
6b740b8d | 526 | || test_bit(Unmerged, &rdev->flags) |
76073054 | 527 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 528 | continue; |
76073054 N |
529 | if (!test_bit(In_sync, &rdev->flags) && |
530 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 531 | continue; |
76073054 N |
532 | if (test_bit(WriteMostly, &rdev->flags)) { |
533 | /* Don't balance among write-mostly, just | |
534 | * use the first as a last resort */ | |
307729c8 N |
535 | if (best_disk < 0) { |
536 | if (is_badblock(rdev, this_sector, sectors, | |
537 | &first_bad, &bad_sectors)) { | |
538 | if (first_bad < this_sector) | |
539 | /* Cannot use this */ | |
540 | continue; | |
541 | best_good_sectors = first_bad - this_sector; | |
542 | } else | |
543 | best_good_sectors = sectors; | |
76073054 | 544 | best_disk = disk; |
307729c8 | 545 | } |
76073054 N |
546 | continue; |
547 | } | |
548 | /* This is a reasonable device to use. It might | |
549 | * even be best. | |
550 | */ | |
d2eb35ac N |
551 | if (is_badblock(rdev, this_sector, sectors, |
552 | &first_bad, &bad_sectors)) { | |
553 | if (best_dist < MaxSector) | |
554 | /* already have a better device */ | |
555 | continue; | |
556 | if (first_bad <= this_sector) { | |
557 | /* cannot read here. If this is the 'primary' | |
558 | * device, then we must not read beyond | |
559 | * bad_sectors from another device.. | |
560 | */ | |
561 | bad_sectors -= (this_sector - first_bad); | |
562 | if (choose_first && sectors > bad_sectors) | |
563 | sectors = bad_sectors; | |
564 | if (best_good_sectors > sectors) | |
565 | best_good_sectors = sectors; | |
566 | ||
567 | } else { | |
568 | sector_t good_sectors = first_bad - this_sector; | |
569 | if (good_sectors > best_good_sectors) { | |
570 | best_good_sectors = good_sectors; | |
571 | best_disk = disk; | |
572 | } | |
573 | if (choose_first) | |
574 | break; | |
575 | } | |
576 | continue; | |
577 | } else | |
578 | best_good_sectors = sectors; | |
579 | ||
76073054 N |
580 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
581 | if (choose_first | |
582 | /* Don't change to another disk for sequential reads */ | |
583 | || conf->next_seq_sect == this_sector | |
584 | || dist == 0 | |
585 | /* If device is idle, use it */ | |
586 | || atomic_read(&rdev->nr_pending) == 0) { | |
587 | best_disk = disk; | |
1da177e4 LT |
588 | break; |
589 | } | |
76073054 N |
590 | if (dist < best_dist) { |
591 | best_dist = dist; | |
592 | best_disk = disk; | |
1da177e4 | 593 | } |
f3ac8bf7 | 594 | } |
1da177e4 | 595 | |
76073054 N |
596 | if (best_disk >= 0) { |
597 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
598 | if (!rdev) |
599 | goto retry; | |
600 | atomic_inc(&rdev->nr_pending); | |
76073054 | 601 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
602 | /* cannot risk returning a device that failed |
603 | * before we inc'ed nr_pending | |
604 | */ | |
03c902e1 | 605 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
606 | goto retry; |
607 | } | |
d2eb35ac | 608 | sectors = best_good_sectors; |
8ddf9efe | 609 | conf->next_seq_sect = this_sector + sectors; |
76073054 | 610 | conf->last_used = best_disk; |
1da177e4 LT |
611 | } |
612 | rcu_read_unlock(); | |
d2eb35ac | 613 | *max_sectors = sectors; |
1da177e4 | 614 | |
76073054 | 615 | return best_disk; |
1da177e4 LT |
616 | } |
617 | ||
6b740b8d N |
618 | static int raid1_mergeable_bvec(struct request_queue *q, |
619 | struct bvec_merge_data *bvm, | |
620 | struct bio_vec *biovec) | |
621 | { | |
622 | struct mddev *mddev = q->queuedata; | |
623 | struct r1conf *conf = mddev->private; | |
624 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); | |
625 | int max = biovec->bv_len; | |
626 | ||
627 | if (mddev->merge_check_needed) { | |
628 | int disk; | |
629 | rcu_read_lock(); | |
630 | for (disk = 0; disk < conf->raid_disks * 2; disk++) { | |
631 | struct md_rdev *rdev = rcu_dereference( | |
632 | conf->mirrors[disk].rdev); | |
633 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
634 | struct request_queue *q = | |
635 | bdev_get_queue(rdev->bdev); | |
636 | if (q->merge_bvec_fn) { | |
637 | bvm->bi_sector = sector + | |
638 | rdev->data_offset; | |
639 | bvm->bi_bdev = rdev->bdev; | |
640 | max = min(max, q->merge_bvec_fn( | |
641 | q, bvm, biovec)); | |
642 | } | |
643 | } | |
644 | } | |
645 | rcu_read_unlock(); | |
646 | } | |
647 | return max; | |
648 | ||
649 | } | |
650 | ||
fd01b88c | 651 | int md_raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 652 | { |
e8096360 | 653 | struct r1conf *conf = mddev->private; |
0d129228 N |
654 | int i, ret = 0; |
655 | ||
34db0cd6 N |
656 | if ((bits & (1 << BDI_async_congested)) && |
657 | conf->pending_count >= max_queued_requests) | |
658 | return 1; | |
659 | ||
0d129228 | 660 | rcu_read_lock(); |
f53e29fc | 661 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 662 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 663 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 664 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 665 | |
1ed7242e JB |
666 | BUG_ON(!q); |
667 | ||
0d129228 N |
668 | /* Note the '|| 1' - when read_balance prefers |
669 | * non-congested targets, it can be removed | |
670 | */ | |
91a9e99d | 671 | if ((bits & (1<<BDI_async_congested)) || 1) |
0d129228 N |
672 | ret |= bdi_congested(&q->backing_dev_info, bits); |
673 | else | |
674 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
675 | } | |
676 | } | |
677 | rcu_read_unlock(); | |
678 | return ret; | |
679 | } | |
1ed7242e | 680 | EXPORT_SYMBOL_GPL(md_raid1_congested); |
0d129228 | 681 | |
1ed7242e JB |
682 | static int raid1_congested(void *data, int bits) |
683 | { | |
fd01b88c | 684 | struct mddev *mddev = data; |
1ed7242e JB |
685 | |
686 | return mddev_congested(mddev, bits) || | |
687 | md_raid1_congested(mddev, bits); | |
688 | } | |
0d129228 | 689 | |
e8096360 | 690 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
691 | { |
692 | /* Any writes that have been queued but are awaiting | |
693 | * bitmap updates get flushed here. | |
a35e63ef | 694 | */ |
a35e63ef N |
695 | spin_lock_irq(&conf->device_lock); |
696 | ||
697 | if (conf->pending_bio_list.head) { | |
698 | struct bio *bio; | |
699 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 700 | conf->pending_count = 0; |
a35e63ef N |
701 | spin_unlock_irq(&conf->device_lock); |
702 | /* flush any pending bitmap writes to | |
703 | * disk before proceeding w/ I/O */ | |
704 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 705 | wake_up(&conf->wait_barrier); |
a35e63ef N |
706 | |
707 | while (bio) { /* submit pending writes */ | |
708 | struct bio *next = bio->bi_next; | |
709 | bio->bi_next = NULL; | |
710 | generic_make_request(bio); | |
711 | bio = next; | |
712 | } | |
a35e63ef N |
713 | } else |
714 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
715 | } |
716 | ||
17999be4 N |
717 | /* Barriers.... |
718 | * Sometimes we need to suspend IO while we do something else, | |
719 | * either some resync/recovery, or reconfigure the array. | |
720 | * To do this we raise a 'barrier'. | |
721 | * The 'barrier' is a counter that can be raised multiple times | |
722 | * to count how many activities are happening which preclude | |
723 | * normal IO. | |
724 | * We can only raise the barrier if there is no pending IO. | |
725 | * i.e. if nr_pending == 0. | |
726 | * We choose only to raise the barrier if no-one is waiting for the | |
727 | * barrier to go down. This means that as soon as an IO request | |
728 | * is ready, no other operations which require a barrier will start | |
729 | * until the IO request has had a chance. | |
730 | * | |
731 | * So: regular IO calls 'wait_barrier'. When that returns there | |
732 | * is no backgroup IO happening, It must arrange to call | |
733 | * allow_barrier when it has finished its IO. | |
734 | * backgroup IO calls must call raise_barrier. Once that returns | |
735 | * there is no normal IO happeing. It must arrange to call | |
736 | * lower_barrier when the particular background IO completes. | |
1da177e4 LT |
737 | */ |
738 | #define RESYNC_DEPTH 32 | |
739 | ||
e8096360 | 740 | static void raise_barrier(struct r1conf *conf) |
1da177e4 LT |
741 | { |
742 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
743 | |
744 | /* Wait until no block IO is waiting */ | |
745 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
c3b328ac | 746 | conf->resync_lock, ); |
17999be4 N |
747 | |
748 | /* block any new IO from starting */ | |
749 | conf->barrier++; | |
750 | ||
046abeed | 751 | /* Now wait for all pending IO to complete */ |
17999be4 N |
752 | wait_event_lock_irq(conf->wait_barrier, |
753 | !conf->nr_pending && conf->barrier < RESYNC_DEPTH, | |
c3b328ac | 754 | conf->resync_lock, ); |
17999be4 N |
755 | |
756 | spin_unlock_irq(&conf->resync_lock); | |
757 | } | |
758 | ||
e8096360 | 759 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
760 | { |
761 | unsigned long flags; | |
709ae487 | 762 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
763 | spin_lock_irqsave(&conf->resync_lock, flags); |
764 | conf->barrier--; | |
765 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
766 | wake_up(&conf->wait_barrier); | |
767 | } | |
768 | ||
e8096360 | 769 | static void wait_barrier(struct r1conf *conf) |
17999be4 N |
770 | { |
771 | spin_lock_irq(&conf->resync_lock); | |
772 | if (conf->barrier) { | |
773 | conf->nr_waiting++; | |
d6b42dcb N |
774 | /* Wait for the barrier to drop. |
775 | * However if there are already pending | |
776 | * requests (preventing the barrier from | |
777 | * rising completely), and the | |
778 | * pre-process bio queue isn't empty, | |
779 | * then don't wait, as we need to empty | |
780 | * that queue to get the nr_pending | |
781 | * count down. | |
782 | */ | |
783 | wait_event_lock_irq(conf->wait_barrier, | |
784 | !conf->barrier || | |
785 | (conf->nr_pending && | |
786 | current->bio_list && | |
787 | !bio_list_empty(current->bio_list)), | |
17999be4 | 788 | conf->resync_lock, |
d6b42dcb | 789 | ); |
17999be4 | 790 | conf->nr_waiting--; |
1da177e4 | 791 | } |
17999be4 | 792 | conf->nr_pending++; |
1da177e4 LT |
793 | spin_unlock_irq(&conf->resync_lock); |
794 | } | |
795 | ||
e8096360 | 796 | static void allow_barrier(struct r1conf *conf) |
17999be4 N |
797 | { |
798 | unsigned long flags; | |
799 | spin_lock_irqsave(&conf->resync_lock, flags); | |
800 | conf->nr_pending--; | |
801 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
802 | wake_up(&conf->wait_barrier); | |
803 | } | |
804 | ||
e8096360 | 805 | static void freeze_array(struct r1conf *conf) |
ddaf22ab N |
806 | { |
807 | /* stop syncio and normal IO and wait for everything to | |
808 | * go quite. | |
809 | * We increment barrier and nr_waiting, and then | |
1c830532 N |
810 | * wait until nr_pending match nr_queued+1 |
811 | * This is called in the context of one normal IO request | |
812 | * that has failed. Thus any sync request that might be pending | |
813 | * will be blocked by nr_pending, and we need to wait for | |
814 | * pending IO requests to complete or be queued for re-try. | |
815 | * Thus the number queued (nr_queued) plus this request (1) | |
816 | * must match the number of pending IOs (nr_pending) before | |
817 | * we continue. | |
ddaf22ab N |
818 | */ |
819 | spin_lock_irq(&conf->resync_lock); | |
820 | conf->barrier++; | |
821 | conf->nr_waiting++; | |
822 | wait_event_lock_irq(conf->wait_barrier, | |
1c830532 | 823 | conf->nr_pending == conf->nr_queued+1, |
ddaf22ab | 824 | conf->resync_lock, |
c3b328ac | 825 | flush_pending_writes(conf)); |
ddaf22ab N |
826 | spin_unlock_irq(&conf->resync_lock); |
827 | } | |
e8096360 | 828 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
829 | { |
830 | /* reverse the effect of the freeze */ | |
831 | spin_lock_irq(&conf->resync_lock); | |
832 | conf->barrier--; | |
833 | conf->nr_waiting--; | |
834 | wake_up(&conf->wait_barrier); | |
835 | spin_unlock_irq(&conf->resync_lock); | |
836 | } | |
837 | ||
17999be4 | 838 | |
4e78064f | 839 | /* duplicate the data pages for behind I/O |
4e78064f | 840 | */ |
9f2c9d12 | 841 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
842 | { |
843 | int i; | |
844 | struct bio_vec *bvec; | |
2ca68f5e | 845 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 846 | GFP_NOIO); |
2ca68f5e | 847 | if (unlikely(!bvecs)) |
af6d7b76 | 848 | return; |
4b6d287f | 849 | |
4b6d287f | 850 | bio_for_each_segment(bvec, bio, i) { |
2ca68f5e N |
851 | bvecs[i] = *bvec; |
852 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
853 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 854 | goto do_sync_io; |
2ca68f5e N |
855 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
856 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
857 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
858 | kunmap(bvec->bv_page); |
859 | } | |
2ca68f5e | 860 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
861 | r1_bio->behind_page_count = bio->bi_vcnt; |
862 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
863 | return; | |
4b6d287f N |
864 | |
865 | do_sync_io: | |
af6d7b76 | 866 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
867 | if (bvecs[i].bv_page) |
868 | put_page(bvecs[i].bv_page); | |
869 | kfree(bvecs); | |
36a4e1fe | 870 | pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); |
4b6d287f N |
871 | } |
872 | ||
f54a9d0e N |
873 | struct raid1_plug_cb { |
874 | struct blk_plug_cb cb; | |
875 | struct bio_list pending; | |
876 | int pending_cnt; | |
877 | }; | |
878 | ||
879 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
880 | { | |
881 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
882 | cb); | |
883 | struct mddev *mddev = plug->cb.data; | |
884 | struct r1conf *conf = mddev->private; | |
885 | struct bio *bio; | |
886 | ||
887 | if (from_schedule) { | |
888 | spin_lock_irq(&conf->device_lock); | |
889 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
890 | conf->pending_count += plug->pending_cnt; | |
891 | spin_unlock_irq(&conf->device_lock); | |
892 | md_wakeup_thread(mddev->thread); | |
893 | kfree(plug); | |
894 | return; | |
895 | } | |
896 | ||
897 | /* we aren't scheduling, so we can do the write-out directly. */ | |
898 | bio = bio_list_get(&plug->pending); | |
899 | bitmap_unplug(mddev->bitmap); | |
900 | wake_up(&conf->wait_barrier); | |
901 | ||
902 | while (bio) { /* submit pending writes */ | |
903 | struct bio *next = bio->bi_next; | |
904 | bio->bi_next = NULL; | |
905 | generic_make_request(bio); | |
906 | bio = next; | |
907 | } | |
908 | kfree(plug); | |
909 | } | |
910 | ||
b4fdcb02 | 911 | static void make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 912 | { |
e8096360 | 913 | struct r1conf *conf = mddev->private; |
0f6d02d5 | 914 | struct mirror_info *mirror; |
9f2c9d12 | 915 | struct r1bio *r1_bio; |
1da177e4 | 916 | struct bio *read_bio; |
1f68f0c4 | 917 | int i, disks; |
84255d10 | 918 | struct bitmap *bitmap; |
191ea9b2 | 919 | unsigned long flags; |
a362357b | 920 | const int rw = bio_data_dir(bio); |
2c7d46ec | 921 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 922 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
3cb03002 | 923 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
924 | struct blk_plug_cb *cb; |
925 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
926 | int first_clone; |
927 | int sectors_handled; | |
928 | int max_sectors; | |
191ea9b2 | 929 | |
1da177e4 LT |
930 | /* |
931 | * Register the new request and wait if the reconstruction | |
932 | * thread has put up a bar for new requests. | |
933 | * Continue immediately if no resync is active currently. | |
934 | */ | |
62de608d | 935 | |
3d310eb7 N |
936 | md_write_start(mddev, bio); /* wait on superblock update early */ |
937 | ||
6eef4b21 N |
938 | if (bio_data_dir(bio) == WRITE && |
939 | bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo && | |
940 | bio->bi_sector < mddev->suspend_hi) { | |
941 | /* As the suspend_* range is controlled by | |
942 | * userspace, we want an interruptible | |
943 | * wait. | |
944 | */ | |
945 | DEFINE_WAIT(w); | |
946 | for (;;) { | |
947 | flush_signals(current); | |
948 | prepare_to_wait(&conf->wait_barrier, | |
949 | &w, TASK_INTERRUPTIBLE); | |
950 | if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo || | |
951 | bio->bi_sector >= mddev->suspend_hi) | |
952 | break; | |
953 | schedule(); | |
954 | } | |
955 | finish_wait(&conf->wait_barrier, &w); | |
956 | } | |
62de608d | 957 | |
17999be4 | 958 | wait_barrier(conf); |
1da177e4 | 959 | |
84255d10 N |
960 | bitmap = mddev->bitmap; |
961 | ||
1da177e4 LT |
962 | /* |
963 | * make_request() can abort the operation when READA is being | |
964 | * used and no empty request is available. | |
965 | * | |
966 | */ | |
967 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
968 | ||
969 | r1_bio->master_bio = bio; | |
970 | r1_bio->sectors = bio->bi_size >> 9; | |
191ea9b2 | 971 | r1_bio->state = 0; |
1da177e4 LT |
972 | r1_bio->mddev = mddev; |
973 | r1_bio->sector = bio->bi_sector; | |
974 | ||
d2eb35ac N |
975 | /* We might need to issue multiple reads to different |
976 | * devices if there are bad blocks around, so we keep | |
977 | * track of the number of reads in bio->bi_phys_segments. | |
978 | * If this is 0, there is only one r1_bio and no locking | |
979 | * will be needed when requests complete. If it is | |
980 | * non-zero, then it is the number of not-completed requests. | |
981 | */ | |
982 | bio->bi_phys_segments = 0; | |
983 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
984 | ||
a362357b | 985 | if (rw == READ) { |
1da177e4 LT |
986 | /* |
987 | * read balancing logic: | |
988 | */ | |
d2eb35ac N |
989 | int rdisk; |
990 | ||
991 | read_again: | |
992 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1da177e4 LT |
993 | |
994 | if (rdisk < 0) { | |
995 | /* couldn't find anywhere to read from */ | |
996 | raid_end_bio_io(r1_bio); | |
5a7bbad2 | 997 | return; |
1da177e4 LT |
998 | } |
999 | mirror = conf->mirrors + rdisk; | |
1000 | ||
e555190d N |
1001 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1002 | bitmap) { | |
1003 | /* Reading from a write-mostly device must | |
1004 | * take care not to over-take any writes | |
1005 | * that are 'behind' | |
1006 | */ | |
1007 | wait_event(bitmap->behind_wait, | |
1008 | atomic_read(&bitmap->behind_writes) == 0); | |
1009 | } | |
1da177e4 LT |
1010 | r1_bio->read_disk = rdisk; |
1011 | ||
a167f663 | 1012 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
d2eb35ac N |
1013 | md_trim_bio(read_bio, r1_bio->sector - bio->bi_sector, |
1014 | max_sectors); | |
1da177e4 LT |
1015 | |
1016 | r1_bio->bios[rdisk] = read_bio; | |
1017 | ||
1018 | read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; | |
1019 | read_bio->bi_bdev = mirror->rdev->bdev; | |
1020 | read_bio->bi_end_io = raid1_end_read_request; | |
7b6d91da | 1021 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1022 | read_bio->bi_private = r1_bio; |
1023 | ||
d2eb35ac N |
1024 | if (max_sectors < r1_bio->sectors) { |
1025 | /* could not read all from this device, so we will | |
1026 | * need another r1_bio. | |
1027 | */ | |
d2eb35ac N |
1028 | |
1029 | sectors_handled = (r1_bio->sector + max_sectors | |
1030 | - bio->bi_sector); | |
1031 | r1_bio->sectors = max_sectors; | |
1032 | spin_lock_irq(&conf->device_lock); | |
1033 | if (bio->bi_phys_segments == 0) | |
1034 | bio->bi_phys_segments = 2; | |
1035 | else | |
1036 | bio->bi_phys_segments++; | |
1037 | spin_unlock_irq(&conf->device_lock); | |
1038 | /* Cannot call generic_make_request directly | |
1039 | * as that will be queued in __make_request | |
1040 | * and subsequent mempool_alloc might block waiting | |
1041 | * for it. So hand bio over to raid1d. | |
1042 | */ | |
1043 | reschedule_retry(r1_bio); | |
1044 | ||
1045 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1046 | ||
1047 | r1_bio->master_bio = bio; | |
1048 | r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled; | |
1049 | r1_bio->state = 0; | |
1050 | r1_bio->mddev = mddev; | |
1051 | r1_bio->sector = bio->bi_sector + sectors_handled; | |
1052 | goto read_again; | |
1053 | } else | |
1054 | generic_make_request(read_bio); | |
5a7bbad2 | 1055 | return; |
1da177e4 LT |
1056 | } |
1057 | ||
1058 | /* | |
1059 | * WRITE: | |
1060 | */ | |
34db0cd6 N |
1061 | if (conf->pending_count >= max_queued_requests) { |
1062 | md_wakeup_thread(mddev->thread); | |
1063 | wait_event(conf->wait_barrier, | |
1064 | conf->pending_count < max_queued_requests); | |
1065 | } | |
1f68f0c4 | 1066 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1067 | * inc refcount on their rdev. Record them by setting |
1068 | * bios[x] to bio | |
1f68f0c4 N |
1069 | * If there are known/acknowledged bad blocks on any device on |
1070 | * which we have seen a write error, we want to avoid writing those | |
1071 | * blocks. | |
1072 | * This potentially requires several writes to write around | |
1073 | * the bad blocks. Each set of writes gets it's own r1bio | |
1074 | * with a set of bios attached. | |
1da177e4 | 1075 | */ |
c3b328ac | 1076 | |
8f19ccb2 | 1077 | disks = conf->raid_disks * 2; |
6bfe0b49 DW |
1078 | retry_write: |
1079 | blocked_rdev = NULL; | |
1da177e4 | 1080 | rcu_read_lock(); |
1f68f0c4 | 1081 | max_sectors = r1_bio->sectors; |
1da177e4 | 1082 | for (i = 0; i < disks; i++) { |
3cb03002 | 1083 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1084 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1085 | atomic_inc(&rdev->nr_pending); | |
1086 | blocked_rdev = rdev; | |
1087 | break; | |
1088 | } | |
1f68f0c4 | 1089 | r1_bio->bios[i] = NULL; |
6b740b8d N |
1090 | if (!rdev || test_bit(Faulty, &rdev->flags) |
1091 | || test_bit(Unmerged, &rdev->flags)) { | |
8f19ccb2 N |
1092 | if (i < conf->raid_disks) |
1093 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1094 | continue; |
1095 | } | |
1096 | ||
1097 | atomic_inc(&rdev->nr_pending); | |
1098 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1099 | sector_t first_bad; | |
1100 | int bad_sectors; | |
1101 | int is_bad; | |
1102 | ||
1103 | is_bad = is_badblock(rdev, r1_bio->sector, | |
1104 | max_sectors, | |
1105 | &first_bad, &bad_sectors); | |
1106 | if (is_bad < 0) { | |
1107 | /* mustn't write here until the bad block is | |
1108 | * acknowledged*/ | |
1109 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1110 | blocked_rdev = rdev; | |
1111 | break; | |
1112 | } | |
1113 | if (is_bad && first_bad <= r1_bio->sector) { | |
1114 | /* Cannot write here at all */ | |
1115 | bad_sectors -= (r1_bio->sector - first_bad); | |
1116 | if (bad_sectors < max_sectors) | |
1117 | /* mustn't write more than bad_sectors | |
1118 | * to other devices yet | |
1119 | */ | |
1120 | max_sectors = bad_sectors; | |
03c902e1 | 1121 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1122 | /* We don't set R1BIO_Degraded as that |
1123 | * only applies if the disk is | |
1124 | * missing, so it might be re-added, | |
1125 | * and we want to know to recover this | |
1126 | * chunk. | |
1127 | * In this case the device is here, | |
1128 | * and the fact that this chunk is not | |
1129 | * in-sync is recorded in the bad | |
1130 | * block log | |
1131 | */ | |
1132 | continue; | |
964147d5 | 1133 | } |
1f68f0c4 N |
1134 | if (is_bad) { |
1135 | int good_sectors = first_bad - r1_bio->sector; | |
1136 | if (good_sectors < max_sectors) | |
1137 | max_sectors = good_sectors; | |
1138 | } | |
1139 | } | |
1140 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1141 | } |
1142 | rcu_read_unlock(); | |
1143 | ||
6bfe0b49 DW |
1144 | if (unlikely(blocked_rdev)) { |
1145 | /* Wait for this device to become unblocked */ | |
1146 | int j; | |
1147 | ||
1148 | for (j = 0; j < i; j++) | |
1149 | if (r1_bio->bios[j]) | |
1150 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1151 | r1_bio->state = 0; |
6bfe0b49 DW |
1152 | allow_barrier(conf); |
1153 | md_wait_for_blocked_rdev(blocked_rdev, mddev); | |
1154 | wait_barrier(conf); | |
1155 | goto retry_write; | |
1156 | } | |
1157 | ||
1f68f0c4 N |
1158 | if (max_sectors < r1_bio->sectors) { |
1159 | /* We are splitting this write into multiple parts, so | |
1160 | * we need to prepare for allocating another r1_bio. | |
1161 | */ | |
1162 | r1_bio->sectors = max_sectors; | |
1163 | spin_lock_irq(&conf->device_lock); | |
1164 | if (bio->bi_phys_segments == 0) | |
1165 | bio->bi_phys_segments = 2; | |
1166 | else | |
1167 | bio->bi_phys_segments++; | |
1168 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1169 | } |
1f68f0c4 | 1170 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector; |
4b6d287f | 1171 | |
4e78064f | 1172 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1173 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1174 | |
1f68f0c4 | 1175 | first_clone = 1; |
1da177e4 LT |
1176 | for (i = 0; i < disks; i++) { |
1177 | struct bio *mbio; | |
1178 | if (!r1_bio->bios[i]) | |
1179 | continue; | |
1180 | ||
a167f663 | 1181 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
1f68f0c4 N |
1182 | md_trim_bio(mbio, r1_bio->sector - bio->bi_sector, max_sectors); |
1183 | ||
1184 | if (first_clone) { | |
1185 | /* do behind I/O ? | |
1186 | * Not if there are too many, or cannot | |
1187 | * allocate memory, or a reader on WriteMostly | |
1188 | * is waiting for behind writes to flush */ | |
1189 | if (bitmap && | |
1190 | (atomic_read(&bitmap->behind_writes) | |
1191 | < mddev->bitmap_info.max_write_behind) && | |
1192 | !waitqueue_active(&bitmap->behind_wait)) | |
1193 | alloc_behind_pages(mbio, r1_bio); | |
1194 | ||
1195 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1196 | r1_bio->sectors, | |
1197 | test_bit(R1BIO_BehindIO, | |
1198 | &r1_bio->state)); | |
1199 | first_clone = 0; | |
1200 | } | |
2ca68f5e | 1201 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1202 | struct bio_vec *bvec; |
1203 | int j; | |
1204 | ||
1205 | /* Yes, I really want the '__' version so that | |
1206 | * we clear any unused pointer in the io_vec, rather | |
1207 | * than leave them unchanged. This is important | |
1208 | * because when we come to free the pages, we won't | |
046abeed | 1209 | * know the original bi_idx, so we just free |
4b6d287f N |
1210 | * them all |
1211 | */ | |
1212 | __bio_for_each_segment(bvec, mbio, j, 0) | |
2ca68f5e | 1213 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1214 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1215 | atomic_inc(&r1_bio->behind_remaining); | |
1216 | } | |
1217 | ||
1f68f0c4 N |
1218 | r1_bio->bios[i] = mbio; |
1219 | ||
1220 | mbio->bi_sector = (r1_bio->sector + | |
1221 | conf->mirrors[i].rdev->data_offset); | |
1222 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1223 | mbio->bi_end_io = raid1_end_write_request; | |
1224 | mbio->bi_rw = WRITE | do_flush_fua | do_sync; | |
1225 | mbio->bi_private = r1_bio; | |
1226 | ||
1da177e4 | 1227 | atomic_inc(&r1_bio->remaining); |
f54a9d0e N |
1228 | |
1229 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); | |
1230 | if (cb) | |
1231 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1232 | else | |
1233 | plug = NULL; | |
4e78064f | 1234 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1235 | if (plug) { |
1236 | bio_list_add(&plug->pending, mbio); | |
1237 | plug->pending_cnt++; | |
1238 | } else { | |
1239 | bio_list_add(&conf->pending_bio_list, mbio); | |
1240 | conf->pending_count++; | |
1241 | } | |
4e78064f | 1242 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1243 | if (!plug) |
b357f04a | 1244 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1245 | } |
079fa166 N |
1246 | /* Mustn't call r1_bio_write_done before this next test, |
1247 | * as it could result in the bio being freed. | |
1248 | */ | |
1f68f0c4 | 1249 | if (sectors_handled < (bio->bi_size >> 9)) { |
079fa166 | 1250 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1251 | /* We need another r1_bio. It has already been counted |
1252 | * in bio->bi_phys_segments | |
1253 | */ | |
1254 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1255 | r1_bio->master_bio = bio; | |
1256 | r1_bio->sectors = (bio->bi_size >> 9) - sectors_handled; | |
1257 | r1_bio->state = 0; | |
1258 | r1_bio->mddev = mddev; | |
1259 | r1_bio->sector = bio->bi_sector + sectors_handled; | |
1260 | goto retry_write; | |
1261 | } | |
1262 | ||
079fa166 N |
1263 | r1_bio_write_done(r1_bio); |
1264 | ||
1265 | /* In case raid1d snuck in to freeze_array */ | |
1266 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1267 | } |
1268 | ||
fd01b88c | 1269 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1270 | { |
e8096360 | 1271 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1272 | int i; |
1273 | ||
1274 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1275 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1276 | rcu_read_lock(); |
1277 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1278 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1279 | seq_printf(seq, "%s", |
ddac7c7e N |
1280 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1281 | } | |
1282 | rcu_read_unlock(); | |
1da177e4 LT |
1283 | seq_printf(seq, "]"); |
1284 | } | |
1285 | ||
1286 | ||
fd01b88c | 1287 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1288 | { |
1289 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1290 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1291 | |
1292 | /* | |
1293 | * If it is not operational, then we have already marked it as dead | |
1294 | * else if it is the last working disks, ignore the error, let the | |
1295 | * next level up know. | |
1296 | * else mark the drive as failed | |
1297 | */ | |
b2d444d7 | 1298 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1299 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1300 | /* |
1301 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1302 | * normal single drive. |
1303 | * However don't try a recovery from this drive as | |
1304 | * it is very likely to fail. | |
1da177e4 | 1305 | */ |
5389042f | 1306 | conf->recovery_disabled = mddev->recovery_disabled; |
1da177e4 | 1307 | return; |
4044ba58 | 1308 | } |
de393cde | 1309 | set_bit(Blocked, &rdev->flags); |
c04be0aa N |
1310 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1311 | unsigned long flags; | |
1312 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1313 | mddev->degraded++; |
dd00a99e | 1314 | set_bit(Faulty, &rdev->flags); |
c04be0aa | 1315 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1316 | /* |
1317 | * if recovery is running, make sure it aborts. | |
1318 | */ | |
dfc70645 | 1319 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
dd00a99e N |
1320 | } else |
1321 | set_bit(Faulty, &rdev->flags); | |
850b2b42 | 1322 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
1323 | printk(KERN_ALERT |
1324 | "md/raid1:%s: Disk failure on %s, disabling device.\n" | |
1325 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
9dd1e2fa N |
1326 | mdname(mddev), bdevname(rdev->bdev, b), |
1327 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1328 | } |
1329 | ||
e8096360 | 1330 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1331 | { |
1332 | int i; | |
1da177e4 | 1333 | |
9dd1e2fa | 1334 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
1da177e4 | 1335 | if (!conf) { |
9dd1e2fa | 1336 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1337 | return; |
1338 | } | |
9dd1e2fa | 1339 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1340 | conf->raid_disks); |
1341 | ||
ddac7c7e | 1342 | rcu_read_lock(); |
1da177e4 LT |
1343 | for (i = 0; i < conf->raid_disks; i++) { |
1344 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1345 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1346 | if (rdev) |
9dd1e2fa | 1347 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
ddac7c7e N |
1348 | i, !test_bit(In_sync, &rdev->flags), |
1349 | !test_bit(Faulty, &rdev->flags), | |
1350 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1351 | } |
ddac7c7e | 1352 | rcu_read_unlock(); |
1da177e4 LT |
1353 | } |
1354 | ||
e8096360 | 1355 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1356 | { |
17999be4 N |
1357 | wait_barrier(conf); |
1358 | allow_barrier(conf); | |
1da177e4 LT |
1359 | |
1360 | mempool_destroy(conf->r1buf_pool); | |
1361 | conf->r1buf_pool = NULL; | |
1362 | } | |
1363 | ||
fd01b88c | 1364 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1365 | { |
1366 | int i; | |
e8096360 | 1367 | struct r1conf *conf = mddev->private; |
6b965620 N |
1368 | int count = 0; |
1369 | unsigned long flags; | |
1da177e4 LT |
1370 | |
1371 | /* | |
1372 | * Find all failed disks within the RAID1 configuration | |
ddac7c7e N |
1373 | * and mark them readable. |
1374 | * Called under mddev lock, so rcu protection not needed. | |
1da177e4 LT |
1375 | */ |
1376 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1377 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1378 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1379 | if (repl | |
1380 | && repl->recovery_offset == MaxSector | |
1381 | && !test_bit(Faulty, &repl->flags) | |
1382 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1383 | /* replacement has just become active */ | |
1384 | if (!rdev || | |
1385 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1386 | count++; | |
1387 | if (rdev) { | |
1388 | /* Replaced device not technically | |
1389 | * faulty, but we need to be sure | |
1390 | * it gets removed and never re-added | |
1391 | */ | |
1392 | set_bit(Faulty, &rdev->flags); | |
1393 | sysfs_notify_dirent_safe( | |
1394 | rdev->sysfs_state); | |
1395 | } | |
1396 | } | |
ddac7c7e N |
1397 | if (rdev |
1398 | && !test_bit(Faulty, &rdev->flags) | |
c04be0aa | 1399 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1400 | count++; |
654e8b5a | 1401 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1402 | } |
1403 | } | |
6b965620 N |
1404 | spin_lock_irqsave(&conf->device_lock, flags); |
1405 | mddev->degraded -= count; | |
1406 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1407 | |
1408 | print_conf(conf); | |
6b965620 | 1409 | return count; |
1da177e4 LT |
1410 | } |
1411 | ||
1412 | ||
fd01b88c | 1413 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1414 | { |
e8096360 | 1415 | struct r1conf *conf = mddev->private; |
199050ea | 1416 | int err = -EEXIST; |
41158c7e | 1417 | int mirror = 0; |
0f6d02d5 | 1418 | struct mirror_info *p; |
6c2fce2e | 1419 | int first = 0; |
30194636 | 1420 | int last = conf->raid_disks - 1; |
6b740b8d | 1421 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
1da177e4 | 1422 | |
5389042f N |
1423 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1424 | return -EBUSY; | |
1425 | ||
6c2fce2e NB |
1426 | if (rdev->raid_disk >= 0) |
1427 | first = last = rdev->raid_disk; | |
1428 | ||
6b740b8d N |
1429 | if (q->merge_bvec_fn) { |
1430 | set_bit(Unmerged, &rdev->flags); | |
1431 | mddev->merge_check_needed = 1; | |
1432 | } | |
1433 | ||
7ef449d1 N |
1434 | for (mirror = first; mirror <= last; mirror++) { |
1435 | p = conf->mirrors+mirror; | |
1436 | if (!p->rdev) { | |
1da177e4 | 1437 | |
8f6c2e4b MP |
1438 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
1439 | rdev->data_offset << 9); | |
1da177e4 LT |
1440 | |
1441 | p->head_position = 0; | |
1442 | rdev->raid_disk = mirror; | |
199050ea | 1443 | err = 0; |
6aea114a N |
1444 | /* As all devices are equivalent, we don't need a full recovery |
1445 | * if this was recently any drive of the array | |
1446 | */ | |
1447 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1448 | conf->fullsync = 1; |
d6065f7b | 1449 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1450 | break; |
1451 | } | |
7ef449d1 N |
1452 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1453 | p[conf->raid_disks].rdev == NULL) { | |
1454 | /* Add this device as a replacement */ | |
1455 | clear_bit(In_sync, &rdev->flags); | |
1456 | set_bit(Replacement, &rdev->flags); | |
1457 | rdev->raid_disk = mirror; | |
1458 | err = 0; | |
1459 | conf->fullsync = 1; | |
1460 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1461 | break; | |
1462 | } | |
1463 | } | |
6b740b8d N |
1464 | if (err == 0 && test_bit(Unmerged, &rdev->flags)) { |
1465 | /* Some requests might not have seen this new | |
1466 | * merge_bvec_fn. We must wait for them to complete | |
1467 | * before merging the device fully. | |
1468 | * First we make sure any code which has tested | |
1469 | * our function has submitted the request, then | |
1470 | * we wait for all outstanding requests to complete. | |
1471 | */ | |
1472 | synchronize_sched(); | |
1473 | raise_barrier(conf); | |
1474 | lower_barrier(conf); | |
1475 | clear_bit(Unmerged, &rdev->flags); | |
1476 | } | |
ac5e7113 | 1477 | md_integrity_add_rdev(rdev, mddev); |
1da177e4 | 1478 | print_conf(conf); |
199050ea | 1479 | return err; |
1da177e4 LT |
1480 | } |
1481 | ||
b8321b68 | 1482 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1483 | { |
e8096360 | 1484 | struct r1conf *conf = mddev->private; |
1da177e4 | 1485 | int err = 0; |
b8321b68 | 1486 | int number = rdev->raid_disk; |
0f6d02d5 | 1487 | struct mirror_info *p = conf->mirrors+ number; |
1da177e4 | 1488 | |
b014f14c N |
1489 | if (rdev != p->rdev) |
1490 | p = conf->mirrors + conf->raid_disks + number; | |
1491 | ||
1da177e4 | 1492 | print_conf(conf); |
b8321b68 | 1493 | if (rdev == p->rdev) { |
b2d444d7 | 1494 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1495 | atomic_read(&rdev->nr_pending)) { |
1496 | err = -EBUSY; | |
1497 | goto abort; | |
1498 | } | |
046abeed | 1499 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1500 | * is not possible. |
1501 | */ | |
1502 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1503 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1504 | mddev->degraded < conf->raid_disks) { |
1505 | err = -EBUSY; | |
1506 | goto abort; | |
1507 | } | |
1da177e4 | 1508 | p->rdev = NULL; |
fbd568a3 | 1509 | synchronize_rcu(); |
1da177e4 LT |
1510 | if (atomic_read(&rdev->nr_pending)) { |
1511 | /* lost the race, try later */ | |
1512 | err = -EBUSY; | |
1513 | p->rdev = rdev; | |
ac5e7113 | 1514 | goto abort; |
8c7a2c2b N |
1515 | } else if (conf->mirrors[conf->raid_disks + number].rdev) { |
1516 | /* We just removed a device that is being replaced. | |
1517 | * Move down the replacement. We drain all IO before | |
1518 | * doing this to avoid confusion. | |
1519 | */ | |
1520 | struct md_rdev *repl = | |
1521 | conf->mirrors[conf->raid_disks + number].rdev; | |
1522 | raise_barrier(conf); | |
1523 | clear_bit(Replacement, &repl->flags); | |
1524 | p->rdev = repl; | |
1525 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
1526 | lower_barrier(conf); | |
1527 | clear_bit(WantReplacement, &rdev->flags); | |
1528 | } else | |
b014f14c | 1529 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1530 | err = md_integrity_register(mddev); |
1da177e4 LT |
1531 | } |
1532 | abort: | |
1533 | ||
1534 | print_conf(conf); | |
1535 | return err; | |
1536 | } | |
1537 | ||
1538 | ||
6712ecf8 | 1539 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1540 | { |
9f2c9d12 | 1541 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1542 | |
0fc280f6 | 1543 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1544 | |
1da177e4 LT |
1545 | /* |
1546 | * we have read a block, now it needs to be re-written, | |
1547 | * or re-read if the read failed. | |
1548 | * We don't do much here, just schedule handling by raid1d | |
1549 | */ | |
69382e85 | 1550 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1da177e4 | 1551 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1552 | |
1553 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1554 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1555 | } |
1556 | ||
6712ecf8 | 1557 | static void end_sync_write(struct bio *bio, int error) |
1da177e4 LT |
1558 | { |
1559 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 1560 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1561 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1562 | struct r1conf *conf = mddev->private; |
1da177e4 | 1563 | int mirror=0; |
4367af55 N |
1564 | sector_t first_bad; |
1565 | int bad_sectors; | |
1da177e4 | 1566 | |
ba3ae3be NK |
1567 | mirror = find_bio_disk(r1_bio, bio); |
1568 | ||
6b1117d5 | 1569 | if (!uptodate) { |
57dab0bd | 1570 | sector_t sync_blocks = 0; |
6b1117d5 N |
1571 | sector_t s = r1_bio->sector; |
1572 | long sectors_to_go = r1_bio->sectors; | |
1573 | /* make sure these bits doesn't get cleared. */ | |
1574 | do { | |
5e3db645 | 1575 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1576 | &sync_blocks, 1); |
1577 | s += sync_blocks; | |
1578 | sectors_to_go -= sync_blocks; | |
1579 | } while (sectors_to_go > 0); | |
d8f05d29 N |
1580 | set_bit(WriteErrorSeen, |
1581 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
1582 | if (!test_and_set_bit(WantReplacement, |
1583 | &conf->mirrors[mirror].rdev->flags)) | |
1584 | set_bit(MD_RECOVERY_NEEDED, & | |
1585 | mddev->recovery); | |
d8f05d29 | 1586 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 N |
1587 | } else if (is_badblock(conf->mirrors[mirror].rdev, |
1588 | r1_bio->sector, | |
1589 | r1_bio->sectors, | |
3a9f28a5 N |
1590 | &first_bad, &bad_sectors) && |
1591 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1592 | r1_bio->sector, | |
1593 | r1_bio->sectors, | |
1594 | &first_bad, &bad_sectors) | |
1595 | ) | |
4367af55 | 1596 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1597 | |
1da177e4 | 1598 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1599 | int s = r1_bio->sectors; |
d8f05d29 N |
1600 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1601 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1602 | reschedule_retry(r1_bio); |
1603 | else { | |
1604 | put_buf(r1_bio); | |
1605 | md_done_sync(mddev, s, uptodate); | |
1606 | } | |
1da177e4 | 1607 | } |
1da177e4 LT |
1608 | } |
1609 | ||
3cb03002 | 1610 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1611 | int sectors, struct page *page, int rw) |
1612 | { | |
1613 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
1614 | /* success */ | |
1615 | return 1; | |
19d67169 | 1616 | if (rw == WRITE) { |
d8f05d29 | 1617 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1618 | if (!test_and_set_bit(WantReplacement, |
1619 | &rdev->flags)) | |
1620 | set_bit(MD_RECOVERY_NEEDED, & | |
1621 | rdev->mddev->recovery); | |
1622 | } | |
d8f05d29 N |
1623 | /* need to record an error - either for the block or the device */ |
1624 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1625 | md_error(rdev->mddev, rdev); | |
1626 | return 0; | |
1627 | } | |
1628 | ||
9f2c9d12 | 1629 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1630 | { |
a68e5870 N |
1631 | /* Try some synchronous reads of other devices to get |
1632 | * good data, much like with normal read errors. Only | |
1633 | * read into the pages we already have so we don't | |
1634 | * need to re-issue the read request. | |
1635 | * We don't need to freeze the array, because being in an | |
1636 | * active sync request, there is no normal IO, and | |
1637 | * no overlapping syncs. | |
06f60385 N |
1638 | * We don't need to check is_badblock() again as we |
1639 | * made sure that anything with a bad block in range | |
1640 | * will have bi_end_io clear. | |
a68e5870 | 1641 | */ |
fd01b88c | 1642 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1643 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1644 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1645 | sector_t sect = r1_bio->sector; | |
1646 | int sectors = r1_bio->sectors; | |
1647 | int idx = 0; | |
1648 | ||
1649 | while(sectors) { | |
1650 | int s = sectors; | |
1651 | int d = r1_bio->read_disk; | |
1652 | int success = 0; | |
3cb03002 | 1653 | struct md_rdev *rdev; |
78d7f5f7 | 1654 | int start; |
a68e5870 N |
1655 | |
1656 | if (s > (PAGE_SIZE>>9)) | |
1657 | s = PAGE_SIZE >> 9; | |
1658 | do { | |
1659 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1660 | /* No rcu protection needed here devices | |
1661 | * can only be removed when no resync is | |
1662 | * active, and resync is currently active | |
1663 | */ | |
1664 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1665 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 N |
1666 | bio->bi_io_vec[idx].bv_page, |
1667 | READ, false)) { | |
1668 | success = 1; | |
1669 | break; | |
1670 | } | |
1671 | } | |
1672 | d++; | |
8f19ccb2 | 1673 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1674 | d = 0; |
1675 | } while (!success && d != r1_bio->read_disk); | |
1676 | ||
78d7f5f7 | 1677 | if (!success) { |
a68e5870 | 1678 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1679 | int abort = 0; |
1680 | /* Cannot read from anywhere, this block is lost. | |
1681 | * Record a bad block on each device. If that doesn't | |
1682 | * work just disable and interrupt the recovery. | |
1683 | * Don't fail devices as that won't really help. | |
1684 | */ | |
a68e5870 N |
1685 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
1686 | " for block %llu\n", | |
1687 | mdname(mddev), | |
1688 | bdevname(bio->bi_bdev, b), | |
1689 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1690 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1691 | rdev = conf->mirrors[d].rdev; |
1692 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1693 | continue; | |
1694 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1695 | abort = 1; | |
1696 | } | |
1697 | if (abort) { | |
d890fa2b N |
1698 | conf->recovery_disabled = |
1699 | mddev->recovery_disabled; | |
3a9f28a5 N |
1700 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1701 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1702 | put_buf(r1_bio); | |
1703 | return 0; | |
1704 | } | |
1705 | /* Try next page */ | |
1706 | sectors -= s; | |
1707 | sect += s; | |
1708 | idx++; | |
1709 | continue; | |
d11c171e | 1710 | } |
78d7f5f7 N |
1711 | |
1712 | start = d; | |
1713 | /* write it back and re-read */ | |
1714 | while (d != r1_bio->read_disk) { | |
1715 | if (d == 0) | |
8f19ccb2 | 1716 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1717 | d--; |
1718 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1719 | continue; | |
1720 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1721 | if (r1_sync_page_io(rdev, sect, s, |
1722 | bio->bi_io_vec[idx].bv_page, | |
1723 | WRITE) == 0) { | |
78d7f5f7 N |
1724 | r1_bio->bios[d]->bi_end_io = NULL; |
1725 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1726 | } |
78d7f5f7 N |
1727 | } |
1728 | d = start; | |
1729 | while (d != r1_bio->read_disk) { | |
1730 | if (d == 0) | |
8f19ccb2 | 1731 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1732 | d--; |
1733 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1734 | continue; | |
1735 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1736 | if (r1_sync_page_io(rdev, sect, s, |
1737 | bio->bi_io_vec[idx].bv_page, | |
1738 | READ) != 0) | |
9d3d8011 | 1739 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1740 | } |
a68e5870 N |
1741 | sectors -= s; |
1742 | sect += s; | |
1743 | idx ++; | |
1744 | } | |
78d7f5f7 | 1745 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
7ca78d57 | 1746 | set_bit(BIO_UPTODATE, &bio->bi_flags); |
a68e5870 N |
1747 | return 1; |
1748 | } | |
1749 | ||
9f2c9d12 | 1750 | static int process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1751 | { |
1752 | /* We have read all readable devices. If we haven't | |
1753 | * got the block, then there is no hope left. | |
1754 | * If we have, then we want to do a comparison | |
1755 | * and skip the write if everything is the same. | |
1756 | * If any blocks failed to read, then we need to | |
1757 | * attempt an over-write | |
1758 | */ | |
fd01b88c | 1759 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1760 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1761 | int primary; |
1762 | int i; | |
f4380a91 | 1763 | int vcnt; |
a68e5870 | 1764 | |
8f19ccb2 | 1765 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 N |
1766 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
1767 | test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { | |
1768 | r1_bio->bios[primary]->bi_end_io = NULL; | |
1769 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
1770 | break; | |
1771 | } | |
1772 | r1_bio->read_disk = primary; | |
f4380a91 | 1773 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); |
8f19ccb2 | 1774 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 1775 | int j; |
78d7f5f7 N |
1776 | struct bio *pbio = r1_bio->bios[primary]; |
1777 | struct bio *sbio = r1_bio->bios[i]; | |
1778 | int size; | |
a68e5870 | 1779 | |
78d7f5f7 N |
1780 | if (r1_bio->bios[i]->bi_end_io != end_sync_read) |
1781 | continue; | |
1782 | ||
1783 | if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { | |
1784 | for (j = vcnt; j-- ; ) { | |
1785 | struct page *p, *s; | |
1786 | p = pbio->bi_io_vec[j].bv_page; | |
1787 | s = sbio->bi_io_vec[j].bv_page; | |
1788 | if (memcmp(page_address(p), | |
1789 | page_address(s), | |
5020ad7d | 1790 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 1791 | break; |
69382e85 | 1792 | } |
78d7f5f7 N |
1793 | } else |
1794 | j = 0; | |
1795 | if (j >= 0) | |
1796 | mddev->resync_mismatches += r1_bio->sectors; | |
1797 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) | |
1798 | && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { | |
1799 | /* No need to write to this device. */ | |
1800 | sbio->bi_end_io = NULL; | |
1801 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
1802 | continue; | |
1803 | } | |
1804 | /* fixup the bio for reuse */ | |
1805 | sbio->bi_vcnt = vcnt; | |
1806 | sbio->bi_size = r1_bio->sectors << 9; | |
1807 | sbio->bi_idx = 0; | |
1808 | sbio->bi_phys_segments = 0; | |
1809 | sbio->bi_flags &= ~(BIO_POOL_MASK - 1); | |
1810 | sbio->bi_flags |= 1 << BIO_UPTODATE; | |
1811 | sbio->bi_next = NULL; | |
1812 | sbio->bi_sector = r1_bio->sector + | |
1813 | conf->mirrors[i].rdev->data_offset; | |
1814 | sbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1815 | size = sbio->bi_size; | |
1816 | for (j = 0; j < vcnt ; j++) { | |
1817 | struct bio_vec *bi; | |
1818 | bi = &sbio->bi_io_vec[j]; | |
1819 | bi->bv_offset = 0; | |
1820 | if (size > PAGE_SIZE) | |
1821 | bi->bv_len = PAGE_SIZE; | |
1822 | else | |
1823 | bi->bv_len = size; | |
1824 | size -= PAGE_SIZE; | |
1825 | memcpy(page_address(bi->bv_page), | |
1826 | page_address(pbio->bi_io_vec[j].bv_page), | |
1827 | PAGE_SIZE); | |
69382e85 | 1828 | } |
78d7f5f7 | 1829 | } |
a68e5870 N |
1830 | return 0; |
1831 | } | |
1832 | ||
9f2c9d12 | 1833 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 1834 | { |
e8096360 | 1835 | struct r1conf *conf = mddev->private; |
a68e5870 | 1836 | int i; |
8f19ccb2 | 1837 | int disks = conf->raid_disks * 2; |
a68e5870 N |
1838 | struct bio *bio, *wbio; |
1839 | ||
1840 | bio = r1_bio->bios[r1_bio->read_disk]; | |
1841 | ||
a68e5870 N |
1842 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
1843 | /* ouch - failed to read all of that. */ | |
1844 | if (!fix_sync_read_error(r1_bio)) | |
1845 | return; | |
7ca78d57 N |
1846 | |
1847 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
1848 | if (process_checks(r1_bio) < 0) | |
1849 | return; | |
d11c171e N |
1850 | /* |
1851 | * schedule writes | |
1852 | */ | |
1da177e4 LT |
1853 | atomic_set(&r1_bio->remaining, 1); |
1854 | for (i = 0; i < disks ; i++) { | |
1855 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
1856 | if (wbio->bi_end_io == NULL || |
1857 | (wbio->bi_end_io == end_sync_read && | |
1858 | (i == r1_bio->read_disk || | |
1859 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
1860 | continue; |
1861 | ||
3e198f78 N |
1862 | wbio->bi_rw = WRITE; |
1863 | wbio->bi_end_io = end_sync_write; | |
1da177e4 LT |
1864 | atomic_inc(&r1_bio->remaining); |
1865 | md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); | |
191ea9b2 | 1866 | |
1da177e4 LT |
1867 | generic_make_request(wbio); |
1868 | } | |
1869 | ||
1870 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 1871 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
1872 | int s = r1_bio->sectors; |
1873 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
1874 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
1875 | reschedule_retry(r1_bio); | |
1876 | else { | |
1877 | put_buf(r1_bio); | |
1878 | md_done_sync(mddev, s, 1); | |
1879 | } | |
1da177e4 LT |
1880 | } |
1881 | } | |
1882 | ||
1883 | /* | |
1884 | * This is a kernel thread which: | |
1885 | * | |
1886 | * 1. Retries failed read operations on working mirrors. | |
1887 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 1888 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
1889 | */ |
1890 | ||
e8096360 | 1891 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
1892 | sector_t sect, int sectors) |
1893 | { | |
fd01b88c | 1894 | struct mddev *mddev = conf->mddev; |
867868fb N |
1895 | while(sectors) { |
1896 | int s = sectors; | |
1897 | int d = read_disk; | |
1898 | int success = 0; | |
1899 | int start; | |
3cb03002 | 1900 | struct md_rdev *rdev; |
867868fb N |
1901 | |
1902 | if (s > (PAGE_SIZE>>9)) | |
1903 | s = PAGE_SIZE >> 9; | |
1904 | ||
1905 | do { | |
1906 | /* Note: no rcu protection needed here | |
1907 | * as this is synchronous in the raid1d thread | |
1908 | * which is the thread that might remove | |
1909 | * a device. If raid1d ever becomes multi-threaded.... | |
1910 | */ | |
d2eb35ac N |
1911 | sector_t first_bad; |
1912 | int bad_sectors; | |
1913 | ||
867868fb N |
1914 | rdev = conf->mirrors[d].rdev; |
1915 | if (rdev && | |
da8840a7 | 1916 | (test_bit(In_sync, &rdev->flags) || |
1917 | (!test_bit(Faulty, &rdev->flags) && | |
1918 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac N |
1919 | is_badblock(rdev, sect, s, |
1920 | &first_bad, &bad_sectors) == 0 && | |
ccebd4c4 JB |
1921 | sync_page_io(rdev, sect, s<<9, |
1922 | conf->tmppage, READ, false)) | |
867868fb N |
1923 | success = 1; |
1924 | else { | |
1925 | d++; | |
8f19ccb2 | 1926 | if (d == conf->raid_disks * 2) |
867868fb N |
1927 | d = 0; |
1928 | } | |
1929 | } while (!success && d != read_disk); | |
1930 | ||
1931 | if (!success) { | |
d8f05d29 | 1932 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 1933 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
1934 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
1935 | md_error(mddev, rdev); | |
867868fb N |
1936 | break; |
1937 | } | |
1938 | /* write it back and re-read */ | |
1939 | start = d; | |
1940 | while (d != read_disk) { | |
1941 | if (d==0) | |
8f19ccb2 | 1942 | d = conf->raid_disks * 2; |
867868fb N |
1943 | d--; |
1944 | rdev = conf->mirrors[d].rdev; | |
1945 | if (rdev && | |
d8f05d29 N |
1946 | test_bit(In_sync, &rdev->flags)) |
1947 | r1_sync_page_io(rdev, sect, s, | |
1948 | conf->tmppage, WRITE); | |
867868fb N |
1949 | } |
1950 | d = start; | |
1951 | while (d != read_disk) { | |
1952 | char b[BDEVNAME_SIZE]; | |
1953 | if (d==0) | |
8f19ccb2 | 1954 | d = conf->raid_disks * 2; |
867868fb N |
1955 | d--; |
1956 | rdev = conf->mirrors[d].rdev; | |
1957 | if (rdev && | |
1958 | test_bit(In_sync, &rdev->flags)) { | |
d8f05d29 N |
1959 | if (r1_sync_page_io(rdev, sect, s, |
1960 | conf->tmppage, READ)) { | |
867868fb N |
1961 | atomic_add(s, &rdev->corrected_errors); |
1962 | printk(KERN_INFO | |
9dd1e2fa | 1963 | "md/raid1:%s: read error corrected " |
867868fb N |
1964 | "(%d sectors at %llu on %s)\n", |
1965 | mdname(mddev), s, | |
969b755a RD |
1966 | (unsigned long long)(sect + |
1967 | rdev->data_offset), | |
867868fb N |
1968 | bdevname(rdev->bdev, b)); |
1969 | } | |
1970 | } | |
1971 | } | |
1972 | sectors -= s; | |
1973 | sect += s; | |
1974 | } | |
1975 | } | |
1976 | ||
cd5ff9a1 N |
1977 | static void bi_complete(struct bio *bio, int error) |
1978 | { | |
1979 | complete((struct completion *)bio->bi_private); | |
1980 | } | |
1981 | ||
1982 | static int submit_bio_wait(int rw, struct bio *bio) | |
1983 | { | |
1984 | struct completion event; | |
1985 | rw |= REQ_SYNC; | |
1986 | ||
1987 | init_completion(&event); | |
1988 | bio->bi_private = &event; | |
1989 | bio->bi_end_io = bi_complete; | |
1990 | submit_bio(rw, bio); | |
1991 | wait_for_completion(&event); | |
1992 | ||
1993 | return test_bit(BIO_UPTODATE, &bio->bi_flags); | |
1994 | } | |
1995 | ||
9f2c9d12 | 1996 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 1997 | { |
fd01b88c | 1998 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1999 | struct r1conf *conf = mddev->private; |
3cb03002 | 2000 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2001 | int vcnt, idx; |
2002 | struct bio_vec *vec; | |
2003 | ||
2004 | /* bio has the data to be written to device 'i' where | |
2005 | * we just recently had a write error. | |
2006 | * We repeatedly clone the bio and trim down to one block, | |
2007 | * then try the write. Where the write fails we record | |
2008 | * a bad block. | |
2009 | * It is conceivable that the bio doesn't exactly align with | |
2010 | * blocks. We must handle this somehow. | |
2011 | * | |
2012 | * We currently own a reference on the rdev. | |
2013 | */ | |
2014 | ||
2015 | int block_sectors; | |
2016 | sector_t sector; | |
2017 | int sectors; | |
2018 | int sect_to_write = r1_bio->sectors; | |
2019 | int ok = 1; | |
2020 | ||
2021 | if (rdev->badblocks.shift < 0) | |
2022 | return 0; | |
2023 | ||
2024 | block_sectors = 1 << rdev->badblocks.shift; | |
2025 | sector = r1_bio->sector; | |
2026 | sectors = ((sector + block_sectors) | |
2027 | & ~(sector_t)(block_sectors - 1)) | |
2028 | - sector; | |
2029 | ||
2030 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
2031 | vcnt = r1_bio->behind_page_count; | |
2032 | vec = r1_bio->behind_bvecs; | |
2033 | idx = 0; | |
2034 | while (vec[idx].bv_page == NULL) | |
2035 | idx++; | |
2036 | } else { | |
2037 | vcnt = r1_bio->master_bio->bi_vcnt; | |
2038 | vec = r1_bio->master_bio->bi_io_vec; | |
2039 | idx = r1_bio->master_bio->bi_idx; | |
2040 | } | |
2041 | while (sect_to_write) { | |
2042 | struct bio *wbio; | |
2043 | if (sectors > sect_to_write) | |
2044 | sectors = sect_to_write; | |
2045 | /* Write at 'sector' for 'sectors'*/ | |
2046 | ||
2047 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2048 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2049 | wbio->bi_sector = r1_bio->sector; | |
2050 | wbio->bi_rw = WRITE; | |
2051 | wbio->bi_vcnt = vcnt; | |
2052 | wbio->bi_size = r1_bio->sectors << 9; | |
2053 | wbio->bi_idx = idx; | |
2054 | ||
2055 | md_trim_bio(wbio, sector - r1_bio->sector, sectors); | |
2056 | wbio->bi_sector += rdev->data_offset; | |
2057 | wbio->bi_bdev = rdev->bdev; | |
2058 | if (submit_bio_wait(WRITE, wbio) == 0) | |
2059 | /* failure! */ | |
2060 | ok = rdev_set_badblocks(rdev, sector, | |
2061 | sectors, 0) | |
2062 | && ok; | |
2063 | ||
2064 | bio_put(wbio); | |
2065 | sect_to_write -= sectors; | |
2066 | sector += sectors; | |
2067 | sectors = block_sectors; | |
2068 | } | |
2069 | return ok; | |
2070 | } | |
2071 | ||
e8096360 | 2072 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2073 | { |
2074 | int m; | |
2075 | int s = r1_bio->sectors; | |
8f19ccb2 | 2076 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2077 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2078 | struct bio *bio = r1_bio->bios[m]; |
2079 | if (bio->bi_end_io == NULL) | |
2080 | continue; | |
2081 | if (test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
2082 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { | |
c6563a8c | 2083 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce N |
2084 | } |
2085 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
2086 | test_bit(R1BIO_WriteError, &r1_bio->state)) { | |
2087 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2088 | md_error(conf->mddev, rdev); | |
2089 | } | |
2090 | } | |
2091 | put_buf(r1_bio); | |
2092 | md_done_sync(conf->mddev, s, 1); | |
2093 | } | |
2094 | ||
e8096360 | 2095 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2096 | { |
2097 | int m; | |
8f19ccb2 | 2098 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2099 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2100 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2101 | rdev_clear_badblocks(rdev, |
2102 | r1_bio->sector, | |
c6563a8c | 2103 | r1_bio->sectors, 0); |
62096bce N |
2104 | rdev_dec_pending(rdev, conf->mddev); |
2105 | } else if (r1_bio->bios[m] != NULL) { | |
2106 | /* This drive got a write error. We need to | |
2107 | * narrow down and record precise write | |
2108 | * errors. | |
2109 | */ | |
2110 | if (!narrow_write_error(r1_bio, m)) { | |
2111 | md_error(conf->mddev, | |
2112 | conf->mirrors[m].rdev); | |
2113 | /* an I/O failed, we can't clear the bitmap */ | |
2114 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2115 | } | |
2116 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2117 | conf->mddev); | |
2118 | } | |
2119 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2120 | close_write(r1_bio); | |
2121 | raid_end_bio_io(r1_bio); | |
2122 | } | |
2123 | ||
e8096360 | 2124 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2125 | { |
2126 | int disk; | |
2127 | int max_sectors; | |
fd01b88c | 2128 | struct mddev *mddev = conf->mddev; |
62096bce N |
2129 | struct bio *bio; |
2130 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2131 | struct md_rdev *rdev; |
62096bce N |
2132 | |
2133 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2134 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2135 | * the block and we can fix it. | |
2136 | * We freeze all other IO, and try reading the block from | |
2137 | * other devices. When we find one, we re-write | |
2138 | * and check it that fixes the read error. | |
2139 | * This is all done synchronously while the array is | |
2140 | * frozen | |
2141 | */ | |
2142 | if (mddev->ro == 0) { | |
2143 | freeze_array(conf); | |
2144 | fix_read_error(conf, r1_bio->read_disk, | |
2145 | r1_bio->sector, r1_bio->sectors); | |
2146 | unfreeze_array(conf); | |
2147 | } else | |
2148 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); | |
2149 | ||
2150 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2151 | bdevname(bio->bi_bdev, b); | |
2152 | read_more: | |
2153 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2154 | if (disk == -1) { | |
2155 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" | |
2156 | " read error for block %llu\n", | |
2157 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
2158 | raid_end_bio_io(r1_bio); | |
2159 | } else { | |
2160 | const unsigned long do_sync | |
2161 | = r1_bio->master_bio->bi_rw & REQ_SYNC; | |
2162 | if (bio) { | |
2163 | r1_bio->bios[r1_bio->read_disk] = | |
2164 | mddev->ro ? IO_BLOCKED : NULL; | |
2165 | bio_put(bio); | |
2166 | } | |
2167 | r1_bio->read_disk = disk; | |
2168 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2169 | md_trim_bio(bio, r1_bio->sector - bio->bi_sector, max_sectors); | |
2170 | r1_bio->bios[r1_bio->read_disk] = bio; | |
2171 | rdev = conf->mirrors[disk].rdev; | |
2172 | printk_ratelimited(KERN_ERR | |
2173 | "md/raid1:%s: redirecting sector %llu" | |
2174 | " to other mirror: %s\n", | |
2175 | mdname(mddev), | |
2176 | (unsigned long long)r1_bio->sector, | |
2177 | bdevname(rdev->bdev, b)); | |
2178 | bio->bi_sector = r1_bio->sector + rdev->data_offset; | |
2179 | bio->bi_bdev = rdev->bdev; | |
2180 | bio->bi_end_io = raid1_end_read_request; | |
2181 | bio->bi_rw = READ | do_sync; | |
2182 | bio->bi_private = r1_bio; | |
2183 | if (max_sectors < r1_bio->sectors) { | |
2184 | /* Drat - have to split this up more */ | |
2185 | struct bio *mbio = r1_bio->master_bio; | |
2186 | int sectors_handled = (r1_bio->sector + max_sectors | |
2187 | - mbio->bi_sector); | |
2188 | r1_bio->sectors = max_sectors; | |
2189 | spin_lock_irq(&conf->device_lock); | |
2190 | if (mbio->bi_phys_segments == 0) | |
2191 | mbio->bi_phys_segments = 2; | |
2192 | else | |
2193 | mbio->bi_phys_segments++; | |
2194 | spin_unlock_irq(&conf->device_lock); | |
2195 | generic_make_request(bio); | |
2196 | bio = NULL; | |
2197 | ||
2198 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2199 | ||
2200 | r1_bio->master_bio = mbio; | |
2201 | r1_bio->sectors = (mbio->bi_size >> 9) | |
2202 | - sectors_handled; | |
2203 | r1_bio->state = 0; | |
2204 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2205 | r1_bio->mddev = mddev; | |
2206 | r1_bio->sector = mbio->bi_sector + sectors_handled; | |
2207 | ||
2208 | goto read_more; | |
2209 | } else | |
2210 | generic_make_request(bio); | |
2211 | } | |
2212 | } | |
2213 | ||
fd01b88c | 2214 | static void raid1d(struct mddev *mddev) |
1da177e4 | 2215 | { |
9f2c9d12 | 2216 | struct r1bio *r1_bio; |
1da177e4 | 2217 | unsigned long flags; |
e8096360 | 2218 | struct r1conf *conf = mddev->private; |
1da177e4 | 2219 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2220 | struct blk_plug plug; |
1da177e4 LT |
2221 | |
2222 | md_check_recovery(mddev); | |
e1dfa0a2 N |
2223 | |
2224 | blk_start_plug(&plug); | |
1da177e4 | 2225 | for (;;) { |
191ea9b2 | 2226 | |
0021b7bc | 2227 | flush_pending_writes(conf); |
191ea9b2 | 2228 | |
a35e63ef N |
2229 | spin_lock_irqsave(&conf->device_lock, flags); |
2230 | if (list_empty(head)) { | |
2231 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2232 | break; |
a35e63ef | 2233 | } |
9f2c9d12 | 2234 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2235 | list_del(head->prev); |
ddaf22ab | 2236 | conf->nr_queued--; |
1da177e4 LT |
2237 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2238 | ||
2239 | mddev = r1_bio->mddev; | |
070ec55d | 2240 | conf = mddev->private; |
4367af55 | 2241 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2242 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2243 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2244 | handle_sync_write_finished(conf, r1_bio); | |
2245 | else | |
4367af55 | 2246 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2247 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2248 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2249 | handle_write_finished(conf, r1_bio); | |
2250 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2251 | handle_read_error(conf, r1_bio); | |
2252 | else | |
d2eb35ac N |
2253 | /* just a partial read to be scheduled from separate |
2254 | * context | |
2255 | */ | |
2256 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2257 | |
1d9d5241 | 2258 | cond_resched(); |
de393cde N |
2259 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2260 | md_check_recovery(mddev); | |
1da177e4 | 2261 | } |
e1dfa0a2 | 2262 | blk_finish_plug(&plug); |
1da177e4 LT |
2263 | } |
2264 | ||
2265 | ||
e8096360 | 2266 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2267 | { |
2268 | int buffs; | |
2269 | ||
2270 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2271 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2272 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2273 | conf->poolinfo); | |
2274 | if (!conf->r1buf_pool) | |
2275 | return -ENOMEM; | |
2276 | conf->next_resync = 0; | |
2277 | return 0; | |
2278 | } | |
2279 | ||
2280 | /* | |
2281 | * perform a "sync" on one "block" | |
2282 | * | |
2283 | * We need to make sure that no normal I/O request - particularly write | |
2284 | * requests - conflict with active sync requests. | |
2285 | * | |
2286 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2287 | * that can be installed to exclude normal IO requests. | |
2288 | */ | |
2289 | ||
fd01b88c | 2290 | static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 | 2291 | { |
e8096360 | 2292 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2293 | struct r1bio *r1_bio; |
1da177e4 LT |
2294 | struct bio *bio; |
2295 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2296 | int disk = -1; |
1da177e4 | 2297 | int i; |
3e198f78 N |
2298 | int wonly = -1; |
2299 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2300 | sector_t sync_blocks; |
e3b9703e | 2301 | int still_degraded = 0; |
06f60385 N |
2302 | int good_sectors = RESYNC_SECTORS; |
2303 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2304 | |
2305 | if (!conf->r1buf_pool) | |
2306 | if (init_resync(conf)) | |
57afd89f | 2307 | return 0; |
1da177e4 | 2308 | |
58c0fed4 | 2309 | max_sector = mddev->dev_sectors; |
1da177e4 | 2310 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2311 | /* If we aborted, we need to abort the |
2312 | * sync on the 'current' bitmap chunk (there will | |
2313 | * only be one in raid1 resync. | |
2314 | * We can find the current addess in mddev->curr_resync | |
2315 | */ | |
6a806c51 N |
2316 | if (mddev->curr_resync < max_sector) /* aborted */ |
2317 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2318 | &sync_blocks, 1); |
6a806c51 | 2319 | else /* completed sync */ |
191ea9b2 | 2320 | conf->fullsync = 0; |
6a806c51 N |
2321 | |
2322 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 LT |
2323 | close_sync(conf); |
2324 | return 0; | |
2325 | } | |
2326 | ||
07d84d10 N |
2327 | if (mddev->bitmap == NULL && |
2328 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2329 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2330 | conf->fullsync == 0) { |
2331 | *skipped = 1; | |
2332 | return max_sector - sector_nr; | |
2333 | } | |
6394cca5 N |
2334 | /* before building a request, check if we can skip these blocks.. |
2335 | * This call the bitmap_start_sync doesn't actually record anything | |
2336 | */ | |
e3b9703e | 2337 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2338 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2339 | /* We can skip this block, and probably several more */ |
2340 | *skipped = 1; | |
2341 | return sync_blocks; | |
2342 | } | |
1da177e4 | 2343 | /* |
17999be4 N |
2344 | * If there is non-resync activity waiting for a turn, |
2345 | * and resync is going fast enough, | |
2346 | * then let it though before starting on this new sync request. | |
1da177e4 | 2347 | */ |
17999be4 | 2348 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 2349 | msleep_interruptible(1000); |
17999be4 | 2350 | |
b47490c9 | 2351 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
1c4588e9 | 2352 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 N |
2353 | raise_barrier(conf); |
2354 | ||
2355 | conf->next_resync = sector_nr; | |
1da177e4 | 2356 | |
3e198f78 | 2357 | rcu_read_lock(); |
1da177e4 | 2358 | /* |
3e198f78 N |
2359 | * If we get a correctably read error during resync or recovery, |
2360 | * we might want to read from a different device. So we | |
2361 | * flag all drives that could conceivably be read from for READ, | |
2362 | * and any others (which will be non-In_sync devices) for WRITE. | |
2363 | * If a read fails, we try reading from something else for which READ | |
2364 | * is OK. | |
1da177e4 | 2365 | */ |
1da177e4 | 2366 | |
1da177e4 LT |
2367 | r1_bio->mddev = mddev; |
2368 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2369 | r1_bio->state = 0; |
1da177e4 | 2370 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2371 | |
8f19ccb2 | 2372 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2373 | struct md_rdev *rdev; |
1da177e4 LT |
2374 | bio = r1_bio->bios[i]; |
2375 | ||
2376 | /* take from bio_init */ | |
2377 | bio->bi_next = NULL; | |
db8d9d35 | 2378 | bio->bi_flags &= ~(BIO_POOL_MASK-1); |
1da177e4 | 2379 | bio->bi_flags |= 1 << BIO_UPTODATE; |
802ba064 | 2380 | bio->bi_rw = READ; |
1da177e4 LT |
2381 | bio->bi_vcnt = 0; |
2382 | bio->bi_idx = 0; | |
2383 | bio->bi_phys_segments = 0; | |
1da177e4 LT |
2384 | bio->bi_size = 0; |
2385 | bio->bi_end_io = NULL; | |
2386 | bio->bi_private = NULL; | |
2387 | ||
3e198f78 N |
2388 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2389 | if (rdev == NULL || | |
06f60385 | 2390 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2391 | if (i < conf->raid_disks) |
2392 | still_degraded = 1; | |
3e198f78 | 2393 | } else if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
2394 | bio->bi_rw = WRITE; |
2395 | bio->bi_end_io = end_sync_write; | |
2396 | write_targets ++; | |
3e198f78 N |
2397 | } else { |
2398 | /* may need to read from here */ | |
06f60385 N |
2399 | sector_t first_bad = MaxSector; |
2400 | int bad_sectors; | |
2401 | ||
2402 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2403 | &first_bad, &bad_sectors)) { | |
2404 | if (first_bad > sector_nr) | |
2405 | good_sectors = first_bad - sector_nr; | |
2406 | else { | |
2407 | bad_sectors -= (sector_nr - first_bad); | |
2408 | if (min_bad == 0 || | |
2409 | min_bad > bad_sectors) | |
2410 | min_bad = bad_sectors; | |
2411 | } | |
2412 | } | |
2413 | if (sector_nr < first_bad) { | |
2414 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2415 | if (wonly < 0) | |
2416 | wonly = i; | |
2417 | } else { | |
2418 | if (disk < 0) | |
2419 | disk = i; | |
2420 | } | |
2421 | bio->bi_rw = READ; | |
2422 | bio->bi_end_io = end_sync_read; | |
2423 | read_targets++; | |
3e198f78 | 2424 | } |
3e198f78 | 2425 | } |
06f60385 N |
2426 | if (bio->bi_end_io) { |
2427 | atomic_inc(&rdev->nr_pending); | |
2428 | bio->bi_sector = sector_nr + rdev->data_offset; | |
2429 | bio->bi_bdev = rdev->bdev; | |
2430 | bio->bi_private = r1_bio; | |
2431 | } | |
1da177e4 | 2432 | } |
3e198f78 N |
2433 | rcu_read_unlock(); |
2434 | if (disk < 0) | |
2435 | disk = wonly; | |
2436 | r1_bio->read_disk = disk; | |
191ea9b2 | 2437 | |
06f60385 N |
2438 | if (read_targets == 0 && min_bad > 0) { |
2439 | /* These sectors are bad on all InSync devices, so we | |
2440 | * need to mark them bad on all write targets | |
2441 | */ | |
2442 | int ok = 1; | |
8f19ccb2 | 2443 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2444 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2445 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2446 | ok = rdev_set_badblocks(rdev, sector_nr, |
2447 | min_bad, 0 | |
2448 | ) && ok; | |
2449 | } | |
2450 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2451 | *skipped = 1; | |
2452 | put_buf(r1_bio); | |
2453 | ||
2454 | if (!ok) { | |
2455 | /* Cannot record the badblocks, so need to | |
2456 | * abort the resync. | |
2457 | * If there are multiple read targets, could just | |
2458 | * fail the really bad ones ??? | |
2459 | */ | |
2460 | conf->recovery_disabled = mddev->recovery_disabled; | |
2461 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2462 | return 0; | |
2463 | } else | |
2464 | return min_bad; | |
2465 | ||
2466 | } | |
2467 | if (min_bad > 0 && min_bad < good_sectors) { | |
2468 | /* only resync enough to reach the next bad->good | |
2469 | * transition */ | |
2470 | good_sectors = min_bad; | |
2471 | } | |
2472 | ||
3e198f78 N |
2473 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2474 | /* extra read targets are also write targets */ | |
2475 | write_targets += read_targets-1; | |
2476 | ||
2477 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2478 | /* There is nowhere to write, so all non-sync |
2479 | * drives must be failed - so we are finished | |
2480 | */ | |
57afd89f N |
2481 | sector_t rv = max_sector - sector_nr; |
2482 | *skipped = 1; | |
1da177e4 | 2483 | put_buf(r1_bio); |
1da177e4 LT |
2484 | return rv; |
2485 | } | |
2486 | ||
c6207277 N |
2487 | if (max_sector > mddev->resync_max) |
2488 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2489 | if (max_sector > sector_nr + good_sectors) |
2490 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2491 | nr_sectors = 0; |
289e99e8 | 2492 | sync_blocks = 0; |
1da177e4 LT |
2493 | do { |
2494 | struct page *page; | |
2495 | int len = PAGE_SIZE; | |
2496 | if (sector_nr + (len>>9) > max_sector) | |
2497 | len = (max_sector - sector_nr) << 9; | |
2498 | if (len == 0) | |
2499 | break; | |
6a806c51 N |
2500 | if (sync_blocks == 0) { |
2501 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2502 | &sync_blocks, still_degraded) && |
2503 | !conf->fullsync && | |
2504 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2505 | break; |
9e77c485 | 2506 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
7571ae88 | 2507 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2508 | len = sync_blocks<<9; |
ab7a30c7 | 2509 | } |
191ea9b2 | 2510 | |
8f19ccb2 | 2511 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2512 | bio = r1_bio->bios[i]; |
2513 | if (bio->bi_end_io) { | |
d11c171e | 2514 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2515 | if (bio_add_page(bio, page, len, 0) == 0) { |
2516 | /* stop here */ | |
d11c171e | 2517 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2518 | while (i > 0) { |
2519 | i--; | |
2520 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2521 | if (bio->bi_end_io==NULL) |
2522 | continue; | |
1da177e4 LT |
2523 | /* remove last page from this bio */ |
2524 | bio->bi_vcnt--; | |
2525 | bio->bi_size -= len; | |
2526 | bio->bi_flags &= ~(1<< BIO_SEG_VALID); | |
2527 | } | |
2528 | goto bio_full; | |
2529 | } | |
2530 | } | |
2531 | } | |
2532 | nr_sectors += len>>9; | |
2533 | sector_nr += len>>9; | |
191ea9b2 | 2534 | sync_blocks -= (len>>9); |
1da177e4 LT |
2535 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2536 | bio_full: | |
1da177e4 LT |
2537 | r1_bio->sectors = nr_sectors; |
2538 | ||
d11c171e N |
2539 | /* For a user-requested sync, we read all readable devices and do a |
2540 | * compare | |
2541 | */ | |
2542 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2543 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2544 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2545 | bio = r1_bio->bios[i]; |
2546 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2547 | read_targets--; |
ddac7c7e | 2548 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
2549 | generic_make_request(bio); |
2550 | } | |
2551 | } | |
2552 | } else { | |
2553 | atomic_set(&r1_bio->remaining, 1); | |
2554 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2555 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 2556 | generic_make_request(bio); |
1da177e4 | 2557 | |
d11c171e | 2558 | } |
1da177e4 LT |
2559 | return nr_sectors; |
2560 | } | |
2561 | ||
fd01b88c | 2562 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2563 | { |
2564 | if (sectors) | |
2565 | return sectors; | |
2566 | ||
2567 | return mddev->dev_sectors; | |
2568 | } | |
2569 | ||
e8096360 | 2570 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2571 | { |
e8096360 | 2572 | struct r1conf *conf; |
709ae487 | 2573 | int i; |
0f6d02d5 | 2574 | struct mirror_info *disk; |
3cb03002 | 2575 | struct md_rdev *rdev; |
709ae487 | 2576 | int err = -ENOMEM; |
1da177e4 | 2577 | |
e8096360 | 2578 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2579 | if (!conf) |
709ae487 | 2580 | goto abort; |
1da177e4 | 2581 | |
8f19ccb2 N |
2582 | conf->mirrors = kzalloc(sizeof(struct mirror_info) |
2583 | * mddev->raid_disks * 2, | |
1da177e4 LT |
2584 | GFP_KERNEL); |
2585 | if (!conf->mirrors) | |
709ae487 | 2586 | goto abort; |
1da177e4 | 2587 | |
ddaf22ab N |
2588 | conf->tmppage = alloc_page(GFP_KERNEL); |
2589 | if (!conf->tmppage) | |
709ae487 | 2590 | goto abort; |
ddaf22ab | 2591 | |
709ae487 | 2592 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2593 | if (!conf->poolinfo) |
709ae487 | 2594 | goto abort; |
8f19ccb2 | 2595 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2596 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2597 | r1bio_pool_free, | |
2598 | conf->poolinfo); | |
2599 | if (!conf->r1bio_pool) | |
709ae487 N |
2600 | goto abort; |
2601 | ||
ed9bfdf1 | 2602 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2603 | |
c19d5798 | 2604 | err = -EINVAL; |
e7e72bf6 | 2605 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2606 | rdev_for_each(rdev, mddev) { |
aba336bd | 2607 | struct request_queue *q; |
709ae487 | 2608 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2609 | if (disk_idx >= mddev->raid_disks |
2610 | || disk_idx < 0) | |
2611 | continue; | |
c19d5798 N |
2612 | if (test_bit(Replacement, &rdev->flags)) |
2613 | disk = conf->mirrors + conf->raid_disks + disk_idx; | |
2614 | else | |
2615 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2616 | |
c19d5798 N |
2617 | if (disk->rdev) |
2618 | goto abort; | |
1da177e4 | 2619 | disk->rdev = rdev; |
aba336bd N |
2620 | q = bdev_get_queue(rdev->bdev); |
2621 | if (q->merge_bvec_fn) | |
2622 | mddev->merge_check_needed = 1; | |
1da177e4 LT |
2623 | |
2624 | disk->head_position = 0; | |
1da177e4 LT |
2625 | } |
2626 | conf->raid_disks = mddev->raid_disks; | |
2627 | conf->mddev = mddev; | |
1da177e4 | 2628 | INIT_LIST_HEAD(&conf->retry_list); |
1da177e4 LT |
2629 | |
2630 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2631 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2632 | |
191ea9b2 | 2633 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2634 | conf->pending_count = 0; |
d890fa2b | 2635 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2636 | |
c19d5798 | 2637 | err = -EIO; |
709ae487 | 2638 | conf->last_used = -1; |
8f19ccb2 | 2639 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2640 | |
2641 | disk = conf->mirrors + i; | |
2642 | ||
c19d5798 N |
2643 | if (i < conf->raid_disks && |
2644 | disk[conf->raid_disks].rdev) { | |
2645 | /* This slot has a replacement. */ | |
2646 | if (!disk->rdev) { | |
2647 | /* No original, just make the replacement | |
2648 | * a recovering spare | |
2649 | */ | |
2650 | disk->rdev = | |
2651 | disk[conf->raid_disks].rdev; | |
2652 | disk[conf->raid_disks].rdev = NULL; | |
2653 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2654 | /* Original is not in_sync - bad */ | |
2655 | goto abort; | |
2656 | } | |
2657 | ||
5fd6c1dc N |
2658 | if (!disk->rdev || |
2659 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2660 | disk->head_position = 0; |
4f0a5e01 JB |
2661 | if (disk->rdev && |
2662 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2663 | conf->fullsync = 1; |
709ae487 N |
2664 | } else if (conf->last_used < 0) |
2665 | /* | |
2666 | * The first working device is used as a | |
2667 | * starting point to read balancing. | |
2668 | */ | |
2669 | conf->last_used = i; | |
1da177e4 | 2670 | } |
709ae487 | 2671 | |
709ae487 | 2672 | if (conf->last_used < 0) { |
9dd1e2fa | 2673 | printk(KERN_ERR "md/raid1:%s: no operational mirrors\n", |
709ae487 N |
2674 | mdname(mddev)); |
2675 | goto abort; | |
2676 | } | |
2677 | err = -ENOMEM; | |
0232605d | 2678 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
709ae487 N |
2679 | if (!conf->thread) { |
2680 | printk(KERN_ERR | |
9dd1e2fa | 2681 | "md/raid1:%s: couldn't allocate thread\n", |
709ae487 N |
2682 | mdname(mddev)); |
2683 | goto abort; | |
11ce99e6 | 2684 | } |
1da177e4 | 2685 | |
709ae487 N |
2686 | return conf; |
2687 | ||
2688 | abort: | |
2689 | if (conf) { | |
2690 | if (conf->r1bio_pool) | |
2691 | mempool_destroy(conf->r1bio_pool); | |
2692 | kfree(conf->mirrors); | |
2693 | safe_put_page(conf->tmppage); | |
2694 | kfree(conf->poolinfo); | |
2695 | kfree(conf); | |
2696 | } | |
2697 | return ERR_PTR(err); | |
2698 | } | |
2699 | ||
5220ea1e | 2700 | static int stop(struct mddev *mddev); |
fd01b88c | 2701 | static int run(struct mddev *mddev) |
709ae487 | 2702 | { |
e8096360 | 2703 | struct r1conf *conf; |
709ae487 | 2704 | int i; |
3cb03002 | 2705 | struct md_rdev *rdev; |
5220ea1e | 2706 | int ret; |
709ae487 N |
2707 | |
2708 | if (mddev->level != 1) { | |
9dd1e2fa | 2709 | printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", |
709ae487 N |
2710 | mdname(mddev), mddev->level); |
2711 | return -EIO; | |
2712 | } | |
2713 | if (mddev->reshape_position != MaxSector) { | |
9dd1e2fa | 2714 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
709ae487 N |
2715 | mdname(mddev)); |
2716 | return -EIO; | |
2717 | } | |
1da177e4 | 2718 | /* |
709ae487 N |
2719 | * copy the already verified devices into our private RAID1 |
2720 | * bookkeeping area. [whatever we allocate in run(), | |
2721 | * should be freed in stop()] | |
1da177e4 | 2722 | */ |
709ae487 N |
2723 | if (mddev->private == NULL) |
2724 | conf = setup_conf(mddev); | |
2725 | else | |
2726 | conf = mddev->private; | |
1da177e4 | 2727 | |
709ae487 N |
2728 | if (IS_ERR(conf)) |
2729 | return PTR_ERR(conf); | |
1da177e4 | 2730 | |
dafb20fa | 2731 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
2732 | if (!mddev->gendisk) |
2733 | continue; | |
709ae487 N |
2734 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2735 | rdev->data_offset << 9); | |
1da177e4 | 2736 | } |
191ea9b2 | 2737 | |
709ae487 N |
2738 | mddev->degraded = 0; |
2739 | for (i=0; i < conf->raid_disks; i++) | |
2740 | if (conf->mirrors[i].rdev == NULL || | |
2741 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
2742 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
2743 | mddev->degraded++; | |
2744 | ||
2745 | if (conf->raid_disks - mddev->degraded == 1) | |
2746 | mddev->recovery_cp = MaxSector; | |
2747 | ||
8c6ac868 | 2748 | if (mddev->recovery_cp != MaxSector) |
9dd1e2fa | 2749 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
8c6ac868 AN |
2750 | " -- starting background reconstruction\n", |
2751 | mdname(mddev)); | |
1da177e4 | 2752 | printk(KERN_INFO |
9dd1e2fa | 2753 | "md/raid1:%s: active with %d out of %d mirrors\n", |
1da177e4 LT |
2754 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
2755 | mddev->raid_disks); | |
709ae487 | 2756 | |
1da177e4 LT |
2757 | /* |
2758 | * Ok, everything is just fine now | |
2759 | */ | |
709ae487 N |
2760 | mddev->thread = conf->thread; |
2761 | conf->thread = NULL; | |
2762 | mddev->private = conf; | |
2763 | ||
1f403624 | 2764 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2765 | |
1ed7242e JB |
2766 | if (mddev->queue) { |
2767 | mddev->queue->backing_dev_info.congested_fn = raid1_congested; | |
2768 | mddev->queue->backing_dev_info.congested_data = mddev; | |
6b740b8d | 2769 | blk_queue_merge_bvec(mddev->queue, raid1_mergeable_bvec); |
1ed7242e | 2770 | } |
5220ea1e | 2771 | |
2772 | ret = md_integrity_register(mddev); | |
2773 | if (ret) | |
2774 | stop(mddev); | |
2775 | return ret; | |
1da177e4 LT |
2776 | } |
2777 | ||
fd01b88c | 2778 | static int stop(struct mddev *mddev) |
1da177e4 | 2779 | { |
e8096360 | 2780 | struct r1conf *conf = mddev->private; |
4b6d287f | 2781 | struct bitmap *bitmap = mddev->bitmap; |
4b6d287f N |
2782 | |
2783 | /* wait for behind writes to complete */ | |
e555190d | 2784 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { |
9dd1e2fa N |
2785 | printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n", |
2786 | mdname(mddev)); | |
4b6d287f | 2787 | /* need to kick something here to make sure I/O goes? */ |
e555190d N |
2788 | wait_event(bitmap->behind_wait, |
2789 | atomic_read(&bitmap->behind_writes) == 0); | |
4b6d287f | 2790 | } |
1da177e4 | 2791 | |
409c57f3 N |
2792 | raise_barrier(conf); |
2793 | lower_barrier(conf); | |
2794 | ||
01f96c0a | 2795 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
2796 | if (conf->r1bio_pool) |
2797 | mempool_destroy(conf->r1bio_pool); | |
990a8baf JJ |
2798 | kfree(conf->mirrors); |
2799 | kfree(conf->poolinfo); | |
1da177e4 LT |
2800 | kfree(conf); |
2801 | mddev->private = NULL; | |
2802 | return 0; | |
2803 | } | |
2804 | ||
fd01b88c | 2805 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
2806 | { |
2807 | /* no resync is happening, and there is enough space | |
2808 | * on all devices, so we can resize. | |
2809 | * We need to make sure resync covers any new space. | |
2810 | * If the array is shrinking we should possibly wait until | |
2811 | * any io in the removed space completes, but it hardly seems | |
2812 | * worth it. | |
2813 | */ | |
a4a6125a N |
2814 | sector_t newsize = raid1_size(mddev, sectors, 0); |
2815 | if (mddev->external_size && | |
2816 | mddev->array_sectors > newsize) | |
b522adcd | 2817 | return -EINVAL; |
a4a6125a N |
2818 | if (mddev->bitmap) { |
2819 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
2820 | if (ret) | |
2821 | return ret; | |
2822 | } | |
2823 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 2824 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 2825 | revalidate_disk(mddev->gendisk); |
b522adcd | 2826 | if (sectors > mddev->dev_sectors && |
b098636c | 2827 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 2828 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
2829 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
2830 | } | |
b522adcd | 2831 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 2832 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
2833 | return 0; |
2834 | } | |
2835 | ||
fd01b88c | 2836 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
2837 | { |
2838 | /* We need to: | |
2839 | * 1/ resize the r1bio_pool | |
2840 | * 2/ resize conf->mirrors | |
2841 | * | |
2842 | * We allocate a new r1bio_pool if we can. | |
2843 | * Then raise a device barrier and wait until all IO stops. | |
2844 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
2845 | * |
2846 | * At the same time, we "pack" the devices so that all the missing | |
2847 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
2848 | */ |
2849 | mempool_t *newpool, *oldpool; | |
2850 | struct pool_info *newpoolinfo; | |
0f6d02d5 | 2851 | struct mirror_info *newmirrors; |
e8096360 | 2852 | struct r1conf *conf = mddev->private; |
63c70c4f | 2853 | int cnt, raid_disks; |
c04be0aa | 2854 | unsigned long flags; |
b5470dc5 | 2855 | int d, d2, err; |
1da177e4 | 2856 | |
63c70c4f | 2857 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 2858 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
2859 | mddev->layout != mddev->new_layout || |
2860 | mddev->level != mddev->new_level) { | |
664e7c41 | 2861 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
2862 | mddev->new_layout = mddev->layout; |
2863 | mddev->new_level = mddev->level; | |
2864 | return -EINVAL; | |
2865 | } | |
2866 | ||
b5470dc5 DW |
2867 | err = md_allow_write(mddev); |
2868 | if (err) | |
2869 | return err; | |
2a2275d6 | 2870 | |
63c70c4f N |
2871 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
2872 | ||
6ea9c07c N |
2873 | if (raid_disks < conf->raid_disks) { |
2874 | cnt=0; | |
2875 | for (d= 0; d < conf->raid_disks; d++) | |
2876 | if (conf->mirrors[d].rdev) | |
2877 | cnt++; | |
2878 | if (cnt > raid_disks) | |
1da177e4 | 2879 | return -EBUSY; |
6ea9c07c | 2880 | } |
1da177e4 LT |
2881 | |
2882 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
2883 | if (!newpoolinfo) | |
2884 | return -ENOMEM; | |
2885 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 2886 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
2887 | |
2888 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
2889 | r1bio_pool_free, newpoolinfo); | |
2890 | if (!newpool) { | |
2891 | kfree(newpoolinfo); | |
2892 | return -ENOMEM; | |
2893 | } | |
8f19ccb2 N |
2894 | newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks * 2, |
2895 | GFP_KERNEL); | |
1da177e4 LT |
2896 | if (!newmirrors) { |
2897 | kfree(newpoolinfo); | |
2898 | mempool_destroy(newpool); | |
2899 | return -ENOMEM; | |
2900 | } | |
1da177e4 | 2901 | |
17999be4 | 2902 | raise_barrier(conf); |
1da177e4 LT |
2903 | |
2904 | /* ok, everything is stopped */ | |
2905 | oldpool = conf->r1bio_pool; | |
2906 | conf->r1bio_pool = newpool; | |
6ea9c07c | 2907 | |
a88aa786 | 2908 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 2909 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 2910 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 2911 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 2912 | rdev->raid_disk = d2; |
36fad858 NK |
2913 | sysfs_unlink_rdev(mddev, rdev); |
2914 | if (sysfs_link_rdev(mddev, rdev)) | |
a88aa786 | 2915 | printk(KERN_WARNING |
36fad858 NK |
2916 | "md/raid1:%s: cannot register rd%d\n", |
2917 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 2918 | } |
a88aa786 N |
2919 | if (rdev) |
2920 | newmirrors[d2++].rdev = rdev; | |
2921 | } | |
1da177e4 LT |
2922 | kfree(conf->mirrors); |
2923 | conf->mirrors = newmirrors; | |
2924 | kfree(conf->poolinfo); | |
2925 | conf->poolinfo = newpoolinfo; | |
2926 | ||
c04be0aa | 2927 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 2928 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 2929 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 2930 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 2931 | mddev->delta_disks = 0; |
1da177e4 | 2932 | |
6ea9c07c | 2933 | conf->last_used = 0; /* just make sure it is in-range */ |
17999be4 | 2934 | lower_barrier(conf); |
1da177e4 LT |
2935 | |
2936 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2937 | md_wakeup_thread(mddev->thread); | |
2938 | ||
2939 | mempool_destroy(oldpool); | |
2940 | return 0; | |
2941 | } | |
2942 | ||
fd01b88c | 2943 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 2944 | { |
e8096360 | 2945 | struct r1conf *conf = mddev->private; |
36fa3063 N |
2946 | |
2947 | switch(state) { | |
6eef4b21 N |
2948 | case 2: /* wake for suspend */ |
2949 | wake_up(&conf->wait_barrier); | |
2950 | break; | |
9e6603da | 2951 | case 1: |
17999be4 | 2952 | raise_barrier(conf); |
36fa3063 | 2953 | break; |
9e6603da | 2954 | case 0: |
17999be4 | 2955 | lower_barrier(conf); |
36fa3063 N |
2956 | break; |
2957 | } | |
36fa3063 N |
2958 | } |
2959 | ||
fd01b88c | 2960 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
2961 | { |
2962 | /* raid1 can take over: | |
2963 | * raid5 with 2 devices, any layout or chunk size | |
2964 | */ | |
2965 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 2966 | struct r1conf *conf; |
709ae487 N |
2967 | mddev->new_level = 1; |
2968 | mddev->new_layout = 0; | |
2969 | mddev->new_chunk_sectors = 0; | |
2970 | conf = setup_conf(mddev); | |
2971 | if (!IS_ERR(conf)) | |
2972 | conf->barrier = 1; | |
2973 | return conf; | |
2974 | } | |
2975 | return ERR_PTR(-EINVAL); | |
2976 | } | |
1da177e4 | 2977 | |
84fc4b56 | 2978 | static struct md_personality raid1_personality = |
1da177e4 LT |
2979 | { |
2980 | .name = "raid1", | |
2604b703 | 2981 | .level = 1, |
1da177e4 LT |
2982 | .owner = THIS_MODULE, |
2983 | .make_request = make_request, | |
2984 | .run = run, | |
2985 | .stop = stop, | |
2986 | .status = status, | |
2987 | .error_handler = error, | |
2988 | .hot_add_disk = raid1_add_disk, | |
2989 | .hot_remove_disk= raid1_remove_disk, | |
2990 | .spare_active = raid1_spare_active, | |
2991 | .sync_request = sync_request, | |
2992 | .resize = raid1_resize, | |
80c3a6ce | 2993 | .size = raid1_size, |
63c70c4f | 2994 | .check_reshape = raid1_reshape, |
36fa3063 | 2995 | .quiesce = raid1_quiesce, |
709ae487 | 2996 | .takeover = raid1_takeover, |
1da177e4 LT |
2997 | }; |
2998 | ||
2999 | static int __init raid_init(void) | |
3000 | { | |
2604b703 | 3001 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3002 | } |
3003 | ||
3004 | static void raid_exit(void) | |
3005 | { | |
2604b703 | 3006 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3007 | } |
3008 | ||
3009 | module_init(raid_init); | |
3010 | module_exit(raid_exit); | |
3011 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3012 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3013 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3014 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3015 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3016 | |
3017 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |