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 | ||
473e87ce JB |
49 | /* when we get a read error on a read-only array, we redirect to another |
50 | * device without failing the first device, or trying to over-write to | |
51 | * correct the read error. To keep track of bad blocks on a per-bio | |
52 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
53 | */ | |
54 | #define IO_BLOCKED ((struct bio *)1) | |
55 | /* When we successfully write to a known bad-block, we need to remove the | |
56 | * bad-block marking which must be done from process context. So we record | |
57 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
58 | */ | |
59 | #define IO_MADE_GOOD ((struct bio *)2) | |
60 | ||
61 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
62 | ||
34db0cd6 N |
63 | /* When there are this many requests queue to be written by |
64 | * the raid1 thread, we become 'congested' to provide back-pressure | |
65 | * for writeback. | |
66 | */ | |
67 | static int max_queued_requests = 1024; | |
1da177e4 | 68 | |
79ef3a8a | 69 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
70 | sector_t bi_sector); | |
e8096360 | 71 | static void lower_barrier(struct r1conf *conf); |
1da177e4 | 72 | |
dd0fc66f | 73 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
74 | { |
75 | struct pool_info *pi = data; | |
9f2c9d12 | 76 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
77 | |
78 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 79 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
80 | } |
81 | ||
82 | static void r1bio_pool_free(void *r1_bio, void *data) | |
83 | { | |
84 | kfree(r1_bio); | |
85 | } | |
86 | ||
87 | #define RESYNC_BLOCK_SIZE (64*1024) | |
8e005f7c | 88 | #define RESYNC_DEPTH 32 |
1da177e4 LT |
89 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
90 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
8e005f7c | 91 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
92 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
93 | #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) | |
1da177e4 | 94 | |
dd0fc66f | 95 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
96 | { |
97 | struct pool_info *pi = data; | |
9f2c9d12 | 98 | struct r1bio *r1_bio; |
1da177e4 LT |
99 | struct bio *bio; |
100 | int i, j; | |
101 | ||
102 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 103 | if (!r1_bio) |
1da177e4 | 104 | return NULL; |
1da177e4 LT |
105 | |
106 | /* | |
107 | * Allocate bios : 1 for reading, n-1 for writing | |
108 | */ | |
109 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 110 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
111 | if (!bio) |
112 | goto out_free_bio; | |
113 | r1_bio->bios[j] = bio; | |
114 | } | |
115 | /* | |
116 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
117 | * the first bio. |
118 | * If this is a user-requested check/repair, allocate | |
119 | * RESYNC_PAGES for each bio. | |
1da177e4 | 120 | */ |
d11c171e N |
121 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
122 | j = pi->raid_disks; | |
123 | else | |
124 | j = 1; | |
125 | while(j--) { | |
126 | bio = r1_bio->bios[j]; | |
a0787606 | 127 | bio->bi_vcnt = RESYNC_PAGES; |
d11c171e | 128 | |
a0787606 KO |
129 | if (bio_alloc_pages(bio, gfp_flags)) |
130 | goto out_free_bio; | |
d11c171e N |
131 | } |
132 | /* If not user-requests, copy the page pointers to all bios */ | |
133 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
134 | for (i=0; i<RESYNC_PAGES ; i++) | |
135 | for (j=1; j<pi->raid_disks; j++) | |
136 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
137 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
138 | } |
139 | ||
140 | r1_bio->master_bio = NULL; | |
141 | ||
142 | return r1_bio; | |
143 | ||
1da177e4 | 144 | out_free_bio: |
8f19ccb2 | 145 | while (++j < pi->raid_disks) |
1da177e4 LT |
146 | bio_put(r1_bio->bios[j]); |
147 | r1bio_pool_free(r1_bio, data); | |
148 | return NULL; | |
149 | } | |
150 | ||
151 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
152 | { | |
153 | struct pool_info *pi = data; | |
d11c171e | 154 | int i,j; |
9f2c9d12 | 155 | struct r1bio *r1bio = __r1_bio; |
1da177e4 | 156 | |
d11c171e N |
157 | for (i = 0; i < RESYNC_PAGES; i++) |
158 | for (j = pi->raid_disks; j-- ;) { | |
159 | if (j == 0 || | |
160 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
161 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 162 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 163 | } |
1da177e4 LT |
164 | for (i=0 ; i < pi->raid_disks; i++) |
165 | bio_put(r1bio->bios[i]); | |
166 | ||
167 | r1bio_pool_free(r1bio, data); | |
168 | } | |
169 | ||
e8096360 | 170 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
171 | { |
172 | int i; | |
173 | ||
8f19ccb2 | 174 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 175 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 176 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
177 | bio_put(*bio); |
178 | *bio = NULL; | |
179 | } | |
180 | } | |
181 | ||
9f2c9d12 | 182 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 183 | { |
e8096360 | 184 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 185 | |
1da177e4 LT |
186 | put_all_bios(conf, r1_bio); |
187 | mempool_free(r1_bio, conf->r1bio_pool); | |
188 | } | |
189 | ||
9f2c9d12 | 190 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 191 | { |
e8096360 | 192 | struct r1conf *conf = r1_bio->mddev->private; |
3e198f78 N |
193 | int i; |
194 | ||
8f19ccb2 | 195 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
196 | struct bio *bio = r1_bio->bios[i]; |
197 | if (bio->bi_end_io) | |
198 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
199 | } | |
1da177e4 LT |
200 | |
201 | mempool_free(r1_bio, conf->r1buf_pool); | |
202 | ||
17999be4 | 203 | lower_barrier(conf); |
1da177e4 LT |
204 | } |
205 | ||
9f2c9d12 | 206 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
207 | { |
208 | unsigned long flags; | |
fd01b88c | 209 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 210 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
211 | |
212 | spin_lock_irqsave(&conf->device_lock, flags); | |
213 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 214 | conf->nr_queued ++; |
1da177e4 LT |
215 | spin_unlock_irqrestore(&conf->device_lock, flags); |
216 | ||
17999be4 | 217 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
218 | md_wakeup_thread(mddev->thread); |
219 | } | |
220 | ||
221 | /* | |
222 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
223 | * operation and are ready to return a success/failure code to the buffer | |
224 | * cache layer. | |
225 | */ | |
9f2c9d12 | 226 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
227 | { |
228 | struct bio *bio = r1_bio->master_bio; | |
229 | int done; | |
e8096360 | 230 | struct r1conf *conf = r1_bio->mddev->private; |
79ef3a8a | 231 | sector_t start_next_window = r1_bio->start_next_window; |
232 | sector_t bi_sector = bio->bi_sector; | |
d2eb35ac N |
233 | |
234 | if (bio->bi_phys_segments) { | |
235 | unsigned long flags; | |
236 | spin_lock_irqsave(&conf->device_lock, flags); | |
237 | bio->bi_phys_segments--; | |
238 | done = (bio->bi_phys_segments == 0); | |
239 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
79ef3a8a | 240 | /* |
241 | * make_request() might be waiting for | |
242 | * bi_phys_segments to decrease | |
243 | */ | |
244 | wake_up(&conf->wait_barrier); | |
d2eb35ac N |
245 | } else |
246 | done = 1; | |
247 | ||
248 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
249 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
250 | if (done) { | |
251 | bio_endio(bio, 0); | |
252 | /* | |
253 | * Wake up any possible resync thread that waits for the device | |
254 | * to go idle. | |
255 | */ | |
79ef3a8a | 256 | allow_barrier(conf, start_next_window, bi_sector); |
d2eb35ac N |
257 | } |
258 | } | |
259 | ||
9f2c9d12 | 260 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
261 | { |
262 | struct bio *bio = r1_bio->master_bio; | |
263 | ||
4b6d287f N |
264 | /* if nobody has done the final endio yet, do it now */ |
265 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
266 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
267 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
268 | (unsigned long long) bio->bi_sector, | |
269 | (unsigned long long) bio->bi_sector + | |
aa8b57aa | 270 | bio_sectors(bio) - 1); |
4b6d287f | 271 | |
d2eb35ac | 272 | call_bio_endio(r1_bio); |
4b6d287f | 273 | } |
1da177e4 LT |
274 | free_r1bio(r1_bio); |
275 | } | |
276 | ||
277 | /* | |
278 | * Update disk head position estimator based on IRQ completion info. | |
279 | */ | |
9f2c9d12 | 280 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 281 | { |
e8096360 | 282 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
283 | |
284 | conf->mirrors[disk].head_position = | |
285 | r1_bio->sector + (r1_bio->sectors); | |
286 | } | |
287 | ||
ba3ae3be NK |
288 | /* |
289 | * Find the disk number which triggered given bio | |
290 | */ | |
9f2c9d12 | 291 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
292 | { |
293 | int mirror; | |
30194636 N |
294 | struct r1conf *conf = r1_bio->mddev->private; |
295 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 296 | |
8f19ccb2 | 297 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
298 | if (r1_bio->bios[mirror] == bio) |
299 | break; | |
300 | ||
8f19ccb2 | 301 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
302 | update_head_pos(mirror, r1_bio); |
303 | ||
304 | return mirror; | |
305 | } | |
306 | ||
6712ecf8 | 307 | static void raid1_end_read_request(struct bio *bio, int error) |
1da177e4 LT |
308 | { |
309 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 310 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 311 | int mirror; |
e8096360 | 312 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 313 | |
1da177e4 LT |
314 | mirror = r1_bio->read_disk; |
315 | /* | |
316 | * this branch is our 'one mirror IO has finished' event handler: | |
317 | */ | |
ddaf22ab N |
318 | update_head_pos(mirror, r1_bio); |
319 | ||
dd00a99e N |
320 | if (uptodate) |
321 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
322 | else { | |
323 | /* If all other devices have failed, we want to return | |
324 | * the error upwards rather than fail the last device. | |
325 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 326 | */ |
dd00a99e N |
327 | unsigned long flags; |
328 | spin_lock_irqsave(&conf->device_lock, flags); | |
329 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
330 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
331 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))) | |
332 | uptodate = 1; | |
333 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
334 | } | |
1da177e4 | 335 | |
7ad4d4a6 | 336 | if (uptodate) { |
1da177e4 | 337 | raid_end_bio_io(r1_bio); |
7ad4d4a6 N |
338 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
339 | } else { | |
1da177e4 LT |
340 | /* |
341 | * oops, read error: | |
342 | */ | |
343 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
344 | printk_ratelimited( |
345 | KERN_ERR "md/raid1:%s: %s: " | |
346 | "rescheduling sector %llu\n", | |
347 | mdname(conf->mddev), | |
348 | bdevname(conf->mirrors[mirror].rdev->bdev, | |
349 | b), | |
350 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 351 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 352 | reschedule_retry(r1_bio); |
7ad4d4a6 | 353 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 354 | } |
1da177e4 LT |
355 | } |
356 | ||
9f2c9d12 | 357 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
358 | { |
359 | /* it really is the end of this request */ | |
360 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
361 | /* free extra copy of the data pages */ | |
362 | int i = r1_bio->behind_page_count; | |
363 | while (i--) | |
364 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
365 | kfree(r1_bio->behind_bvecs); | |
366 | r1_bio->behind_bvecs = NULL; | |
367 | } | |
368 | /* clear the bitmap if all writes complete successfully */ | |
369 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
370 | r1_bio->sectors, | |
371 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
372 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
373 | md_write_end(r1_bio->mddev); | |
374 | } | |
375 | ||
9f2c9d12 | 376 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 377 | { |
cd5ff9a1 N |
378 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
379 | return; | |
380 | ||
381 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
382 | reschedule_retry(r1_bio); | |
383 | else { | |
384 | close_write(r1_bio); | |
4367af55 N |
385 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
386 | reschedule_retry(r1_bio); | |
387 | else | |
388 | raid_end_bio_io(r1_bio); | |
4e78064f N |
389 | } |
390 | } | |
391 | ||
6712ecf8 | 392 | static void raid1_end_write_request(struct bio *bio, int error) |
1da177e4 LT |
393 | { |
394 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 395 | struct r1bio *r1_bio = bio->bi_private; |
a9701a30 | 396 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 397 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 398 | struct bio *to_put = NULL; |
1da177e4 | 399 | |
ba3ae3be | 400 | mirror = find_bio_disk(r1_bio, bio); |
1da177e4 | 401 | |
e9c7469b TH |
402 | /* |
403 | * 'one mirror IO has finished' event handler: | |
404 | */ | |
e9c7469b | 405 | if (!uptodate) { |
cd5ff9a1 N |
406 | set_bit(WriteErrorSeen, |
407 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
408 | if (!test_and_set_bit(WantReplacement, |
409 | &conf->mirrors[mirror].rdev->flags)) | |
410 | set_bit(MD_RECOVERY_NEEDED, & | |
411 | conf->mddev->recovery); | |
412 | ||
cd5ff9a1 | 413 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 | 414 | } else { |
1da177e4 | 415 | /* |
e9c7469b TH |
416 | * Set R1BIO_Uptodate in our master bio, so that we |
417 | * will return a good error code for to the higher | |
418 | * levels even if IO on some other mirrored buffer | |
419 | * fails. | |
420 | * | |
421 | * The 'master' represents the composite IO operation | |
422 | * to user-side. So if something waits for IO, then it | |
423 | * will wait for the 'master' bio. | |
1da177e4 | 424 | */ |
4367af55 N |
425 | sector_t first_bad; |
426 | int bad_sectors; | |
427 | ||
cd5ff9a1 N |
428 | r1_bio->bios[mirror] = NULL; |
429 | to_put = bio; | |
3056e3ae AL |
430 | /* |
431 | * Do not set R1BIO_Uptodate if the current device is | |
432 | * rebuilding or Faulty. This is because we cannot use | |
433 | * such device for properly reading the data back (we could | |
434 | * potentially use it, if the current write would have felt | |
435 | * before rdev->recovery_offset, but for simplicity we don't | |
436 | * check this here. | |
437 | */ | |
438 | if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) && | |
439 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)) | |
440 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
e9c7469b | 441 | |
4367af55 N |
442 | /* Maybe we can clear some bad blocks. */ |
443 | if (is_badblock(conf->mirrors[mirror].rdev, | |
444 | r1_bio->sector, r1_bio->sectors, | |
445 | &first_bad, &bad_sectors)) { | |
446 | r1_bio->bios[mirror] = IO_MADE_GOOD; | |
447 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
448 | } | |
449 | } | |
450 | ||
e9c7469b TH |
451 | if (behind) { |
452 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) | |
453 | atomic_dec(&r1_bio->behind_remaining); | |
454 | ||
455 | /* | |
456 | * In behind mode, we ACK the master bio once the I/O | |
457 | * has safely reached all non-writemostly | |
458 | * disks. Setting the Returned bit ensures that this | |
459 | * gets done only once -- we don't ever want to return | |
460 | * -EIO here, instead we'll wait | |
461 | */ | |
462 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
463 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
464 | /* Maybe we can return now */ | |
465 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
466 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
467 | pr_debug("raid1: behind end write sectors" |
468 | " %llu-%llu\n", | |
469 | (unsigned long long) mbio->bi_sector, | |
470 | (unsigned long long) mbio->bi_sector + | |
aa8b57aa | 471 | bio_sectors(mbio) - 1); |
d2eb35ac | 472 | call_bio_endio(r1_bio); |
4b6d287f N |
473 | } |
474 | } | |
475 | } | |
4367af55 N |
476 | if (r1_bio->bios[mirror] == NULL) |
477 | rdev_dec_pending(conf->mirrors[mirror].rdev, | |
478 | conf->mddev); | |
e9c7469b | 479 | |
1da177e4 | 480 | /* |
1da177e4 LT |
481 | * Let's see if all mirrored write operations have finished |
482 | * already. | |
483 | */ | |
af6d7b76 | 484 | r1_bio_write_done(r1_bio); |
c70810b3 | 485 | |
04b857f7 N |
486 | if (to_put) |
487 | bio_put(to_put); | |
1da177e4 LT |
488 | } |
489 | ||
490 | ||
491 | /* | |
492 | * This routine returns the disk from which the requested read should | |
493 | * be done. There is a per-array 'next expected sequential IO' sector | |
494 | * number - if this matches on the next IO then we use the last disk. | |
495 | * There is also a per-disk 'last know head position' sector that is | |
496 | * maintained from IRQ contexts, both the normal and the resync IO | |
497 | * completion handlers update this position correctly. If there is no | |
498 | * perfect sequential match then we pick the disk whose head is closest. | |
499 | * | |
500 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
501 | * because position is mirror, not device based. | |
502 | * | |
503 | * The rdev for the device selected will have nr_pending incremented. | |
504 | */ | |
e8096360 | 505 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 506 | { |
af3a2cd6 | 507 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
508 | int sectors; |
509 | int best_good_sectors; | |
9dedf603 SL |
510 | int best_disk, best_dist_disk, best_pending_disk; |
511 | int has_nonrot_disk; | |
be4d3280 | 512 | int disk; |
76073054 | 513 | sector_t best_dist; |
9dedf603 | 514 | unsigned int min_pending; |
3cb03002 | 515 | struct md_rdev *rdev; |
f3ac8bf7 | 516 | int choose_first; |
12cee5a8 | 517 | int choose_next_idle; |
1da177e4 LT |
518 | |
519 | rcu_read_lock(); | |
520 | /* | |
8ddf9efe | 521 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
522 | * device if no resync is going on, or below the resync window. |
523 | * We take the first readable disk when above the resync window. | |
524 | */ | |
525 | retry: | |
d2eb35ac | 526 | sectors = r1_bio->sectors; |
76073054 | 527 | best_disk = -1; |
9dedf603 | 528 | best_dist_disk = -1; |
76073054 | 529 | best_dist = MaxSector; |
9dedf603 SL |
530 | best_pending_disk = -1; |
531 | min_pending = UINT_MAX; | |
d2eb35ac | 532 | best_good_sectors = 0; |
9dedf603 | 533 | has_nonrot_disk = 0; |
12cee5a8 | 534 | choose_next_idle = 0; |
d2eb35ac | 535 | |
1da177e4 | 536 | if (conf->mddev->recovery_cp < MaxSector && |
be4d3280 | 537 | (this_sector + sectors >= conf->next_resync)) |
f3ac8bf7 | 538 | choose_first = 1; |
be4d3280 | 539 | else |
f3ac8bf7 | 540 | choose_first = 0; |
1da177e4 | 541 | |
be4d3280 | 542 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 543 | sector_t dist; |
d2eb35ac N |
544 | sector_t first_bad; |
545 | int bad_sectors; | |
9dedf603 | 546 | unsigned int pending; |
12cee5a8 | 547 | bool nonrot; |
d2eb35ac | 548 | |
f3ac8bf7 N |
549 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
550 | if (r1_bio->bios[disk] == IO_BLOCKED | |
551 | || rdev == NULL | |
6b740b8d | 552 | || test_bit(Unmerged, &rdev->flags) |
76073054 | 553 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 554 | continue; |
76073054 N |
555 | if (!test_bit(In_sync, &rdev->flags) && |
556 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 557 | continue; |
76073054 N |
558 | if (test_bit(WriteMostly, &rdev->flags)) { |
559 | /* Don't balance among write-mostly, just | |
560 | * use the first as a last resort */ | |
307729c8 N |
561 | if (best_disk < 0) { |
562 | if (is_badblock(rdev, this_sector, sectors, | |
563 | &first_bad, &bad_sectors)) { | |
564 | if (first_bad < this_sector) | |
565 | /* Cannot use this */ | |
566 | continue; | |
567 | best_good_sectors = first_bad - this_sector; | |
568 | } else | |
569 | best_good_sectors = sectors; | |
76073054 | 570 | best_disk = disk; |
307729c8 | 571 | } |
76073054 N |
572 | continue; |
573 | } | |
574 | /* This is a reasonable device to use. It might | |
575 | * even be best. | |
576 | */ | |
d2eb35ac N |
577 | if (is_badblock(rdev, this_sector, sectors, |
578 | &first_bad, &bad_sectors)) { | |
579 | if (best_dist < MaxSector) | |
580 | /* already have a better device */ | |
581 | continue; | |
582 | if (first_bad <= this_sector) { | |
583 | /* cannot read here. If this is the 'primary' | |
584 | * device, then we must not read beyond | |
585 | * bad_sectors from another device.. | |
586 | */ | |
587 | bad_sectors -= (this_sector - first_bad); | |
588 | if (choose_first && sectors > bad_sectors) | |
589 | sectors = bad_sectors; | |
590 | if (best_good_sectors > sectors) | |
591 | best_good_sectors = sectors; | |
592 | ||
593 | } else { | |
594 | sector_t good_sectors = first_bad - this_sector; | |
595 | if (good_sectors > best_good_sectors) { | |
596 | best_good_sectors = good_sectors; | |
597 | best_disk = disk; | |
598 | } | |
599 | if (choose_first) | |
600 | break; | |
601 | } | |
602 | continue; | |
603 | } else | |
604 | best_good_sectors = sectors; | |
605 | ||
12cee5a8 SL |
606 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
607 | has_nonrot_disk |= nonrot; | |
9dedf603 | 608 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 609 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 610 | if (choose_first) { |
76073054 | 611 | best_disk = disk; |
1da177e4 LT |
612 | break; |
613 | } | |
12cee5a8 SL |
614 | /* Don't change to another disk for sequential reads */ |
615 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
616 | || dist == 0) { | |
617 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
618 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
619 | ||
620 | best_disk = disk; | |
621 | /* | |
622 | * If buffered sequential IO size exceeds optimal | |
623 | * iosize, check if there is idle disk. If yes, choose | |
624 | * the idle disk. read_balance could already choose an | |
625 | * idle disk before noticing it's a sequential IO in | |
626 | * this disk. This doesn't matter because this disk | |
627 | * will idle, next time it will be utilized after the | |
628 | * first disk has IO size exceeds optimal iosize. In | |
629 | * this way, iosize of the first disk will be optimal | |
630 | * iosize at least. iosize of the second disk might be | |
631 | * small, but not a big deal since when the second disk | |
632 | * starts IO, the first disk is likely still busy. | |
633 | */ | |
634 | if (nonrot && opt_iosize > 0 && | |
635 | mirror->seq_start != MaxSector && | |
636 | mirror->next_seq_sect > opt_iosize && | |
637 | mirror->next_seq_sect - opt_iosize >= | |
638 | mirror->seq_start) { | |
639 | choose_next_idle = 1; | |
640 | continue; | |
641 | } | |
642 | break; | |
643 | } | |
644 | /* If device is idle, use it */ | |
645 | if (pending == 0) { | |
646 | best_disk = disk; | |
647 | break; | |
648 | } | |
649 | ||
650 | if (choose_next_idle) | |
651 | continue; | |
9dedf603 SL |
652 | |
653 | if (min_pending > pending) { | |
654 | min_pending = pending; | |
655 | best_pending_disk = disk; | |
656 | } | |
657 | ||
76073054 N |
658 | if (dist < best_dist) { |
659 | best_dist = dist; | |
9dedf603 | 660 | best_dist_disk = disk; |
1da177e4 | 661 | } |
f3ac8bf7 | 662 | } |
1da177e4 | 663 | |
9dedf603 SL |
664 | /* |
665 | * If all disks are rotational, choose the closest disk. If any disk is | |
666 | * non-rotational, choose the disk with less pending request even the | |
667 | * disk is rotational, which might/might not be optimal for raids with | |
668 | * mixed ratation/non-rotational disks depending on workload. | |
669 | */ | |
670 | if (best_disk == -1) { | |
671 | if (has_nonrot_disk) | |
672 | best_disk = best_pending_disk; | |
673 | else | |
674 | best_disk = best_dist_disk; | |
675 | } | |
676 | ||
76073054 N |
677 | if (best_disk >= 0) { |
678 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
679 | if (!rdev) |
680 | goto retry; | |
681 | atomic_inc(&rdev->nr_pending); | |
76073054 | 682 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
683 | /* cannot risk returning a device that failed |
684 | * before we inc'ed nr_pending | |
685 | */ | |
03c902e1 | 686 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
687 | goto retry; |
688 | } | |
d2eb35ac | 689 | sectors = best_good_sectors; |
12cee5a8 SL |
690 | |
691 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
692 | conf->mirrors[best_disk].seq_start = this_sector; | |
693 | ||
be4d3280 | 694 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
695 | } |
696 | rcu_read_unlock(); | |
d2eb35ac | 697 | *max_sectors = sectors; |
1da177e4 | 698 | |
76073054 | 699 | return best_disk; |
1da177e4 LT |
700 | } |
701 | ||
6b740b8d N |
702 | static int raid1_mergeable_bvec(struct request_queue *q, |
703 | struct bvec_merge_data *bvm, | |
704 | struct bio_vec *biovec) | |
705 | { | |
706 | struct mddev *mddev = q->queuedata; | |
707 | struct r1conf *conf = mddev->private; | |
708 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); | |
709 | int max = biovec->bv_len; | |
710 | ||
711 | if (mddev->merge_check_needed) { | |
712 | int disk; | |
713 | rcu_read_lock(); | |
714 | for (disk = 0; disk < conf->raid_disks * 2; disk++) { | |
715 | struct md_rdev *rdev = rcu_dereference( | |
716 | conf->mirrors[disk].rdev); | |
717 | if (rdev && !test_bit(Faulty, &rdev->flags)) { | |
718 | struct request_queue *q = | |
719 | bdev_get_queue(rdev->bdev); | |
720 | if (q->merge_bvec_fn) { | |
721 | bvm->bi_sector = sector + | |
722 | rdev->data_offset; | |
723 | bvm->bi_bdev = rdev->bdev; | |
724 | max = min(max, q->merge_bvec_fn( | |
725 | q, bvm, biovec)); | |
726 | } | |
727 | } | |
728 | } | |
729 | rcu_read_unlock(); | |
730 | } | |
731 | return max; | |
732 | ||
733 | } | |
734 | ||
fd01b88c | 735 | int md_raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 736 | { |
e8096360 | 737 | struct r1conf *conf = mddev->private; |
0d129228 N |
738 | int i, ret = 0; |
739 | ||
34db0cd6 N |
740 | if ((bits & (1 << BDI_async_congested)) && |
741 | conf->pending_count >= max_queued_requests) | |
742 | return 1; | |
743 | ||
0d129228 | 744 | rcu_read_lock(); |
f53e29fc | 745 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 746 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 747 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 748 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 749 | |
1ed7242e JB |
750 | BUG_ON(!q); |
751 | ||
0d129228 N |
752 | /* Note the '|| 1' - when read_balance prefers |
753 | * non-congested targets, it can be removed | |
754 | */ | |
91a9e99d | 755 | if ((bits & (1<<BDI_async_congested)) || 1) |
0d129228 N |
756 | ret |= bdi_congested(&q->backing_dev_info, bits); |
757 | else | |
758 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
759 | } | |
760 | } | |
761 | rcu_read_unlock(); | |
762 | return ret; | |
763 | } | |
1ed7242e | 764 | EXPORT_SYMBOL_GPL(md_raid1_congested); |
0d129228 | 765 | |
1ed7242e JB |
766 | static int raid1_congested(void *data, int bits) |
767 | { | |
fd01b88c | 768 | struct mddev *mddev = data; |
1ed7242e JB |
769 | |
770 | return mddev_congested(mddev, bits) || | |
771 | md_raid1_congested(mddev, bits); | |
772 | } | |
0d129228 | 773 | |
e8096360 | 774 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
775 | { |
776 | /* Any writes that have been queued but are awaiting | |
777 | * bitmap updates get flushed here. | |
a35e63ef | 778 | */ |
a35e63ef N |
779 | spin_lock_irq(&conf->device_lock); |
780 | ||
781 | if (conf->pending_bio_list.head) { | |
782 | struct bio *bio; | |
783 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 784 | conf->pending_count = 0; |
a35e63ef N |
785 | spin_unlock_irq(&conf->device_lock); |
786 | /* flush any pending bitmap writes to | |
787 | * disk before proceeding w/ I/O */ | |
788 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 789 | wake_up(&conf->wait_barrier); |
a35e63ef N |
790 | |
791 | while (bio) { /* submit pending writes */ | |
792 | struct bio *next = bio->bi_next; | |
793 | bio->bi_next = NULL; | |
2ff8cc2c SL |
794 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
795 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
796 | /* Just ignore it */ | |
797 | bio_endio(bio, 0); | |
798 | else | |
799 | generic_make_request(bio); | |
a35e63ef N |
800 | bio = next; |
801 | } | |
a35e63ef N |
802 | } else |
803 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
804 | } |
805 | ||
17999be4 N |
806 | /* Barriers.... |
807 | * Sometimes we need to suspend IO while we do something else, | |
808 | * either some resync/recovery, or reconfigure the array. | |
809 | * To do this we raise a 'barrier'. | |
810 | * The 'barrier' is a counter that can be raised multiple times | |
811 | * to count how many activities are happening which preclude | |
812 | * normal IO. | |
813 | * We can only raise the barrier if there is no pending IO. | |
814 | * i.e. if nr_pending == 0. | |
815 | * We choose only to raise the barrier if no-one is waiting for the | |
816 | * barrier to go down. This means that as soon as an IO request | |
817 | * is ready, no other operations which require a barrier will start | |
818 | * until the IO request has had a chance. | |
819 | * | |
820 | * So: regular IO calls 'wait_barrier'. When that returns there | |
821 | * is no backgroup IO happening, It must arrange to call | |
822 | * allow_barrier when it has finished its IO. | |
823 | * backgroup IO calls must call raise_barrier. Once that returns | |
824 | * there is no normal IO happeing. It must arrange to call | |
825 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 826 | */ |
e8096360 | 827 | static void raise_barrier(struct r1conf *conf) |
1da177e4 LT |
828 | { |
829 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
830 | |
831 | /* Wait until no block IO is waiting */ | |
832 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
eed8c02e | 833 | conf->resync_lock); |
17999be4 N |
834 | |
835 | /* block any new IO from starting */ | |
836 | conf->barrier++; | |
837 | ||
79ef3a8a | 838 | /* For these conditions we must wait: |
839 | * A: while the array is in frozen state | |
840 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach | |
841 | * the max count which allowed. | |
842 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning | |
843 | * next resync will reach to the window which normal bios are | |
844 | * handling. | |
845 | */ | |
17999be4 | 846 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 847 | !conf->array_frozen && |
79ef3a8a | 848 | conf->barrier < RESYNC_DEPTH && |
849 | (conf->start_next_window >= | |
850 | conf->next_resync + RESYNC_SECTORS), | |
eed8c02e | 851 | conf->resync_lock); |
17999be4 N |
852 | |
853 | spin_unlock_irq(&conf->resync_lock); | |
854 | } | |
855 | ||
e8096360 | 856 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
857 | { |
858 | unsigned long flags; | |
709ae487 | 859 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
860 | spin_lock_irqsave(&conf->resync_lock, flags); |
861 | conf->barrier--; | |
862 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
863 | wake_up(&conf->wait_barrier); | |
864 | } | |
865 | ||
79ef3a8a | 866 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
17999be4 | 867 | { |
79ef3a8a | 868 | bool wait = false; |
869 | ||
870 | if (conf->array_frozen || !bio) | |
871 | wait = true; | |
872 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { | |
873 | if (conf->next_resync < RESYNC_WINDOW_SECTORS) | |
874 | wait = true; | |
875 | else if ((conf->next_resync - RESYNC_WINDOW_SECTORS | |
876 | >= bio_end_sector(bio)) || | |
877 | (conf->next_resync + NEXT_NORMALIO_DISTANCE | |
878 | <= bio->bi_sector)) | |
879 | wait = false; | |
880 | else | |
881 | wait = true; | |
882 | } | |
883 | ||
884 | return wait; | |
885 | } | |
886 | ||
887 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) | |
888 | { | |
889 | sector_t sector = 0; | |
890 | ||
17999be4 | 891 | spin_lock_irq(&conf->resync_lock); |
79ef3a8a | 892 | if (need_to_wait_for_sync(conf, bio)) { |
17999be4 | 893 | conf->nr_waiting++; |
d6b42dcb N |
894 | /* Wait for the barrier to drop. |
895 | * However if there are already pending | |
896 | * requests (preventing the barrier from | |
897 | * rising completely), and the | |
898 | * pre-process bio queue isn't empty, | |
899 | * then don't wait, as we need to empty | |
900 | * that queue to get the nr_pending | |
901 | * count down. | |
902 | */ | |
903 | wait_event_lock_irq(conf->wait_barrier, | |
b364e3d0 | 904 | !conf->array_frozen && |
905 | (!conf->barrier || | |
79ef3a8a | 906 | ((conf->start_next_window < |
907 | conf->next_resync + RESYNC_SECTORS) && | |
d6b42dcb | 908 | current->bio_list && |
b364e3d0 | 909 | !bio_list_empty(current->bio_list))), |
eed8c02e | 910 | conf->resync_lock); |
17999be4 | 911 | conf->nr_waiting--; |
1da177e4 | 912 | } |
79ef3a8a | 913 | |
914 | if (bio && bio_data_dir(bio) == WRITE) { | |
915 | if (conf->next_resync + NEXT_NORMALIO_DISTANCE | |
916 | <= bio->bi_sector) { | |
917 | if (conf->start_next_window == MaxSector) | |
918 | conf->start_next_window = | |
919 | conf->next_resync + | |
920 | NEXT_NORMALIO_DISTANCE; | |
921 | ||
922 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) | |
923 | <= bio->bi_sector) | |
924 | conf->next_window_requests++; | |
925 | else | |
926 | conf->current_window_requests++; | |
927 | } | |
928 | if (bio->bi_sector >= conf->start_next_window) | |
929 | sector = conf->start_next_window; | |
930 | } | |
931 | ||
17999be4 | 932 | conf->nr_pending++; |
1da177e4 | 933 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 934 | return sector; |
1da177e4 LT |
935 | } |
936 | ||
79ef3a8a | 937 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
938 | sector_t bi_sector) | |
17999be4 N |
939 | { |
940 | unsigned long flags; | |
79ef3a8a | 941 | |
17999be4 N |
942 | spin_lock_irqsave(&conf->resync_lock, flags); |
943 | conf->nr_pending--; | |
79ef3a8a | 944 | if (start_next_window) { |
945 | if (start_next_window == conf->start_next_window) { | |
946 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE | |
947 | <= bi_sector) | |
948 | conf->next_window_requests--; | |
949 | else | |
950 | conf->current_window_requests--; | |
951 | } else | |
952 | conf->current_window_requests--; | |
953 | ||
954 | if (!conf->current_window_requests) { | |
955 | if (conf->next_window_requests) { | |
956 | conf->current_window_requests = | |
957 | conf->next_window_requests; | |
958 | conf->next_window_requests = 0; | |
959 | conf->start_next_window += | |
960 | NEXT_NORMALIO_DISTANCE; | |
961 | } else | |
962 | conf->start_next_window = MaxSector; | |
963 | } | |
964 | } | |
17999be4 N |
965 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
966 | wake_up(&conf->wait_barrier); | |
967 | } | |
968 | ||
e2d59925 | 969 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab N |
970 | { |
971 | /* stop syncio and normal IO and wait for everything to | |
972 | * go quite. | |
b364e3d0 | 973 | * We wait until nr_pending match nr_queued+extra |
1c830532 N |
974 | * This is called in the context of one normal IO request |
975 | * that has failed. Thus any sync request that might be pending | |
976 | * will be blocked by nr_pending, and we need to wait for | |
977 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 978 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
979 | * must match the number of pending IOs (nr_pending) before |
980 | * we continue. | |
ddaf22ab N |
981 | */ |
982 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 983 | conf->array_frozen = 1; |
eed8c02e | 984 | wait_event_lock_irq_cmd(conf->wait_barrier, |
e2d59925 | 985 | conf->nr_pending == conf->nr_queued+extra, |
eed8c02e LC |
986 | conf->resync_lock, |
987 | flush_pending_writes(conf)); | |
ddaf22ab N |
988 | spin_unlock_irq(&conf->resync_lock); |
989 | } | |
e8096360 | 990 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
991 | { |
992 | /* reverse the effect of the freeze */ | |
993 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 994 | conf->array_frozen = 0; |
ddaf22ab N |
995 | wake_up(&conf->wait_barrier); |
996 | spin_unlock_irq(&conf->resync_lock); | |
997 | } | |
998 | ||
17999be4 | 999 | |
4e78064f | 1000 | /* duplicate the data pages for behind I/O |
4e78064f | 1001 | */ |
9f2c9d12 | 1002 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
1003 | { |
1004 | int i; | |
1005 | struct bio_vec *bvec; | |
2ca68f5e | 1006 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 1007 | GFP_NOIO); |
2ca68f5e | 1008 | if (unlikely(!bvecs)) |
af6d7b76 | 1009 | return; |
4b6d287f | 1010 | |
cb34e057 | 1011 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
1012 | bvecs[i] = *bvec; |
1013 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
1014 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 1015 | goto do_sync_io; |
2ca68f5e N |
1016 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
1017 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
1018 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
1019 | kunmap(bvec->bv_page); |
1020 | } | |
2ca68f5e | 1021 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
1022 | r1_bio->behind_page_count = bio->bi_vcnt; |
1023 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
1024 | return; | |
4b6d287f N |
1025 | |
1026 | do_sync_io: | |
af6d7b76 | 1027 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
1028 | if (bvecs[i].bv_page) |
1029 | put_page(bvecs[i].bv_page); | |
1030 | kfree(bvecs); | |
36a4e1fe | 1031 | pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size); |
4b6d287f N |
1032 | } |
1033 | ||
f54a9d0e N |
1034 | struct raid1_plug_cb { |
1035 | struct blk_plug_cb cb; | |
1036 | struct bio_list pending; | |
1037 | int pending_cnt; | |
1038 | }; | |
1039 | ||
1040 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1041 | { | |
1042 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1043 | cb); | |
1044 | struct mddev *mddev = plug->cb.data; | |
1045 | struct r1conf *conf = mddev->private; | |
1046 | struct bio *bio; | |
1047 | ||
874807a8 | 1048 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1049 | spin_lock_irq(&conf->device_lock); |
1050 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1051 | conf->pending_count += plug->pending_cnt; | |
1052 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1053 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1054 | md_wakeup_thread(mddev->thread); |
1055 | kfree(plug); | |
1056 | return; | |
1057 | } | |
1058 | ||
1059 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1060 | bio = bio_list_get(&plug->pending); | |
1061 | bitmap_unplug(mddev->bitmap); | |
1062 | wake_up(&conf->wait_barrier); | |
1063 | ||
1064 | while (bio) { /* submit pending writes */ | |
1065 | struct bio *next = bio->bi_next; | |
1066 | bio->bi_next = NULL; | |
32f9f570 SL |
1067 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
1068 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
1069 | /* Just ignore it */ | |
1070 | bio_endio(bio, 0); | |
1071 | else | |
1072 | generic_make_request(bio); | |
f54a9d0e N |
1073 | bio = next; |
1074 | } | |
1075 | kfree(plug); | |
1076 | } | |
1077 | ||
b4fdcb02 | 1078 | static void make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 1079 | { |
e8096360 | 1080 | struct r1conf *conf = mddev->private; |
0eaf822c | 1081 | struct raid1_info *mirror; |
9f2c9d12 | 1082 | struct r1bio *r1_bio; |
1da177e4 | 1083 | struct bio *read_bio; |
1f68f0c4 | 1084 | int i, disks; |
84255d10 | 1085 | struct bitmap *bitmap; |
191ea9b2 | 1086 | unsigned long flags; |
a362357b | 1087 | const int rw = bio_data_dir(bio); |
2c7d46ec | 1088 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 1089 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
2ff8cc2c SL |
1090 | const unsigned long do_discard = (bio->bi_rw |
1091 | & (REQ_DISCARD | REQ_SECURE)); | |
c8dc9c65 | 1092 | const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME); |
3cb03002 | 1093 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1094 | struct blk_plug_cb *cb; |
1095 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
1096 | int first_clone; |
1097 | int sectors_handled; | |
1098 | int max_sectors; | |
79ef3a8a | 1099 | sector_t start_next_window; |
191ea9b2 | 1100 | |
1da177e4 LT |
1101 | /* |
1102 | * Register the new request and wait if the reconstruction | |
1103 | * thread has put up a bar for new requests. | |
1104 | * Continue immediately if no resync is active currently. | |
1105 | */ | |
62de608d | 1106 | |
3d310eb7 N |
1107 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1108 | ||
6eef4b21 | 1109 | if (bio_data_dir(bio) == WRITE && |
f73a1c7d | 1110 | bio_end_sector(bio) > mddev->suspend_lo && |
6eef4b21 N |
1111 | bio->bi_sector < mddev->suspend_hi) { |
1112 | /* As the suspend_* range is controlled by | |
1113 | * userspace, we want an interruptible | |
1114 | * wait. | |
1115 | */ | |
1116 | DEFINE_WAIT(w); | |
1117 | for (;;) { | |
1118 | flush_signals(current); | |
1119 | prepare_to_wait(&conf->wait_barrier, | |
1120 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1121 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
6eef4b21 N |
1122 | bio->bi_sector >= mddev->suspend_hi) |
1123 | break; | |
1124 | schedule(); | |
1125 | } | |
1126 | finish_wait(&conf->wait_barrier, &w); | |
1127 | } | |
62de608d | 1128 | |
79ef3a8a | 1129 | start_next_window = wait_barrier(conf, bio); |
1da177e4 | 1130 | |
84255d10 N |
1131 | bitmap = mddev->bitmap; |
1132 | ||
1da177e4 LT |
1133 | /* |
1134 | * make_request() can abort the operation when READA is being | |
1135 | * used and no empty request is available. | |
1136 | * | |
1137 | */ | |
1138 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1139 | ||
1140 | r1_bio->master_bio = bio; | |
aa8b57aa | 1141 | r1_bio->sectors = bio_sectors(bio); |
191ea9b2 | 1142 | r1_bio->state = 0; |
1da177e4 LT |
1143 | r1_bio->mddev = mddev; |
1144 | r1_bio->sector = bio->bi_sector; | |
1145 | ||
d2eb35ac N |
1146 | /* We might need to issue multiple reads to different |
1147 | * devices if there are bad blocks around, so we keep | |
1148 | * track of the number of reads in bio->bi_phys_segments. | |
1149 | * If this is 0, there is only one r1_bio and no locking | |
1150 | * will be needed when requests complete. If it is | |
1151 | * non-zero, then it is the number of not-completed requests. | |
1152 | */ | |
1153 | bio->bi_phys_segments = 0; | |
1154 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
1155 | ||
a362357b | 1156 | if (rw == READ) { |
1da177e4 LT |
1157 | /* |
1158 | * read balancing logic: | |
1159 | */ | |
d2eb35ac N |
1160 | int rdisk; |
1161 | ||
1162 | read_again: | |
1163 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1da177e4 LT |
1164 | |
1165 | if (rdisk < 0) { | |
1166 | /* couldn't find anywhere to read from */ | |
1167 | raid_end_bio_io(r1_bio); | |
5a7bbad2 | 1168 | return; |
1da177e4 LT |
1169 | } |
1170 | mirror = conf->mirrors + rdisk; | |
1171 | ||
e555190d N |
1172 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1173 | bitmap) { | |
1174 | /* Reading from a write-mostly device must | |
1175 | * take care not to over-take any writes | |
1176 | * that are 'behind' | |
1177 | */ | |
1178 | wait_event(bitmap->behind_wait, | |
1179 | atomic_read(&bitmap->behind_writes) == 0); | |
1180 | } | |
1da177e4 LT |
1181 | r1_bio->read_disk = rdisk; |
1182 | ||
a167f663 | 1183 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
6678d83f KO |
1184 | bio_trim(read_bio, r1_bio->sector - bio->bi_sector, |
1185 | max_sectors); | |
1da177e4 LT |
1186 | |
1187 | r1_bio->bios[rdisk] = read_bio; | |
1188 | ||
1189 | read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; | |
1190 | read_bio->bi_bdev = mirror->rdev->bdev; | |
1191 | read_bio->bi_end_io = raid1_end_read_request; | |
7b6d91da | 1192 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1193 | read_bio->bi_private = r1_bio; |
1194 | ||
d2eb35ac N |
1195 | if (max_sectors < r1_bio->sectors) { |
1196 | /* could not read all from this device, so we will | |
1197 | * need another r1_bio. | |
1198 | */ | |
d2eb35ac N |
1199 | |
1200 | sectors_handled = (r1_bio->sector + max_sectors | |
1201 | - bio->bi_sector); | |
1202 | r1_bio->sectors = max_sectors; | |
1203 | spin_lock_irq(&conf->device_lock); | |
1204 | if (bio->bi_phys_segments == 0) | |
1205 | bio->bi_phys_segments = 2; | |
1206 | else | |
1207 | bio->bi_phys_segments++; | |
1208 | spin_unlock_irq(&conf->device_lock); | |
1209 | /* Cannot call generic_make_request directly | |
1210 | * as that will be queued in __make_request | |
1211 | * and subsequent mempool_alloc might block waiting | |
1212 | * for it. So hand bio over to raid1d. | |
1213 | */ | |
1214 | reschedule_retry(r1_bio); | |
1215 | ||
1216 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1217 | ||
1218 | r1_bio->master_bio = bio; | |
aa8b57aa | 1219 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
d2eb35ac N |
1220 | r1_bio->state = 0; |
1221 | r1_bio->mddev = mddev; | |
1222 | r1_bio->sector = bio->bi_sector + sectors_handled; | |
1223 | goto read_again; | |
1224 | } else | |
1225 | generic_make_request(read_bio); | |
5a7bbad2 | 1226 | return; |
1da177e4 LT |
1227 | } |
1228 | ||
1229 | /* | |
1230 | * WRITE: | |
1231 | */ | |
34db0cd6 N |
1232 | if (conf->pending_count >= max_queued_requests) { |
1233 | md_wakeup_thread(mddev->thread); | |
1234 | wait_event(conf->wait_barrier, | |
1235 | conf->pending_count < max_queued_requests); | |
1236 | } | |
1f68f0c4 | 1237 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1238 | * inc refcount on their rdev. Record them by setting |
1239 | * bios[x] to bio | |
1f68f0c4 N |
1240 | * If there are known/acknowledged bad blocks on any device on |
1241 | * which we have seen a write error, we want to avoid writing those | |
1242 | * blocks. | |
1243 | * This potentially requires several writes to write around | |
1244 | * the bad blocks. Each set of writes gets it's own r1bio | |
1245 | * with a set of bios attached. | |
1da177e4 | 1246 | */ |
c3b328ac | 1247 | |
8f19ccb2 | 1248 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1249 | retry_write: |
79ef3a8a | 1250 | r1_bio->start_next_window = start_next_window; |
6bfe0b49 | 1251 | blocked_rdev = NULL; |
1da177e4 | 1252 | rcu_read_lock(); |
1f68f0c4 | 1253 | max_sectors = r1_bio->sectors; |
1da177e4 | 1254 | for (i = 0; i < disks; i++) { |
3cb03002 | 1255 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1256 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1257 | atomic_inc(&rdev->nr_pending); | |
1258 | blocked_rdev = rdev; | |
1259 | break; | |
1260 | } | |
1f68f0c4 | 1261 | r1_bio->bios[i] = NULL; |
6b740b8d N |
1262 | if (!rdev || test_bit(Faulty, &rdev->flags) |
1263 | || test_bit(Unmerged, &rdev->flags)) { | |
8f19ccb2 N |
1264 | if (i < conf->raid_disks) |
1265 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1266 | continue; |
1267 | } | |
1268 | ||
1269 | atomic_inc(&rdev->nr_pending); | |
1270 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1271 | sector_t first_bad; | |
1272 | int bad_sectors; | |
1273 | int is_bad; | |
1274 | ||
1275 | is_bad = is_badblock(rdev, r1_bio->sector, | |
1276 | max_sectors, | |
1277 | &first_bad, &bad_sectors); | |
1278 | if (is_bad < 0) { | |
1279 | /* mustn't write here until the bad block is | |
1280 | * acknowledged*/ | |
1281 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1282 | blocked_rdev = rdev; | |
1283 | break; | |
1284 | } | |
1285 | if (is_bad && first_bad <= r1_bio->sector) { | |
1286 | /* Cannot write here at all */ | |
1287 | bad_sectors -= (r1_bio->sector - first_bad); | |
1288 | if (bad_sectors < max_sectors) | |
1289 | /* mustn't write more than bad_sectors | |
1290 | * to other devices yet | |
1291 | */ | |
1292 | max_sectors = bad_sectors; | |
03c902e1 | 1293 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1294 | /* We don't set R1BIO_Degraded as that |
1295 | * only applies if the disk is | |
1296 | * missing, so it might be re-added, | |
1297 | * and we want to know to recover this | |
1298 | * chunk. | |
1299 | * In this case the device is here, | |
1300 | * and the fact that this chunk is not | |
1301 | * in-sync is recorded in the bad | |
1302 | * block log | |
1303 | */ | |
1304 | continue; | |
964147d5 | 1305 | } |
1f68f0c4 N |
1306 | if (is_bad) { |
1307 | int good_sectors = first_bad - r1_bio->sector; | |
1308 | if (good_sectors < max_sectors) | |
1309 | max_sectors = good_sectors; | |
1310 | } | |
1311 | } | |
1312 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1313 | } |
1314 | rcu_read_unlock(); | |
1315 | ||
6bfe0b49 DW |
1316 | if (unlikely(blocked_rdev)) { |
1317 | /* Wait for this device to become unblocked */ | |
1318 | int j; | |
79ef3a8a | 1319 | sector_t old = start_next_window; |
6bfe0b49 DW |
1320 | |
1321 | for (j = 0; j < i; j++) | |
1322 | if (r1_bio->bios[j]) | |
1323 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1324 | r1_bio->state = 0; |
79ef3a8a | 1325 | allow_barrier(conf, start_next_window, bio->bi_sector); |
6bfe0b49 | 1326 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
79ef3a8a | 1327 | start_next_window = wait_barrier(conf, bio); |
1328 | /* | |
1329 | * We must make sure the multi r1bios of bio have | |
1330 | * the same value of bi_phys_segments | |
1331 | */ | |
1332 | if (bio->bi_phys_segments && old && | |
1333 | old != start_next_window) | |
1334 | /* Wait for the former r1bio(s) to complete */ | |
1335 | wait_event(conf->wait_barrier, | |
1336 | bio->bi_phys_segments == 1); | |
6bfe0b49 DW |
1337 | goto retry_write; |
1338 | } | |
1339 | ||
1f68f0c4 N |
1340 | if (max_sectors < r1_bio->sectors) { |
1341 | /* We are splitting this write into multiple parts, so | |
1342 | * we need to prepare for allocating another r1_bio. | |
1343 | */ | |
1344 | r1_bio->sectors = max_sectors; | |
1345 | spin_lock_irq(&conf->device_lock); | |
1346 | if (bio->bi_phys_segments == 0) | |
1347 | bio->bi_phys_segments = 2; | |
1348 | else | |
1349 | bio->bi_phys_segments++; | |
1350 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1351 | } |
1f68f0c4 | 1352 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector; |
4b6d287f | 1353 | |
4e78064f | 1354 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1355 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1356 | |
1f68f0c4 | 1357 | first_clone = 1; |
1da177e4 LT |
1358 | for (i = 0; i < disks; i++) { |
1359 | struct bio *mbio; | |
1360 | if (!r1_bio->bios[i]) | |
1361 | continue; | |
1362 | ||
a167f663 | 1363 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
6678d83f | 1364 | bio_trim(mbio, r1_bio->sector - bio->bi_sector, max_sectors); |
1f68f0c4 N |
1365 | |
1366 | if (first_clone) { | |
1367 | /* do behind I/O ? | |
1368 | * Not if there are too many, or cannot | |
1369 | * allocate memory, or a reader on WriteMostly | |
1370 | * is waiting for behind writes to flush */ | |
1371 | if (bitmap && | |
1372 | (atomic_read(&bitmap->behind_writes) | |
1373 | < mddev->bitmap_info.max_write_behind) && | |
1374 | !waitqueue_active(&bitmap->behind_wait)) | |
1375 | alloc_behind_pages(mbio, r1_bio); | |
1376 | ||
1377 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1378 | r1_bio->sectors, | |
1379 | test_bit(R1BIO_BehindIO, | |
1380 | &r1_bio->state)); | |
1381 | first_clone = 0; | |
1382 | } | |
2ca68f5e | 1383 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1384 | struct bio_vec *bvec; |
1385 | int j; | |
1386 | ||
cb34e057 KO |
1387 | /* |
1388 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1389 | */ |
d74c6d51 | 1390 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1391 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1392 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1393 | atomic_inc(&r1_bio->behind_remaining); | |
1394 | } | |
1395 | ||
1f68f0c4 N |
1396 | r1_bio->bios[i] = mbio; |
1397 | ||
1398 | mbio->bi_sector = (r1_bio->sector + | |
1399 | conf->mirrors[i].rdev->data_offset); | |
1400 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1401 | mbio->bi_end_io = raid1_end_write_request; | |
c8dc9c65 JL |
1402 | mbio->bi_rw = |
1403 | WRITE | do_flush_fua | do_sync | do_discard | do_same; | |
1f68f0c4 N |
1404 | mbio->bi_private = r1_bio; |
1405 | ||
1da177e4 | 1406 | atomic_inc(&r1_bio->remaining); |
f54a9d0e N |
1407 | |
1408 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); | |
1409 | if (cb) | |
1410 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1411 | else | |
1412 | plug = NULL; | |
4e78064f | 1413 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1414 | if (plug) { |
1415 | bio_list_add(&plug->pending, mbio); | |
1416 | plug->pending_cnt++; | |
1417 | } else { | |
1418 | bio_list_add(&conf->pending_bio_list, mbio); | |
1419 | conf->pending_count++; | |
1420 | } | |
4e78064f | 1421 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1422 | if (!plug) |
b357f04a | 1423 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1424 | } |
079fa166 N |
1425 | /* Mustn't call r1_bio_write_done before this next test, |
1426 | * as it could result in the bio being freed. | |
1427 | */ | |
aa8b57aa | 1428 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1429 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1430 | /* We need another r1_bio. It has already been counted |
1431 | * in bio->bi_phys_segments | |
1432 | */ | |
1433 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1434 | r1_bio->master_bio = bio; | |
aa8b57aa | 1435 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
1f68f0c4 N |
1436 | r1_bio->state = 0; |
1437 | r1_bio->mddev = mddev; | |
1438 | r1_bio->sector = bio->bi_sector + sectors_handled; | |
1439 | goto retry_write; | |
1440 | } | |
1441 | ||
079fa166 N |
1442 | r1_bio_write_done(r1_bio); |
1443 | ||
1444 | /* In case raid1d snuck in to freeze_array */ | |
1445 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1446 | } |
1447 | ||
fd01b88c | 1448 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1449 | { |
e8096360 | 1450 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1451 | int i; |
1452 | ||
1453 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1454 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1455 | rcu_read_lock(); |
1456 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1457 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1458 | seq_printf(seq, "%s", |
ddac7c7e N |
1459 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1460 | } | |
1461 | rcu_read_unlock(); | |
1da177e4 LT |
1462 | seq_printf(seq, "]"); |
1463 | } | |
1464 | ||
1465 | ||
fd01b88c | 1466 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1467 | { |
1468 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1469 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1470 | |
1471 | /* | |
1472 | * If it is not operational, then we have already marked it as dead | |
1473 | * else if it is the last working disks, ignore the error, let the | |
1474 | * next level up know. | |
1475 | * else mark the drive as failed | |
1476 | */ | |
b2d444d7 | 1477 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1478 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1479 | /* |
1480 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1481 | * normal single drive. |
1482 | * However don't try a recovery from this drive as | |
1483 | * it is very likely to fail. | |
1da177e4 | 1484 | */ |
5389042f | 1485 | conf->recovery_disabled = mddev->recovery_disabled; |
1da177e4 | 1486 | return; |
4044ba58 | 1487 | } |
de393cde | 1488 | set_bit(Blocked, &rdev->flags); |
c04be0aa N |
1489 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1490 | unsigned long flags; | |
1491 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 1492 | mddev->degraded++; |
dd00a99e | 1493 | set_bit(Faulty, &rdev->flags); |
c04be0aa | 1494 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
1495 | /* |
1496 | * if recovery is running, make sure it aborts. | |
1497 | */ | |
dfc70645 | 1498 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
dd00a99e N |
1499 | } else |
1500 | set_bit(Faulty, &rdev->flags); | |
850b2b42 | 1501 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
067032bc JP |
1502 | printk(KERN_ALERT |
1503 | "md/raid1:%s: Disk failure on %s, disabling device.\n" | |
1504 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
9dd1e2fa N |
1505 | mdname(mddev), bdevname(rdev->bdev, b), |
1506 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1507 | } |
1508 | ||
e8096360 | 1509 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1510 | { |
1511 | int i; | |
1da177e4 | 1512 | |
9dd1e2fa | 1513 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
1da177e4 | 1514 | if (!conf) { |
9dd1e2fa | 1515 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1516 | return; |
1517 | } | |
9dd1e2fa | 1518 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1519 | conf->raid_disks); |
1520 | ||
ddac7c7e | 1521 | rcu_read_lock(); |
1da177e4 LT |
1522 | for (i = 0; i < conf->raid_disks; i++) { |
1523 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1524 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1525 | if (rdev) |
9dd1e2fa | 1526 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
ddac7c7e N |
1527 | i, !test_bit(In_sync, &rdev->flags), |
1528 | !test_bit(Faulty, &rdev->flags), | |
1529 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1530 | } |
ddac7c7e | 1531 | rcu_read_unlock(); |
1da177e4 LT |
1532 | } |
1533 | ||
e8096360 | 1534 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1535 | { |
79ef3a8a | 1536 | wait_barrier(conf, NULL); |
1537 | allow_barrier(conf, 0, 0); | |
1da177e4 LT |
1538 | |
1539 | mempool_destroy(conf->r1buf_pool); | |
1540 | conf->r1buf_pool = NULL; | |
79ef3a8a | 1541 | |
1542 | conf->next_resync = 0; | |
1543 | conf->start_next_window = MaxSector; | |
1da177e4 LT |
1544 | } |
1545 | ||
fd01b88c | 1546 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1547 | { |
1548 | int i; | |
e8096360 | 1549 | struct r1conf *conf = mddev->private; |
6b965620 N |
1550 | int count = 0; |
1551 | unsigned long flags; | |
1da177e4 LT |
1552 | |
1553 | /* | |
1554 | * Find all failed disks within the RAID1 configuration | |
ddac7c7e N |
1555 | * and mark them readable. |
1556 | * Called under mddev lock, so rcu protection not needed. | |
1da177e4 LT |
1557 | */ |
1558 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1559 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1560 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1561 | if (repl | |
1562 | && repl->recovery_offset == MaxSector | |
1563 | && !test_bit(Faulty, &repl->flags) | |
1564 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1565 | /* replacement has just become active */ | |
1566 | if (!rdev || | |
1567 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1568 | count++; | |
1569 | if (rdev) { | |
1570 | /* Replaced device not technically | |
1571 | * faulty, but we need to be sure | |
1572 | * it gets removed and never re-added | |
1573 | */ | |
1574 | set_bit(Faulty, &rdev->flags); | |
1575 | sysfs_notify_dirent_safe( | |
1576 | rdev->sysfs_state); | |
1577 | } | |
1578 | } | |
ddac7c7e | 1579 | if (rdev |
61e4947c | 1580 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1581 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1582 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1583 | count++; |
654e8b5a | 1584 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1585 | } |
1586 | } | |
6b965620 N |
1587 | spin_lock_irqsave(&conf->device_lock, flags); |
1588 | mddev->degraded -= count; | |
1589 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1590 | |
1591 | print_conf(conf); | |
6b965620 | 1592 | return count; |
1da177e4 LT |
1593 | } |
1594 | ||
1595 | ||
fd01b88c | 1596 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1597 | { |
e8096360 | 1598 | struct r1conf *conf = mddev->private; |
199050ea | 1599 | int err = -EEXIST; |
41158c7e | 1600 | int mirror = 0; |
0eaf822c | 1601 | struct raid1_info *p; |
6c2fce2e | 1602 | int first = 0; |
30194636 | 1603 | int last = conf->raid_disks - 1; |
6b740b8d | 1604 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
1da177e4 | 1605 | |
5389042f N |
1606 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1607 | return -EBUSY; | |
1608 | ||
6c2fce2e NB |
1609 | if (rdev->raid_disk >= 0) |
1610 | first = last = rdev->raid_disk; | |
1611 | ||
6b740b8d N |
1612 | if (q->merge_bvec_fn) { |
1613 | set_bit(Unmerged, &rdev->flags); | |
1614 | mddev->merge_check_needed = 1; | |
1615 | } | |
1616 | ||
7ef449d1 N |
1617 | for (mirror = first; mirror <= last; mirror++) { |
1618 | p = conf->mirrors+mirror; | |
1619 | if (!p->rdev) { | |
1da177e4 | 1620 | |
9092c02d JB |
1621 | if (mddev->gendisk) |
1622 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1623 | rdev->data_offset << 9); | |
1da177e4 LT |
1624 | |
1625 | p->head_position = 0; | |
1626 | rdev->raid_disk = mirror; | |
199050ea | 1627 | err = 0; |
6aea114a N |
1628 | /* As all devices are equivalent, we don't need a full recovery |
1629 | * if this was recently any drive of the array | |
1630 | */ | |
1631 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1632 | conf->fullsync = 1; |
d6065f7b | 1633 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1634 | break; |
1635 | } | |
7ef449d1 N |
1636 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1637 | p[conf->raid_disks].rdev == NULL) { | |
1638 | /* Add this device as a replacement */ | |
1639 | clear_bit(In_sync, &rdev->flags); | |
1640 | set_bit(Replacement, &rdev->flags); | |
1641 | rdev->raid_disk = mirror; | |
1642 | err = 0; | |
1643 | conf->fullsync = 1; | |
1644 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1645 | break; | |
1646 | } | |
1647 | } | |
6b740b8d N |
1648 | if (err == 0 && test_bit(Unmerged, &rdev->flags)) { |
1649 | /* Some requests might not have seen this new | |
1650 | * merge_bvec_fn. We must wait for them to complete | |
1651 | * before merging the device fully. | |
1652 | * First we make sure any code which has tested | |
1653 | * our function has submitted the request, then | |
1654 | * we wait for all outstanding requests to complete. | |
1655 | */ | |
1656 | synchronize_sched(); | |
e2d59925 N |
1657 | freeze_array(conf, 0); |
1658 | unfreeze_array(conf); | |
6b740b8d N |
1659 | clear_bit(Unmerged, &rdev->flags); |
1660 | } | |
ac5e7113 | 1661 | md_integrity_add_rdev(rdev, mddev); |
9092c02d | 1662 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1663 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1664 | print_conf(conf); |
199050ea | 1665 | return err; |
1da177e4 LT |
1666 | } |
1667 | ||
b8321b68 | 1668 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1669 | { |
e8096360 | 1670 | struct r1conf *conf = mddev->private; |
1da177e4 | 1671 | int err = 0; |
b8321b68 | 1672 | int number = rdev->raid_disk; |
0eaf822c | 1673 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1674 | |
b014f14c N |
1675 | if (rdev != p->rdev) |
1676 | p = conf->mirrors + conf->raid_disks + number; | |
1677 | ||
1da177e4 | 1678 | print_conf(conf); |
b8321b68 | 1679 | if (rdev == p->rdev) { |
b2d444d7 | 1680 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1681 | atomic_read(&rdev->nr_pending)) { |
1682 | err = -EBUSY; | |
1683 | goto abort; | |
1684 | } | |
046abeed | 1685 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1686 | * is not possible. |
1687 | */ | |
1688 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1689 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1690 | mddev->degraded < conf->raid_disks) { |
1691 | err = -EBUSY; | |
1692 | goto abort; | |
1693 | } | |
1da177e4 | 1694 | p->rdev = NULL; |
fbd568a3 | 1695 | synchronize_rcu(); |
1da177e4 LT |
1696 | if (atomic_read(&rdev->nr_pending)) { |
1697 | /* lost the race, try later */ | |
1698 | err = -EBUSY; | |
1699 | p->rdev = rdev; | |
ac5e7113 | 1700 | goto abort; |
8c7a2c2b N |
1701 | } else if (conf->mirrors[conf->raid_disks + number].rdev) { |
1702 | /* We just removed a device that is being replaced. | |
1703 | * Move down the replacement. We drain all IO before | |
1704 | * doing this to avoid confusion. | |
1705 | */ | |
1706 | struct md_rdev *repl = | |
1707 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1708 | freeze_array(conf, 0); |
8c7a2c2b N |
1709 | clear_bit(Replacement, &repl->flags); |
1710 | p->rdev = repl; | |
1711 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1712 | unfreeze_array(conf); |
8c7a2c2b N |
1713 | clear_bit(WantReplacement, &rdev->flags); |
1714 | } else | |
b014f14c | 1715 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1716 | err = md_integrity_register(mddev); |
1da177e4 LT |
1717 | } |
1718 | abort: | |
1719 | ||
1720 | print_conf(conf); | |
1721 | return err; | |
1722 | } | |
1723 | ||
1724 | ||
6712ecf8 | 1725 | static void end_sync_read(struct bio *bio, int error) |
1da177e4 | 1726 | { |
9f2c9d12 | 1727 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1728 | |
0fc280f6 | 1729 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1730 | |
1da177e4 LT |
1731 | /* |
1732 | * we have read a block, now it needs to be re-written, | |
1733 | * or re-read if the read failed. | |
1734 | * We don't do much here, just schedule handling by raid1d | |
1735 | */ | |
69382e85 | 1736 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1da177e4 | 1737 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1738 | |
1739 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1740 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1741 | } |
1742 | ||
6712ecf8 | 1743 | static void end_sync_write(struct bio *bio, int error) |
1da177e4 LT |
1744 | { |
1745 | int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9f2c9d12 | 1746 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1747 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1748 | struct r1conf *conf = mddev->private; |
1da177e4 | 1749 | int mirror=0; |
4367af55 N |
1750 | sector_t first_bad; |
1751 | int bad_sectors; | |
1da177e4 | 1752 | |
ba3ae3be NK |
1753 | mirror = find_bio_disk(r1_bio, bio); |
1754 | ||
6b1117d5 | 1755 | if (!uptodate) { |
57dab0bd | 1756 | sector_t sync_blocks = 0; |
6b1117d5 N |
1757 | sector_t s = r1_bio->sector; |
1758 | long sectors_to_go = r1_bio->sectors; | |
1759 | /* make sure these bits doesn't get cleared. */ | |
1760 | do { | |
5e3db645 | 1761 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1762 | &sync_blocks, 1); |
1763 | s += sync_blocks; | |
1764 | sectors_to_go -= sync_blocks; | |
1765 | } while (sectors_to_go > 0); | |
d8f05d29 N |
1766 | set_bit(WriteErrorSeen, |
1767 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
1768 | if (!test_and_set_bit(WantReplacement, |
1769 | &conf->mirrors[mirror].rdev->flags)) | |
1770 | set_bit(MD_RECOVERY_NEEDED, & | |
1771 | mddev->recovery); | |
d8f05d29 | 1772 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 N |
1773 | } else if (is_badblock(conf->mirrors[mirror].rdev, |
1774 | r1_bio->sector, | |
1775 | r1_bio->sectors, | |
3a9f28a5 N |
1776 | &first_bad, &bad_sectors) && |
1777 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1778 | r1_bio->sector, | |
1779 | r1_bio->sectors, | |
1780 | &first_bad, &bad_sectors) | |
1781 | ) | |
4367af55 | 1782 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1783 | |
1da177e4 | 1784 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1785 | int s = r1_bio->sectors; |
d8f05d29 N |
1786 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1787 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1788 | reschedule_retry(r1_bio); |
1789 | else { | |
1790 | put_buf(r1_bio); | |
1791 | md_done_sync(mddev, s, uptodate); | |
1792 | } | |
1da177e4 | 1793 | } |
1da177e4 LT |
1794 | } |
1795 | ||
3cb03002 | 1796 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1797 | int sectors, struct page *page, int rw) |
1798 | { | |
1799 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
1800 | /* success */ | |
1801 | return 1; | |
19d67169 | 1802 | if (rw == WRITE) { |
d8f05d29 | 1803 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1804 | if (!test_and_set_bit(WantReplacement, |
1805 | &rdev->flags)) | |
1806 | set_bit(MD_RECOVERY_NEEDED, & | |
1807 | rdev->mddev->recovery); | |
1808 | } | |
d8f05d29 N |
1809 | /* need to record an error - either for the block or the device */ |
1810 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1811 | md_error(rdev->mddev, rdev); | |
1812 | return 0; | |
1813 | } | |
1814 | ||
9f2c9d12 | 1815 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1816 | { |
a68e5870 N |
1817 | /* Try some synchronous reads of other devices to get |
1818 | * good data, much like with normal read errors. Only | |
1819 | * read into the pages we already have so we don't | |
1820 | * need to re-issue the read request. | |
1821 | * We don't need to freeze the array, because being in an | |
1822 | * active sync request, there is no normal IO, and | |
1823 | * no overlapping syncs. | |
06f60385 N |
1824 | * We don't need to check is_badblock() again as we |
1825 | * made sure that anything with a bad block in range | |
1826 | * will have bi_end_io clear. | |
a68e5870 | 1827 | */ |
fd01b88c | 1828 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1829 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1830 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1831 | sector_t sect = r1_bio->sector; | |
1832 | int sectors = r1_bio->sectors; | |
1833 | int idx = 0; | |
1834 | ||
1835 | while(sectors) { | |
1836 | int s = sectors; | |
1837 | int d = r1_bio->read_disk; | |
1838 | int success = 0; | |
3cb03002 | 1839 | struct md_rdev *rdev; |
78d7f5f7 | 1840 | int start; |
a68e5870 N |
1841 | |
1842 | if (s > (PAGE_SIZE>>9)) | |
1843 | s = PAGE_SIZE >> 9; | |
1844 | do { | |
1845 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1846 | /* No rcu protection needed here devices | |
1847 | * can only be removed when no resync is | |
1848 | * active, and resync is currently active | |
1849 | */ | |
1850 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1851 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 N |
1852 | bio->bi_io_vec[idx].bv_page, |
1853 | READ, false)) { | |
1854 | success = 1; | |
1855 | break; | |
1856 | } | |
1857 | } | |
1858 | d++; | |
8f19ccb2 | 1859 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1860 | d = 0; |
1861 | } while (!success && d != r1_bio->read_disk); | |
1862 | ||
78d7f5f7 | 1863 | if (!success) { |
a68e5870 | 1864 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1865 | int abort = 0; |
1866 | /* Cannot read from anywhere, this block is lost. | |
1867 | * Record a bad block on each device. If that doesn't | |
1868 | * work just disable and interrupt the recovery. | |
1869 | * Don't fail devices as that won't really help. | |
1870 | */ | |
a68e5870 N |
1871 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
1872 | " for block %llu\n", | |
1873 | mdname(mddev), | |
1874 | bdevname(bio->bi_bdev, b), | |
1875 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1876 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1877 | rdev = conf->mirrors[d].rdev; |
1878 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1879 | continue; | |
1880 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1881 | abort = 1; | |
1882 | } | |
1883 | if (abort) { | |
d890fa2b N |
1884 | conf->recovery_disabled = |
1885 | mddev->recovery_disabled; | |
3a9f28a5 N |
1886 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1887 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1888 | put_buf(r1_bio); | |
1889 | return 0; | |
1890 | } | |
1891 | /* Try next page */ | |
1892 | sectors -= s; | |
1893 | sect += s; | |
1894 | idx++; | |
1895 | continue; | |
d11c171e | 1896 | } |
78d7f5f7 N |
1897 | |
1898 | start = d; | |
1899 | /* write it back and re-read */ | |
1900 | while (d != r1_bio->read_disk) { | |
1901 | if (d == 0) | |
8f19ccb2 | 1902 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1903 | d--; |
1904 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1905 | continue; | |
1906 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1907 | if (r1_sync_page_io(rdev, sect, s, |
1908 | bio->bi_io_vec[idx].bv_page, | |
1909 | WRITE) == 0) { | |
78d7f5f7 N |
1910 | r1_bio->bios[d]->bi_end_io = NULL; |
1911 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1912 | } |
78d7f5f7 N |
1913 | } |
1914 | d = start; | |
1915 | while (d != r1_bio->read_disk) { | |
1916 | if (d == 0) | |
8f19ccb2 | 1917 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1918 | d--; |
1919 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1920 | continue; | |
1921 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1922 | if (r1_sync_page_io(rdev, sect, s, |
1923 | bio->bi_io_vec[idx].bv_page, | |
1924 | READ) != 0) | |
9d3d8011 | 1925 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1926 | } |
a68e5870 N |
1927 | sectors -= s; |
1928 | sect += s; | |
1929 | idx ++; | |
1930 | } | |
78d7f5f7 | 1931 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
7ca78d57 | 1932 | set_bit(BIO_UPTODATE, &bio->bi_flags); |
a68e5870 N |
1933 | return 1; |
1934 | } | |
1935 | ||
9f2c9d12 | 1936 | static int process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1937 | { |
1938 | /* We have read all readable devices. If we haven't | |
1939 | * got the block, then there is no hope left. | |
1940 | * If we have, then we want to do a comparison | |
1941 | * and skip the write if everything is the same. | |
1942 | * If any blocks failed to read, then we need to | |
1943 | * attempt an over-write | |
1944 | */ | |
fd01b88c | 1945 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1946 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1947 | int primary; |
1948 | int i; | |
f4380a91 | 1949 | int vcnt; |
a68e5870 | 1950 | |
30bc9b53 N |
1951 | /* Fix variable parts of all bios */ |
1952 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
1953 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
1954 | int j; | |
1955 | int size; | |
1956 | struct bio *b = r1_bio->bios[i]; | |
1957 | if (b->bi_end_io != end_sync_read) | |
1958 | continue; | |
1959 | /* fixup the bio for reuse */ | |
1960 | bio_reset(b); | |
1961 | b->bi_vcnt = vcnt; | |
1962 | b->bi_size = r1_bio->sectors << 9; | |
1963 | b->bi_sector = r1_bio->sector + | |
1964 | conf->mirrors[i].rdev->data_offset; | |
1965 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1966 | b->bi_end_io = end_sync_read; | |
1967 | b->bi_private = r1_bio; | |
1968 | ||
1969 | size = b->bi_size; | |
1970 | for (j = 0; j < vcnt ; j++) { | |
1971 | struct bio_vec *bi; | |
1972 | bi = &b->bi_io_vec[j]; | |
1973 | bi->bv_offset = 0; | |
1974 | if (size > PAGE_SIZE) | |
1975 | bi->bv_len = PAGE_SIZE; | |
1976 | else | |
1977 | bi->bv_len = size; | |
1978 | size -= PAGE_SIZE; | |
1979 | } | |
1980 | } | |
8f19ccb2 | 1981 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 N |
1982 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
1983 | test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) { | |
1984 | r1_bio->bios[primary]->bi_end_io = NULL; | |
1985 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
1986 | break; | |
1987 | } | |
1988 | r1_bio->read_disk = primary; | |
8f19ccb2 | 1989 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 1990 | int j; |
78d7f5f7 N |
1991 | struct bio *pbio = r1_bio->bios[primary]; |
1992 | struct bio *sbio = r1_bio->bios[i]; | |
a68e5870 | 1993 | |
2aabaa65 | 1994 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 N |
1995 | continue; |
1996 | ||
1997 | if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) { | |
1998 | for (j = vcnt; j-- ; ) { | |
1999 | struct page *p, *s; | |
2000 | p = pbio->bi_io_vec[j].bv_page; | |
2001 | s = sbio->bi_io_vec[j].bv_page; | |
2002 | if (memcmp(page_address(p), | |
2003 | page_address(s), | |
5020ad7d | 2004 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 2005 | break; |
69382e85 | 2006 | } |
78d7f5f7 N |
2007 | } else |
2008 | j = 0; | |
2009 | if (j >= 0) | |
7f7583d4 | 2010 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 N |
2011 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
2012 | && test_bit(BIO_UPTODATE, &sbio->bi_flags))) { | |
2013 | /* No need to write to this device. */ | |
2014 | sbio->bi_end_io = NULL; | |
2015 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2016 | continue; | |
2017 | } | |
d3b45c2a KO |
2018 | |
2019 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2020 | } |
a68e5870 N |
2021 | return 0; |
2022 | } | |
2023 | ||
9f2c9d12 | 2024 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2025 | { |
e8096360 | 2026 | struct r1conf *conf = mddev->private; |
a68e5870 | 2027 | int i; |
8f19ccb2 | 2028 | int disks = conf->raid_disks * 2; |
a68e5870 N |
2029 | struct bio *bio, *wbio; |
2030 | ||
2031 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2032 | ||
a68e5870 N |
2033 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2034 | /* ouch - failed to read all of that. */ | |
2035 | if (!fix_sync_read_error(r1_bio)) | |
2036 | return; | |
7ca78d57 N |
2037 | |
2038 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
2039 | if (process_checks(r1_bio) < 0) | |
2040 | return; | |
d11c171e N |
2041 | /* |
2042 | * schedule writes | |
2043 | */ | |
1da177e4 LT |
2044 | atomic_set(&r1_bio->remaining, 1); |
2045 | for (i = 0; i < disks ; i++) { | |
2046 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2047 | if (wbio->bi_end_io == NULL || |
2048 | (wbio->bi_end_io == end_sync_read && | |
2049 | (i == r1_bio->read_disk || | |
2050 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
2051 | continue; |
2052 | ||
3e198f78 N |
2053 | wbio->bi_rw = WRITE; |
2054 | wbio->bi_end_io = end_sync_write; | |
1da177e4 | 2055 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2056 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2057 | |
1da177e4 LT |
2058 | generic_make_request(wbio); |
2059 | } | |
2060 | ||
2061 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2062 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2063 | int s = r1_bio->sectors; |
2064 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2065 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2066 | reschedule_retry(r1_bio); | |
2067 | else { | |
2068 | put_buf(r1_bio); | |
2069 | md_done_sync(mddev, s, 1); | |
2070 | } | |
1da177e4 LT |
2071 | } |
2072 | } | |
2073 | ||
2074 | /* | |
2075 | * This is a kernel thread which: | |
2076 | * | |
2077 | * 1. Retries failed read operations on working mirrors. | |
2078 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2079 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2080 | */ |
2081 | ||
e8096360 | 2082 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2083 | sector_t sect, int sectors) |
2084 | { | |
fd01b88c | 2085 | struct mddev *mddev = conf->mddev; |
867868fb N |
2086 | while(sectors) { |
2087 | int s = sectors; | |
2088 | int d = read_disk; | |
2089 | int success = 0; | |
2090 | int start; | |
3cb03002 | 2091 | struct md_rdev *rdev; |
867868fb N |
2092 | |
2093 | if (s > (PAGE_SIZE>>9)) | |
2094 | s = PAGE_SIZE >> 9; | |
2095 | ||
2096 | do { | |
2097 | /* Note: no rcu protection needed here | |
2098 | * as this is synchronous in the raid1d thread | |
2099 | * which is the thread that might remove | |
2100 | * a device. If raid1d ever becomes multi-threaded.... | |
2101 | */ | |
d2eb35ac N |
2102 | sector_t first_bad; |
2103 | int bad_sectors; | |
2104 | ||
867868fb N |
2105 | rdev = conf->mirrors[d].rdev; |
2106 | if (rdev && | |
da8840a7 | 2107 | (test_bit(In_sync, &rdev->flags) || |
2108 | (!test_bit(Faulty, &rdev->flags) && | |
2109 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac N |
2110 | is_badblock(rdev, sect, s, |
2111 | &first_bad, &bad_sectors) == 0 && | |
ccebd4c4 JB |
2112 | sync_page_io(rdev, sect, s<<9, |
2113 | conf->tmppage, READ, false)) | |
867868fb N |
2114 | success = 1; |
2115 | else { | |
2116 | d++; | |
8f19ccb2 | 2117 | if (d == conf->raid_disks * 2) |
867868fb N |
2118 | d = 0; |
2119 | } | |
2120 | } while (!success && d != read_disk); | |
2121 | ||
2122 | if (!success) { | |
d8f05d29 | 2123 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2124 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2125 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2126 | md_error(mddev, rdev); | |
867868fb N |
2127 | break; |
2128 | } | |
2129 | /* write it back and re-read */ | |
2130 | start = d; | |
2131 | while (d != read_disk) { | |
2132 | if (d==0) | |
8f19ccb2 | 2133 | d = conf->raid_disks * 2; |
867868fb N |
2134 | d--; |
2135 | rdev = conf->mirrors[d].rdev; | |
2136 | if (rdev && | |
d8f05d29 N |
2137 | test_bit(In_sync, &rdev->flags)) |
2138 | r1_sync_page_io(rdev, sect, s, | |
2139 | conf->tmppage, WRITE); | |
867868fb N |
2140 | } |
2141 | d = start; | |
2142 | while (d != read_disk) { | |
2143 | char b[BDEVNAME_SIZE]; | |
2144 | if (d==0) | |
8f19ccb2 | 2145 | d = conf->raid_disks * 2; |
867868fb N |
2146 | d--; |
2147 | rdev = conf->mirrors[d].rdev; | |
2148 | if (rdev && | |
2149 | test_bit(In_sync, &rdev->flags)) { | |
d8f05d29 N |
2150 | if (r1_sync_page_io(rdev, sect, s, |
2151 | conf->tmppage, READ)) { | |
867868fb N |
2152 | atomic_add(s, &rdev->corrected_errors); |
2153 | printk(KERN_INFO | |
9dd1e2fa | 2154 | "md/raid1:%s: read error corrected " |
867868fb N |
2155 | "(%d sectors at %llu on %s)\n", |
2156 | mdname(mddev), s, | |
969b755a RD |
2157 | (unsigned long long)(sect + |
2158 | rdev->data_offset), | |
867868fb N |
2159 | bdevname(rdev->bdev, b)); |
2160 | } | |
2161 | } | |
2162 | } | |
2163 | sectors -= s; | |
2164 | sect += s; | |
2165 | } | |
2166 | } | |
2167 | ||
9f2c9d12 | 2168 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2169 | { |
fd01b88c | 2170 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2171 | struct r1conf *conf = mddev->private; |
3cb03002 | 2172 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2173 | |
2174 | /* bio has the data to be written to device 'i' where | |
2175 | * we just recently had a write error. | |
2176 | * We repeatedly clone the bio and trim down to one block, | |
2177 | * then try the write. Where the write fails we record | |
2178 | * a bad block. | |
2179 | * It is conceivable that the bio doesn't exactly align with | |
2180 | * blocks. We must handle this somehow. | |
2181 | * | |
2182 | * We currently own a reference on the rdev. | |
2183 | */ | |
2184 | ||
2185 | int block_sectors; | |
2186 | sector_t sector; | |
2187 | int sectors; | |
2188 | int sect_to_write = r1_bio->sectors; | |
2189 | int ok = 1; | |
2190 | ||
2191 | if (rdev->badblocks.shift < 0) | |
2192 | return 0; | |
2193 | ||
2194 | block_sectors = 1 << rdev->badblocks.shift; | |
2195 | sector = r1_bio->sector; | |
2196 | sectors = ((sector + block_sectors) | |
2197 | & ~(sector_t)(block_sectors - 1)) | |
2198 | - sector; | |
2199 | ||
cd5ff9a1 N |
2200 | while (sect_to_write) { |
2201 | struct bio *wbio; | |
2202 | if (sectors > sect_to_write) | |
2203 | sectors = sect_to_write; | |
2204 | /* Write at 'sector' for 'sectors'*/ | |
2205 | ||
b783863f KO |
2206 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2207 | unsigned vcnt = r1_bio->behind_page_count; | |
2208 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2209 | ||
2210 | while (!vec->bv_page) { | |
2211 | vec++; | |
2212 | vcnt--; | |
2213 | } | |
2214 | ||
2215 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2216 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2217 | ||
2218 | wbio->bi_vcnt = vcnt; | |
2219 | } else { | |
2220 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2221 | } | |
2222 | ||
cd5ff9a1 | 2223 | wbio->bi_rw = WRITE; |
b783863f | 2224 | wbio->bi_sector = r1_bio->sector; |
cd5ff9a1 | 2225 | wbio->bi_size = r1_bio->sectors << 9; |
cd5ff9a1 | 2226 | |
6678d83f | 2227 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
cd5ff9a1 N |
2228 | wbio->bi_sector += rdev->data_offset; |
2229 | wbio->bi_bdev = rdev->bdev; | |
2230 | if (submit_bio_wait(WRITE, wbio) == 0) | |
2231 | /* failure! */ | |
2232 | ok = rdev_set_badblocks(rdev, sector, | |
2233 | sectors, 0) | |
2234 | && ok; | |
2235 | ||
2236 | bio_put(wbio); | |
2237 | sect_to_write -= sectors; | |
2238 | sector += sectors; | |
2239 | sectors = block_sectors; | |
2240 | } | |
2241 | return ok; | |
2242 | } | |
2243 | ||
e8096360 | 2244 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2245 | { |
2246 | int m; | |
2247 | int s = r1_bio->sectors; | |
8f19ccb2 | 2248 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2249 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2250 | struct bio *bio = r1_bio->bios[m]; |
2251 | if (bio->bi_end_io == NULL) | |
2252 | continue; | |
2253 | if (test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
2254 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { | |
c6563a8c | 2255 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce N |
2256 | } |
2257 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags) && | |
2258 | test_bit(R1BIO_WriteError, &r1_bio->state)) { | |
2259 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2260 | md_error(conf->mddev, rdev); | |
2261 | } | |
2262 | } | |
2263 | put_buf(r1_bio); | |
2264 | md_done_sync(conf->mddev, s, 1); | |
2265 | } | |
2266 | ||
e8096360 | 2267 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2268 | { |
2269 | int m; | |
8f19ccb2 | 2270 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2271 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2272 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2273 | rdev_clear_badblocks(rdev, |
2274 | r1_bio->sector, | |
c6563a8c | 2275 | r1_bio->sectors, 0); |
62096bce N |
2276 | rdev_dec_pending(rdev, conf->mddev); |
2277 | } else if (r1_bio->bios[m] != NULL) { | |
2278 | /* This drive got a write error. We need to | |
2279 | * narrow down and record precise write | |
2280 | * errors. | |
2281 | */ | |
2282 | if (!narrow_write_error(r1_bio, m)) { | |
2283 | md_error(conf->mddev, | |
2284 | conf->mirrors[m].rdev); | |
2285 | /* an I/O failed, we can't clear the bitmap */ | |
2286 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2287 | } | |
2288 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2289 | conf->mddev); | |
2290 | } | |
2291 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2292 | close_write(r1_bio); | |
2293 | raid_end_bio_io(r1_bio); | |
2294 | } | |
2295 | ||
e8096360 | 2296 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2297 | { |
2298 | int disk; | |
2299 | int max_sectors; | |
fd01b88c | 2300 | struct mddev *mddev = conf->mddev; |
62096bce N |
2301 | struct bio *bio; |
2302 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2303 | struct md_rdev *rdev; |
62096bce N |
2304 | |
2305 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2306 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2307 | * the block and we can fix it. | |
2308 | * We freeze all other IO, and try reading the block from | |
2309 | * other devices. When we find one, we re-write | |
2310 | * and check it that fixes the read error. | |
2311 | * This is all done synchronously while the array is | |
2312 | * frozen | |
2313 | */ | |
2314 | if (mddev->ro == 0) { | |
e2d59925 | 2315 | freeze_array(conf, 1); |
62096bce N |
2316 | fix_read_error(conf, r1_bio->read_disk, |
2317 | r1_bio->sector, r1_bio->sectors); | |
2318 | unfreeze_array(conf); | |
2319 | } else | |
2320 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); | |
7ad4d4a6 | 2321 | rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); |
62096bce N |
2322 | |
2323 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2324 | bdevname(bio->bi_bdev, b); | |
2325 | read_more: | |
2326 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2327 | if (disk == -1) { | |
2328 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" | |
2329 | " read error for block %llu\n", | |
2330 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
2331 | raid_end_bio_io(r1_bio); | |
2332 | } else { | |
2333 | const unsigned long do_sync | |
2334 | = r1_bio->master_bio->bi_rw & REQ_SYNC; | |
2335 | if (bio) { | |
2336 | r1_bio->bios[r1_bio->read_disk] = | |
2337 | mddev->ro ? IO_BLOCKED : NULL; | |
2338 | bio_put(bio); | |
2339 | } | |
2340 | r1_bio->read_disk = disk; | |
2341 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
6678d83f | 2342 | bio_trim(bio, r1_bio->sector - bio->bi_sector, max_sectors); |
62096bce N |
2343 | r1_bio->bios[r1_bio->read_disk] = bio; |
2344 | rdev = conf->mirrors[disk].rdev; | |
2345 | printk_ratelimited(KERN_ERR | |
2346 | "md/raid1:%s: redirecting sector %llu" | |
2347 | " to other mirror: %s\n", | |
2348 | mdname(mddev), | |
2349 | (unsigned long long)r1_bio->sector, | |
2350 | bdevname(rdev->bdev, b)); | |
2351 | bio->bi_sector = r1_bio->sector + rdev->data_offset; | |
2352 | bio->bi_bdev = rdev->bdev; | |
2353 | bio->bi_end_io = raid1_end_read_request; | |
2354 | bio->bi_rw = READ | do_sync; | |
2355 | bio->bi_private = r1_bio; | |
2356 | if (max_sectors < r1_bio->sectors) { | |
2357 | /* Drat - have to split this up more */ | |
2358 | struct bio *mbio = r1_bio->master_bio; | |
2359 | int sectors_handled = (r1_bio->sector + max_sectors | |
2360 | - mbio->bi_sector); | |
2361 | r1_bio->sectors = max_sectors; | |
2362 | spin_lock_irq(&conf->device_lock); | |
2363 | if (mbio->bi_phys_segments == 0) | |
2364 | mbio->bi_phys_segments = 2; | |
2365 | else | |
2366 | mbio->bi_phys_segments++; | |
2367 | spin_unlock_irq(&conf->device_lock); | |
2368 | generic_make_request(bio); | |
2369 | bio = NULL; | |
2370 | ||
2371 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2372 | ||
2373 | r1_bio->master_bio = mbio; | |
aa8b57aa | 2374 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
62096bce N |
2375 | r1_bio->state = 0; |
2376 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2377 | r1_bio->mddev = mddev; | |
2378 | r1_bio->sector = mbio->bi_sector + sectors_handled; | |
2379 | ||
2380 | goto read_more; | |
2381 | } else | |
2382 | generic_make_request(bio); | |
2383 | } | |
2384 | } | |
2385 | ||
4ed8731d | 2386 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2387 | { |
4ed8731d | 2388 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2389 | struct r1bio *r1_bio; |
1da177e4 | 2390 | unsigned long flags; |
e8096360 | 2391 | struct r1conf *conf = mddev->private; |
1da177e4 | 2392 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2393 | struct blk_plug plug; |
1da177e4 LT |
2394 | |
2395 | md_check_recovery(mddev); | |
e1dfa0a2 N |
2396 | |
2397 | blk_start_plug(&plug); | |
1da177e4 | 2398 | for (;;) { |
191ea9b2 | 2399 | |
0021b7bc | 2400 | flush_pending_writes(conf); |
191ea9b2 | 2401 | |
a35e63ef N |
2402 | spin_lock_irqsave(&conf->device_lock, flags); |
2403 | if (list_empty(head)) { | |
2404 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2405 | break; |
a35e63ef | 2406 | } |
9f2c9d12 | 2407 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2408 | list_del(head->prev); |
ddaf22ab | 2409 | conf->nr_queued--; |
1da177e4 LT |
2410 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2411 | ||
2412 | mddev = r1_bio->mddev; | |
070ec55d | 2413 | conf = mddev->private; |
4367af55 | 2414 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2415 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2416 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2417 | handle_sync_write_finished(conf, r1_bio); | |
2418 | else | |
4367af55 | 2419 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2420 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2421 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2422 | handle_write_finished(conf, r1_bio); | |
2423 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2424 | handle_read_error(conf, r1_bio); | |
2425 | else | |
d2eb35ac N |
2426 | /* just a partial read to be scheduled from separate |
2427 | * context | |
2428 | */ | |
2429 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2430 | |
1d9d5241 | 2431 | cond_resched(); |
de393cde N |
2432 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2433 | md_check_recovery(mddev); | |
1da177e4 | 2434 | } |
e1dfa0a2 | 2435 | blk_finish_plug(&plug); |
1da177e4 LT |
2436 | } |
2437 | ||
2438 | ||
e8096360 | 2439 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2440 | { |
2441 | int buffs; | |
2442 | ||
2443 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2444 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2445 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2446 | conf->poolinfo); | |
2447 | if (!conf->r1buf_pool) | |
2448 | return -ENOMEM; | |
2449 | conf->next_resync = 0; | |
2450 | return 0; | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * perform a "sync" on one "block" | |
2455 | * | |
2456 | * We need to make sure that no normal I/O request - particularly write | |
2457 | * requests - conflict with active sync requests. | |
2458 | * | |
2459 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2460 | * that can be installed to exclude normal IO requests. | |
2461 | */ | |
2462 | ||
fd01b88c | 2463 | static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 | 2464 | { |
e8096360 | 2465 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2466 | struct r1bio *r1_bio; |
1da177e4 LT |
2467 | struct bio *bio; |
2468 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2469 | int disk = -1; |
1da177e4 | 2470 | int i; |
3e198f78 N |
2471 | int wonly = -1; |
2472 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2473 | sector_t sync_blocks; |
e3b9703e | 2474 | int still_degraded = 0; |
06f60385 N |
2475 | int good_sectors = RESYNC_SECTORS; |
2476 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2477 | |
2478 | if (!conf->r1buf_pool) | |
2479 | if (init_resync(conf)) | |
57afd89f | 2480 | return 0; |
1da177e4 | 2481 | |
58c0fed4 | 2482 | max_sector = mddev->dev_sectors; |
1da177e4 | 2483 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2484 | /* If we aborted, we need to abort the |
2485 | * sync on the 'current' bitmap chunk (there will | |
2486 | * only be one in raid1 resync. | |
2487 | * We can find the current addess in mddev->curr_resync | |
2488 | */ | |
6a806c51 N |
2489 | if (mddev->curr_resync < max_sector) /* aborted */ |
2490 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2491 | &sync_blocks, 1); |
6a806c51 | 2492 | else /* completed sync */ |
191ea9b2 | 2493 | conf->fullsync = 0; |
6a806c51 N |
2494 | |
2495 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 LT |
2496 | close_sync(conf); |
2497 | return 0; | |
2498 | } | |
2499 | ||
07d84d10 N |
2500 | if (mddev->bitmap == NULL && |
2501 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2502 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2503 | conf->fullsync == 0) { |
2504 | *skipped = 1; | |
2505 | return max_sector - sector_nr; | |
2506 | } | |
6394cca5 N |
2507 | /* before building a request, check if we can skip these blocks.. |
2508 | * This call the bitmap_start_sync doesn't actually record anything | |
2509 | */ | |
e3b9703e | 2510 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2511 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2512 | /* We can skip this block, and probably several more */ |
2513 | *skipped = 1; | |
2514 | return sync_blocks; | |
2515 | } | |
1da177e4 | 2516 | /* |
17999be4 N |
2517 | * If there is non-resync activity waiting for a turn, |
2518 | * and resync is going fast enough, | |
2519 | * then let it though before starting on this new sync request. | |
1da177e4 | 2520 | */ |
17999be4 | 2521 | if (!go_faster && conf->nr_waiting) |
1da177e4 | 2522 | msleep_interruptible(1000); |
17999be4 | 2523 | |
b47490c9 | 2524 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
1c4588e9 | 2525 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 N |
2526 | raise_barrier(conf); |
2527 | ||
2528 | conf->next_resync = sector_nr; | |
1da177e4 | 2529 | |
3e198f78 | 2530 | rcu_read_lock(); |
1da177e4 | 2531 | /* |
3e198f78 N |
2532 | * If we get a correctably read error during resync or recovery, |
2533 | * we might want to read from a different device. So we | |
2534 | * flag all drives that could conceivably be read from for READ, | |
2535 | * and any others (which will be non-In_sync devices) for WRITE. | |
2536 | * If a read fails, we try reading from something else for which READ | |
2537 | * is OK. | |
1da177e4 | 2538 | */ |
1da177e4 | 2539 | |
1da177e4 LT |
2540 | r1_bio->mddev = mddev; |
2541 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2542 | r1_bio->state = 0; |
1da177e4 | 2543 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2544 | |
8f19ccb2 | 2545 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2546 | struct md_rdev *rdev; |
1da177e4 | 2547 | bio = r1_bio->bios[i]; |
2aabaa65 | 2548 | bio_reset(bio); |
1da177e4 | 2549 | |
3e198f78 N |
2550 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2551 | if (rdev == NULL || | |
06f60385 | 2552 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2553 | if (i < conf->raid_disks) |
2554 | still_degraded = 1; | |
3e198f78 | 2555 | } else if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
2556 | bio->bi_rw = WRITE; |
2557 | bio->bi_end_io = end_sync_write; | |
2558 | write_targets ++; | |
3e198f78 N |
2559 | } else { |
2560 | /* may need to read from here */ | |
06f60385 N |
2561 | sector_t first_bad = MaxSector; |
2562 | int bad_sectors; | |
2563 | ||
2564 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2565 | &first_bad, &bad_sectors)) { | |
2566 | if (first_bad > sector_nr) | |
2567 | good_sectors = first_bad - sector_nr; | |
2568 | else { | |
2569 | bad_sectors -= (sector_nr - first_bad); | |
2570 | if (min_bad == 0 || | |
2571 | min_bad > bad_sectors) | |
2572 | min_bad = bad_sectors; | |
2573 | } | |
2574 | } | |
2575 | if (sector_nr < first_bad) { | |
2576 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2577 | if (wonly < 0) | |
2578 | wonly = i; | |
2579 | } else { | |
2580 | if (disk < 0) | |
2581 | disk = i; | |
2582 | } | |
2583 | bio->bi_rw = READ; | |
2584 | bio->bi_end_io = end_sync_read; | |
2585 | read_targets++; | |
d57368af AL |
2586 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2587 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2588 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2589 | /* | |
2590 | * The device is suitable for reading (InSync), | |
2591 | * but has bad block(s) here. Let's try to correct them, | |
2592 | * if we are doing resync or repair. Otherwise, leave | |
2593 | * this device alone for this sync request. | |
2594 | */ | |
2595 | bio->bi_rw = WRITE; | |
2596 | bio->bi_end_io = end_sync_write; | |
2597 | write_targets++; | |
3e198f78 | 2598 | } |
3e198f78 | 2599 | } |
06f60385 N |
2600 | if (bio->bi_end_io) { |
2601 | atomic_inc(&rdev->nr_pending); | |
2602 | bio->bi_sector = sector_nr + rdev->data_offset; | |
2603 | bio->bi_bdev = rdev->bdev; | |
2604 | bio->bi_private = r1_bio; | |
2605 | } | |
1da177e4 | 2606 | } |
3e198f78 N |
2607 | rcu_read_unlock(); |
2608 | if (disk < 0) | |
2609 | disk = wonly; | |
2610 | r1_bio->read_disk = disk; | |
191ea9b2 | 2611 | |
06f60385 N |
2612 | if (read_targets == 0 && min_bad > 0) { |
2613 | /* These sectors are bad on all InSync devices, so we | |
2614 | * need to mark them bad on all write targets | |
2615 | */ | |
2616 | int ok = 1; | |
8f19ccb2 | 2617 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2618 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2619 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2620 | ok = rdev_set_badblocks(rdev, sector_nr, |
2621 | min_bad, 0 | |
2622 | ) && ok; | |
2623 | } | |
2624 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2625 | *skipped = 1; | |
2626 | put_buf(r1_bio); | |
2627 | ||
2628 | if (!ok) { | |
2629 | /* Cannot record the badblocks, so need to | |
2630 | * abort the resync. | |
2631 | * If there are multiple read targets, could just | |
2632 | * fail the really bad ones ??? | |
2633 | */ | |
2634 | conf->recovery_disabled = mddev->recovery_disabled; | |
2635 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2636 | return 0; | |
2637 | } else | |
2638 | return min_bad; | |
2639 | ||
2640 | } | |
2641 | if (min_bad > 0 && min_bad < good_sectors) { | |
2642 | /* only resync enough to reach the next bad->good | |
2643 | * transition */ | |
2644 | good_sectors = min_bad; | |
2645 | } | |
2646 | ||
3e198f78 N |
2647 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2648 | /* extra read targets are also write targets */ | |
2649 | write_targets += read_targets-1; | |
2650 | ||
2651 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2652 | /* There is nowhere to write, so all non-sync |
2653 | * drives must be failed - so we are finished | |
2654 | */ | |
b7219ccb N |
2655 | sector_t rv; |
2656 | if (min_bad > 0) | |
2657 | max_sector = sector_nr + min_bad; | |
2658 | rv = max_sector - sector_nr; | |
57afd89f | 2659 | *skipped = 1; |
1da177e4 | 2660 | put_buf(r1_bio); |
1da177e4 LT |
2661 | return rv; |
2662 | } | |
2663 | ||
c6207277 N |
2664 | if (max_sector > mddev->resync_max) |
2665 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2666 | if (max_sector > sector_nr + good_sectors) |
2667 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2668 | nr_sectors = 0; |
289e99e8 | 2669 | sync_blocks = 0; |
1da177e4 LT |
2670 | do { |
2671 | struct page *page; | |
2672 | int len = PAGE_SIZE; | |
2673 | if (sector_nr + (len>>9) > max_sector) | |
2674 | len = (max_sector - sector_nr) << 9; | |
2675 | if (len == 0) | |
2676 | break; | |
6a806c51 N |
2677 | if (sync_blocks == 0) { |
2678 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2679 | &sync_blocks, still_degraded) && |
2680 | !conf->fullsync && | |
2681 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2682 | break; |
9e77c485 | 2683 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
7571ae88 | 2684 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2685 | len = sync_blocks<<9; |
ab7a30c7 | 2686 | } |
191ea9b2 | 2687 | |
8f19ccb2 | 2688 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2689 | bio = r1_bio->bios[i]; |
2690 | if (bio->bi_end_io) { | |
d11c171e | 2691 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2692 | if (bio_add_page(bio, page, len, 0) == 0) { |
2693 | /* stop here */ | |
d11c171e | 2694 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2695 | while (i > 0) { |
2696 | i--; | |
2697 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2698 | if (bio->bi_end_io==NULL) |
2699 | continue; | |
1da177e4 LT |
2700 | /* remove last page from this bio */ |
2701 | bio->bi_vcnt--; | |
2702 | bio->bi_size -= len; | |
2703 | bio->bi_flags &= ~(1<< BIO_SEG_VALID); | |
2704 | } | |
2705 | goto bio_full; | |
2706 | } | |
2707 | } | |
2708 | } | |
2709 | nr_sectors += len>>9; | |
2710 | sector_nr += len>>9; | |
191ea9b2 | 2711 | sync_blocks -= (len>>9); |
1da177e4 LT |
2712 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2713 | bio_full: | |
1da177e4 LT |
2714 | r1_bio->sectors = nr_sectors; |
2715 | ||
d11c171e N |
2716 | /* For a user-requested sync, we read all readable devices and do a |
2717 | * compare | |
2718 | */ | |
2719 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2720 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2721 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2722 | bio = r1_bio->bios[i]; |
2723 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2724 | read_targets--; |
ddac7c7e | 2725 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
2726 | generic_make_request(bio); |
2727 | } | |
2728 | } | |
2729 | } else { | |
2730 | atomic_set(&r1_bio->remaining, 1); | |
2731 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2732 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 2733 | generic_make_request(bio); |
1da177e4 | 2734 | |
d11c171e | 2735 | } |
1da177e4 LT |
2736 | return nr_sectors; |
2737 | } | |
2738 | ||
fd01b88c | 2739 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2740 | { |
2741 | if (sectors) | |
2742 | return sectors; | |
2743 | ||
2744 | return mddev->dev_sectors; | |
2745 | } | |
2746 | ||
e8096360 | 2747 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2748 | { |
e8096360 | 2749 | struct r1conf *conf; |
709ae487 | 2750 | int i; |
0eaf822c | 2751 | struct raid1_info *disk; |
3cb03002 | 2752 | struct md_rdev *rdev; |
709ae487 | 2753 | int err = -ENOMEM; |
1da177e4 | 2754 | |
e8096360 | 2755 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2756 | if (!conf) |
709ae487 | 2757 | goto abort; |
1da177e4 | 2758 | |
0eaf822c | 2759 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2760 | * mddev->raid_disks * 2, |
1da177e4 LT |
2761 | GFP_KERNEL); |
2762 | if (!conf->mirrors) | |
709ae487 | 2763 | goto abort; |
1da177e4 | 2764 | |
ddaf22ab N |
2765 | conf->tmppage = alloc_page(GFP_KERNEL); |
2766 | if (!conf->tmppage) | |
709ae487 | 2767 | goto abort; |
ddaf22ab | 2768 | |
709ae487 | 2769 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2770 | if (!conf->poolinfo) |
709ae487 | 2771 | goto abort; |
8f19ccb2 | 2772 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2773 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2774 | r1bio_pool_free, | |
2775 | conf->poolinfo); | |
2776 | if (!conf->r1bio_pool) | |
709ae487 N |
2777 | goto abort; |
2778 | ||
ed9bfdf1 | 2779 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2780 | |
c19d5798 | 2781 | err = -EINVAL; |
e7e72bf6 | 2782 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2783 | rdev_for_each(rdev, mddev) { |
aba336bd | 2784 | struct request_queue *q; |
709ae487 | 2785 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2786 | if (disk_idx >= mddev->raid_disks |
2787 | || disk_idx < 0) | |
2788 | continue; | |
c19d5798 | 2789 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2790 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2791 | else |
2792 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2793 | |
c19d5798 N |
2794 | if (disk->rdev) |
2795 | goto abort; | |
1da177e4 | 2796 | disk->rdev = rdev; |
aba336bd N |
2797 | q = bdev_get_queue(rdev->bdev); |
2798 | if (q->merge_bvec_fn) | |
2799 | mddev->merge_check_needed = 1; | |
1da177e4 LT |
2800 | |
2801 | disk->head_position = 0; | |
12cee5a8 | 2802 | disk->seq_start = MaxSector; |
1da177e4 LT |
2803 | } |
2804 | conf->raid_disks = mddev->raid_disks; | |
2805 | conf->mddev = mddev; | |
1da177e4 | 2806 | INIT_LIST_HEAD(&conf->retry_list); |
1da177e4 LT |
2807 | |
2808 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2809 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2810 | |
191ea9b2 | 2811 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2812 | conf->pending_count = 0; |
d890fa2b | 2813 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2814 | |
79ef3a8a | 2815 | conf->start_next_window = MaxSector; |
2816 | conf->current_window_requests = conf->next_window_requests = 0; | |
2817 | ||
c19d5798 | 2818 | err = -EIO; |
8f19ccb2 | 2819 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2820 | |
2821 | disk = conf->mirrors + i; | |
2822 | ||
c19d5798 N |
2823 | if (i < conf->raid_disks && |
2824 | disk[conf->raid_disks].rdev) { | |
2825 | /* This slot has a replacement. */ | |
2826 | if (!disk->rdev) { | |
2827 | /* No original, just make the replacement | |
2828 | * a recovering spare | |
2829 | */ | |
2830 | disk->rdev = | |
2831 | disk[conf->raid_disks].rdev; | |
2832 | disk[conf->raid_disks].rdev = NULL; | |
2833 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2834 | /* Original is not in_sync - bad */ | |
2835 | goto abort; | |
2836 | } | |
2837 | ||
5fd6c1dc N |
2838 | if (!disk->rdev || |
2839 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2840 | disk->head_position = 0; |
4f0a5e01 JB |
2841 | if (disk->rdev && |
2842 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2843 | conf->fullsync = 1; |
be4d3280 | 2844 | } |
1da177e4 | 2845 | } |
709ae487 | 2846 | |
709ae487 | 2847 | err = -ENOMEM; |
0232605d | 2848 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
709ae487 N |
2849 | if (!conf->thread) { |
2850 | printk(KERN_ERR | |
9dd1e2fa | 2851 | "md/raid1:%s: couldn't allocate thread\n", |
709ae487 N |
2852 | mdname(mddev)); |
2853 | goto abort; | |
11ce99e6 | 2854 | } |
1da177e4 | 2855 | |
709ae487 N |
2856 | return conf; |
2857 | ||
2858 | abort: | |
2859 | if (conf) { | |
2860 | if (conf->r1bio_pool) | |
2861 | mempool_destroy(conf->r1bio_pool); | |
2862 | kfree(conf->mirrors); | |
2863 | safe_put_page(conf->tmppage); | |
2864 | kfree(conf->poolinfo); | |
2865 | kfree(conf); | |
2866 | } | |
2867 | return ERR_PTR(err); | |
2868 | } | |
2869 | ||
5220ea1e | 2870 | static int stop(struct mddev *mddev); |
fd01b88c | 2871 | static int run(struct mddev *mddev) |
709ae487 | 2872 | { |
e8096360 | 2873 | struct r1conf *conf; |
709ae487 | 2874 | int i; |
3cb03002 | 2875 | struct md_rdev *rdev; |
5220ea1e | 2876 | int ret; |
2ff8cc2c | 2877 | bool discard_supported = false; |
709ae487 N |
2878 | |
2879 | if (mddev->level != 1) { | |
9dd1e2fa | 2880 | printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", |
709ae487 N |
2881 | mdname(mddev), mddev->level); |
2882 | return -EIO; | |
2883 | } | |
2884 | if (mddev->reshape_position != MaxSector) { | |
9dd1e2fa | 2885 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
709ae487 N |
2886 | mdname(mddev)); |
2887 | return -EIO; | |
2888 | } | |
1da177e4 | 2889 | /* |
709ae487 N |
2890 | * copy the already verified devices into our private RAID1 |
2891 | * bookkeeping area. [whatever we allocate in run(), | |
2892 | * should be freed in stop()] | |
1da177e4 | 2893 | */ |
709ae487 N |
2894 | if (mddev->private == NULL) |
2895 | conf = setup_conf(mddev); | |
2896 | else | |
2897 | conf = mddev->private; | |
1da177e4 | 2898 | |
709ae487 N |
2899 | if (IS_ERR(conf)) |
2900 | return PTR_ERR(conf); | |
1da177e4 | 2901 | |
c8dc9c65 | 2902 | if (mddev->queue) |
5026d7a9 PA |
2903 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
2904 | ||
dafb20fa | 2905 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
2906 | if (!mddev->gendisk) |
2907 | continue; | |
709ae487 N |
2908 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2909 | rdev->data_offset << 9); | |
2ff8cc2c SL |
2910 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2911 | discard_supported = true; | |
1da177e4 | 2912 | } |
191ea9b2 | 2913 | |
709ae487 N |
2914 | mddev->degraded = 0; |
2915 | for (i=0; i < conf->raid_disks; i++) | |
2916 | if (conf->mirrors[i].rdev == NULL || | |
2917 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
2918 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
2919 | mddev->degraded++; | |
2920 | ||
2921 | if (conf->raid_disks - mddev->degraded == 1) | |
2922 | mddev->recovery_cp = MaxSector; | |
2923 | ||
8c6ac868 | 2924 | if (mddev->recovery_cp != MaxSector) |
9dd1e2fa | 2925 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
8c6ac868 AN |
2926 | " -- starting background reconstruction\n", |
2927 | mdname(mddev)); | |
1da177e4 | 2928 | printk(KERN_INFO |
9dd1e2fa | 2929 | "md/raid1:%s: active with %d out of %d mirrors\n", |
1da177e4 LT |
2930 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
2931 | mddev->raid_disks); | |
709ae487 | 2932 | |
1da177e4 LT |
2933 | /* |
2934 | * Ok, everything is just fine now | |
2935 | */ | |
709ae487 N |
2936 | mddev->thread = conf->thread; |
2937 | conf->thread = NULL; | |
2938 | mddev->private = conf; | |
2939 | ||
1f403624 | 2940 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2941 | |
1ed7242e JB |
2942 | if (mddev->queue) { |
2943 | mddev->queue->backing_dev_info.congested_fn = raid1_congested; | |
2944 | mddev->queue->backing_dev_info.congested_data = mddev; | |
6b740b8d | 2945 | blk_queue_merge_bvec(mddev->queue, raid1_mergeable_bvec); |
2ff8cc2c SL |
2946 | |
2947 | if (discard_supported) | |
2948 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
2949 | mddev->queue); | |
2950 | else | |
2951 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
2952 | mddev->queue); | |
1ed7242e | 2953 | } |
5220ea1e | 2954 | |
2955 | ret = md_integrity_register(mddev); | |
2956 | if (ret) | |
2957 | stop(mddev); | |
2958 | return ret; | |
1da177e4 LT |
2959 | } |
2960 | ||
fd01b88c | 2961 | static int stop(struct mddev *mddev) |
1da177e4 | 2962 | { |
e8096360 | 2963 | struct r1conf *conf = mddev->private; |
4b6d287f | 2964 | struct bitmap *bitmap = mddev->bitmap; |
4b6d287f N |
2965 | |
2966 | /* wait for behind writes to complete */ | |
e555190d | 2967 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { |
9dd1e2fa N |
2968 | printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n", |
2969 | mdname(mddev)); | |
4b6d287f | 2970 | /* need to kick something here to make sure I/O goes? */ |
e555190d N |
2971 | wait_event(bitmap->behind_wait, |
2972 | atomic_read(&bitmap->behind_writes) == 0); | |
4b6d287f | 2973 | } |
1da177e4 | 2974 | |
07169fd4 | 2975 | freeze_array(conf, 0); |
2976 | unfreeze_array(conf); | |
409c57f3 | 2977 | |
01f96c0a | 2978 | md_unregister_thread(&mddev->thread); |
1da177e4 LT |
2979 | if (conf->r1bio_pool) |
2980 | mempool_destroy(conf->r1bio_pool); | |
990a8baf | 2981 | kfree(conf->mirrors); |
0fea7ed8 | 2982 | safe_put_page(conf->tmppage); |
990a8baf | 2983 | kfree(conf->poolinfo); |
1da177e4 LT |
2984 | kfree(conf); |
2985 | mddev->private = NULL; | |
2986 | return 0; | |
2987 | } | |
2988 | ||
fd01b88c | 2989 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
2990 | { |
2991 | /* no resync is happening, and there is enough space | |
2992 | * on all devices, so we can resize. | |
2993 | * We need to make sure resync covers any new space. | |
2994 | * If the array is shrinking we should possibly wait until | |
2995 | * any io in the removed space completes, but it hardly seems | |
2996 | * worth it. | |
2997 | */ | |
a4a6125a N |
2998 | sector_t newsize = raid1_size(mddev, sectors, 0); |
2999 | if (mddev->external_size && | |
3000 | mddev->array_sectors > newsize) | |
b522adcd | 3001 | return -EINVAL; |
a4a6125a N |
3002 | if (mddev->bitmap) { |
3003 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3004 | if (ret) | |
3005 | return ret; | |
3006 | } | |
3007 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 3008 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 3009 | revalidate_disk(mddev->gendisk); |
b522adcd | 3010 | if (sectors > mddev->dev_sectors && |
b098636c | 3011 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3012 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3013 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3014 | } | |
b522adcd | 3015 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3016 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3017 | return 0; |
3018 | } | |
3019 | ||
fd01b88c | 3020 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3021 | { |
3022 | /* We need to: | |
3023 | * 1/ resize the r1bio_pool | |
3024 | * 2/ resize conf->mirrors | |
3025 | * | |
3026 | * We allocate a new r1bio_pool if we can. | |
3027 | * Then raise a device barrier and wait until all IO stops. | |
3028 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3029 | * |
3030 | * At the same time, we "pack" the devices so that all the missing | |
3031 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3032 | */ |
3033 | mempool_t *newpool, *oldpool; | |
3034 | struct pool_info *newpoolinfo; | |
0eaf822c | 3035 | struct raid1_info *newmirrors; |
e8096360 | 3036 | struct r1conf *conf = mddev->private; |
63c70c4f | 3037 | int cnt, raid_disks; |
c04be0aa | 3038 | unsigned long flags; |
b5470dc5 | 3039 | int d, d2, err; |
1da177e4 | 3040 | |
63c70c4f | 3041 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3042 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3043 | mddev->layout != mddev->new_layout || |
3044 | mddev->level != mddev->new_level) { | |
664e7c41 | 3045 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3046 | mddev->new_layout = mddev->layout; |
3047 | mddev->new_level = mddev->level; | |
3048 | return -EINVAL; | |
3049 | } | |
3050 | ||
b5470dc5 DW |
3051 | err = md_allow_write(mddev); |
3052 | if (err) | |
3053 | return err; | |
2a2275d6 | 3054 | |
63c70c4f N |
3055 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3056 | ||
6ea9c07c N |
3057 | if (raid_disks < conf->raid_disks) { |
3058 | cnt=0; | |
3059 | for (d= 0; d < conf->raid_disks; d++) | |
3060 | if (conf->mirrors[d].rdev) | |
3061 | cnt++; | |
3062 | if (cnt > raid_disks) | |
1da177e4 | 3063 | return -EBUSY; |
6ea9c07c | 3064 | } |
1da177e4 LT |
3065 | |
3066 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3067 | if (!newpoolinfo) | |
3068 | return -ENOMEM; | |
3069 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3070 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3071 | |
3072 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3073 | r1bio_pool_free, newpoolinfo); | |
3074 | if (!newpool) { | |
3075 | kfree(newpoolinfo); | |
3076 | return -ENOMEM; | |
3077 | } | |
0eaf822c | 3078 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3079 | GFP_KERNEL); |
1da177e4 LT |
3080 | if (!newmirrors) { |
3081 | kfree(newpoolinfo); | |
3082 | mempool_destroy(newpool); | |
3083 | return -ENOMEM; | |
3084 | } | |
1da177e4 | 3085 | |
e2d59925 | 3086 | freeze_array(conf, 0); |
1da177e4 LT |
3087 | |
3088 | /* ok, everything is stopped */ | |
3089 | oldpool = conf->r1bio_pool; | |
3090 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3091 | |
a88aa786 | 3092 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3093 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3094 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3095 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3096 | rdev->raid_disk = d2; |
36fad858 NK |
3097 | sysfs_unlink_rdev(mddev, rdev); |
3098 | if (sysfs_link_rdev(mddev, rdev)) | |
a88aa786 | 3099 | printk(KERN_WARNING |
36fad858 NK |
3100 | "md/raid1:%s: cannot register rd%d\n", |
3101 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3102 | } |
a88aa786 N |
3103 | if (rdev) |
3104 | newmirrors[d2++].rdev = rdev; | |
3105 | } | |
1da177e4 LT |
3106 | kfree(conf->mirrors); |
3107 | conf->mirrors = newmirrors; | |
3108 | kfree(conf->poolinfo); | |
3109 | conf->poolinfo = newpoolinfo; | |
3110 | ||
c04be0aa | 3111 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3112 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3113 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3114 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3115 | mddev->delta_disks = 0; |
1da177e4 | 3116 | |
e2d59925 | 3117 | unfreeze_array(conf); |
1da177e4 LT |
3118 | |
3119 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3120 | md_wakeup_thread(mddev->thread); | |
3121 | ||
3122 | mempool_destroy(oldpool); | |
3123 | return 0; | |
3124 | } | |
3125 | ||
fd01b88c | 3126 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3127 | { |
e8096360 | 3128 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3129 | |
3130 | switch(state) { | |
6eef4b21 N |
3131 | case 2: /* wake for suspend */ |
3132 | wake_up(&conf->wait_barrier); | |
3133 | break; | |
9e6603da | 3134 | case 1: |
07169fd4 | 3135 | freeze_array(conf, 0); |
36fa3063 | 3136 | break; |
9e6603da | 3137 | case 0: |
07169fd4 | 3138 | unfreeze_array(conf); |
36fa3063 N |
3139 | break; |
3140 | } | |
36fa3063 N |
3141 | } |
3142 | ||
fd01b88c | 3143 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3144 | { |
3145 | /* raid1 can take over: | |
3146 | * raid5 with 2 devices, any layout or chunk size | |
3147 | */ | |
3148 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3149 | struct r1conf *conf; |
709ae487 N |
3150 | mddev->new_level = 1; |
3151 | mddev->new_layout = 0; | |
3152 | mddev->new_chunk_sectors = 0; | |
3153 | conf = setup_conf(mddev); | |
3154 | if (!IS_ERR(conf)) | |
07169fd4 | 3155 | /* Array must appear to be quiesced */ |
3156 | conf->array_frozen = 1; | |
709ae487 N |
3157 | return conf; |
3158 | } | |
3159 | return ERR_PTR(-EINVAL); | |
3160 | } | |
1da177e4 | 3161 | |
84fc4b56 | 3162 | static struct md_personality raid1_personality = |
1da177e4 LT |
3163 | { |
3164 | .name = "raid1", | |
2604b703 | 3165 | .level = 1, |
1da177e4 LT |
3166 | .owner = THIS_MODULE, |
3167 | .make_request = make_request, | |
3168 | .run = run, | |
3169 | .stop = stop, | |
3170 | .status = status, | |
3171 | .error_handler = error, | |
3172 | .hot_add_disk = raid1_add_disk, | |
3173 | .hot_remove_disk= raid1_remove_disk, | |
3174 | .spare_active = raid1_spare_active, | |
3175 | .sync_request = sync_request, | |
3176 | .resize = raid1_resize, | |
80c3a6ce | 3177 | .size = raid1_size, |
63c70c4f | 3178 | .check_reshape = raid1_reshape, |
36fa3063 | 3179 | .quiesce = raid1_quiesce, |
709ae487 | 3180 | .takeover = raid1_takeover, |
1da177e4 LT |
3181 | }; |
3182 | ||
3183 | static int __init raid_init(void) | |
3184 | { | |
2604b703 | 3185 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3186 | } |
3187 | ||
3188 | static void raid_exit(void) | |
3189 | { | |
2604b703 | 3190 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3191 | } |
3192 | ||
3193 | module_init(raid_init); | |
3194 | module_exit(raid_exit); | |
3195 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3196 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3197 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3198 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3199 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3200 | |
3201 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |