]>
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) | |
c40f341f GR |
93 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
94 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
8e005f7c | 95 | #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) |
1da177e4 | 96 | |
dd0fc66f | 97 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
98 | { |
99 | struct pool_info *pi = data; | |
9f2c9d12 | 100 | struct r1bio *r1_bio; |
1da177e4 | 101 | struct bio *bio; |
da1aab3d | 102 | int need_pages; |
1da177e4 LT |
103 | int i, j; |
104 | ||
105 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 106 | if (!r1_bio) |
1da177e4 | 107 | return NULL; |
1da177e4 LT |
108 | |
109 | /* | |
110 | * Allocate bios : 1 for reading, n-1 for writing | |
111 | */ | |
112 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 113 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
114 | if (!bio) |
115 | goto out_free_bio; | |
116 | r1_bio->bios[j] = bio; | |
117 | } | |
118 | /* | |
119 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
120 | * the first bio. |
121 | * If this is a user-requested check/repair, allocate | |
122 | * RESYNC_PAGES for each bio. | |
1da177e4 | 123 | */ |
d11c171e | 124 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 125 | need_pages = pi->raid_disks; |
d11c171e | 126 | else |
da1aab3d N |
127 | need_pages = 1; |
128 | for (j = 0; j < need_pages; j++) { | |
d11c171e | 129 | bio = r1_bio->bios[j]; |
a0787606 | 130 | bio->bi_vcnt = RESYNC_PAGES; |
d11c171e | 131 | |
a0787606 | 132 | if (bio_alloc_pages(bio, gfp_flags)) |
da1aab3d | 133 | goto out_free_pages; |
d11c171e N |
134 | } |
135 | /* If not user-requests, copy the page pointers to all bios */ | |
136 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
137 | for (i=0; i<RESYNC_PAGES ; i++) | |
138 | for (j=1; j<pi->raid_disks; j++) | |
139 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
140 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
141 | } |
142 | ||
143 | r1_bio->master_bio = NULL; | |
144 | ||
145 | return r1_bio; | |
146 | ||
da1aab3d N |
147 | out_free_pages: |
148 | while (--j >= 0) { | |
149 | struct bio_vec *bv; | |
150 | ||
151 | bio_for_each_segment_all(bv, r1_bio->bios[j], i) | |
152 | __free_page(bv->bv_page); | |
153 | } | |
154 | ||
1da177e4 | 155 | out_free_bio: |
8f19ccb2 | 156 | while (++j < pi->raid_disks) |
1da177e4 LT |
157 | bio_put(r1_bio->bios[j]); |
158 | r1bio_pool_free(r1_bio, data); | |
159 | return NULL; | |
160 | } | |
161 | ||
162 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
163 | { | |
164 | struct pool_info *pi = data; | |
d11c171e | 165 | int i,j; |
9f2c9d12 | 166 | struct r1bio *r1bio = __r1_bio; |
1da177e4 | 167 | |
d11c171e N |
168 | for (i = 0; i < RESYNC_PAGES; i++) |
169 | for (j = pi->raid_disks; j-- ;) { | |
170 | if (j == 0 || | |
171 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
172 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 173 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 174 | } |
1da177e4 LT |
175 | for (i=0 ; i < pi->raid_disks; i++) |
176 | bio_put(r1bio->bios[i]); | |
177 | ||
178 | r1bio_pool_free(r1bio, data); | |
179 | } | |
180 | ||
e8096360 | 181 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
182 | { |
183 | int i; | |
184 | ||
8f19ccb2 | 185 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 186 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 187 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
188 | bio_put(*bio); |
189 | *bio = NULL; | |
190 | } | |
191 | } | |
192 | ||
9f2c9d12 | 193 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 194 | { |
e8096360 | 195 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 196 | |
1da177e4 LT |
197 | put_all_bios(conf, r1_bio); |
198 | mempool_free(r1_bio, conf->r1bio_pool); | |
199 | } | |
200 | ||
9f2c9d12 | 201 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 202 | { |
e8096360 | 203 | struct r1conf *conf = r1_bio->mddev->private; |
3e198f78 N |
204 | int i; |
205 | ||
8f19ccb2 | 206 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
207 | struct bio *bio = r1_bio->bios[i]; |
208 | if (bio->bi_end_io) | |
209 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
210 | } | |
1da177e4 LT |
211 | |
212 | mempool_free(r1_bio, conf->r1buf_pool); | |
213 | ||
17999be4 | 214 | lower_barrier(conf); |
1da177e4 LT |
215 | } |
216 | ||
9f2c9d12 | 217 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
218 | { |
219 | unsigned long flags; | |
fd01b88c | 220 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 221 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
222 | |
223 | spin_lock_irqsave(&conf->device_lock, flags); | |
224 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 225 | conf->nr_queued ++; |
1da177e4 LT |
226 | spin_unlock_irqrestore(&conf->device_lock, flags); |
227 | ||
17999be4 | 228 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
229 | md_wakeup_thread(mddev->thread); |
230 | } | |
231 | ||
232 | /* | |
233 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
234 | * operation and are ready to return a success/failure code to the buffer | |
235 | * cache layer. | |
236 | */ | |
9f2c9d12 | 237 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
238 | { |
239 | struct bio *bio = r1_bio->master_bio; | |
240 | int done; | |
e8096360 | 241 | struct r1conf *conf = r1_bio->mddev->private; |
79ef3a8a | 242 | sector_t start_next_window = r1_bio->start_next_window; |
4f024f37 | 243 | sector_t bi_sector = bio->bi_iter.bi_sector; |
d2eb35ac N |
244 | |
245 | if (bio->bi_phys_segments) { | |
246 | unsigned long flags; | |
247 | spin_lock_irqsave(&conf->device_lock, flags); | |
248 | bio->bi_phys_segments--; | |
249 | done = (bio->bi_phys_segments == 0); | |
250 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
79ef3a8a | 251 | /* |
252 | * make_request() might be waiting for | |
253 | * bi_phys_segments to decrease | |
254 | */ | |
255 | wake_up(&conf->wait_barrier); | |
d2eb35ac N |
256 | } else |
257 | done = 1; | |
258 | ||
259 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4246a0b6 CH |
260 | bio->bi_error = -EIO; |
261 | ||
d2eb35ac | 262 | if (done) { |
4246a0b6 | 263 | bio_endio(bio); |
d2eb35ac N |
264 | /* |
265 | * Wake up any possible resync thread that waits for the device | |
266 | * to go idle. | |
267 | */ | |
79ef3a8a | 268 | allow_barrier(conf, start_next_window, bi_sector); |
d2eb35ac N |
269 | } |
270 | } | |
271 | ||
9f2c9d12 | 272 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
273 | { |
274 | struct bio *bio = r1_bio->master_bio; | |
275 | ||
4b6d287f N |
276 | /* if nobody has done the final endio yet, do it now */ |
277 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
278 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
279 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
280 | (unsigned long long) bio->bi_iter.bi_sector, |
281 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 282 | |
d2eb35ac | 283 | call_bio_endio(r1_bio); |
4b6d287f | 284 | } |
1da177e4 LT |
285 | free_r1bio(r1_bio); |
286 | } | |
287 | ||
288 | /* | |
289 | * Update disk head position estimator based on IRQ completion info. | |
290 | */ | |
9f2c9d12 | 291 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 292 | { |
e8096360 | 293 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
294 | |
295 | conf->mirrors[disk].head_position = | |
296 | r1_bio->sector + (r1_bio->sectors); | |
297 | } | |
298 | ||
ba3ae3be NK |
299 | /* |
300 | * Find the disk number which triggered given bio | |
301 | */ | |
9f2c9d12 | 302 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
303 | { |
304 | int mirror; | |
30194636 N |
305 | struct r1conf *conf = r1_bio->mddev->private; |
306 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 307 | |
8f19ccb2 | 308 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
309 | if (r1_bio->bios[mirror] == bio) |
310 | break; | |
311 | ||
8f19ccb2 | 312 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
313 | update_head_pos(mirror, r1_bio); |
314 | ||
315 | return mirror; | |
316 | } | |
317 | ||
4246a0b6 | 318 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 319 | { |
4246a0b6 | 320 | int uptodate = !bio->bi_error; |
9f2c9d12 | 321 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 322 | int mirror; |
e8096360 | 323 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 324 | |
1da177e4 LT |
325 | mirror = r1_bio->read_disk; |
326 | /* | |
327 | * this branch is our 'one mirror IO has finished' event handler: | |
328 | */ | |
ddaf22ab N |
329 | update_head_pos(mirror, r1_bio); |
330 | ||
dd00a99e N |
331 | if (uptodate) |
332 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
333 | else { | |
334 | /* If all other devices have failed, we want to return | |
335 | * the error upwards rather than fail the last device. | |
336 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 337 | */ |
dd00a99e N |
338 | unsigned long flags; |
339 | spin_lock_irqsave(&conf->device_lock, flags); | |
340 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
341 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
34cab6f4 | 342 | test_bit(In_sync, &conf->mirrors[mirror].rdev->flags))) |
dd00a99e N |
343 | uptodate = 1; |
344 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
345 | } | |
1da177e4 | 346 | |
7ad4d4a6 | 347 | if (uptodate) { |
1da177e4 | 348 | raid_end_bio_io(r1_bio); |
7ad4d4a6 N |
349 | rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); |
350 | } else { | |
1da177e4 LT |
351 | /* |
352 | * oops, read error: | |
353 | */ | |
354 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
355 | printk_ratelimited( |
356 | KERN_ERR "md/raid1:%s: %s: " | |
357 | "rescheduling sector %llu\n", | |
358 | mdname(conf->mddev), | |
359 | bdevname(conf->mirrors[mirror].rdev->bdev, | |
360 | b), | |
361 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 362 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 363 | reschedule_retry(r1_bio); |
7ad4d4a6 | 364 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 365 | } |
1da177e4 LT |
366 | } |
367 | ||
9f2c9d12 | 368 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
369 | { |
370 | /* it really is the end of this request */ | |
371 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
372 | /* free extra copy of the data pages */ | |
373 | int i = r1_bio->behind_page_count; | |
374 | while (i--) | |
375 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
376 | kfree(r1_bio->behind_bvecs); | |
377 | r1_bio->behind_bvecs = NULL; | |
378 | } | |
379 | /* clear the bitmap if all writes complete successfully */ | |
380 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
381 | r1_bio->sectors, | |
382 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
383 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
384 | md_write_end(r1_bio->mddev); | |
385 | } | |
386 | ||
9f2c9d12 | 387 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 388 | { |
cd5ff9a1 N |
389 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
390 | return; | |
391 | ||
392 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
393 | reschedule_retry(r1_bio); | |
394 | else { | |
395 | close_write(r1_bio); | |
4367af55 N |
396 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
397 | reschedule_retry(r1_bio); | |
398 | else | |
399 | raid_end_bio_io(r1_bio); | |
4e78064f N |
400 | } |
401 | } | |
402 | ||
4246a0b6 | 403 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 404 | { |
9f2c9d12 | 405 | struct r1bio *r1_bio = bio->bi_private; |
a9701a30 | 406 | int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 407 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 408 | struct bio *to_put = NULL; |
1da177e4 | 409 | |
ba3ae3be | 410 | mirror = find_bio_disk(r1_bio, bio); |
1da177e4 | 411 | |
e9c7469b TH |
412 | /* |
413 | * 'one mirror IO has finished' event handler: | |
414 | */ | |
4246a0b6 | 415 | if (bio->bi_error) { |
cd5ff9a1 N |
416 | set_bit(WriteErrorSeen, |
417 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
418 | if (!test_and_set_bit(WantReplacement, |
419 | &conf->mirrors[mirror].rdev->flags)) | |
420 | set_bit(MD_RECOVERY_NEEDED, & | |
421 | conf->mddev->recovery); | |
422 | ||
cd5ff9a1 | 423 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 | 424 | } else { |
1da177e4 | 425 | /* |
e9c7469b TH |
426 | * Set R1BIO_Uptodate in our master bio, so that we |
427 | * will return a good error code for to the higher | |
428 | * levels even if IO on some other mirrored buffer | |
429 | * fails. | |
430 | * | |
431 | * The 'master' represents the composite IO operation | |
432 | * to user-side. So if something waits for IO, then it | |
433 | * will wait for the 'master' bio. | |
1da177e4 | 434 | */ |
4367af55 N |
435 | sector_t first_bad; |
436 | int bad_sectors; | |
437 | ||
cd5ff9a1 N |
438 | r1_bio->bios[mirror] = NULL; |
439 | to_put = bio; | |
3056e3ae AL |
440 | /* |
441 | * Do not set R1BIO_Uptodate if the current device is | |
442 | * rebuilding or Faulty. This is because we cannot use | |
443 | * such device for properly reading the data back (we could | |
444 | * potentially use it, if the current write would have felt | |
445 | * before rdev->recovery_offset, but for simplicity we don't | |
446 | * check this here. | |
447 | */ | |
448 | if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) && | |
449 | !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)) | |
450 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
e9c7469b | 451 | |
4367af55 N |
452 | /* Maybe we can clear some bad blocks. */ |
453 | if (is_badblock(conf->mirrors[mirror].rdev, | |
454 | r1_bio->sector, r1_bio->sectors, | |
455 | &first_bad, &bad_sectors)) { | |
456 | r1_bio->bios[mirror] = IO_MADE_GOOD; | |
457 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
458 | } | |
459 | } | |
460 | ||
e9c7469b TH |
461 | if (behind) { |
462 | if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags)) | |
463 | atomic_dec(&r1_bio->behind_remaining); | |
464 | ||
465 | /* | |
466 | * In behind mode, we ACK the master bio once the I/O | |
467 | * has safely reached all non-writemostly | |
468 | * disks. Setting the Returned bit ensures that this | |
469 | * gets done only once -- we don't ever want to return | |
470 | * -EIO here, instead we'll wait | |
471 | */ | |
472 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
473 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
474 | /* Maybe we can return now */ | |
475 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
476 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
477 | pr_debug("raid1: behind end write sectors" |
478 | " %llu-%llu\n", | |
4f024f37 KO |
479 | (unsigned long long) mbio->bi_iter.bi_sector, |
480 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 481 | call_bio_endio(r1_bio); |
4b6d287f N |
482 | } |
483 | } | |
484 | } | |
4367af55 N |
485 | if (r1_bio->bios[mirror] == NULL) |
486 | rdev_dec_pending(conf->mirrors[mirror].rdev, | |
487 | conf->mddev); | |
e9c7469b | 488 | |
1da177e4 | 489 | /* |
1da177e4 LT |
490 | * Let's see if all mirrored write operations have finished |
491 | * already. | |
492 | */ | |
af6d7b76 | 493 | r1_bio_write_done(r1_bio); |
c70810b3 | 494 | |
04b857f7 N |
495 | if (to_put) |
496 | bio_put(to_put); | |
1da177e4 LT |
497 | } |
498 | ||
1da177e4 LT |
499 | /* |
500 | * This routine returns the disk from which the requested read should | |
501 | * be done. There is a per-array 'next expected sequential IO' sector | |
502 | * number - if this matches on the next IO then we use the last disk. | |
503 | * There is also a per-disk 'last know head position' sector that is | |
504 | * maintained from IRQ contexts, both the normal and the resync IO | |
505 | * completion handlers update this position correctly. If there is no | |
506 | * perfect sequential match then we pick the disk whose head is closest. | |
507 | * | |
508 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
509 | * because position is mirror, not device based. | |
510 | * | |
511 | * The rdev for the device selected will have nr_pending incremented. | |
512 | */ | |
e8096360 | 513 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 514 | { |
af3a2cd6 | 515 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
516 | int sectors; |
517 | int best_good_sectors; | |
9dedf603 SL |
518 | int best_disk, best_dist_disk, best_pending_disk; |
519 | int has_nonrot_disk; | |
be4d3280 | 520 | int disk; |
76073054 | 521 | sector_t best_dist; |
9dedf603 | 522 | unsigned int min_pending; |
3cb03002 | 523 | struct md_rdev *rdev; |
f3ac8bf7 | 524 | int choose_first; |
12cee5a8 | 525 | int choose_next_idle; |
1da177e4 LT |
526 | |
527 | rcu_read_lock(); | |
528 | /* | |
8ddf9efe | 529 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
530 | * device if no resync is going on, or below the resync window. |
531 | * We take the first readable disk when above the resync window. | |
532 | */ | |
533 | retry: | |
d2eb35ac | 534 | sectors = r1_bio->sectors; |
76073054 | 535 | best_disk = -1; |
9dedf603 | 536 | best_dist_disk = -1; |
76073054 | 537 | best_dist = MaxSector; |
9dedf603 SL |
538 | best_pending_disk = -1; |
539 | min_pending = UINT_MAX; | |
d2eb35ac | 540 | best_good_sectors = 0; |
9dedf603 | 541 | has_nonrot_disk = 0; |
12cee5a8 | 542 | choose_next_idle = 0; |
d2eb35ac | 543 | |
7d49ffcf GR |
544 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
545 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 546 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
547 | this_sector + sectors))) |
548 | choose_first = 1; | |
549 | else | |
550 | choose_first = 0; | |
1da177e4 | 551 | |
be4d3280 | 552 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 553 | sector_t dist; |
d2eb35ac N |
554 | sector_t first_bad; |
555 | int bad_sectors; | |
9dedf603 | 556 | unsigned int pending; |
12cee5a8 | 557 | bool nonrot; |
d2eb35ac | 558 | |
f3ac8bf7 N |
559 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
560 | if (r1_bio->bios[disk] == IO_BLOCKED | |
561 | || rdev == NULL | |
76073054 | 562 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 563 | continue; |
76073054 N |
564 | if (!test_bit(In_sync, &rdev->flags) && |
565 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 566 | continue; |
76073054 N |
567 | if (test_bit(WriteMostly, &rdev->flags)) { |
568 | /* Don't balance among write-mostly, just | |
569 | * use the first as a last resort */ | |
d1901ef0 | 570 | if (best_dist_disk < 0) { |
307729c8 N |
571 | if (is_badblock(rdev, this_sector, sectors, |
572 | &first_bad, &bad_sectors)) { | |
573 | if (first_bad < this_sector) | |
574 | /* Cannot use this */ | |
575 | continue; | |
576 | best_good_sectors = first_bad - this_sector; | |
577 | } else | |
578 | best_good_sectors = sectors; | |
d1901ef0 TH |
579 | best_dist_disk = disk; |
580 | best_pending_disk = disk; | |
307729c8 | 581 | } |
76073054 N |
582 | continue; |
583 | } | |
584 | /* This is a reasonable device to use. It might | |
585 | * even be best. | |
586 | */ | |
d2eb35ac N |
587 | if (is_badblock(rdev, this_sector, sectors, |
588 | &first_bad, &bad_sectors)) { | |
589 | if (best_dist < MaxSector) | |
590 | /* already have a better device */ | |
591 | continue; | |
592 | if (first_bad <= this_sector) { | |
593 | /* cannot read here. If this is the 'primary' | |
594 | * device, then we must not read beyond | |
595 | * bad_sectors from another device.. | |
596 | */ | |
597 | bad_sectors -= (this_sector - first_bad); | |
598 | if (choose_first && sectors > bad_sectors) | |
599 | sectors = bad_sectors; | |
600 | if (best_good_sectors > sectors) | |
601 | best_good_sectors = sectors; | |
602 | ||
603 | } else { | |
604 | sector_t good_sectors = first_bad - this_sector; | |
605 | if (good_sectors > best_good_sectors) { | |
606 | best_good_sectors = good_sectors; | |
607 | best_disk = disk; | |
608 | } | |
609 | if (choose_first) | |
610 | break; | |
611 | } | |
612 | continue; | |
613 | } else | |
614 | best_good_sectors = sectors; | |
615 | ||
12cee5a8 SL |
616 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
617 | has_nonrot_disk |= nonrot; | |
9dedf603 | 618 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 619 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 620 | if (choose_first) { |
76073054 | 621 | best_disk = disk; |
1da177e4 LT |
622 | break; |
623 | } | |
12cee5a8 SL |
624 | /* Don't change to another disk for sequential reads */ |
625 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
626 | || dist == 0) { | |
627 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
628 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
629 | ||
630 | best_disk = disk; | |
631 | /* | |
632 | * If buffered sequential IO size exceeds optimal | |
633 | * iosize, check if there is idle disk. If yes, choose | |
634 | * the idle disk. read_balance could already choose an | |
635 | * idle disk before noticing it's a sequential IO in | |
636 | * this disk. This doesn't matter because this disk | |
637 | * will idle, next time it will be utilized after the | |
638 | * first disk has IO size exceeds optimal iosize. In | |
639 | * this way, iosize of the first disk will be optimal | |
640 | * iosize at least. iosize of the second disk might be | |
641 | * small, but not a big deal since when the second disk | |
642 | * starts IO, the first disk is likely still busy. | |
643 | */ | |
644 | if (nonrot && opt_iosize > 0 && | |
645 | mirror->seq_start != MaxSector && | |
646 | mirror->next_seq_sect > opt_iosize && | |
647 | mirror->next_seq_sect - opt_iosize >= | |
648 | mirror->seq_start) { | |
649 | choose_next_idle = 1; | |
650 | continue; | |
651 | } | |
652 | break; | |
653 | } | |
654 | /* If device is idle, use it */ | |
655 | if (pending == 0) { | |
656 | best_disk = disk; | |
657 | break; | |
658 | } | |
659 | ||
660 | if (choose_next_idle) | |
661 | continue; | |
9dedf603 SL |
662 | |
663 | if (min_pending > pending) { | |
664 | min_pending = pending; | |
665 | best_pending_disk = disk; | |
666 | } | |
667 | ||
76073054 N |
668 | if (dist < best_dist) { |
669 | best_dist = dist; | |
9dedf603 | 670 | best_dist_disk = disk; |
1da177e4 | 671 | } |
f3ac8bf7 | 672 | } |
1da177e4 | 673 | |
9dedf603 SL |
674 | /* |
675 | * If all disks are rotational, choose the closest disk. If any disk is | |
676 | * non-rotational, choose the disk with less pending request even the | |
677 | * disk is rotational, which might/might not be optimal for raids with | |
678 | * mixed ratation/non-rotational disks depending on workload. | |
679 | */ | |
680 | if (best_disk == -1) { | |
681 | if (has_nonrot_disk) | |
682 | best_disk = best_pending_disk; | |
683 | else | |
684 | best_disk = best_dist_disk; | |
685 | } | |
686 | ||
76073054 N |
687 | if (best_disk >= 0) { |
688 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
689 | if (!rdev) |
690 | goto retry; | |
691 | atomic_inc(&rdev->nr_pending); | |
76073054 | 692 | if (test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
693 | /* cannot risk returning a device that failed |
694 | * before we inc'ed nr_pending | |
695 | */ | |
03c902e1 | 696 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
697 | goto retry; |
698 | } | |
d2eb35ac | 699 | sectors = best_good_sectors; |
12cee5a8 SL |
700 | |
701 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
702 | conf->mirrors[best_disk].seq_start = this_sector; | |
703 | ||
be4d3280 | 704 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
705 | } |
706 | rcu_read_unlock(); | |
d2eb35ac | 707 | *max_sectors = sectors; |
1da177e4 | 708 | |
76073054 | 709 | return best_disk; |
1da177e4 LT |
710 | } |
711 | ||
5c675f83 | 712 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 713 | { |
e8096360 | 714 | struct r1conf *conf = mddev->private; |
0d129228 N |
715 | int i, ret = 0; |
716 | ||
4452226e | 717 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
718 | conf->pending_count >= max_queued_requests) |
719 | return 1; | |
720 | ||
0d129228 | 721 | rcu_read_lock(); |
f53e29fc | 722 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 723 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 724 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 725 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 726 | |
1ed7242e JB |
727 | BUG_ON(!q); |
728 | ||
0d129228 N |
729 | /* Note the '|| 1' - when read_balance prefers |
730 | * non-congested targets, it can be removed | |
731 | */ | |
4452226e | 732 | if ((bits & (1 << WB_async_congested)) || 1) |
0d129228 N |
733 | ret |= bdi_congested(&q->backing_dev_info, bits); |
734 | else | |
735 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
736 | } | |
737 | } | |
738 | rcu_read_unlock(); | |
739 | return ret; | |
740 | } | |
0d129228 | 741 | |
e8096360 | 742 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
743 | { |
744 | /* Any writes that have been queued but are awaiting | |
745 | * bitmap updates get flushed here. | |
a35e63ef | 746 | */ |
a35e63ef N |
747 | spin_lock_irq(&conf->device_lock); |
748 | ||
749 | if (conf->pending_bio_list.head) { | |
750 | struct bio *bio; | |
751 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 752 | conf->pending_count = 0; |
a35e63ef N |
753 | spin_unlock_irq(&conf->device_lock); |
754 | /* flush any pending bitmap writes to | |
755 | * disk before proceeding w/ I/O */ | |
756 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 757 | wake_up(&conf->wait_barrier); |
a35e63ef N |
758 | |
759 | while (bio) { /* submit pending writes */ | |
760 | struct bio *next = bio->bi_next; | |
761 | bio->bi_next = NULL; | |
2ff8cc2c SL |
762 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
763 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
764 | /* Just ignore it */ | |
4246a0b6 | 765 | bio_endio(bio); |
2ff8cc2c SL |
766 | else |
767 | generic_make_request(bio); | |
a35e63ef N |
768 | bio = next; |
769 | } | |
a35e63ef N |
770 | } else |
771 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
772 | } |
773 | ||
17999be4 N |
774 | /* Barriers.... |
775 | * Sometimes we need to suspend IO while we do something else, | |
776 | * either some resync/recovery, or reconfigure the array. | |
777 | * To do this we raise a 'barrier'. | |
778 | * The 'barrier' is a counter that can be raised multiple times | |
779 | * to count how many activities are happening which preclude | |
780 | * normal IO. | |
781 | * We can only raise the barrier if there is no pending IO. | |
782 | * i.e. if nr_pending == 0. | |
783 | * We choose only to raise the barrier if no-one is waiting for the | |
784 | * barrier to go down. This means that as soon as an IO request | |
785 | * is ready, no other operations which require a barrier will start | |
786 | * until the IO request has had a chance. | |
787 | * | |
788 | * So: regular IO calls 'wait_barrier'. When that returns there | |
789 | * is no backgroup IO happening, It must arrange to call | |
790 | * allow_barrier when it has finished its IO. | |
791 | * backgroup IO calls must call raise_barrier. Once that returns | |
792 | * there is no normal IO happeing. It must arrange to call | |
793 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 794 | */ |
c2fd4c94 | 795 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 LT |
796 | { |
797 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
798 | |
799 | /* Wait until no block IO is waiting */ | |
800 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
eed8c02e | 801 | conf->resync_lock); |
17999be4 N |
802 | |
803 | /* block any new IO from starting */ | |
804 | conf->barrier++; | |
c2fd4c94 | 805 | conf->next_resync = sector_nr; |
17999be4 | 806 | |
79ef3a8a | 807 | /* For these conditions we must wait: |
808 | * A: while the array is in frozen state | |
809 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach | |
810 | * the max count which allowed. | |
811 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning | |
812 | * next resync will reach to the window which normal bios are | |
813 | * handling. | |
2f73d3c5 | 814 | * D: while there are any active requests in the current window. |
79ef3a8a | 815 | */ |
17999be4 | 816 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 817 | !conf->array_frozen && |
79ef3a8a | 818 | conf->barrier < RESYNC_DEPTH && |
2f73d3c5 | 819 | conf->current_window_requests == 0 && |
79ef3a8a | 820 | (conf->start_next_window >= |
821 | conf->next_resync + RESYNC_SECTORS), | |
eed8c02e | 822 | conf->resync_lock); |
17999be4 | 823 | |
34e97f17 | 824 | conf->nr_pending++; |
17999be4 N |
825 | spin_unlock_irq(&conf->resync_lock); |
826 | } | |
827 | ||
e8096360 | 828 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
829 | { |
830 | unsigned long flags; | |
709ae487 | 831 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
832 | spin_lock_irqsave(&conf->resync_lock, flags); |
833 | conf->barrier--; | |
34e97f17 | 834 | conf->nr_pending--; |
17999be4 N |
835 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
836 | wake_up(&conf->wait_barrier); | |
837 | } | |
838 | ||
79ef3a8a | 839 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
17999be4 | 840 | { |
79ef3a8a | 841 | bool wait = false; |
842 | ||
843 | if (conf->array_frozen || !bio) | |
844 | wait = true; | |
845 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { | |
23554960 N |
846 | if ((conf->mddev->curr_resync_completed |
847 | >= bio_end_sector(bio)) || | |
848 | (conf->next_resync + NEXT_NORMALIO_DISTANCE | |
849 | <= bio->bi_iter.bi_sector)) | |
79ef3a8a | 850 | wait = false; |
851 | else | |
852 | wait = true; | |
853 | } | |
854 | ||
855 | return wait; | |
856 | } | |
857 | ||
858 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) | |
859 | { | |
860 | sector_t sector = 0; | |
861 | ||
17999be4 | 862 | spin_lock_irq(&conf->resync_lock); |
79ef3a8a | 863 | if (need_to_wait_for_sync(conf, bio)) { |
17999be4 | 864 | conf->nr_waiting++; |
d6b42dcb N |
865 | /* Wait for the barrier to drop. |
866 | * However if there are already pending | |
867 | * requests (preventing the barrier from | |
868 | * rising completely), and the | |
5965b642 | 869 | * per-process bio queue isn't empty, |
d6b42dcb | 870 | * then don't wait, as we need to empty |
5965b642 N |
871 | * that queue to allow conf->start_next_window |
872 | * to increase. | |
d6b42dcb N |
873 | */ |
874 | wait_event_lock_irq(conf->wait_barrier, | |
b364e3d0 | 875 | !conf->array_frozen && |
876 | (!conf->barrier || | |
5965b642 N |
877 | ((conf->start_next_window < |
878 | conf->next_resync + RESYNC_SECTORS) && | |
879 | current->bio_list && | |
880 | !bio_list_empty(current->bio_list))), | |
eed8c02e | 881 | conf->resync_lock); |
17999be4 | 882 | conf->nr_waiting--; |
1da177e4 | 883 | } |
79ef3a8a | 884 | |
885 | if (bio && bio_data_dir(bio) == WRITE) { | |
e8ff8bf0 | 886 | if (bio->bi_iter.bi_sector >= conf->next_resync) { |
79ef3a8a | 887 | if (conf->start_next_window == MaxSector) |
888 | conf->start_next_window = | |
889 | conf->next_resync + | |
890 | NEXT_NORMALIO_DISTANCE; | |
891 | ||
892 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) | |
4f024f37 | 893 | <= bio->bi_iter.bi_sector) |
79ef3a8a | 894 | conf->next_window_requests++; |
895 | else | |
896 | conf->current_window_requests++; | |
79ef3a8a | 897 | sector = conf->start_next_window; |
41a336e0 | 898 | } |
79ef3a8a | 899 | } |
900 | ||
17999be4 | 901 | conf->nr_pending++; |
1da177e4 | 902 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 903 | return sector; |
1da177e4 LT |
904 | } |
905 | ||
79ef3a8a | 906 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
907 | sector_t bi_sector) | |
17999be4 N |
908 | { |
909 | unsigned long flags; | |
79ef3a8a | 910 | |
17999be4 N |
911 | spin_lock_irqsave(&conf->resync_lock, flags); |
912 | conf->nr_pending--; | |
79ef3a8a | 913 | if (start_next_window) { |
914 | if (start_next_window == conf->start_next_window) { | |
915 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE | |
916 | <= bi_sector) | |
917 | conf->next_window_requests--; | |
918 | else | |
919 | conf->current_window_requests--; | |
920 | } else | |
921 | conf->current_window_requests--; | |
922 | ||
923 | if (!conf->current_window_requests) { | |
924 | if (conf->next_window_requests) { | |
925 | conf->current_window_requests = | |
926 | conf->next_window_requests; | |
927 | conf->next_window_requests = 0; | |
928 | conf->start_next_window += | |
929 | NEXT_NORMALIO_DISTANCE; | |
930 | } else | |
931 | conf->start_next_window = MaxSector; | |
932 | } | |
933 | } | |
17999be4 N |
934 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
935 | wake_up(&conf->wait_barrier); | |
936 | } | |
937 | ||
e2d59925 | 938 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab N |
939 | { |
940 | /* stop syncio and normal IO and wait for everything to | |
941 | * go quite. | |
b364e3d0 | 942 | * We wait until nr_pending match nr_queued+extra |
1c830532 N |
943 | * This is called in the context of one normal IO request |
944 | * that has failed. Thus any sync request that might be pending | |
945 | * will be blocked by nr_pending, and we need to wait for | |
946 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 947 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
948 | * must match the number of pending IOs (nr_pending) before |
949 | * we continue. | |
ddaf22ab N |
950 | */ |
951 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 952 | conf->array_frozen = 1; |
eed8c02e | 953 | wait_event_lock_irq_cmd(conf->wait_barrier, |
e2d59925 | 954 | conf->nr_pending == conf->nr_queued+extra, |
eed8c02e LC |
955 | conf->resync_lock, |
956 | flush_pending_writes(conf)); | |
ddaf22ab N |
957 | spin_unlock_irq(&conf->resync_lock); |
958 | } | |
e8096360 | 959 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
960 | { |
961 | /* reverse the effect of the freeze */ | |
962 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 963 | conf->array_frozen = 0; |
ddaf22ab N |
964 | wake_up(&conf->wait_barrier); |
965 | spin_unlock_irq(&conf->resync_lock); | |
966 | } | |
967 | ||
f72ffdd6 | 968 | /* duplicate the data pages for behind I/O |
4e78064f | 969 | */ |
9f2c9d12 | 970 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
971 | { |
972 | int i; | |
973 | struct bio_vec *bvec; | |
2ca68f5e | 974 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 975 | GFP_NOIO); |
2ca68f5e | 976 | if (unlikely(!bvecs)) |
af6d7b76 | 977 | return; |
4b6d287f | 978 | |
cb34e057 | 979 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
980 | bvecs[i] = *bvec; |
981 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
982 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 983 | goto do_sync_io; |
2ca68f5e N |
984 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
985 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
986 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
987 | kunmap(bvec->bv_page); |
988 | } | |
2ca68f5e | 989 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
990 | r1_bio->behind_page_count = bio->bi_vcnt; |
991 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
992 | return; | |
4b6d287f N |
993 | |
994 | do_sync_io: | |
af6d7b76 | 995 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
996 | if (bvecs[i].bv_page) |
997 | put_page(bvecs[i].bv_page); | |
998 | kfree(bvecs); | |
4f024f37 KO |
999 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1000 | bio->bi_iter.bi_size); | |
4b6d287f N |
1001 | } |
1002 | ||
f54a9d0e N |
1003 | struct raid1_plug_cb { |
1004 | struct blk_plug_cb cb; | |
1005 | struct bio_list pending; | |
1006 | int pending_cnt; | |
1007 | }; | |
1008 | ||
1009 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1010 | { | |
1011 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1012 | cb); | |
1013 | struct mddev *mddev = plug->cb.data; | |
1014 | struct r1conf *conf = mddev->private; | |
1015 | struct bio *bio; | |
1016 | ||
874807a8 | 1017 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1018 | spin_lock_irq(&conf->device_lock); |
1019 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1020 | conf->pending_count += plug->pending_cnt; | |
1021 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1022 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1023 | md_wakeup_thread(mddev->thread); |
1024 | kfree(plug); | |
1025 | return; | |
1026 | } | |
1027 | ||
1028 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1029 | bio = bio_list_get(&plug->pending); | |
1030 | bitmap_unplug(mddev->bitmap); | |
1031 | wake_up(&conf->wait_barrier); | |
1032 | ||
1033 | while (bio) { /* submit pending writes */ | |
1034 | struct bio *next = bio->bi_next; | |
1035 | bio->bi_next = NULL; | |
32f9f570 SL |
1036 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
1037 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
1038 | /* Just ignore it */ | |
4246a0b6 | 1039 | bio_endio(bio); |
32f9f570 SL |
1040 | else |
1041 | generic_make_request(bio); | |
f54a9d0e N |
1042 | bio = next; |
1043 | } | |
1044 | kfree(plug); | |
1045 | } | |
1046 | ||
849674e4 | 1047 | static void raid1_make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 1048 | { |
e8096360 | 1049 | struct r1conf *conf = mddev->private; |
0eaf822c | 1050 | struct raid1_info *mirror; |
9f2c9d12 | 1051 | struct r1bio *r1_bio; |
1da177e4 | 1052 | struct bio *read_bio; |
1f68f0c4 | 1053 | int i, disks; |
84255d10 | 1054 | struct bitmap *bitmap; |
191ea9b2 | 1055 | unsigned long flags; |
a362357b | 1056 | const int rw = bio_data_dir(bio); |
2c7d46ec | 1057 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 1058 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
2ff8cc2c SL |
1059 | const unsigned long do_discard = (bio->bi_rw |
1060 | & (REQ_DISCARD | REQ_SECURE)); | |
c8dc9c65 | 1061 | const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME); |
3cb03002 | 1062 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1063 | struct blk_plug_cb *cb; |
1064 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
1065 | int first_clone; |
1066 | int sectors_handled; | |
1067 | int max_sectors; | |
79ef3a8a | 1068 | sector_t start_next_window; |
191ea9b2 | 1069 | |
1da177e4 LT |
1070 | /* |
1071 | * Register the new request and wait if the reconstruction | |
1072 | * thread has put up a bar for new requests. | |
1073 | * Continue immediately if no resync is active currently. | |
1074 | */ | |
62de608d | 1075 | |
3d310eb7 N |
1076 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1077 | ||
6eef4b21 | 1078 | if (bio_data_dir(bio) == WRITE && |
589a1c49 GR |
1079 | ((bio_end_sector(bio) > mddev->suspend_lo && |
1080 | bio->bi_iter.bi_sector < mddev->suspend_hi) || | |
1081 | (mddev_is_clustered(mddev) && | |
90382ed9 GR |
1082 | md_cluster_ops->area_resyncing(mddev, WRITE, |
1083 | bio->bi_iter.bi_sector, bio_end_sector(bio))))) { | |
6eef4b21 N |
1084 | /* As the suspend_* range is controlled by |
1085 | * userspace, we want an interruptible | |
1086 | * wait. | |
1087 | */ | |
1088 | DEFINE_WAIT(w); | |
1089 | for (;;) { | |
1090 | flush_signals(current); | |
1091 | prepare_to_wait(&conf->wait_barrier, | |
1092 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1093 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
589a1c49 GR |
1094 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
1095 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1096 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
589a1c49 | 1097 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) |
6eef4b21 N |
1098 | break; |
1099 | schedule(); | |
1100 | } | |
1101 | finish_wait(&conf->wait_barrier, &w); | |
1102 | } | |
62de608d | 1103 | |
79ef3a8a | 1104 | start_next_window = wait_barrier(conf, bio); |
1da177e4 | 1105 | |
84255d10 N |
1106 | bitmap = mddev->bitmap; |
1107 | ||
1da177e4 LT |
1108 | /* |
1109 | * make_request() can abort the operation when READA is being | |
1110 | * used and no empty request is available. | |
1111 | * | |
1112 | */ | |
1113 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1114 | ||
1115 | r1_bio->master_bio = bio; | |
aa8b57aa | 1116 | r1_bio->sectors = bio_sectors(bio); |
191ea9b2 | 1117 | r1_bio->state = 0; |
1da177e4 | 1118 | r1_bio->mddev = mddev; |
4f024f37 | 1119 | r1_bio->sector = bio->bi_iter.bi_sector; |
1da177e4 | 1120 | |
d2eb35ac N |
1121 | /* We might need to issue multiple reads to different |
1122 | * devices if there are bad blocks around, so we keep | |
1123 | * track of the number of reads in bio->bi_phys_segments. | |
1124 | * If this is 0, there is only one r1_bio and no locking | |
1125 | * will be needed when requests complete. If it is | |
1126 | * non-zero, then it is the number of not-completed requests. | |
1127 | */ | |
1128 | bio->bi_phys_segments = 0; | |
b7c44ed9 | 1129 | bio_clear_flag(bio, BIO_SEG_VALID); |
d2eb35ac | 1130 | |
a362357b | 1131 | if (rw == READ) { |
1da177e4 LT |
1132 | /* |
1133 | * read balancing logic: | |
1134 | */ | |
d2eb35ac N |
1135 | int rdisk; |
1136 | ||
1137 | read_again: | |
1138 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1da177e4 LT |
1139 | |
1140 | if (rdisk < 0) { | |
1141 | /* couldn't find anywhere to read from */ | |
1142 | raid_end_bio_io(r1_bio); | |
5a7bbad2 | 1143 | return; |
1da177e4 LT |
1144 | } |
1145 | mirror = conf->mirrors + rdisk; | |
1146 | ||
e555190d N |
1147 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1148 | bitmap) { | |
1149 | /* Reading from a write-mostly device must | |
1150 | * take care not to over-take any writes | |
1151 | * that are 'behind' | |
1152 | */ | |
1153 | wait_event(bitmap->behind_wait, | |
1154 | atomic_read(&bitmap->behind_writes) == 0); | |
1155 | } | |
1da177e4 | 1156 | r1_bio->read_disk = rdisk; |
f0cc9a05 | 1157 | r1_bio->start_next_window = 0; |
1da177e4 | 1158 | |
a167f663 | 1159 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1160 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1161 | max_sectors); |
1da177e4 LT |
1162 | |
1163 | r1_bio->bios[rdisk] = read_bio; | |
1164 | ||
4f024f37 KO |
1165 | read_bio->bi_iter.bi_sector = r1_bio->sector + |
1166 | mirror->rdev->data_offset; | |
1da177e4 LT |
1167 | read_bio->bi_bdev = mirror->rdev->bdev; |
1168 | read_bio->bi_end_io = raid1_end_read_request; | |
7b6d91da | 1169 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1170 | read_bio->bi_private = r1_bio; |
1171 | ||
d2eb35ac N |
1172 | if (max_sectors < r1_bio->sectors) { |
1173 | /* could not read all from this device, so we will | |
1174 | * need another r1_bio. | |
1175 | */ | |
d2eb35ac N |
1176 | |
1177 | sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 1178 | - bio->bi_iter.bi_sector); |
d2eb35ac N |
1179 | r1_bio->sectors = max_sectors; |
1180 | spin_lock_irq(&conf->device_lock); | |
1181 | if (bio->bi_phys_segments == 0) | |
1182 | bio->bi_phys_segments = 2; | |
1183 | else | |
1184 | bio->bi_phys_segments++; | |
1185 | spin_unlock_irq(&conf->device_lock); | |
1186 | /* Cannot call generic_make_request directly | |
1187 | * as that will be queued in __make_request | |
1188 | * and subsequent mempool_alloc might block waiting | |
1189 | * for it. So hand bio over to raid1d. | |
1190 | */ | |
1191 | reschedule_retry(r1_bio); | |
1192 | ||
1193 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1194 | ||
1195 | r1_bio->master_bio = bio; | |
aa8b57aa | 1196 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
d2eb35ac N |
1197 | r1_bio->state = 0; |
1198 | r1_bio->mddev = mddev; | |
4f024f37 KO |
1199 | r1_bio->sector = bio->bi_iter.bi_sector + |
1200 | sectors_handled; | |
d2eb35ac N |
1201 | goto read_again; |
1202 | } else | |
1203 | generic_make_request(read_bio); | |
5a7bbad2 | 1204 | return; |
1da177e4 LT |
1205 | } |
1206 | ||
1207 | /* | |
1208 | * WRITE: | |
1209 | */ | |
34db0cd6 N |
1210 | if (conf->pending_count >= max_queued_requests) { |
1211 | md_wakeup_thread(mddev->thread); | |
1212 | wait_event(conf->wait_barrier, | |
1213 | conf->pending_count < max_queued_requests); | |
1214 | } | |
1f68f0c4 | 1215 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1216 | * inc refcount on their rdev. Record them by setting |
1217 | * bios[x] to bio | |
1f68f0c4 N |
1218 | * If there are known/acknowledged bad blocks on any device on |
1219 | * which we have seen a write error, we want to avoid writing those | |
1220 | * blocks. | |
1221 | * This potentially requires several writes to write around | |
1222 | * the bad blocks. Each set of writes gets it's own r1bio | |
1223 | * with a set of bios attached. | |
1da177e4 | 1224 | */ |
c3b328ac | 1225 | |
8f19ccb2 | 1226 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1227 | retry_write: |
79ef3a8a | 1228 | r1_bio->start_next_window = start_next_window; |
6bfe0b49 | 1229 | blocked_rdev = NULL; |
1da177e4 | 1230 | rcu_read_lock(); |
1f68f0c4 | 1231 | max_sectors = r1_bio->sectors; |
1da177e4 | 1232 | for (i = 0; i < disks; i++) { |
3cb03002 | 1233 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1234 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1235 | atomic_inc(&rdev->nr_pending); | |
1236 | blocked_rdev = rdev; | |
1237 | break; | |
1238 | } | |
1f68f0c4 | 1239 | r1_bio->bios[i] = NULL; |
8ae12666 | 1240 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1241 | if (i < conf->raid_disks) |
1242 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1243 | continue; |
1244 | } | |
1245 | ||
1246 | atomic_inc(&rdev->nr_pending); | |
1247 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1248 | sector_t first_bad; | |
1249 | int bad_sectors; | |
1250 | int is_bad; | |
1251 | ||
1252 | is_bad = is_badblock(rdev, r1_bio->sector, | |
1253 | max_sectors, | |
1254 | &first_bad, &bad_sectors); | |
1255 | if (is_bad < 0) { | |
1256 | /* mustn't write here until the bad block is | |
1257 | * acknowledged*/ | |
1258 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1259 | blocked_rdev = rdev; | |
1260 | break; | |
1261 | } | |
1262 | if (is_bad && first_bad <= r1_bio->sector) { | |
1263 | /* Cannot write here at all */ | |
1264 | bad_sectors -= (r1_bio->sector - first_bad); | |
1265 | if (bad_sectors < max_sectors) | |
1266 | /* mustn't write more than bad_sectors | |
1267 | * to other devices yet | |
1268 | */ | |
1269 | max_sectors = bad_sectors; | |
03c902e1 | 1270 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1271 | /* We don't set R1BIO_Degraded as that |
1272 | * only applies if the disk is | |
1273 | * missing, so it might be re-added, | |
1274 | * and we want to know to recover this | |
1275 | * chunk. | |
1276 | * In this case the device is here, | |
1277 | * and the fact that this chunk is not | |
1278 | * in-sync is recorded in the bad | |
1279 | * block log | |
1280 | */ | |
1281 | continue; | |
964147d5 | 1282 | } |
1f68f0c4 N |
1283 | if (is_bad) { |
1284 | int good_sectors = first_bad - r1_bio->sector; | |
1285 | if (good_sectors < max_sectors) | |
1286 | max_sectors = good_sectors; | |
1287 | } | |
1288 | } | |
1289 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1290 | } |
1291 | rcu_read_unlock(); | |
1292 | ||
6bfe0b49 DW |
1293 | if (unlikely(blocked_rdev)) { |
1294 | /* Wait for this device to become unblocked */ | |
1295 | int j; | |
79ef3a8a | 1296 | sector_t old = start_next_window; |
6bfe0b49 DW |
1297 | |
1298 | for (j = 0; j < i; j++) | |
1299 | if (r1_bio->bios[j]) | |
1300 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1301 | r1_bio->state = 0; |
4f024f37 | 1302 | allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); |
6bfe0b49 | 1303 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
79ef3a8a | 1304 | start_next_window = wait_barrier(conf, bio); |
1305 | /* | |
1306 | * We must make sure the multi r1bios of bio have | |
1307 | * the same value of bi_phys_segments | |
1308 | */ | |
1309 | if (bio->bi_phys_segments && old && | |
1310 | old != start_next_window) | |
1311 | /* Wait for the former r1bio(s) to complete */ | |
1312 | wait_event(conf->wait_barrier, | |
1313 | bio->bi_phys_segments == 1); | |
6bfe0b49 DW |
1314 | goto retry_write; |
1315 | } | |
1316 | ||
1f68f0c4 N |
1317 | if (max_sectors < r1_bio->sectors) { |
1318 | /* We are splitting this write into multiple parts, so | |
1319 | * we need to prepare for allocating another r1_bio. | |
1320 | */ | |
1321 | r1_bio->sectors = max_sectors; | |
1322 | spin_lock_irq(&conf->device_lock); | |
1323 | if (bio->bi_phys_segments == 0) | |
1324 | bio->bi_phys_segments = 2; | |
1325 | else | |
1326 | bio->bi_phys_segments++; | |
1327 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1328 | } |
4f024f37 | 1329 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
4b6d287f | 1330 | |
4e78064f | 1331 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1332 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1333 | |
1f68f0c4 | 1334 | first_clone = 1; |
1da177e4 LT |
1335 | for (i = 0; i < disks; i++) { |
1336 | struct bio *mbio; | |
1337 | if (!r1_bio->bios[i]) | |
1338 | continue; | |
1339 | ||
a167f663 | 1340 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1341 | bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
1f68f0c4 N |
1342 | |
1343 | if (first_clone) { | |
1344 | /* do behind I/O ? | |
1345 | * Not if there are too many, or cannot | |
1346 | * allocate memory, or a reader on WriteMostly | |
1347 | * is waiting for behind writes to flush */ | |
1348 | if (bitmap && | |
1349 | (atomic_read(&bitmap->behind_writes) | |
1350 | < mddev->bitmap_info.max_write_behind) && | |
1351 | !waitqueue_active(&bitmap->behind_wait)) | |
1352 | alloc_behind_pages(mbio, r1_bio); | |
1353 | ||
1354 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1355 | r1_bio->sectors, | |
1356 | test_bit(R1BIO_BehindIO, | |
1357 | &r1_bio->state)); | |
1358 | first_clone = 0; | |
1359 | } | |
2ca68f5e | 1360 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1361 | struct bio_vec *bvec; |
1362 | int j; | |
1363 | ||
cb34e057 KO |
1364 | /* |
1365 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1366 | */ |
d74c6d51 | 1367 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1368 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1369 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1370 | atomic_inc(&r1_bio->behind_remaining); | |
1371 | } | |
1372 | ||
1f68f0c4 N |
1373 | r1_bio->bios[i] = mbio; |
1374 | ||
4f024f37 | 1375 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 N |
1376 | conf->mirrors[i].rdev->data_offset); |
1377 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1378 | mbio->bi_end_io = raid1_end_write_request; | |
c8dc9c65 JL |
1379 | mbio->bi_rw = |
1380 | WRITE | do_flush_fua | do_sync | do_discard | do_same; | |
1f68f0c4 N |
1381 | mbio->bi_private = r1_bio; |
1382 | ||
1da177e4 | 1383 | atomic_inc(&r1_bio->remaining); |
f54a9d0e N |
1384 | |
1385 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); | |
1386 | if (cb) | |
1387 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1388 | else | |
1389 | plug = NULL; | |
4e78064f | 1390 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1391 | if (plug) { |
1392 | bio_list_add(&plug->pending, mbio); | |
1393 | plug->pending_cnt++; | |
1394 | } else { | |
1395 | bio_list_add(&conf->pending_bio_list, mbio); | |
1396 | conf->pending_count++; | |
1397 | } | |
4e78064f | 1398 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1399 | if (!plug) |
b357f04a | 1400 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1401 | } |
079fa166 N |
1402 | /* Mustn't call r1_bio_write_done before this next test, |
1403 | * as it could result in the bio being freed. | |
1404 | */ | |
aa8b57aa | 1405 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1406 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1407 | /* We need another r1_bio. It has already been counted |
1408 | * in bio->bi_phys_segments | |
1409 | */ | |
1410 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1411 | r1_bio->master_bio = bio; | |
aa8b57aa | 1412 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
1f68f0c4 N |
1413 | r1_bio->state = 0; |
1414 | r1_bio->mddev = mddev; | |
4f024f37 | 1415 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
1f68f0c4 N |
1416 | goto retry_write; |
1417 | } | |
1418 | ||
079fa166 N |
1419 | r1_bio_write_done(r1_bio); |
1420 | ||
1421 | /* In case raid1d snuck in to freeze_array */ | |
1422 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1423 | } |
1424 | ||
849674e4 | 1425 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1426 | { |
e8096360 | 1427 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1428 | int i; |
1429 | ||
1430 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1431 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1432 | rcu_read_lock(); |
1433 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1434 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1435 | seq_printf(seq, "%s", |
ddac7c7e N |
1436 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1437 | } | |
1438 | rcu_read_unlock(); | |
1da177e4 LT |
1439 | seq_printf(seq, "]"); |
1440 | } | |
1441 | ||
849674e4 | 1442 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1443 | { |
1444 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1445 | struct r1conf *conf = mddev->private; |
423f04d6 | 1446 | unsigned long flags; |
1da177e4 LT |
1447 | |
1448 | /* | |
1449 | * If it is not operational, then we have already marked it as dead | |
1450 | * else if it is the last working disks, ignore the error, let the | |
1451 | * next level up know. | |
1452 | * else mark the drive as failed | |
1453 | */ | |
b2d444d7 | 1454 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1455 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1456 | /* |
1457 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1458 | * normal single drive. |
1459 | * However don't try a recovery from this drive as | |
1460 | * it is very likely to fail. | |
1da177e4 | 1461 | */ |
5389042f | 1462 | conf->recovery_disabled = mddev->recovery_disabled; |
1da177e4 | 1463 | return; |
4044ba58 | 1464 | } |
de393cde | 1465 | set_bit(Blocked, &rdev->flags); |
423f04d6 | 1466 | spin_lock_irqsave(&conf->device_lock, flags); |
c04be0aa | 1467 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1468 | mddev->degraded++; |
dd00a99e | 1469 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1470 | } else |
1471 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1472 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1473 | /* |
1474 | * if recovery is running, make sure it aborts. | |
1475 | */ | |
1476 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
850b2b42 | 1477 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
55ce74d4 | 1478 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
067032bc JP |
1479 | printk(KERN_ALERT |
1480 | "md/raid1:%s: Disk failure on %s, disabling device.\n" | |
1481 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
9dd1e2fa N |
1482 | mdname(mddev), bdevname(rdev->bdev, b), |
1483 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1484 | } |
1485 | ||
e8096360 | 1486 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1487 | { |
1488 | int i; | |
1da177e4 | 1489 | |
9dd1e2fa | 1490 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
1da177e4 | 1491 | if (!conf) { |
9dd1e2fa | 1492 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1493 | return; |
1494 | } | |
9dd1e2fa | 1495 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1496 | conf->raid_disks); |
1497 | ||
ddac7c7e | 1498 | rcu_read_lock(); |
1da177e4 LT |
1499 | for (i = 0; i < conf->raid_disks; i++) { |
1500 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1501 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1502 | if (rdev) |
9dd1e2fa | 1503 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
ddac7c7e N |
1504 | i, !test_bit(In_sync, &rdev->flags), |
1505 | !test_bit(Faulty, &rdev->flags), | |
1506 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1507 | } |
ddac7c7e | 1508 | rcu_read_unlock(); |
1da177e4 LT |
1509 | } |
1510 | ||
e8096360 | 1511 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1512 | { |
79ef3a8a | 1513 | wait_barrier(conf, NULL); |
1514 | allow_barrier(conf, 0, 0); | |
1da177e4 LT |
1515 | |
1516 | mempool_destroy(conf->r1buf_pool); | |
1517 | conf->r1buf_pool = NULL; | |
79ef3a8a | 1518 | |
669cc7ba | 1519 | spin_lock_irq(&conf->resync_lock); |
e8ff8bf0 | 1520 | conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE; |
79ef3a8a | 1521 | conf->start_next_window = MaxSector; |
669cc7ba N |
1522 | conf->current_window_requests += |
1523 | conf->next_window_requests; | |
1524 | conf->next_window_requests = 0; | |
1525 | spin_unlock_irq(&conf->resync_lock); | |
1da177e4 LT |
1526 | } |
1527 | ||
fd01b88c | 1528 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1529 | { |
1530 | int i; | |
e8096360 | 1531 | struct r1conf *conf = mddev->private; |
6b965620 N |
1532 | int count = 0; |
1533 | unsigned long flags; | |
1da177e4 LT |
1534 | |
1535 | /* | |
f72ffdd6 | 1536 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1537 | * and mark them readable. |
1538 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1539 | * device_lock used to avoid races with raid1_end_read_request |
1540 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1541 | */ |
423f04d6 | 1542 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1543 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1544 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1545 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1546 | if (repl | |
1aee41f6 | 1547 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1548 | && repl->recovery_offset == MaxSector |
1549 | && !test_bit(Faulty, &repl->flags) | |
1550 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1551 | /* replacement has just become active */ | |
1552 | if (!rdev || | |
1553 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1554 | count++; | |
1555 | if (rdev) { | |
1556 | /* Replaced device not technically | |
1557 | * faulty, but we need to be sure | |
1558 | * it gets removed and never re-added | |
1559 | */ | |
1560 | set_bit(Faulty, &rdev->flags); | |
1561 | sysfs_notify_dirent_safe( | |
1562 | rdev->sysfs_state); | |
1563 | } | |
1564 | } | |
ddac7c7e | 1565 | if (rdev |
61e4947c | 1566 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1567 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1568 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1569 | count++; |
654e8b5a | 1570 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1571 | } |
1572 | } | |
6b965620 N |
1573 | mddev->degraded -= count; |
1574 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1575 | |
1576 | print_conf(conf); | |
6b965620 | 1577 | return count; |
1da177e4 LT |
1578 | } |
1579 | ||
fd01b88c | 1580 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1581 | { |
e8096360 | 1582 | struct r1conf *conf = mddev->private; |
199050ea | 1583 | int err = -EEXIST; |
41158c7e | 1584 | int mirror = 0; |
0eaf822c | 1585 | struct raid1_info *p; |
6c2fce2e | 1586 | int first = 0; |
30194636 | 1587 | int last = conf->raid_disks - 1; |
1da177e4 | 1588 | |
5389042f N |
1589 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1590 | return -EBUSY; | |
1591 | ||
1501efad DW |
1592 | if (md_integrity_add_rdev(rdev, mddev)) |
1593 | return -ENXIO; | |
1594 | ||
6c2fce2e NB |
1595 | if (rdev->raid_disk >= 0) |
1596 | first = last = rdev->raid_disk; | |
1597 | ||
70bcecdb GR |
1598 | /* |
1599 | * find the disk ... but prefer rdev->saved_raid_disk | |
1600 | * if possible. | |
1601 | */ | |
1602 | if (rdev->saved_raid_disk >= 0 && | |
1603 | rdev->saved_raid_disk >= first && | |
1604 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) | |
1605 | first = last = rdev->saved_raid_disk; | |
1606 | ||
7ef449d1 N |
1607 | for (mirror = first; mirror <= last; mirror++) { |
1608 | p = conf->mirrors+mirror; | |
1609 | if (!p->rdev) { | |
1da177e4 | 1610 | |
9092c02d JB |
1611 | if (mddev->gendisk) |
1612 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1613 | rdev->data_offset << 9); | |
1da177e4 LT |
1614 | |
1615 | p->head_position = 0; | |
1616 | rdev->raid_disk = mirror; | |
199050ea | 1617 | err = 0; |
6aea114a N |
1618 | /* As all devices are equivalent, we don't need a full recovery |
1619 | * if this was recently any drive of the array | |
1620 | */ | |
1621 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1622 | conf->fullsync = 1; |
d6065f7b | 1623 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1624 | break; |
1625 | } | |
7ef449d1 N |
1626 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1627 | p[conf->raid_disks].rdev == NULL) { | |
1628 | /* Add this device as a replacement */ | |
1629 | clear_bit(In_sync, &rdev->flags); | |
1630 | set_bit(Replacement, &rdev->flags); | |
1631 | rdev->raid_disk = mirror; | |
1632 | err = 0; | |
1633 | conf->fullsync = 1; | |
1634 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1635 | break; | |
1636 | } | |
1637 | } | |
9092c02d | 1638 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1639 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1640 | print_conf(conf); |
199050ea | 1641 | return err; |
1da177e4 LT |
1642 | } |
1643 | ||
b8321b68 | 1644 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1645 | { |
e8096360 | 1646 | struct r1conf *conf = mddev->private; |
1da177e4 | 1647 | int err = 0; |
b8321b68 | 1648 | int number = rdev->raid_disk; |
0eaf822c | 1649 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1650 | |
b014f14c N |
1651 | if (rdev != p->rdev) |
1652 | p = conf->mirrors + conf->raid_disks + number; | |
1653 | ||
1da177e4 | 1654 | print_conf(conf); |
b8321b68 | 1655 | if (rdev == p->rdev) { |
b2d444d7 | 1656 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1657 | atomic_read(&rdev->nr_pending)) { |
1658 | err = -EBUSY; | |
1659 | goto abort; | |
1660 | } | |
046abeed | 1661 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1662 | * is not possible. |
1663 | */ | |
1664 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1665 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1666 | mddev->degraded < conf->raid_disks) { |
1667 | err = -EBUSY; | |
1668 | goto abort; | |
1669 | } | |
1da177e4 | 1670 | p->rdev = NULL; |
fbd568a3 | 1671 | synchronize_rcu(); |
1da177e4 LT |
1672 | if (atomic_read(&rdev->nr_pending)) { |
1673 | /* lost the race, try later */ | |
1674 | err = -EBUSY; | |
1675 | p->rdev = rdev; | |
ac5e7113 | 1676 | goto abort; |
8c7a2c2b N |
1677 | } else if (conf->mirrors[conf->raid_disks + number].rdev) { |
1678 | /* We just removed a device that is being replaced. | |
1679 | * Move down the replacement. We drain all IO before | |
1680 | * doing this to avoid confusion. | |
1681 | */ | |
1682 | struct md_rdev *repl = | |
1683 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1684 | freeze_array(conf, 0); |
8c7a2c2b N |
1685 | clear_bit(Replacement, &repl->flags); |
1686 | p->rdev = repl; | |
1687 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1688 | unfreeze_array(conf); |
8c7a2c2b N |
1689 | clear_bit(WantReplacement, &rdev->flags); |
1690 | } else | |
b014f14c | 1691 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1692 | err = md_integrity_register(mddev); |
1da177e4 LT |
1693 | } |
1694 | abort: | |
1695 | ||
1696 | print_conf(conf); | |
1697 | return err; | |
1698 | } | |
1699 | ||
4246a0b6 | 1700 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1701 | { |
9f2c9d12 | 1702 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1703 | |
0fc280f6 | 1704 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1705 | |
1da177e4 LT |
1706 | /* |
1707 | * we have read a block, now it needs to be re-written, | |
1708 | * or re-read if the read failed. | |
1709 | * We don't do much here, just schedule handling by raid1d | |
1710 | */ | |
4246a0b6 | 1711 | if (!bio->bi_error) |
1da177e4 | 1712 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1713 | |
1714 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1715 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1716 | } |
1717 | ||
4246a0b6 | 1718 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1719 | { |
4246a0b6 | 1720 | int uptodate = !bio->bi_error; |
9f2c9d12 | 1721 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1722 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1723 | struct r1conf *conf = mddev->private; |
1da177e4 | 1724 | int mirror=0; |
4367af55 N |
1725 | sector_t first_bad; |
1726 | int bad_sectors; | |
1da177e4 | 1727 | |
ba3ae3be NK |
1728 | mirror = find_bio_disk(r1_bio, bio); |
1729 | ||
6b1117d5 | 1730 | if (!uptodate) { |
57dab0bd | 1731 | sector_t sync_blocks = 0; |
6b1117d5 N |
1732 | sector_t s = r1_bio->sector; |
1733 | long sectors_to_go = r1_bio->sectors; | |
1734 | /* make sure these bits doesn't get cleared. */ | |
1735 | do { | |
5e3db645 | 1736 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1737 | &sync_blocks, 1); |
1738 | s += sync_blocks; | |
1739 | sectors_to_go -= sync_blocks; | |
1740 | } while (sectors_to_go > 0); | |
d8f05d29 N |
1741 | set_bit(WriteErrorSeen, |
1742 | &conf->mirrors[mirror].rdev->flags); | |
19d67169 N |
1743 | if (!test_and_set_bit(WantReplacement, |
1744 | &conf->mirrors[mirror].rdev->flags)) | |
1745 | set_bit(MD_RECOVERY_NEEDED, & | |
1746 | mddev->recovery); | |
d8f05d29 | 1747 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 N |
1748 | } else if (is_badblock(conf->mirrors[mirror].rdev, |
1749 | r1_bio->sector, | |
1750 | r1_bio->sectors, | |
3a9f28a5 N |
1751 | &first_bad, &bad_sectors) && |
1752 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1753 | r1_bio->sector, | |
1754 | r1_bio->sectors, | |
1755 | &first_bad, &bad_sectors) | |
1756 | ) | |
4367af55 | 1757 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1758 | |
1da177e4 | 1759 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1760 | int s = r1_bio->sectors; |
d8f05d29 N |
1761 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1762 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1763 | reschedule_retry(r1_bio); |
1764 | else { | |
1765 | put_buf(r1_bio); | |
1766 | md_done_sync(mddev, s, uptodate); | |
1767 | } | |
1da177e4 | 1768 | } |
1da177e4 LT |
1769 | } |
1770 | ||
3cb03002 | 1771 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1772 | int sectors, struct page *page, int rw) |
1773 | { | |
1774 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
1775 | /* success */ | |
1776 | return 1; | |
19d67169 | 1777 | if (rw == WRITE) { |
d8f05d29 | 1778 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1779 | if (!test_and_set_bit(WantReplacement, |
1780 | &rdev->flags)) | |
1781 | set_bit(MD_RECOVERY_NEEDED, & | |
1782 | rdev->mddev->recovery); | |
1783 | } | |
d8f05d29 N |
1784 | /* need to record an error - either for the block or the device */ |
1785 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1786 | md_error(rdev->mddev, rdev); | |
1787 | return 0; | |
1788 | } | |
1789 | ||
9f2c9d12 | 1790 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1791 | { |
a68e5870 N |
1792 | /* Try some synchronous reads of other devices to get |
1793 | * good data, much like with normal read errors. Only | |
1794 | * read into the pages we already have so we don't | |
1795 | * need to re-issue the read request. | |
1796 | * We don't need to freeze the array, because being in an | |
1797 | * active sync request, there is no normal IO, and | |
1798 | * no overlapping syncs. | |
06f60385 N |
1799 | * We don't need to check is_badblock() again as we |
1800 | * made sure that anything with a bad block in range | |
1801 | * will have bi_end_io clear. | |
a68e5870 | 1802 | */ |
fd01b88c | 1803 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1804 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1805 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1806 | sector_t sect = r1_bio->sector; | |
1807 | int sectors = r1_bio->sectors; | |
1808 | int idx = 0; | |
1809 | ||
1810 | while(sectors) { | |
1811 | int s = sectors; | |
1812 | int d = r1_bio->read_disk; | |
1813 | int success = 0; | |
3cb03002 | 1814 | struct md_rdev *rdev; |
78d7f5f7 | 1815 | int start; |
a68e5870 N |
1816 | |
1817 | if (s > (PAGE_SIZE>>9)) | |
1818 | s = PAGE_SIZE >> 9; | |
1819 | do { | |
1820 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1821 | /* No rcu protection needed here devices | |
1822 | * can only be removed when no resync is | |
1823 | * active, and resync is currently active | |
1824 | */ | |
1825 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1826 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 N |
1827 | bio->bi_io_vec[idx].bv_page, |
1828 | READ, false)) { | |
1829 | success = 1; | |
1830 | break; | |
1831 | } | |
1832 | } | |
1833 | d++; | |
8f19ccb2 | 1834 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1835 | d = 0; |
1836 | } while (!success && d != r1_bio->read_disk); | |
1837 | ||
78d7f5f7 | 1838 | if (!success) { |
a68e5870 | 1839 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1840 | int abort = 0; |
1841 | /* Cannot read from anywhere, this block is lost. | |
1842 | * Record a bad block on each device. If that doesn't | |
1843 | * work just disable and interrupt the recovery. | |
1844 | * Don't fail devices as that won't really help. | |
1845 | */ | |
a68e5870 N |
1846 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
1847 | " for block %llu\n", | |
1848 | mdname(mddev), | |
1849 | bdevname(bio->bi_bdev, b), | |
1850 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1851 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1852 | rdev = conf->mirrors[d].rdev; |
1853 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1854 | continue; | |
1855 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1856 | abort = 1; | |
1857 | } | |
1858 | if (abort) { | |
d890fa2b N |
1859 | conf->recovery_disabled = |
1860 | mddev->recovery_disabled; | |
3a9f28a5 N |
1861 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1862 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1863 | put_buf(r1_bio); | |
1864 | return 0; | |
1865 | } | |
1866 | /* Try next page */ | |
1867 | sectors -= s; | |
1868 | sect += s; | |
1869 | idx++; | |
1870 | continue; | |
d11c171e | 1871 | } |
78d7f5f7 N |
1872 | |
1873 | start = d; | |
1874 | /* write it back and re-read */ | |
1875 | while (d != r1_bio->read_disk) { | |
1876 | if (d == 0) | |
8f19ccb2 | 1877 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1878 | d--; |
1879 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1880 | continue; | |
1881 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1882 | if (r1_sync_page_io(rdev, sect, s, |
1883 | bio->bi_io_vec[idx].bv_page, | |
1884 | WRITE) == 0) { | |
78d7f5f7 N |
1885 | r1_bio->bios[d]->bi_end_io = NULL; |
1886 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1887 | } |
78d7f5f7 N |
1888 | } |
1889 | d = start; | |
1890 | while (d != r1_bio->read_disk) { | |
1891 | if (d == 0) | |
8f19ccb2 | 1892 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1893 | d--; |
1894 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1895 | continue; | |
1896 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1897 | if (r1_sync_page_io(rdev, sect, s, |
1898 | bio->bi_io_vec[idx].bv_page, | |
1899 | READ) != 0) | |
9d3d8011 | 1900 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1901 | } |
a68e5870 N |
1902 | sectors -= s; |
1903 | sect += s; | |
1904 | idx ++; | |
1905 | } | |
78d7f5f7 | 1906 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4246a0b6 | 1907 | bio->bi_error = 0; |
a68e5870 N |
1908 | return 1; |
1909 | } | |
1910 | ||
c95e6385 | 1911 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1912 | { |
1913 | /* We have read all readable devices. If we haven't | |
1914 | * got the block, then there is no hope left. | |
1915 | * If we have, then we want to do a comparison | |
1916 | * and skip the write if everything is the same. | |
1917 | * If any blocks failed to read, then we need to | |
1918 | * attempt an over-write | |
1919 | */ | |
fd01b88c | 1920 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1921 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1922 | int primary; |
1923 | int i; | |
f4380a91 | 1924 | int vcnt; |
a68e5870 | 1925 | |
30bc9b53 N |
1926 | /* Fix variable parts of all bios */ |
1927 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
1928 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
1929 | int j; | |
1930 | int size; | |
4246a0b6 | 1931 | int error; |
30bc9b53 N |
1932 | struct bio *b = r1_bio->bios[i]; |
1933 | if (b->bi_end_io != end_sync_read) | |
1934 | continue; | |
4246a0b6 CH |
1935 | /* fixup the bio for reuse, but preserve errno */ |
1936 | error = b->bi_error; | |
30bc9b53 | 1937 | bio_reset(b); |
4246a0b6 | 1938 | b->bi_error = error; |
30bc9b53 | 1939 | b->bi_vcnt = vcnt; |
4f024f37 KO |
1940 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
1941 | b->bi_iter.bi_sector = r1_bio->sector + | |
30bc9b53 N |
1942 | conf->mirrors[i].rdev->data_offset; |
1943 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1944 | b->bi_end_io = end_sync_read; | |
1945 | b->bi_private = r1_bio; | |
1946 | ||
4f024f37 | 1947 | size = b->bi_iter.bi_size; |
30bc9b53 N |
1948 | for (j = 0; j < vcnt ; j++) { |
1949 | struct bio_vec *bi; | |
1950 | bi = &b->bi_io_vec[j]; | |
1951 | bi->bv_offset = 0; | |
1952 | if (size > PAGE_SIZE) | |
1953 | bi->bv_len = PAGE_SIZE; | |
1954 | else | |
1955 | bi->bv_len = size; | |
1956 | size -= PAGE_SIZE; | |
1957 | } | |
1958 | } | |
8f19ccb2 | 1959 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 1960 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4246a0b6 | 1961 | !r1_bio->bios[primary]->bi_error) { |
a68e5870 N |
1962 | r1_bio->bios[primary]->bi_end_io = NULL; |
1963 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
1964 | break; | |
1965 | } | |
1966 | r1_bio->read_disk = primary; | |
8f19ccb2 | 1967 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 1968 | int j; |
78d7f5f7 N |
1969 | struct bio *pbio = r1_bio->bios[primary]; |
1970 | struct bio *sbio = r1_bio->bios[i]; | |
4246a0b6 | 1971 | int error = sbio->bi_error; |
a68e5870 | 1972 | |
2aabaa65 | 1973 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 1974 | continue; |
4246a0b6 CH |
1975 | /* Now we can 'fixup' the error value */ |
1976 | sbio->bi_error = 0; | |
78d7f5f7 | 1977 | |
4246a0b6 | 1978 | if (!error) { |
78d7f5f7 N |
1979 | for (j = vcnt; j-- ; ) { |
1980 | struct page *p, *s; | |
1981 | p = pbio->bi_io_vec[j].bv_page; | |
1982 | s = sbio->bi_io_vec[j].bv_page; | |
1983 | if (memcmp(page_address(p), | |
1984 | page_address(s), | |
5020ad7d | 1985 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 1986 | break; |
69382e85 | 1987 | } |
78d7f5f7 N |
1988 | } else |
1989 | j = 0; | |
1990 | if (j >= 0) | |
7f7583d4 | 1991 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 1992 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4246a0b6 | 1993 | && !error)) { |
78d7f5f7 N |
1994 | /* No need to write to this device. */ |
1995 | sbio->bi_end_io = NULL; | |
1996 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
1997 | continue; | |
1998 | } | |
d3b45c2a KO |
1999 | |
2000 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2001 | } |
a68e5870 N |
2002 | } |
2003 | ||
9f2c9d12 | 2004 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2005 | { |
e8096360 | 2006 | struct r1conf *conf = mddev->private; |
a68e5870 | 2007 | int i; |
8f19ccb2 | 2008 | int disks = conf->raid_disks * 2; |
a68e5870 N |
2009 | struct bio *bio, *wbio; |
2010 | ||
2011 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2012 | ||
a68e5870 N |
2013 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2014 | /* ouch - failed to read all of that. */ | |
2015 | if (!fix_sync_read_error(r1_bio)) | |
2016 | return; | |
7ca78d57 N |
2017 | |
2018 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2019 | process_checks(r1_bio); |
2020 | ||
d11c171e N |
2021 | /* |
2022 | * schedule writes | |
2023 | */ | |
1da177e4 LT |
2024 | atomic_set(&r1_bio->remaining, 1); |
2025 | for (i = 0; i < disks ; i++) { | |
2026 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2027 | if (wbio->bi_end_io == NULL || |
2028 | (wbio->bi_end_io == end_sync_read && | |
2029 | (i == r1_bio->read_disk || | |
2030 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
2031 | continue; |
2032 | ||
3e198f78 N |
2033 | wbio->bi_rw = WRITE; |
2034 | wbio->bi_end_io = end_sync_write; | |
1da177e4 | 2035 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2036 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2037 | |
1da177e4 LT |
2038 | generic_make_request(wbio); |
2039 | } | |
2040 | ||
2041 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2042 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2043 | int s = r1_bio->sectors; |
2044 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2045 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2046 | reschedule_retry(r1_bio); | |
2047 | else { | |
2048 | put_buf(r1_bio); | |
2049 | md_done_sync(mddev, s, 1); | |
2050 | } | |
1da177e4 LT |
2051 | } |
2052 | } | |
2053 | ||
2054 | /* | |
2055 | * This is a kernel thread which: | |
2056 | * | |
2057 | * 1. Retries failed read operations on working mirrors. | |
2058 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2059 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2060 | */ |
2061 | ||
e8096360 | 2062 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2063 | sector_t sect, int sectors) |
2064 | { | |
fd01b88c | 2065 | struct mddev *mddev = conf->mddev; |
867868fb N |
2066 | while(sectors) { |
2067 | int s = sectors; | |
2068 | int d = read_disk; | |
2069 | int success = 0; | |
2070 | int start; | |
3cb03002 | 2071 | struct md_rdev *rdev; |
867868fb N |
2072 | |
2073 | if (s > (PAGE_SIZE>>9)) | |
2074 | s = PAGE_SIZE >> 9; | |
2075 | ||
2076 | do { | |
2077 | /* Note: no rcu protection needed here | |
2078 | * as this is synchronous in the raid1d thread | |
2079 | * which is the thread that might remove | |
2080 | * a device. If raid1d ever becomes multi-threaded.... | |
2081 | */ | |
d2eb35ac N |
2082 | sector_t first_bad; |
2083 | int bad_sectors; | |
2084 | ||
867868fb N |
2085 | rdev = conf->mirrors[d].rdev; |
2086 | if (rdev && | |
da8840a7 | 2087 | (test_bit(In_sync, &rdev->flags) || |
2088 | (!test_bit(Faulty, &rdev->flags) && | |
2089 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac N |
2090 | is_badblock(rdev, sect, s, |
2091 | &first_bad, &bad_sectors) == 0 && | |
ccebd4c4 JB |
2092 | sync_page_io(rdev, sect, s<<9, |
2093 | conf->tmppage, READ, false)) | |
867868fb N |
2094 | success = 1; |
2095 | else { | |
2096 | d++; | |
8f19ccb2 | 2097 | if (d == conf->raid_disks * 2) |
867868fb N |
2098 | d = 0; |
2099 | } | |
2100 | } while (!success && d != read_disk); | |
2101 | ||
2102 | if (!success) { | |
d8f05d29 | 2103 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2104 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2105 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2106 | md_error(mddev, rdev); | |
867868fb N |
2107 | break; |
2108 | } | |
2109 | /* write it back and re-read */ | |
2110 | start = d; | |
2111 | while (d != read_disk) { | |
2112 | if (d==0) | |
8f19ccb2 | 2113 | d = conf->raid_disks * 2; |
867868fb N |
2114 | d--; |
2115 | rdev = conf->mirrors[d].rdev; | |
2116 | if (rdev && | |
b8cb6b4c | 2117 | !test_bit(Faulty, &rdev->flags)) |
d8f05d29 N |
2118 | r1_sync_page_io(rdev, sect, s, |
2119 | conf->tmppage, WRITE); | |
867868fb N |
2120 | } |
2121 | d = start; | |
2122 | while (d != read_disk) { | |
2123 | char b[BDEVNAME_SIZE]; | |
2124 | if (d==0) | |
8f19ccb2 | 2125 | d = conf->raid_disks * 2; |
867868fb N |
2126 | d--; |
2127 | rdev = conf->mirrors[d].rdev; | |
2128 | if (rdev && | |
b8cb6b4c | 2129 | !test_bit(Faulty, &rdev->flags)) { |
d8f05d29 N |
2130 | if (r1_sync_page_io(rdev, sect, s, |
2131 | conf->tmppage, READ)) { | |
867868fb N |
2132 | atomic_add(s, &rdev->corrected_errors); |
2133 | printk(KERN_INFO | |
9dd1e2fa | 2134 | "md/raid1:%s: read error corrected " |
867868fb N |
2135 | "(%d sectors at %llu on %s)\n", |
2136 | mdname(mddev), s, | |
969b755a RD |
2137 | (unsigned long long)(sect + |
2138 | rdev->data_offset), | |
867868fb N |
2139 | bdevname(rdev->bdev, b)); |
2140 | } | |
2141 | } | |
2142 | } | |
2143 | sectors -= s; | |
2144 | sect += s; | |
2145 | } | |
2146 | } | |
2147 | ||
9f2c9d12 | 2148 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2149 | { |
fd01b88c | 2150 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2151 | struct r1conf *conf = mddev->private; |
3cb03002 | 2152 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2153 | |
2154 | /* bio has the data to be written to device 'i' where | |
2155 | * we just recently had a write error. | |
2156 | * We repeatedly clone the bio and trim down to one block, | |
2157 | * then try the write. Where the write fails we record | |
2158 | * a bad block. | |
2159 | * It is conceivable that the bio doesn't exactly align with | |
2160 | * blocks. We must handle this somehow. | |
2161 | * | |
2162 | * We currently own a reference on the rdev. | |
2163 | */ | |
2164 | ||
2165 | int block_sectors; | |
2166 | sector_t sector; | |
2167 | int sectors; | |
2168 | int sect_to_write = r1_bio->sectors; | |
2169 | int ok = 1; | |
2170 | ||
2171 | if (rdev->badblocks.shift < 0) | |
2172 | return 0; | |
2173 | ||
ab713cdc ND |
2174 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2175 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2176 | sector = r1_bio->sector; |
2177 | sectors = ((sector + block_sectors) | |
2178 | & ~(sector_t)(block_sectors - 1)) | |
2179 | - sector; | |
2180 | ||
cd5ff9a1 N |
2181 | while (sect_to_write) { |
2182 | struct bio *wbio; | |
2183 | if (sectors > sect_to_write) | |
2184 | sectors = sect_to_write; | |
2185 | /* Write at 'sector' for 'sectors'*/ | |
2186 | ||
b783863f KO |
2187 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2188 | unsigned vcnt = r1_bio->behind_page_count; | |
2189 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2190 | ||
2191 | while (!vec->bv_page) { | |
2192 | vec++; | |
2193 | vcnt--; | |
2194 | } | |
2195 | ||
2196 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2197 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2198 | ||
2199 | wbio->bi_vcnt = vcnt; | |
2200 | } else { | |
2201 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2202 | } | |
2203 | ||
cd5ff9a1 | 2204 | wbio->bi_rw = WRITE; |
4f024f37 KO |
2205 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2206 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2207 | |
6678d83f | 2208 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2209 | wbio->bi_iter.bi_sector += rdev->data_offset; |
cd5ff9a1 | 2210 | wbio->bi_bdev = rdev->bdev; |
203d27b0 | 2211 | if (submit_bio_wait(WRITE, wbio) < 0) |
cd5ff9a1 N |
2212 | /* failure! */ |
2213 | ok = rdev_set_badblocks(rdev, sector, | |
2214 | sectors, 0) | |
2215 | && ok; | |
2216 | ||
2217 | bio_put(wbio); | |
2218 | sect_to_write -= sectors; | |
2219 | sector += sectors; | |
2220 | sectors = block_sectors; | |
2221 | } | |
2222 | return ok; | |
2223 | } | |
2224 | ||
e8096360 | 2225 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2226 | { |
2227 | int m; | |
2228 | int s = r1_bio->sectors; | |
8f19ccb2 | 2229 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2230 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2231 | struct bio *bio = r1_bio->bios[m]; |
2232 | if (bio->bi_end_io == NULL) | |
2233 | continue; | |
4246a0b6 | 2234 | if (!bio->bi_error && |
62096bce | 2235 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2236 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2237 | } |
4246a0b6 | 2238 | if (bio->bi_error && |
62096bce N |
2239 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2240 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2241 | md_error(conf->mddev, rdev); | |
2242 | } | |
2243 | } | |
2244 | put_buf(r1_bio); | |
2245 | md_done_sync(conf->mddev, s, 1); | |
2246 | } | |
2247 | ||
e8096360 | 2248 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2249 | { |
2250 | int m; | |
55ce74d4 | 2251 | bool fail = false; |
8f19ccb2 | 2252 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2253 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2254 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2255 | rdev_clear_badblocks(rdev, |
2256 | r1_bio->sector, | |
c6563a8c | 2257 | r1_bio->sectors, 0); |
62096bce N |
2258 | rdev_dec_pending(rdev, conf->mddev); |
2259 | } else if (r1_bio->bios[m] != NULL) { | |
2260 | /* This drive got a write error. We need to | |
2261 | * narrow down and record precise write | |
2262 | * errors. | |
2263 | */ | |
55ce74d4 | 2264 | fail = true; |
62096bce N |
2265 | if (!narrow_write_error(r1_bio, m)) { |
2266 | md_error(conf->mddev, | |
2267 | conf->mirrors[m].rdev); | |
2268 | /* an I/O failed, we can't clear the bitmap */ | |
2269 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2270 | } | |
2271 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2272 | conf->mddev); | |
2273 | } | |
55ce74d4 N |
2274 | if (fail) { |
2275 | spin_lock_irq(&conf->device_lock); | |
2276 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
2277 | spin_unlock_irq(&conf->device_lock); | |
2278 | md_wakeup_thread(conf->mddev->thread); | |
bd8688a1 N |
2279 | } else { |
2280 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2281 | close_write(r1_bio); | |
55ce74d4 | 2282 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2283 | } |
62096bce N |
2284 | } |
2285 | ||
e8096360 | 2286 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2287 | { |
2288 | int disk; | |
2289 | int max_sectors; | |
fd01b88c | 2290 | struct mddev *mddev = conf->mddev; |
62096bce N |
2291 | struct bio *bio; |
2292 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2293 | struct md_rdev *rdev; |
62096bce N |
2294 | |
2295 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2296 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2297 | * the block and we can fix it. | |
2298 | * We freeze all other IO, and try reading the block from | |
2299 | * other devices. When we find one, we re-write | |
2300 | * and check it that fixes the read error. | |
2301 | * This is all done synchronously while the array is | |
2302 | * frozen | |
2303 | */ | |
2304 | if (mddev->ro == 0) { | |
e2d59925 | 2305 | freeze_array(conf, 1); |
62096bce N |
2306 | fix_read_error(conf, r1_bio->read_disk, |
2307 | r1_bio->sector, r1_bio->sectors); | |
2308 | unfreeze_array(conf); | |
2309 | } else | |
2310 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); | |
7ad4d4a6 | 2311 | rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); |
62096bce N |
2312 | |
2313 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2314 | bdevname(bio->bi_bdev, b); | |
2315 | read_more: | |
2316 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2317 | if (disk == -1) { | |
2318 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" | |
2319 | " read error for block %llu\n", | |
2320 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
2321 | raid_end_bio_io(r1_bio); | |
2322 | } else { | |
2323 | const unsigned long do_sync | |
2324 | = r1_bio->master_bio->bi_rw & REQ_SYNC; | |
2325 | if (bio) { | |
2326 | r1_bio->bios[r1_bio->read_disk] = | |
2327 | mddev->ro ? IO_BLOCKED : NULL; | |
2328 | bio_put(bio); | |
2329 | } | |
2330 | r1_bio->read_disk = disk; | |
2331 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
4f024f37 KO |
2332 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
2333 | max_sectors); | |
62096bce N |
2334 | r1_bio->bios[r1_bio->read_disk] = bio; |
2335 | rdev = conf->mirrors[disk].rdev; | |
2336 | printk_ratelimited(KERN_ERR | |
2337 | "md/raid1:%s: redirecting sector %llu" | |
2338 | " to other mirror: %s\n", | |
2339 | mdname(mddev), | |
2340 | (unsigned long long)r1_bio->sector, | |
2341 | bdevname(rdev->bdev, b)); | |
4f024f37 | 2342 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
62096bce N |
2343 | bio->bi_bdev = rdev->bdev; |
2344 | bio->bi_end_io = raid1_end_read_request; | |
2345 | bio->bi_rw = READ | do_sync; | |
2346 | bio->bi_private = r1_bio; | |
2347 | if (max_sectors < r1_bio->sectors) { | |
2348 | /* Drat - have to split this up more */ | |
2349 | struct bio *mbio = r1_bio->master_bio; | |
2350 | int sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 2351 | - mbio->bi_iter.bi_sector); |
62096bce N |
2352 | r1_bio->sectors = max_sectors; |
2353 | spin_lock_irq(&conf->device_lock); | |
2354 | if (mbio->bi_phys_segments == 0) | |
2355 | mbio->bi_phys_segments = 2; | |
2356 | else | |
2357 | mbio->bi_phys_segments++; | |
2358 | spin_unlock_irq(&conf->device_lock); | |
2359 | generic_make_request(bio); | |
2360 | bio = NULL; | |
2361 | ||
2362 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2363 | ||
2364 | r1_bio->master_bio = mbio; | |
aa8b57aa | 2365 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
62096bce N |
2366 | r1_bio->state = 0; |
2367 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2368 | r1_bio->mddev = mddev; | |
4f024f37 KO |
2369 | r1_bio->sector = mbio->bi_iter.bi_sector + |
2370 | sectors_handled; | |
62096bce N |
2371 | |
2372 | goto read_more; | |
2373 | } else | |
2374 | generic_make_request(bio); | |
2375 | } | |
2376 | } | |
2377 | ||
4ed8731d | 2378 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2379 | { |
4ed8731d | 2380 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2381 | struct r1bio *r1_bio; |
1da177e4 | 2382 | unsigned long flags; |
e8096360 | 2383 | struct r1conf *conf = mddev->private; |
1da177e4 | 2384 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2385 | struct blk_plug plug; |
1da177e4 LT |
2386 | |
2387 | md_check_recovery(mddev); | |
e1dfa0a2 | 2388 | |
55ce74d4 N |
2389 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2390 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
2391 | LIST_HEAD(tmp); | |
2392 | spin_lock_irqsave(&conf->device_lock, flags); | |
2393 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
2394 | list_add(&tmp, &conf->bio_end_io_list); | |
2395 | list_del_init(&conf->bio_end_io_list); | |
2396 | } | |
2397 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2398 | while (!list_empty(&tmp)) { | |
a452744b MP |
2399 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2400 | retry_list); | |
55ce74d4 | 2401 | list_del(&r1_bio->retry_list); |
bd8688a1 N |
2402 | if (mddev->degraded) |
2403 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2404 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2405 | close_write(r1_bio); | |
55ce74d4 N |
2406 | raid_end_bio_io(r1_bio); |
2407 | } | |
2408 | } | |
2409 | ||
e1dfa0a2 | 2410 | blk_start_plug(&plug); |
1da177e4 | 2411 | for (;;) { |
191ea9b2 | 2412 | |
0021b7bc | 2413 | flush_pending_writes(conf); |
191ea9b2 | 2414 | |
a35e63ef N |
2415 | spin_lock_irqsave(&conf->device_lock, flags); |
2416 | if (list_empty(head)) { | |
2417 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2418 | break; |
a35e63ef | 2419 | } |
9f2c9d12 | 2420 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2421 | list_del(head->prev); |
ddaf22ab | 2422 | conf->nr_queued--; |
1da177e4 LT |
2423 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2424 | ||
2425 | mddev = r1_bio->mddev; | |
070ec55d | 2426 | conf = mddev->private; |
4367af55 | 2427 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2428 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2429 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2430 | handle_sync_write_finished(conf, r1_bio); | |
2431 | else | |
4367af55 | 2432 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2433 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2434 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2435 | handle_write_finished(conf, r1_bio); | |
2436 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2437 | handle_read_error(conf, r1_bio); | |
2438 | else | |
d2eb35ac N |
2439 | /* just a partial read to be scheduled from separate |
2440 | * context | |
2441 | */ | |
2442 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2443 | |
1d9d5241 | 2444 | cond_resched(); |
de393cde N |
2445 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2446 | md_check_recovery(mddev); | |
1da177e4 | 2447 | } |
e1dfa0a2 | 2448 | blk_finish_plug(&plug); |
1da177e4 LT |
2449 | } |
2450 | ||
e8096360 | 2451 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2452 | { |
2453 | int buffs; | |
2454 | ||
2455 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2456 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2457 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2458 | conf->poolinfo); | |
2459 | if (!conf->r1buf_pool) | |
2460 | return -ENOMEM; | |
2461 | conf->next_resync = 0; | |
2462 | return 0; | |
2463 | } | |
2464 | ||
2465 | /* | |
2466 | * perform a "sync" on one "block" | |
2467 | * | |
2468 | * We need to make sure that no normal I/O request - particularly write | |
2469 | * requests - conflict with active sync requests. | |
2470 | * | |
2471 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2472 | * that can be installed to exclude normal IO requests. | |
2473 | */ | |
2474 | ||
849674e4 SL |
2475 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2476 | int *skipped) | |
1da177e4 | 2477 | { |
e8096360 | 2478 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2479 | struct r1bio *r1_bio; |
1da177e4 LT |
2480 | struct bio *bio; |
2481 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2482 | int disk = -1; |
1da177e4 | 2483 | int i; |
3e198f78 N |
2484 | int wonly = -1; |
2485 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2486 | sector_t sync_blocks; |
e3b9703e | 2487 | int still_degraded = 0; |
06f60385 N |
2488 | int good_sectors = RESYNC_SECTORS; |
2489 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2490 | |
2491 | if (!conf->r1buf_pool) | |
2492 | if (init_resync(conf)) | |
57afd89f | 2493 | return 0; |
1da177e4 | 2494 | |
58c0fed4 | 2495 | max_sector = mddev->dev_sectors; |
1da177e4 | 2496 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2497 | /* If we aborted, we need to abort the |
2498 | * sync on the 'current' bitmap chunk (there will | |
2499 | * only be one in raid1 resync. | |
2500 | * We can find the current addess in mddev->curr_resync | |
2501 | */ | |
6a806c51 N |
2502 | if (mddev->curr_resync < max_sector) /* aborted */ |
2503 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2504 | &sync_blocks, 1); |
6a806c51 | 2505 | else /* completed sync */ |
191ea9b2 | 2506 | conf->fullsync = 0; |
6a806c51 N |
2507 | |
2508 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2509 | close_sync(conf); |
c40f341f GR |
2510 | |
2511 | if (mddev_is_clustered(mddev)) { | |
2512 | conf->cluster_sync_low = 0; | |
2513 | conf->cluster_sync_high = 0; | |
c40f341f | 2514 | } |
1da177e4 LT |
2515 | return 0; |
2516 | } | |
2517 | ||
07d84d10 N |
2518 | if (mddev->bitmap == NULL && |
2519 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2520 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2521 | conf->fullsync == 0) { |
2522 | *skipped = 1; | |
2523 | return max_sector - sector_nr; | |
2524 | } | |
6394cca5 N |
2525 | /* before building a request, check if we can skip these blocks.. |
2526 | * This call the bitmap_start_sync doesn't actually record anything | |
2527 | */ | |
e3b9703e | 2528 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2529 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2530 | /* We can skip this block, and probably several more */ |
2531 | *skipped = 1; | |
2532 | return sync_blocks; | |
2533 | } | |
17999be4 | 2534 | |
c40f341f GR |
2535 | /* we are incrementing sector_nr below. To be safe, we check against |
2536 | * sector_nr + two times RESYNC_SECTORS | |
2537 | */ | |
2538 | ||
2539 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
2540 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
1c4588e9 | 2541 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 | 2542 | |
c2fd4c94 | 2543 | raise_barrier(conf, sector_nr); |
1da177e4 | 2544 | |
3e198f78 | 2545 | rcu_read_lock(); |
1da177e4 | 2546 | /* |
3e198f78 N |
2547 | * If we get a correctably read error during resync or recovery, |
2548 | * we might want to read from a different device. So we | |
2549 | * flag all drives that could conceivably be read from for READ, | |
2550 | * and any others (which will be non-In_sync devices) for WRITE. | |
2551 | * If a read fails, we try reading from something else for which READ | |
2552 | * is OK. | |
1da177e4 | 2553 | */ |
1da177e4 | 2554 | |
1da177e4 LT |
2555 | r1_bio->mddev = mddev; |
2556 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2557 | r1_bio->state = 0; |
1da177e4 | 2558 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2559 | |
8f19ccb2 | 2560 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2561 | struct md_rdev *rdev; |
1da177e4 | 2562 | bio = r1_bio->bios[i]; |
2aabaa65 | 2563 | bio_reset(bio); |
1da177e4 | 2564 | |
3e198f78 N |
2565 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2566 | if (rdev == NULL || | |
06f60385 | 2567 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2568 | if (i < conf->raid_disks) |
2569 | still_degraded = 1; | |
3e198f78 | 2570 | } else if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
2571 | bio->bi_rw = WRITE; |
2572 | bio->bi_end_io = end_sync_write; | |
2573 | write_targets ++; | |
3e198f78 N |
2574 | } else { |
2575 | /* may need to read from here */ | |
06f60385 N |
2576 | sector_t first_bad = MaxSector; |
2577 | int bad_sectors; | |
2578 | ||
2579 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2580 | &first_bad, &bad_sectors)) { | |
2581 | if (first_bad > sector_nr) | |
2582 | good_sectors = first_bad - sector_nr; | |
2583 | else { | |
2584 | bad_sectors -= (sector_nr - first_bad); | |
2585 | if (min_bad == 0 || | |
2586 | min_bad > bad_sectors) | |
2587 | min_bad = bad_sectors; | |
2588 | } | |
2589 | } | |
2590 | if (sector_nr < first_bad) { | |
2591 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2592 | if (wonly < 0) | |
2593 | wonly = i; | |
2594 | } else { | |
2595 | if (disk < 0) | |
2596 | disk = i; | |
2597 | } | |
2598 | bio->bi_rw = READ; | |
2599 | bio->bi_end_io = end_sync_read; | |
2600 | read_targets++; | |
d57368af AL |
2601 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2602 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2603 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2604 | /* | |
2605 | * The device is suitable for reading (InSync), | |
2606 | * but has bad block(s) here. Let's try to correct them, | |
2607 | * if we are doing resync or repair. Otherwise, leave | |
2608 | * this device alone for this sync request. | |
2609 | */ | |
2610 | bio->bi_rw = WRITE; | |
2611 | bio->bi_end_io = end_sync_write; | |
2612 | write_targets++; | |
3e198f78 | 2613 | } |
3e198f78 | 2614 | } |
06f60385 N |
2615 | if (bio->bi_end_io) { |
2616 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2617 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
06f60385 N |
2618 | bio->bi_bdev = rdev->bdev; |
2619 | bio->bi_private = r1_bio; | |
2620 | } | |
1da177e4 | 2621 | } |
3e198f78 N |
2622 | rcu_read_unlock(); |
2623 | if (disk < 0) | |
2624 | disk = wonly; | |
2625 | r1_bio->read_disk = disk; | |
191ea9b2 | 2626 | |
06f60385 N |
2627 | if (read_targets == 0 && min_bad > 0) { |
2628 | /* These sectors are bad on all InSync devices, so we | |
2629 | * need to mark them bad on all write targets | |
2630 | */ | |
2631 | int ok = 1; | |
8f19ccb2 | 2632 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2633 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2634 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2635 | ok = rdev_set_badblocks(rdev, sector_nr, |
2636 | min_bad, 0 | |
2637 | ) && ok; | |
2638 | } | |
2639 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2640 | *skipped = 1; | |
2641 | put_buf(r1_bio); | |
2642 | ||
2643 | if (!ok) { | |
2644 | /* Cannot record the badblocks, so need to | |
2645 | * abort the resync. | |
2646 | * If there are multiple read targets, could just | |
2647 | * fail the really bad ones ??? | |
2648 | */ | |
2649 | conf->recovery_disabled = mddev->recovery_disabled; | |
2650 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2651 | return 0; | |
2652 | } else | |
2653 | return min_bad; | |
2654 | ||
2655 | } | |
2656 | if (min_bad > 0 && min_bad < good_sectors) { | |
2657 | /* only resync enough to reach the next bad->good | |
2658 | * transition */ | |
2659 | good_sectors = min_bad; | |
2660 | } | |
2661 | ||
3e198f78 N |
2662 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2663 | /* extra read targets are also write targets */ | |
2664 | write_targets += read_targets-1; | |
2665 | ||
2666 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2667 | /* There is nowhere to write, so all non-sync |
2668 | * drives must be failed - so we are finished | |
2669 | */ | |
b7219ccb N |
2670 | sector_t rv; |
2671 | if (min_bad > 0) | |
2672 | max_sector = sector_nr + min_bad; | |
2673 | rv = max_sector - sector_nr; | |
57afd89f | 2674 | *skipped = 1; |
1da177e4 | 2675 | put_buf(r1_bio); |
1da177e4 LT |
2676 | return rv; |
2677 | } | |
2678 | ||
c6207277 N |
2679 | if (max_sector > mddev->resync_max) |
2680 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2681 | if (max_sector > sector_nr + good_sectors) |
2682 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2683 | nr_sectors = 0; |
289e99e8 | 2684 | sync_blocks = 0; |
1da177e4 LT |
2685 | do { |
2686 | struct page *page; | |
2687 | int len = PAGE_SIZE; | |
2688 | if (sector_nr + (len>>9) > max_sector) | |
2689 | len = (max_sector - sector_nr) << 9; | |
2690 | if (len == 0) | |
2691 | break; | |
6a806c51 N |
2692 | if (sync_blocks == 0) { |
2693 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2694 | &sync_blocks, still_degraded) && |
2695 | !conf->fullsync && | |
2696 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2697 | break; |
9e77c485 | 2698 | BUG_ON(sync_blocks < (PAGE_SIZE>>9)); |
7571ae88 | 2699 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2700 | len = sync_blocks<<9; |
ab7a30c7 | 2701 | } |
191ea9b2 | 2702 | |
8f19ccb2 | 2703 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2704 | bio = r1_bio->bios[i]; |
2705 | if (bio->bi_end_io) { | |
d11c171e | 2706 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2707 | if (bio_add_page(bio, page, len, 0) == 0) { |
2708 | /* stop here */ | |
d11c171e | 2709 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2710 | while (i > 0) { |
2711 | i--; | |
2712 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2713 | if (bio->bi_end_io==NULL) |
2714 | continue; | |
1da177e4 LT |
2715 | /* remove last page from this bio */ |
2716 | bio->bi_vcnt--; | |
4f024f37 | 2717 | bio->bi_iter.bi_size -= len; |
b7c44ed9 | 2718 | bio_clear_flag(bio, BIO_SEG_VALID); |
1da177e4 LT |
2719 | } |
2720 | goto bio_full; | |
2721 | } | |
2722 | } | |
2723 | } | |
2724 | nr_sectors += len>>9; | |
2725 | sector_nr += len>>9; | |
191ea9b2 | 2726 | sync_blocks -= (len>>9); |
1da177e4 LT |
2727 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2728 | bio_full: | |
1da177e4 LT |
2729 | r1_bio->sectors = nr_sectors; |
2730 | ||
c40f341f GR |
2731 | if (mddev_is_clustered(mddev) && |
2732 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2733 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2734 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2735 | /* Send resync message */ | |
2736 | md_cluster_ops->resync_info_update(mddev, | |
2737 | conf->cluster_sync_low, | |
2738 | conf->cluster_sync_high); | |
2739 | } | |
2740 | ||
d11c171e N |
2741 | /* For a user-requested sync, we read all readable devices and do a |
2742 | * compare | |
2743 | */ | |
2744 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2745 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2746 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2747 | bio = r1_bio->bios[i]; |
2748 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2749 | read_targets--; |
ddac7c7e | 2750 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
2751 | generic_make_request(bio); |
2752 | } | |
2753 | } | |
2754 | } else { | |
2755 | atomic_set(&r1_bio->remaining, 1); | |
2756 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2757 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 2758 | generic_make_request(bio); |
1da177e4 | 2759 | |
d11c171e | 2760 | } |
1da177e4 LT |
2761 | return nr_sectors; |
2762 | } | |
2763 | ||
fd01b88c | 2764 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2765 | { |
2766 | if (sectors) | |
2767 | return sectors; | |
2768 | ||
2769 | return mddev->dev_sectors; | |
2770 | } | |
2771 | ||
e8096360 | 2772 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2773 | { |
e8096360 | 2774 | struct r1conf *conf; |
709ae487 | 2775 | int i; |
0eaf822c | 2776 | struct raid1_info *disk; |
3cb03002 | 2777 | struct md_rdev *rdev; |
709ae487 | 2778 | int err = -ENOMEM; |
1da177e4 | 2779 | |
e8096360 | 2780 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2781 | if (!conf) |
709ae487 | 2782 | goto abort; |
1da177e4 | 2783 | |
0eaf822c | 2784 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2785 | * mddev->raid_disks * 2, |
1da177e4 LT |
2786 | GFP_KERNEL); |
2787 | if (!conf->mirrors) | |
709ae487 | 2788 | goto abort; |
1da177e4 | 2789 | |
ddaf22ab N |
2790 | conf->tmppage = alloc_page(GFP_KERNEL); |
2791 | if (!conf->tmppage) | |
709ae487 | 2792 | goto abort; |
ddaf22ab | 2793 | |
709ae487 | 2794 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2795 | if (!conf->poolinfo) |
709ae487 | 2796 | goto abort; |
8f19ccb2 | 2797 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2798 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2799 | r1bio_pool_free, | |
2800 | conf->poolinfo); | |
2801 | if (!conf->r1bio_pool) | |
709ae487 N |
2802 | goto abort; |
2803 | ||
ed9bfdf1 | 2804 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2805 | |
c19d5798 | 2806 | err = -EINVAL; |
e7e72bf6 | 2807 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2808 | rdev_for_each(rdev, mddev) { |
aba336bd | 2809 | struct request_queue *q; |
709ae487 | 2810 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2811 | if (disk_idx >= mddev->raid_disks |
2812 | || disk_idx < 0) | |
2813 | continue; | |
c19d5798 | 2814 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2815 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2816 | else |
2817 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2818 | |
c19d5798 N |
2819 | if (disk->rdev) |
2820 | goto abort; | |
1da177e4 | 2821 | disk->rdev = rdev; |
aba336bd | 2822 | q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
2823 | |
2824 | disk->head_position = 0; | |
12cee5a8 | 2825 | disk->seq_start = MaxSector; |
1da177e4 LT |
2826 | } |
2827 | conf->raid_disks = mddev->raid_disks; | |
2828 | conf->mddev = mddev; | |
1da177e4 | 2829 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 2830 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
2831 | |
2832 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2833 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2834 | |
191ea9b2 | 2835 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2836 | conf->pending_count = 0; |
d890fa2b | 2837 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2838 | |
79ef3a8a | 2839 | conf->start_next_window = MaxSector; |
2840 | conf->current_window_requests = conf->next_window_requests = 0; | |
2841 | ||
c19d5798 | 2842 | err = -EIO; |
8f19ccb2 | 2843 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2844 | |
2845 | disk = conf->mirrors + i; | |
2846 | ||
c19d5798 N |
2847 | if (i < conf->raid_disks && |
2848 | disk[conf->raid_disks].rdev) { | |
2849 | /* This slot has a replacement. */ | |
2850 | if (!disk->rdev) { | |
2851 | /* No original, just make the replacement | |
2852 | * a recovering spare | |
2853 | */ | |
2854 | disk->rdev = | |
2855 | disk[conf->raid_disks].rdev; | |
2856 | disk[conf->raid_disks].rdev = NULL; | |
2857 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2858 | /* Original is not in_sync - bad */ | |
2859 | goto abort; | |
2860 | } | |
2861 | ||
5fd6c1dc N |
2862 | if (!disk->rdev || |
2863 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2864 | disk->head_position = 0; |
4f0a5e01 JB |
2865 | if (disk->rdev && |
2866 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2867 | conf->fullsync = 1; |
be4d3280 | 2868 | } |
1da177e4 | 2869 | } |
709ae487 | 2870 | |
709ae487 | 2871 | err = -ENOMEM; |
0232605d | 2872 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
709ae487 N |
2873 | if (!conf->thread) { |
2874 | printk(KERN_ERR | |
9dd1e2fa | 2875 | "md/raid1:%s: couldn't allocate thread\n", |
709ae487 N |
2876 | mdname(mddev)); |
2877 | goto abort; | |
11ce99e6 | 2878 | } |
1da177e4 | 2879 | |
709ae487 N |
2880 | return conf; |
2881 | ||
2882 | abort: | |
2883 | if (conf) { | |
644df1a8 | 2884 | mempool_destroy(conf->r1bio_pool); |
709ae487 N |
2885 | kfree(conf->mirrors); |
2886 | safe_put_page(conf->tmppage); | |
2887 | kfree(conf->poolinfo); | |
2888 | kfree(conf); | |
2889 | } | |
2890 | return ERR_PTR(err); | |
2891 | } | |
2892 | ||
afa0f557 | 2893 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 2894 | static int raid1_run(struct mddev *mddev) |
709ae487 | 2895 | { |
e8096360 | 2896 | struct r1conf *conf; |
709ae487 | 2897 | int i; |
3cb03002 | 2898 | struct md_rdev *rdev; |
5220ea1e | 2899 | int ret; |
2ff8cc2c | 2900 | bool discard_supported = false; |
709ae487 N |
2901 | |
2902 | if (mddev->level != 1) { | |
9dd1e2fa | 2903 | printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", |
709ae487 N |
2904 | mdname(mddev), mddev->level); |
2905 | return -EIO; | |
2906 | } | |
2907 | if (mddev->reshape_position != MaxSector) { | |
9dd1e2fa | 2908 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
709ae487 N |
2909 | mdname(mddev)); |
2910 | return -EIO; | |
2911 | } | |
1da177e4 | 2912 | /* |
709ae487 N |
2913 | * copy the already verified devices into our private RAID1 |
2914 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 2915 | * should be freed in raid1_free()] |
1da177e4 | 2916 | */ |
709ae487 N |
2917 | if (mddev->private == NULL) |
2918 | conf = setup_conf(mddev); | |
2919 | else | |
2920 | conf = mddev->private; | |
1da177e4 | 2921 | |
709ae487 N |
2922 | if (IS_ERR(conf)) |
2923 | return PTR_ERR(conf); | |
1da177e4 | 2924 | |
c8dc9c65 | 2925 | if (mddev->queue) |
5026d7a9 PA |
2926 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
2927 | ||
dafb20fa | 2928 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
2929 | if (!mddev->gendisk) |
2930 | continue; | |
709ae487 N |
2931 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2932 | rdev->data_offset << 9); | |
2ff8cc2c SL |
2933 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2934 | discard_supported = true; | |
1da177e4 | 2935 | } |
191ea9b2 | 2936 | |
709ae487 N |
2937 | mddev->degraded = 0; |
2938 | for (i=0; i < conf->raid_disks; i++) | |
2939 | if (conf->mirrors[i].rdev == NULL || | |
2940 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
2941 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
2942 | mddev->degraded++; | |
2943 | ||
2944 | if (conf->raid_disks - mddev->degraded == 1) | |
2945 | mddev->recovery_cp = MaxSector; | |
2946 | ||
8c6ac868 | 2947 | if (mddev->recovery_cp != MaxSector) |
9dd1e2fa | 2948 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
8c6ac868 AN |
2949 | " -- starting background reconstruction\n", |
2950 | mdname(mddev)); | |
f72ffdd6 | 2951 | printk(KERN_INFO |
9dd1e2fa | 2952 | "md/raid1:%s: active with %d out of %d mirrors\n", |
f72ffdd6 | 2953 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 2954 | mddev->raid_disks); |
709ae487 | 2955 | |
1da177e4 LT |
2956 | /* |
2957 | * Ok, everything is just fine now | |
2958 | */ | |
709ae487 N |
2959 | mddev->thread = conf->thread; |
2960 | conf->thread = NULL; | |
2961 | mddev->private = conf; | |
2962 | ||
1f403624 | 2963 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2964 | |
1ed7242e | 2965 | if (mddev->queue) { |
2ff8cc2c SL |
2966 | if (discard_supported) |
2967 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
2968 | mddev->queue); | |
2969 | else | |
2970 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
2971 | mddev->queue); | |
1ed7242e | 2972 | } |
5220ea1e | 2973 | |
2974 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
2975 | if (ret) { |
2976 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 2977 | raid1_free(mddev, conf); |
5aa61f42 | 2978 | } |
5220ea1e | 2979 | return ret; |
1da177e4 LT |
2980 | } |
2981 | ||
afa0f557 | 2982 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 2983 | { |
afa0f557 | 2984 | struct r1conf *conf = priv; |
409c57f3 | 2985 | |
644df1a8 | 2986 | mempool_destroy(conf->r1bio_pool); |
990a8baf | 2987 | kfree(conf->mirrors); |
0fea7ed8 | 2988 | safe_put_page(conf->tmppage); |
990a8baf | 2989 | kfree(conf->poolinfo); |
1da177e4 | 2990 | kfree(conf); |
1da177e4 LT |
2991 | } |
2992 | ||
fd01b88c | 2993 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
2994 | { |
2995 | /* no resync is happening, and there is enough space | |
2996 | * on all devices, so we can resize. | |
2997 | * We need to make sure resync covers any new space. | |
2998 | * If the array is shrinking we should possibly wait until | |
2999 | * any io in the removed space completes, but it hardly seems | |
3000 | * worth it. | |
3001 | */ | |
a4a6125a N |
3002 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3003 | if (mddev->external_size && | |
3004 | mddev->array_sectors > newsize) | |
b522adcd | 3005 | return -EINVAL; |
a4a6125a N |
3006 | if (mddev->bitmap) { |
3007 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3008 | if (ret) | |
3009 | return ret; | |
3010 | } | |
3011 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 3012 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 3013 | revalidate_disk(mddev->gendisk); |
b522adcd | 3014 | if (sectors > mddev->dev_sectors && |
b098636c | 3015 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3016 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3017 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3018 | } | |
b522adcd | 3019 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3020 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3021 | return 0; |
3022 | } | |
3023 | ||
fd01b88c | 3024 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3025 | { |
3026 | /* We need to: | |
3027 | * 1/ resize the r1bio_pool | |
3028 | * 2/ resize conf->mirrors | |
3029 | * | |
3030 | * We allocate a new r1bio_pool if we can. | |
3031 | * Then raise a device barrier and wait until all IO stops. | |
3032 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3033 | * |
3034 | * At the same time, we "pack" the devices so that all the missing | |
3035 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3036 | */ |
3037 | mempool_t *newpool, *oldpool; | |
3038 | struct pool_info *newpoolinfo; | |
0eaf822c | 3039 | struct raid1_info *newmirrors; |
e8096360 | 3040 | struct r1conf *conf = mddev->private; |
63c70c4f | 3041 | int cnt, raid_disks; |
c04be0aa | 3042 | unsigned long flags; |
b5470dc5 | 3043 | int d, d2, err; |
1da177e4 | 3044 | |
63c70c4f | 3045 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3046 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3047 | mddev->layout != mddev->new_layout || |
3048 | mddev->level != mddev->new_level) { | |
664e7c41 | 3049 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3050 | mddev->new_layout = mddev->layout; |
3051 | mddev->new_level = mddev->level; | |
3052 | return -EINVAL; | |
3053 | } | |
3054 | ||
28c1b9fd GR |
3055 | if (!mddev_is_clustered(mddev)) { |
3056 | err = md_allow_write(mddev); | |
3057 | if (err) | |
3058 | return err; | |
3059 | } | |
2a2275d6 | 3060 | |
63c70c4f N |
3061 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3062 | ||
6ea9c07c N |
3063 | if (raid_disks < conf->raid_disks) { |
3064 | cnt=0; | |
3065 | for (d= 0; d < conf->raid_disks; d++) | |
3066 | if (conf->mirrors[d].rdev) | |
3067 | cnt++; | |
3068 | if (cnt > raid_disks) | |
1da177e4 | 3069 | return -EBUSY; |
6ea9c07c | 3070 | } |
1da177e4 LT |
3071 | |
3072 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3073 | if (!newpoolinfo) | |
3074 | return -ENOMEM; | |
3075 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3076 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3077 | |
3078 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3079 | r1bio_pool_free, newpoolinfo); | |
3080 | if (!newpool) { | |
3081 | kfree(newpoolinfo); | |
3082 | return -ENOMEM; | |
3083 | } | |
0eaf822c | 3084 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3085 | GFP_KERNEL); |
1da177e4 LT |
3086 | if (!newmirrors) { |
3087 | kfree(newpoolinfo); | |
3088 | mempool_destroy(newpool); | |
3089 | return -ENOMEM; | |
3090 | } | |
1da177e4 | 3091 | |
e2d59925 | 3092 | freeze_array(conf, 0); |
1da177e4 LT |
3093 | |
3094 | /* ok, everything is stopped */ | |
3095 | oldpool = conf->r1bio_pool; | |
3096 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3097 | |
a88aa786 | 3098 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3099 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3100 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3101 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3102 | rdev->raid_disk = d2; |
36fad858 NK |
3103 | sysfs_unlink_rdev(mddev, rdev); |
3104 | if (sysfs_link_rdev(mddev, rdev)) | |
a88aa786 | 3105 | printk(KERN_WARNING |
36fad858 NK |
3106 | "md/raid1:%s: cannot register rd%d\n", |
3107 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3108 | } |
a88aa786 N |
3109 | if (rdev) |
3110 | newmirrors[d2++].rdev = rdev; | |
3111 | } | |
1da177e4 LT |
3112 | kfree(conf->mirrors); |
3113 | conf->mirrors = newmirrors; | |
3114 | kfree(conf->poolinfo); | |
3115 | conf->poolinfo = newpoolinfo; | |
3116 | ||
c04be0aa | 3117 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3118 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3119 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3120 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3121 | mddev->delta_disks = 0; |
1da177e4 | 3122 | |
e2d59925 | 3123 | unfreeze_array(conf); |
1da177e4 | 3124 | |
985ca973 | 3125 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3126 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3127 | md_wakeup_thread(mddev->thread); | |
3128 | ||
3129 | mempool_destroy(oldpool); | |
3130 | return 0; | |
3131 | } | |
3132 | ||
fd01b88c | 3133 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3134 | { |
e8096360 | 3135 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3136 | |
3137 | switch(state) { | |
6eef4b21 N |
3138 | case 2: /* wake for suspend */ |
3139 | wake_up(&conf->wait_barrier); | |
3140 | break; | |
9e6603da | 3141 | case 1: |
07169fd4 | 3142 | freeze_array(conf, 0); |
36fa3063 | 3143 | break; |
9e6603da | 3144 | case 0: |
07169fd4 | 3145 | unfreeze_array(conf); |
36fa3063 N |
3146 | break; |
3147 | } | |
36fa3063 N |
3148 | } |
3149 | ||
fd01b88c | 3150 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3151 | { |
3152 | /* raid1 can take over: | |
3153 | * raid5 with 2 devices, any layout or chunk size | |
3154 | */ | |
3155 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3156 | struct r1conf *conf; |
709ae487 N |
3157 | mddev->new_level = 1; |
3158 | mddev->new_layout = 0; | |
3159 | mddev->new_chunk_sectors = 0; | |
3160 | conf = setup_conf(mddev); | |
3161 | if (!IS_ERR(conf)) | |
07169fd4 | 3162 | /* Array must appear to be quiesced */ |
3163 | conf->array_frozen = 1; | |
709ae487 N |
3164 | return conf; |
3165 | } | |
3166 | return ERR_PTR(-EINVAL); | |
3167 | } | |
1da177e4 | 3168 | |
84fc4b56 | 3169 | static struct md_personality raid1_personality = |
1da177e4 LT |
3170 | { |
3171 | .name = "raid1", | |
2604b703 | 3172 | .level = 1, |
1da177e4 | 3173 | .owner = THIS_MODULE, |
849674e4 SL |
3174 | .make_request = raid1_make_request, |
3175 | .run = raid1_run, | |
afa0f557 | 3176 | .free = raid1_free, |
849674e4 SL |
3177 | .status = raid1_status, |
3178 | .error_handler = raid1_error, | |
1da177e4 LT |
3179 | .hot_add_disk = raid1_add_disk, |
3180 | .hot_remove_disk= raid1_remove_disk, | |
3181 | .spare_active = raid1_spare_active, | |
849674e4 | 3182 | .sync_request = raid1_sync_request, |
1da177e4 | 3183 | .resize = raid1_resize, |
80c3a6ce | 3184 | .size = raid1_size, |
63c70c4f | 3185 | .check_reshape = raid1_reshape, |
36fa3063 | 3186 | .quiesce = raid1_quiesce, |
709ae487 | 3187 | .takeover = raid1_takeover, |
5c675f83 | 3188 | .congested = raid1_congested, |
1da177e4 LT |
3189 | }; |
3190 | ||
3191 | static int __init raid_init(void) | |
3192 | { | |
2604b703 | 3193 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3194 | } |
3195 | ||
3196 | static void raid_exit(void) | |
3197 | { | |
2604b703 | 3198 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3199 | } |
3200 | ||
3201 | module_init(raid_init); | |
3202 | module_exit(raid_exit); | |
3203 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3204 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3205 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3206 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3207 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3208 | |
3209 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |