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; |
e8096360 | 322 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 323 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 324 | |
1da177e4 LT |
325 | /* |
326 | * this branch is our 'one mirror IO has finished' event handler: | |
327 | */ | |
e5872d58 | 328 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 329 | |
dd00a99e N |
330 | if (uptodate) |
331 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
332 | else { | |
333 | /* If all other devices have failed, we want to return | |
334 | * the error upwards rather than fail the last device. | |
335 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 336 | */ |
dd00a99e N |
337 | unsigned long flags; |
338 | spin_lock_irqsave(&conf->device_lock, flags); | |
339 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
340 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 341 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
342 | uptodate = 1; |
343 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
344 | } | |
1da177e4 | 345 | |
7ad4d4a6 | 346 | if (uptodate) { |
1da177e4 | 347 | raid_end_bio_io(r1_bio); |
e5872d58 | 348 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 349 | } else { |
1da177e4 LT |
350 | /* |
351 | * oops, read error: | |
352 | */ | |
353 | char b[BDEVNAME_SIZE]; | |
8bda470e CD |
354 | printk_ratelimited( |
355 | KERN_ERR "md/raid1:%s: %s: " | |
356 | "rescheduling sector %llu\n", | |
357 | mdname(conf->mddev), | |
e5872d58 | 358 | bdevname(rdev->bdev, |
8bda470e CD |
359 | b), |
360 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 361 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 362 | reschedule_retry(r1_bio); |
7ad4d4a6 | 363 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 364 | } |
1da177e4 LT |
365 | } |
366 | ||
9f2c9d12 | 367 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
368 | { |
369 | /* it really is the end of this request */ | |
370 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
371 | /* free extra copy of the data pages */ | |
372 | int i = r1_bio->behind_page_count; | |
373 | while (i--) | |
374 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
375 | kfree(r1_bio->behind_bvecs); | |
376 | r1_bio->behind_bvecs = NULL; | |
377 | } | |
378 | /* clear the bitmap if all writes complete successfully */ | |
379 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
380 | r1_bio->sectors, | |
381 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
382 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
383 | md_write_end(r1_bio->mddev); | |
384 | } | |
385 | ||
9f2c9d12 | 386 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 387 | { |
cd5ff9a1 N |
388 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
389 | return; | |
390 | ||
391 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
392 | reschedule_retry(r1_bio); | |
393 | else { | |
394 | close_write(r1_bio); | |
4367af55 N |
395 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
396 | reschedule_retry(r1_bio); | |
397 | else | |
398 | raid_end_bio_io(r1_bio); | |
4e78064f N |
399 | } |
400 | } | |
401 | ||
4246a0b6 | 402 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 403 | { |
9f2c9d12 | 404 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 405 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 406 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 407 | struct bio *to_put = NULL; |
e5872d58 N |
408 | int mirror = find_bio_disk(r1_bio, bio); |
409 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
1da177e4 | 410 | |
e9c7469b TH |
411 | /* |
412 | * 'one mirror IO has finished' event handler: | |
413 | */ | |
4246a0b6 | 414 | if (bio->bi_error) { |
e5872d58 N |
415 | set_bit(WriteErrorSeen, &rdev->flags); |
416 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
417 | set_bit(MD_RECOVERY_NEEDED, & |
418 | conf->mddev->recovery); | |
419 | ||
cd5ff9a1 | 420 | set_bit(R1BIO_WriteError, &r1_bio->state); |
4367af55 | 421 | } else { |
1da177e4 | 422 | /* |
e9c7469b TH |
423 | * Set R1BIO_Uptodate in our master bio, so that we |
424 | * will return a good error code for to the higher | |
425 | * levels even if IO on some other mirrored buffer | |
426 | * fails. | |
427 | * | |
428 | * The 'master' represents the composite IO operation | |
429 | * to user-side. So if something waits for IO, then it | |
430 | * will wait for the 'master' bio. | |
1da177e4 | 431 | */ |
4367af55 N |
432 | sector_t first_bad; |
433 | int bad_sectors; | |
434 | ||
cd5ff9a1 N |
435 | r1_bio->bios[mirror] = NULL; |
436 | to_put = bio; | |
3056e3ae AL |
437 | /* |
438 | * Do not set R1BIO_Uptodate if the current device is | |
439 | * rebuilding or Faulty. This is because we cannot use | |
440 | * such device for properly reading the data back (we could | |
441 | * potentially use it, if the current write would have felt | |
442 | * before rdev->recovery_offset, but for simplicity we don't | |
443 | * check this here. | |
444 | */ | |
e5872d58 N |
445 | if (test_bit(In_sync, &rdev->flags) && |
446 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 447 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 448 | |
4367af55 | 449 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 450 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
4367af55 N |
451 | &first_bad, &bad_sectors)) { |
452 | r1_bio->bios[mirror] = IO_MADE_GOOD; | |
453 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
454 | } | |
455 | } | |
456 | ||
e9c7469b | 457 | if (behind) { |
e5872d58 | 458 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
459 | atomic_dec(&r1_bio->behind_remaining); |
460 | ||
461 | /* | |
462 | * In behind mode, we ACK the master bio once the I/O | |
463 | * has safely reached all non-writemostly | |
464 | * disks. Setting the Returned bit ensures that this | |
465 | * gets done only once -- we don't ever want to return | |
466 | * -EIO here, instead we'll wait | |
467 | */ | |
468 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
469 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
470 | /* Maybe we can return now */ | |
471 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
472 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
473 | pr_debug("raid1: behind end write sectors" |
474 | " %llu-%llu\n", | |
4f024f37 KO |
475 | (unsigned long long) mbio->bi_iter.bi_sector, |
476 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 477 | call_bio_endio(r1_bio); |
4b6d287f N |
478 | } |
479 | } | |
480 | } | |
4367af55 | 481 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 482 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 483 | |
1da177e4 | 484 | /* |
1da177e4 LT |
485 | * Let's see if all mirrored write operations have finished |
486 | * already. | |
487 | */ | |
af6d7b76 | 488 | r1_bio_write_done(r1_bio); |
c70810b3 | 489 | |
04b857f7 N |
490 | if (to_put) |
491 | bio_put(to_put); | |
1da177e4 LT |
492 | } |
493 | ||
1da177e4 LT |
494 | /* |
495 | * This routine returns the disk from which the requested read should | |
496 | * be done. There is a per-array 'next expected sequential IO' sector | |
497 | * number - if this matches on the next IO then we use the last disk. | |
498 | * There is also a per-disk 'last know head position' sector that is | |
499 | * maintained from IRQ contexts, both the normal and the resync IO | |
500 | * completion handlers update this position correctly. If there is no | |
501 | * perfect sequential match then we pick the disk whose head is closest. | |
502 | * | |
503 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
504 | * because position is mirror, not device based. | |
505 | * | |
506 | * The rdev for the device selected will have nr_pending incremented. | |
507 | */ | |
e8096360 | 508 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 509 | { |
af3a2cd6 | 510 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
511 | int sectors; |
512 | int best_good_sectors; | |
9dedf603 SL |
513 | int best_disk, best_dist_disk, best_pending_disk; |
514 | int has_nonrot_disk; | |
be4d3280 | 515 | int disk; |
76073054 | 516 | sector_t best_dist; |
9dedf603 | 517 | unsigned int min_pending; |
3cb03002 | 518 | struct md_rdev *rdev; |
f3ac8bf7 | 519 | int choose_first; |
12cee5a8 | 520 | int choose_next_idle; |
1da177e4 LT |
521 | |
522 | rcu_read_lock(); | |
523 | /* | |
8ddf9efe | 524 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
525 | * device if no resync is going on, or below the resync window. |
526 | * We take the first readable disk when above the resync window. | |
527 | */ | |
528 | retry: | |
d2eb35ac | 529 | sectors = r1_bio->sectors; |
76073054 | 530 | best_disk = -1; |
9dedf603 | 531 | best_dist_disk = -1; |
76073054 | 532 | best_dist = MaxSector; |
9dedf603 SL |
533 | best_pending_disk = -1; |
534 | min_pending = UINT_MAX; | |
d2eb35ac | 535 | best_good_sectors = 0; |
9dedf603 | 536 | has_nonrot_disk = 0; |
12cee5a8 | 537 | choose_next_idle = 0; |
d2eb35ac | 538 | |
7d49ffcf GR |
539 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
540 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 541 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
542 | this_sector + sectors))) |
543 | choose_first = 1; | |
544 | else | |
545 | choose_first = 0; | |
1da177e4 | 546 | |
be4d3280 | 547 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 548 | sector_t dist; |
d2eb35ac N |
549 | sector_t first_bad; |
550 | int bad_sectors; | |
9dedf603 | 551 | unsigned int pending; |
12cee5a8 | 552 | bool nonrot; |
d2eb35ac | 553 | |
f3ac8bf7 N |
554 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
555 | if (r1_bio->bios[disk] == IO_BLOCKED | |
556 | || rdev == NULL | |
76073054 | 557 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 558 | continue; |
76073054 N |
559 | if (!test_bit(In_sync, &rdev->flags) && |
560 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 561 | continue; |
76073054 N |
562 | if (test_bit(WriteMostly, &rdev->flags)) { |
563 | /* Don't balance among write-mostly, just | |
564 | * use the first as a last resort */ | |
d1901ef0 | 565 | if (best_dist_disk < 0) { |
307729c8 N |
566 | if (is_badblock(rdev, this_sector, sectors, |
567 | &first_bad, &bad_sectors)) { | |
816b0acf | 568 | if (first_bad <= this_sector) |
307729c8 N |
569 | /* Cannot use this */ |
570 | continue; | |
571 | best_good_sectors = first_bad - this_sector; | |
572 | } else | |
573 | best_good_sectors = sectors; | |
d1901ef0 TH |
574 | best_dist_disk = disk; |
575 | best_pending_disk = disk; | |
307729c8 | 576 | } |
76073054 N |
577 | continue; |
578 | } | |
579 | /* This is a reasonable device to use. It might | |
580 | * even be best. | |
581 | */ | |
d2eb35ac N |
582 | if (is_badblock(rdev, this_sector, sectors, |
583 | &first_bad, &bad_sectors)) { | |
584 | if (best_dist < MaxSector) | |
585 | /* already have a better device */ | |
586 | continue; | |
587 | if (first_bad <= this_sector) { | |
588 | /* cannot read here. If this is the 'primary' | |
589 | * device, then we must not read beyond | |
590 | * bad_sectors from another device.. | |
591 | */ | |
592 | bad_sectors -= (this_sector - first_bad); | |
593 | if (choose_first && sectors > bad_sectors) | |
594 | sectors = bad_sectors; | |
595 | if (best_good_sectors > sectors) | |
596 | best_good_sectors = sectors; | |
597 | ||
598 | } else { | |
599 | sector_t good_sectors = first_bad - this_sector; | |
600 | if (good_sectors > best_good_sectors) { | |
601 | best_good_sectors = good_sectors; | |
602 | best_disk = disk; | |
603 | } | |
604 | if (choose_first) | |
605 | break; | |
606 | } | |
607 | continue; | |
608 | } else | |
609 | best_good_sectors = sectors; | |
610 | ||
12cee5a8 SL |
611 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
612 | has_nonrot_disk |= nonrot; | |
9dedf603 | 613 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 614 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 615 | if (choose_first) { |
76073054 | 616 | best_disk = disk; |
1da177e4 LT |
617 | break; |
618 | } | |
12cee5a8 SL |
619 | /* Don't change to another disk for sequential reads */ |
620 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
621 | || dist == 0) { | |
622 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
623 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
624 | ||
625 | best_disk = disk; | |
626 | /* | |
627 | * If buffered sequential IO size exceeds optimal | |
628 | * iosize, check if there is idle disk. If yes, choose | |
629 | * the idle disk. read_balance could already choose an | |
630 | * idle disk before noticing it's a sequential IO in | |
631 | * this disk. This doesn't matter because this disk | |
632 | * will idle, next time it will be utilized after the | |
633 | * first disk has IO size exceeds optimal iosize. In | |
634 | * this way, iosize of the first disk will be optimal | |
635 | * iosize at least. iosize of the second disk might be | |
636 | * small, but not a big deal since when the second disk | |
637 | * starts IO, the first disk is likely still busy. | |
638 | */ | |
639 | if (nonrot && opt_iosize > 0 && | |
640 | mirror->seq_start != MaxSector && | |
641 | mirror->next_seq_sect > opt_iosize && | |
642 | mirror->next_seq_sect - opt_iosize >= | |
643 | mirror->seq_start) { | |
644 | choose_next_idle = 1; | |
645 | continue; | |
646 | } | |
647 | break; | |
648 | } | |
649 | /* If device is idle, use it */ | |
650 | if (pending == 0) { | |
651 | best_disk = disk; | |
652 | break; | |
653 | } | |
654 | ||
655 | if (choose_next_idle) | |
656 | continue; | |
9dedf603 SL |
657 | |
658 | if (min_pending > pending) { | |
659 | min_pending = pending; | |
660 | best_pending_disk = disk; | |
661 | } | |
662 | ||
76073054 N |
663 | if (dist < best_dist) { |
664 | best_dist = dist; | |
9dedf603 | 665 | best_dist_disk = disk; |
1da177e4 | 666 | } |
f3ac8bf7 | 667 | } |
1da177e4 | 668 | |
9dedf603 SL |
669 | /* |
670 | * If all disks are rotational, choose the closest disk. If any disk is | |
671 | * non-rotational, choose the disk with less pending request even the | |
672 | * disk is rotational, which might/might not be optimal for raids with | |
673 | * mixed ratation/non-rotational disks depending on workload. | |
674 | */ | |
675 | if (best_disk == -1) { | |
676 | if (has_nonrot_disk) | |
677 | best_disk = best_pending_disk; | |
678 | else | |
679 | best_disk = best_dist_disk; | |
680 | } | |
681 | ||
76073054 N |
682 | if (best_disk >= 0) { |
683 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
684 | if (!rdev) |
685 | goto retry; | |
686 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 687 | sectors = best_good_sectors; |
12cee5a8 SL |
688 | |
689 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
690 | conf->mirrors[best_disk].seq_start = this_sector; | |
691 | ||
be4d3280 | 692 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
693 | } |
694 | rcu_read_unlock(); | |
d2eb35ac | 695 | *max_sectors = sectors; |
1da177e4 | 696 | |
76073054 | 697 | return best_disk; |
1da177e4 LT |
698 | } |
699 | ||
5c675f83 | 700 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 701 | { |
e8096360 | 702 | struct r1conf *conf = mddev->private; |
0d129228 N |
703 | int i, ret = 0; |
704 | ||
4452226e | 705 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
706 | conf->pending_count >= max_queued_requests) |
707 | return 1; | |
708 | ||
0d129228 | 709 | rcu_read_lock(); |
f53e29fc | 710 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 711 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 712 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 713 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 714 | |
1ed7242e JB |
715 | BUG_ON(!q); |
716 | ||
0d129228 N |
717 | /* Note the '|| 1' - when read_balance prefers |
718 | * non-congested targets, it can be removed | |
719 | */ | |
4452226e | 720 | if ((bits & (1 << WB_async_congested)) || 1) |
0d129228 N |
721 | ret |= bdi_congested(&q->backing_dev_info, bits); |
722 | else | |
723 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
724 | } | |
725 | } | |
726 | rcu_read_unlock(); | |
727 | return ret; | |
728 | } | |
0d129228 | 729 | |
e8096360 | 730 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
731 | { |
732 | /* Any writes that have been queued but are awaiting | |
733 | * bitmap updates get flushed here. | |
a35e63ef | 734 | */ |
a35e63ef N |
735 | spin_lock_irq(&conf->device_lock); |
736 | ||
737 | if (conf->pending_bio_list.head) { | |
738 | struct bio *bio; | |
739 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 740 | conf->pending_count = 0; |
a35e63ef N |
741 | spin_unlock_irq(&conf->device_lock); |
742 | /* flush any pending bitmap writes to | |
743 | * disk before proceeding w/ I/O */ | |
744 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 745 | wake_up(&conf->wait_barrier); |
a35e63ef N |
746 | |
747 | while (bio) { /* submit pending writes */ | |
748 | struct bio *next = bio->bi_next; | |
749 | bio->bi_next = NULL; | |
2ff8cc2c SL |
750 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
751 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
752 | /* Just ignore it */ | |
4246a0b6 | 753 | bio_endio(bio); |
2ff8cc2c SL |
754 | else |
755 | generic_make_request(bio); | |
a35e63ef N |
756 | bio = next; |
757 | } | |
a35e63ef N |
758 | } else |
759 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
760 | } |
761 | ||
17999be4 N |
762 | /* Barriers.... |
763 | * Sometimes we need to suspend IO while we do something else, | |
764 | * either some resync/recovery, or reconfigure the array. | |
765 | * To do this we raise a 'barrier'. | |
766 | * The 'barrier' is a counter that can be raised multiple times | |
767 | * to count how many activities are happening which preclude | |
768 | * normal IO. | |
769 | * We can only raise the barrier if there is no pending IO. | |
770 | * i.e. if nr_pending == 0. | |
771 | * We choose only to raise the barrier if no-one is waiting for the | |
772 | * barrier to go down. This means that as soon as an IO request | |
773 | * is ready, no other operations which require a barrier will start | |
774 | * until the IO request has had a chance. | |
775 | * | |
776 | * So: regular IO calls 'wait_barrier'. When that returns there | |
777 | * is no backgroup IO happening, It must arrange to call | |
778 | * allow_barrier when it has finished its IO. | |
779 | * backgroup IO calls must call raise_barrier. Once that returns | |
780 | * there is no normal IO happeing. It must arrange to call | |
781 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 782 | */ |
c2fd4c94 | 783 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 LT |
784 | { |
785 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
786 | |
787 | /* Wait until no block IO is waiting */ | |
788 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
eed8c02e | 789 | conf->resync_lock); |
17999be4 N |
790 | |
791 | /* block any new IO from starting */ | |
792 | conf->barrier++; | |
c2fd4c94 | 793 | conf->next_resync = sector_nr; |
17999be4 | 794 | |
79ef3a8a | 795 | /* For these conditions we must wait: |
796 | * A: while the array is in frozen state | |
797 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach | |
798 | * the max count which allowed. | |
799 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning | |
800 | * next resync will reach to the window which normal bios are | |
801 | * handling. | |
2f73d3c5 | 802 | * D: while there are any active requests in the current window. |
79ef3a8a | 803 | */ |
17999be4 | 804 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 805 | !conf->array_frozen && |
79ef3a8a | 806 | conf->barrier < RESYNC_DEPTH && |
2f73d3c5 | 807 | conf->current_window_requests == 0 && |
79ef3a8a | 808 | (conf->start_next_window >= |
809 | conf->next_resync + RESYNC_SECTORS), | |
eed8c02e | 810 | conf->resync_lock); |
17999be4 | 811 | |
34e97f17 | 812 | conf->nr_pending++; |
17999be4 N |
813 | spin_unlock_irq(&conf->resync_lock); |
814 | } | |
815 | ||
e8096360 | 816 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
817 | { |
818 | unsigned long flags; | |
709ae487 | 819 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
820 | spin_lock_irqsave(&conf->resync_lock, flags); |
821 | conf->barrier--; | |
34e97f17 | 822 | conf->nr_pending--; |
17999be4 N |
823 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
824 | wake_up(&conf->wait_barrier); | |
825 | } | |
826 | ||
79ef3a8a | 827 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
17999be4 | 828 | { |
79ef3a8a | 829 | bool wait = false; |
830 | ||
831 | if (conf->array_frozen || !bio) | |
832 | wait = true; | |
833 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { | |
23554960 N |
834 | if ((conf->mddev->curr_resync_completed |
835 | >= bio_end_sector(bio)) || | |
836 | (conf->next_resync + NEXT_NORMALIO_DISTANCE | |
837 | <= bio->bi_iter.bi_sector)) | |
79ef3a8a | 838 | wait = false; |
839 | else | |
840 | wait = true; | |
841 | } | |
842 | ||
843 | return wait; | |
844 | } | |
845 | ||
846 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) | |
847 | { | |
848 | sector_t sector = 0; | |
849 | ||
17999be4 | 850 | spin_lock_irq(&conf->resync_lock); |
79ef3a8a | 851 | if (need_to_wait_for_sync(conf, bio)) { |
17999be4 | 852 | conf->nr_waiting++; |
d6b42dcb N |
853 | /* Wait for the barrier to drop. |
854 | * However if there are already pending | |
855 | * requests (preventing the barrier from | |
856 | * rising completely), and the | |
5965b642 | 857 | * per-process bio queue isn't empty, |
d6b42dcb | 858 | * then don't wait, as we need to empty |
5965b642 N |
859 | * that queue to allow conf->start_next_window |
860 | * to increase. | |
d6b42dcb N |
861 | */ |
862 | wait_event_lock_irq(conf->wait_barrier, | |
b364e3d0 | 863 | !conf->array_frozen && |
864 | (!conf->barrier || | |
5965b642 N |
865 | ((conf->start_next_window < |
866 | conf->next_resync + RESYNC_SECTORS) && | |
867 | current->bio_list && | |
868 | !bio_list_empty(current->bio_list))), | |
eed8c02e | 869 | conf->resync_lock); |
17999be4 | 870 | conf->nr_waiting--; |
1da177e4 | 871 | } |
79ef3a8a | 872 | |
873 | if (bio && bio_data_dir(bio) == WRITE) { | |
e8ff8bf0 | 874 | if (bio->bi_iter.bi_sector >= conf->next_resync) { |
79ef3a8a | 875 | if (conf->start_next_window == MaxSector) |
876 | conf->start_next_window = | |
877 | conf->next_resync + | |
878 | NEXT_NORMALIO_DISTANCE; | |
879 | ||
880 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) | |
4f024f37 | 881 | <= bio->bi_iter.bi_sector) |
79ef3a8a | 882 | conf->next_window_requests++; |
883 | else | |
884 | conf->current_window_requests++; | |
79ef3a8a | 885 | sector = conf->start_next_window; |
41a336e0 | 886 | } |
79ef3a8a | 887 | } |
888 | ||
17999be4 | 889 | conf->nr_pending++; |
1da177e4 | 890 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 891 | return sector; |
1da177e4 LT |
892 | } |
893 | ||
79ef3a8a | 894 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
895 | sector_t bi_sector) | |
17999be4 N |
896 | { |
897 | unsigned long flags; | |
79ef3a8a | 898 | |
17999be4 N |
899 | spin_lock_irqsave(&conf->resync_lock, flags); |
900 | conf->nr_pending--; | |
79ef3a8a | 901 | if (start_next_window) { |
902 | if (start_next_window == conf->start_next_window) { | |
903 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE | |
904 | <= bi_sector) | |
905 | conf->next_window_requests--; | |
906 | else | |
907 | conf->current_window_requests--; | |
908 | } else | |
909 | conf->current_window_requests--; | |
910 | ||
911 | if (!conf->current_window_requests) { | |
912 | if (conf->next_window_requests) { | |
913 | conf->current_window_requests = | |
914 | conf->next_window_requests; | |
915 | conf->next_window_requests = 0; | |
916 | conf->start_next_window += | |
917 | NEXT_NORMALIO_DISTANCE; | |
918 | } else | |
919 | conf->start_next_window = MaxSector; | |
920 | } | |
921 | } | |
17999be4 N |
922 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
923 | wake_up(&conf->wait_barrier); | |
924 | } | |
925 | ||
e2d59925 | 926 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab N |
927 | { |
928 | /* stop syncio and normal IO and wait for everything to | |
929 | * go quite. | |
b364e3d0 | 930 | * We wait until nr_pending match nr_queued+extra |
1c830532 N |
931 | * This is called in the context of one normal IO request |
932 | * that has failed. Thus any sync request that might be pending | |
933 | * will be blocked by nr_pending, and we need to wait for | |
934 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 935 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
936 | * must match the number of pending IOs (nr_pending) before |
937 | * we continue. | |
ddaf22ab N |
938 | */ |
939 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 940 | conf->array_frozen = 1; |
eed8c02e | 941 | wait_event_lock_irq_cmd(conf->wait_barrier, |
e2d59925 | 942 | conf->nr_pending == conf->nr_queued+extra, |
eed8c02e LC |
943 | conf->resync_lock, |
944 | flush_pending_writes(conf)); | |
ddaf22ab N |
945 | spin_unlock_irq(&conf->resync_lock); |
946 | } | |
e8096360 | 947 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
948 | { |
949 | /* reverse the effect of the freeze */ | |
950 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 951 | conf->array_frozen = 0; |
ddaf22ab N |
952 | wake_up(&conf->wait_barrier); |
953 | spin_unlock_irq(&conf->resync_lock); | |
954 | } | |
955 | ||
f72ffdd6 | 956 | /* duplicate the data pages for behind I/O |
4e78064f | 957 | */ |
9f2c9d12 | 958 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
959 | { |
960 | int i; | |
961 | struct bio_vec *bvec; | |
2ca68f5e | 962 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 963 | GFP_NOIO); |
2ca68f5e | 964 | if (unlikely(!bvecs)) |
af6d7b76 | 965 | return; |
4b6d287f | 966 | |
cb34e057 | 967 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
968 | bvecs[i] = *bvec; |
969 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
970 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 971 | goto do_sync_io; |
2ca68f5e N |
972 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
973 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
974 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
975 | kunmap(bvec->bv_page); |
976 | } | |
2ca68f5e | 977 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
978 | r1_bio->behind_page_count = bio->bi_vcnt; |
979 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
980 | return; | |
4b6d287f N |
981 | |
982 | do_sync_io: | |
af6d7b76 | 983 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
984 | if (bvecs[i].bv_page) |
985 | put_page(bvecs[i].bv_page); | |
986 | kfree(bvecs); | |
4f024f37 KO |
987 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
988 | bio->bi_iter.bi_size); | |
4b6d287f N |
989 | } |
990 | ||
f54a9d0e N |
991 | struct raid1_plug_cb { |
992 | struct blk_plug_cb cb; | |
993 | struct bio_list pending; | |
994 | int pending_cnt; | |
995 | }; | |
996 | ||
997 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
998 | { | |
999 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1000 | cb); | |
1001 | struct mddev *mddev = plug->cb.data; | |
1002 | struct r1conf *conf = mddev->private; | |
1003 | struct bio *bio; | |
1004 | ||
874807a8 | 1005 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1006 | spin_lock_irq(&conf->device_lock); |
1007 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1008 | conf->pending_count += plug->pending_cnt; | |
1009 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1010 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1011 | md_wakeup_thread(mddev->thread); |
1012 | kfree(plug); | |
1013 | return; | |
1014 | } | |
1015 | ||
1016 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1017 | bio = bio_list_get(&plug->pending); | |
1018 | bitmap_unplug(mddev->bitmap); | |
1019 | wake_up(&conf->wait_barrier); | |
1020 | ||
1021 | while (bio) { /* submit pending writes */ | |
1022 | struct bio *next = bio->bi_next; | |
1023 | bio->bi_next = NULL; | |
32f9f570 SL |
1024 | if (unlikely((bio->bi_rw & REQ_DISCARD) && |
1025 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
1026 | /* Just ignore it */ | |
4246a0b6 | 1027 | bio_endio(bio); |
32f9f570 SL |
1028 | else |
1029 | generic_make_request(bio); | |
f54a9d0e N |
1030 | bio = next; |
1031 | } | |
1032 | kfree(plug); | |
1033 | } | |
1034 | ||
849674e4 | 1035 | static void raid1_make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 1036 | { |
e8096360 | 1037 | struct r1conf *conf = mddev->private; |
0eaf822c | 1038 | struct raid1_info *mirror; |
9f2c9d12 | 1039 | struct r1bio *r1_bio; |
1da177e4 | 1040 | struct bio *read_bio; |
1f68f0c4 | 1041 | int i, disks; |
84255d10 | 1042 | struct bitmap *bitmap; |
191ea9b2 | 1043 | unsigned long flags; |
a362357b | 1044 | const int rw = bio_data_dir(bio); |
2c7d46ec | 1045 | const unsigned long do_sync = (bio->bi_rw & REQ_SYNC); |
e9c7469b | 1046 | const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA)); |
2ff8cc2c SL |
1047 | const unsigned long do_discard = (bio->bi_rw |
1048 | & (REQ_DISCARD | REQ_SECURE)); | |
c8dc9c65 | 1049 | const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME); |
3cb03002 | 1050 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1051 | struct blk_plug_cb *cb; |
1052 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
1053 | int first_clone; |
1054 | int sectors_handled; | |
1055 | int max_sectors; | |
79ef3a8a | 1056 | sector_t start_next_window; |
191ea9b2 | 1057 | |
1da177e4 LT |
1058 | /* |
1059 | * Register the new request and wait if the reconstruction | |
1060 | * thread has put up a bar for new requests. | |
1061 | * Continue immediately if no resync is active currently. | |
1062 | */ | |
62de608d | 1063 | |
3d310eb7 N |
1064 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1065 | ||
6eef4b21 | 1066 | if (bio_data_dir(bio) == WRITE && |
589a1c49 GR |
1067 | ((bio_end_sector(bio) > mddev->suspend_lo && |
1068 | bio->bi_iter.bi_sector < mddev->suspend_hi) || | |
1069 | (mddev_is_clustered(mddev) && | |
90382ed9 GR |
1070 | md_cluster_ops->area_resyncing(mddev, WRITE, |
1071 | bio->bi_iter.bi_sector, bio_end_sector(bio))))) { | |
6eef4b21 N |
1072 | /* As the suspend_* range is controlled by |
1073 | * userspace, we want an interruptible | |
1074 | * wait. | |
1075 | */ | |
1076 | DEFINE_WAIT(w); | |
1077 | for (;;) { | |
1078 | flush_signals(current); | |
1079 | prepare_to_wait(&conf->wait_barrier, | |
1080 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1081 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
589a1c49 GR |
1082 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
1083 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1084 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
589a1c49 | 1085 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) |
6eef4b21 N |
1086 | break; |
1087 | schedule(); | |
1088 | } | |
1089 | finish_wait(&conf->wait_barrier, &w); | |
1090 | } | |
62de608d | 1091 | |
79ef3a8a | 1092 | start_next_window = wait_barrier(conf, bio); |
1da177e4 | 1093 | |
84255d10 N |
1094 | bitmap = mddev->bitmap; |
1095 | ||
1da177e4 LT |
1096 | /* |
1097 | * make_request() can abort the operation when READA is being | |
1098 | * used and no empty request is available. | |
1099 | * | |
1100 | */ | |
1101 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1102 | ||
1103 | r1_bio->master_bio = bio; | |
aa8b57aa | 1104 | r1_bio->sectors = bio_sectors(bio); |
191ea9b2 | 1105 | r1_bio->state = 0; |
1da177e4 | 1106 | r1_bio->mddev = mddev; |
4f024f37 | 1107 | r1_bio->sector = bio->bi_iter.bi_sector; |
1da177e4 | 1108 | |
d2eb35ac N |
1109 | /* We might need to issue multiple reads to different |
1110 | * devices if there are bad blocks around, so we keep | |
1111 | * track of the number of reads in bio->bi_phys_segments. | |
1112 | * If this is 0, there is only one r1_bio and no locking | |
1113 | * will be needed when requests complete. If it is | |
1114 | * non-zero, then it is the number of not-completed requests. | |
1115 | */ | |
1116 | bio->bi_phys_segments = 0; | |
b7c44ed9 | 1117 | bio_clear_flag(bio, BIO_SEG_VALID); |
d2eb35ac | 1118 | |
a362357b | 1119 | if (rw == READ) { |
1da177e4 LT |
1120 | /* |
1121 | * read balancing logic: | |
1122 | */ | |
d2eb35ac N |
1123 | int rdisk; |
1124 | ||
1125 | read_again: | |
1126 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1da177e4 LT |
1127 | |
1128 | if (rdisk < 0) { | |
1129 | /* couldn't find anywhere to read from */ | |
1130 | raid_end_bio_io(r1_bio); | |
5a7bbad2 | 1131 | return; |
1da177e4 LT |
1132 | } |
1133 | mirror = conf->mirrors + rdisk; | |
1134 | ||
e555190d N |
1135 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1136 | bitmap) { | |
1137 | /* Reading from a write-mostly device must | |
1138 | * take care not to over-take any writes | |
1139 | * that are 'behind' | |
1140 | */ | |
1141 | wait_event(bitmap->behind_wait, | |
1142 | atomic_read(&bitmap->behind_writes) == 0); | |
1143 | } | |
1da177e4 | 1144 | r1_bio->read_disk = rdisk; |
f0cc9a05 | 1145 | r1_bio->start_next_window = 0; |
1da177e4 | 1146 | |
a167f663 | 1147 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1148 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1149 | max_sectors); |
1da177e4 LT |
1150 | |
1151 | r1_bio->bios[rdisk] = read_bio; | |
1152 | ||
4f024f37 KO |
1153 | read_bio->bi_iter.bi_sector = r1_bio->sector + |
1154 | mirror->rdev->data_offset; | |
1da177e4 LT |
1155 | read_bio->bi_bdev = mirror->rdev->bdev; |
1156 | read_bio->bi_end_io = raid1_end_read_request; | |
7b6d91da | 1157 | read_bio->bi_rw = READ | do_sync; |
1da177e4 LT |
1158 | read_bio->bi_private = r1_bio; |
1159 | ||
d2eb35ac N |
1160 | if (max_sectors < r1_bio->sectors) { |
1161 | /* could not read all from this device, so we will | |
1162 | * need another r1_bio. | |
1163 | */ | |
d2eb35ac N |
1164 | |
1165 | sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 1166 | - bio->bi_iter.bi_sector); |
d2eb35ac N |
1167 | r1_bio->sectors = max_sectors; |
1168 | spin_lock_irq(&conf->device_lock); | |
1169 | if (bio->bi_phys_segments == 0) | |
1170 | bio->bi_phys_segments = 2; | |
1171 | else | |
1172 | bio->bi_phys_segments++; | |
1173 | spin_unlock_irq(&conf->device_lock); | |
1174 | /* Cannot call generic_make_request directly | |
1175 | * as that will be queued in __make_request | |
1176 | * and subsequent mempool_alloc might block waiting | |
1177 | * for it. So hand bio over to raid1d. | |
1178 | */ | |
1179 | reschedule_retry(r1_bio); | |
1180 | ||
1181 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1182 | ||
1183 | r1_bio->master_bio = bio; | |
aa8b57aa | 1184 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
d2eb35ac N |
1185 | r1_bio->state = 0; |
1186 | r1_bio->mddev = mddev; | |
4f024f37 KO |
1187 | r1_bio->sector = bio->bi_iter.bi_sector + |
1188 | sectors_handled; | |
d2eb35ac N |
1189 | goto read_again; |
1190 | } else | |
1191 | generic_make_request(read_bio); | |
5a7bbad2 | 1192 | return; |
1da177e4 LT |
1193 | } |
1194 | ||
1195 | /* | |
1196 | * WRITE: | |
1197 | */ | |
34db0cd6 N |
1198 | if (conf->pending_count >= max_queued_requests) { |
1199 | md_wakeup_thread(mddev->thread); | |
1200 | wait_event(conf->wait_barrier, | |
1201 | conf->pending_count < max_queued_requests); | |
1202 | } | |
1f68f0c4 | 1203 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1204 | * inc refcount on their rdev. Record them by setting |
1205 | * bios[x] to bio | |
1f68f0c4 N |
1206 | * If there are known/acknowledged bad blocks on any device on |
1207 | * which we have seen a write error, we want to avoid writing those | |
1208 | * blocks. | |
1209 | * This potentially requires several writes to write around | |
1210 | * the bad blocks. Each set of writes gets it's own r1bio | |
1211 | * with a set of bios attached. | |
1da177e4 | 1212 | */ |
c3b328ac | 1213 | |
8f19ccb2 | 1214 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1215 | retry_write: |
79ef3a8a | 1216 | r1_bio->start_next_window = start_next_window; |
6bfe0b49 | 1217 | blocked_rdev = NULL; |
1da177e4 | 1218 | rcu_read_lock(); |
1f68f0c4 | 1219 | max_sectors = r1_bio->sectors; |
1da177e4 | 1220 | for (i = 0; i < disks; i++) { |
3cb03002 | 1221 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1222 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1223 | atomic_inc(&rdev->nr_pending); | |
1224 | blocked_rdev = rdev; | |
1225 | break; | |
1226 | } | |
1f68f0c4 | 1227 | r1_bio->bios[i] = NULL; |
8ae12666 | 1228 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1229 | if (i < conf->raid_disks) |
1230 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1231 | continue; |
1232 | } | |
1233 | ||
1234 | atomic_inc(&rdev->nr_pending); | |
1235 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1236 | sector_t first_bad; | |
1237 | int bad_sectors; | |
1238 | int is_bad; | |
1239 | ||
1240 | is_bad = is_badblock(rdev, r1_bio->sector, | |
1241 | max_sectors, | |
1242 | &first_bad, &bad_sectors); | |
1243 | if (is_bad < 0) { | |
1244 | /* mustn't write here until the bad block is | |
1245 | * acknowledged*/ | |
1246 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1247 | blocked_rdev = rdev; | |
1248 | break; | |
1249 | } | |
1250 | if (is_bad && first_bad <= r1_bio->sector) { | |
1251 | /* Cannot write here at all */ | |
1252 | bad_sectors -= (r1_bio->sector - first_bad); | |
1253 | if (bad_sectors < max_sectors) | |
1254 | /* mustn't write more than bad_sectors | |
1255 | * to other devices yet | |
1256 | */ | |
1257 | max_sectors = bad_sectors; | |
03c902e1 | 1258 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1259 | /* We don't set R1BIO_Degraded as that |
1260 | * only applies if the disk is | |
1261 | * missing, so it might be re-added, | |
1262 | * and we want to know to recover this | |
1263 | * chunk. | |
1264 | * In this case the device is here, | |
1265 | * and the fact that this chunk is not | |
1266 | * in-sync is recorded in the bad | |
1267 | * block log | |
1268 | */ | |
1269 | continue; | |
964147d5 | 1270 | } |
1f68f0c4 N |
1271 | if (is_bad) { |
1272 | int good_sectors = first_bad - r1_bio->sector; | |
1273 | if (good_sectors < max_sectors) | |
1274 | max_sectors = good_sectors; | |
1275 | } | |
1276 | } | |
1277 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1278 | } |
1279 | rcu_read_unlock(); | |
1280 | ||
6bfe0b49 DW |
1281 | if (unlikely(blocked_rdev)) { |
1282 | /* Wait for this device to become unblocked */ | |
1283 | int j; | |
79ef3a8a | 1284 | sector_t old = start_next_window; |
6bfe0b49 DW |
1285 | |
1286 | for (j = 0; j < i; j++) | |
1287 | if (r1_bio->bios[j]) | |
1288 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1289 | r1_bio->state = 0; |
4f024f37 | 1290 | allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); |
6bfe0b49 | 1291 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
79ef3a8a | 1292 | start_next_window = wait_barrier(conf, bio); |
1293 | /* | |
1294 | * We must make sure the multi r1bios of bio have | |
1295 | * the same value of bi_phys_segments | |
1296 | */ | |
1297 | if (bio->bi_phys_segments && old && | |
1298 | old != start_next_window) | |
1299 | /* Wait for the former r1bio(s) to complete */ | |
1300 | wait_event(conf->wait_barrier, | |
1301 | bio->bi_phys_segments == 1); | |
6bfe0b49 DW |
1302 | goto retry_write; |
1303 | } | |
1304 | ||
1f68f0c4 N |
1305 | if (max_sectors < r1_bio->sectors) { |
1306 | /* We are splitting this write into multiple parts, so | |
1307 | * we need to prepare for allocating another r1_bio. | |
1308 | */ | |
1309 | r1_bio->sectors = max_sectors; | |
1310 | spin_lock_irq(&conf->device_lock); | |
1311 | if (bio->bi_phys_segments == 0) | |
1312 | bio->bi_phys_segments = 2; | |
1313 | else | |
1314 | bio->bi_phys_segments++; | |
1315 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1316 | } |
4f024f37 | 1317 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
4b6d287f | 1318 | |
4e78064f | 1319 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1320 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1321 | |
1f68f0c4 | 1322 | first_clone = 1; |
1da177e4 LT |
1323 | for (i = 0; i < disks; i++) { |
1324 | struct bio *mbio; | |
1325 | if (!r1_bio->bios[i]) | |
1326 | continue; | |
1327 | ||
a167f663 | 1328 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1329 | bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
1f68f0c4 N |
1330 | |
1331 | if (first_clone) { | |
1332 | /* do behind I/O ? | |
1333 | * Not if there are too many, or cannot | |
1334 | * allocate memory, or a reader on WriteMostly | |
1335 | * is waiting for behind writes to flush */ | |
1336 | if (bitmap && | |
1337 | (atomic_read(&bitmap->behind_writes) | |
1338 | < mddev->bitmap_info.max_write_behind) && | |
1339 | !waitqueue_active(&bitmap->behind_wait)) | |
1340 | alloc_behind_pages(mbio, r1_bio); | |
1341 | ||
1342 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1343 | r1_bio->sectors, | |
1344 | test_bit(R1BIO_BehindIO, | |
1345 | &r1_bio->state)); | |
1346 | first_clone = 0; | |
1347 | } | |
2ca68f5e | 1348 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1349 | struct bio_vec *bvec; |
1350 | int j; | |
1351 | ||
cb34e057 KO |
1352 | /* |
1353 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1354 | */ |
d74c6d51 | 1355 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1356 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1357 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1358 | atomic_inc(&r1_bio->behind_remaining); | |
1359 | } | |
1360 | ||
1f68f0c4 N |
1361 | r1_bio->bios[i] = mbio; |
1362 | ||
4f024f37 | 1363 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 N |
1364 | conf->mirrors[i].rdev->data_offset); |
1365 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1366 | mbio->bi_end_io = raid1_end_write_request; | |
c8dc9c65 JL |
1367 | mbio->bi_rw = |
1368 | WRITE | do_flush_fua | do_sync | do_discard | do_same; | |
1f68f0c4 N |
1369 | mbio->bi_private = r1_bio; |
1370 | ||
1da177e4 | 1371 | atomic_inc(&r1_bio->remaining); |
f54a9d0e N |
1372 | |
1373 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); | |
1374 | if (cb) | |
1375 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1376 | else | |
1377 | plug = NULL; | |
4e78064f | 1378 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1379 | if (plug) { |
1380 | bio_list_add(&plug->pending, mbio); | |
1381 | plug->pending_cnt++; | |
1382 | } else { | |
1383 | bio_list_add(&conf->pending_bio_list, mbio); | |
1384 | conf->pending_count++; | |
1385 | } | |
4e78064f | 1386 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1387 | if (!plug) |
b357f04a | 1388 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1389 | } |
079fa166 N |
1390 | /* Mustn't call r1_bio_write_done before this next test, |
1391 | * as it could result in the bio being freed. | |
1392 | */ | |
aa8b57aa | 1393 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1394 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1395 | /* We need another r1_bio. It has already been counted |
1396 | * in bio->bi_phys_segments | |
1397 | */ | |
1398 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1399 | r1_bio->master_bio = bio; | |
aa8b57aa | 1400 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
1f68f0c4 N |
1401 | r1_bio->state = 0; |
1402 | r1_bio->mddev = mddev; | |
4f024f37 | 1403 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
1f68f0c4 N |
1404 | goto retry_write; |
1405 | } | |
1406 | ||
079fa166 N |
1407 | r1_bio_write_done(r1_bio); |
1408 | ||
1409 | /* In case raid1d snuck in to freeze_array */ | |
1410 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1411 | } |
1412 | ||
849674e4 | 1413 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1414 | { |
e8096360 | 1415 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1416 | int i; |
1417 | ||
1418 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1419 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1420 | rcu_read_lock(); |
1421 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1422 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1423 | seq_printf(seq, "%s", |
ddac7c7e N |
1424 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1425 | } | |
1426 | rcu_read_unlock(); | |
1da177e4 LT |
1427 | seq_printf(seq, "]"); |
1428 | } | |
1429 | ||
849674e4 | 1430 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1431 | { |
1432 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1433 | struct r1conf *conf = mddev->private; |
423f04d6 | 1434 | unsigned long flags; |
1da177e4 LT |
1435 | |
1436 | /* | |
1437 | * If it is not operational, then we have already marked it as dead | |
1438 | * else if it is the last working disks, ignore the error, let the | |
1439 | * next level up know. | |
1440 | * else mark the drive as failed | |
1441 | */ | |
b2d444d7 | 1442 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1443 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1444 | /* |
1445 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1446 | * normal single drive. |
1447 | * However don't try a recovery from this drive as | |
1448 | * it is very likely to fail. | |
1da177e4 | 1449 | */ |
5389042f | 1450 | conf->recovery_disabled = mddev->recovery_disabled; |
1da177e4 | 1451 | return; |
4044ba58 | 1452 | } |
de393cde | 1453 | set_bit(Blocked, &rdev->flags); |
423f04d6 | 1454 | spin_lock_irqsave(&conf->device_lock, flags); |
c04be0aa | 1455 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1456 | mddev->degraded++; |
dd00a99e | 1457 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1458 | } else |
1459 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1460 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1461 | /* |
1462 | * if recovery is running, make sure it aborts. | |
1463 | */ | |
1464 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
85ad1d13 GJ |
1465 | set_mask_bits(&mddev->flags, 0, |
1466 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
067032bc JP |
1467 | printk(KERN_ALERT |
1468 | "md/raid1:%s: Disk failure on %s, disabling device.\n" | |
1469 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
9dd1e2fa N |
1470 | mdname(mddev), bdevname(rdev->bdev, b), |
1471 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1472 | } |
1473 | ||
e8096360 | 1474 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1475 | { |
1476 | int i; | |
1da177e4 | 1477 | |
9dd1e2fa | 1478 | printk(KERN_DEBUG "RAID1 conf printout:\n"); |
1da177e4 | 1479 | if (!conf) { |
9dd1e2fa | 1480 | printk(KERN_DEBUG "(!conf)\n"); |
1da177e4 LT |
1481 | return; |
1482 | } | |
9dd1e2fa | 1483 | printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1da177e4 LT |
1484 | conf->raid_disks); |
1485 | ||
ddac7c7e | 1486 | rcu_read_lock(); |
1da177e4 LT |
1487 | for (i = 0; i < conf->raid_disks; i++) { |
1488 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1489 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1490 | if (rdev) |
9dd1e2fa | 1491 | printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n", |
ddac7c7e N |
1492 | i, !test_bit(In_sync, &rdev->flags), |
1493 | !test_bit(Faulty, &rdev->flags), | |
1494 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1495 | } |
ddac7c7e | 1496 | rcu_read_unlock(); |
1da177e4 LT |
1497 | } |
1498 | ||
e8096360 | 1499 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1500 | { |
79ef3a8a | 1501 | wait_barrier(conf, NULL); |
1502 | allow_barrier(conf, 0, 0); | |
1da177e4 LT |
1503 | |
1504 | mempool_destroy(conf->r1buf_pool); | |
1505 | conf->r1buf_pool = NULL; | |
79ef3a8a | 1506 | |
669cc7ba | 1507 | spin_lock_irq(&conf->resync_lock); |
e8ff8bf0 | 1508 | conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE; |
79ef3a8a | 1509 | conf->start_next_window = MaxSector; |
669cc7ba N |
1510 | conf->current_window_requests += |
1511 | conf->next_window_requests; | |
1512 | conf->next_window_requests = 0; | |
1513 | spin_unlock_irq(&conf->resync_lock); | |
1da177e4 LT |
1514 | } |
1515 | ||
fd01b88c | 1516 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1517 | { |
1518 | int i; | |
e8096360 | 1519 | struct r1conf *conf = mddev->private; |
6b965620 N |
1520 | int count = 0; |
1521 | unsigned long flags; | |
1da177e4 LT |
1522 | |
1523 | /* | |
f72ffdd6 | 1524 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1525 | * and mark them readable. |
1526 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1527 | * device_lock used to avoid races with raid1_end_read_request |
1528 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1529 | */ |
423f04d6 | 1530 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1531 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1532 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1533 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1534 | if (repl | |
1aee41f6 | 1535 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1536 | && repl->recovery_offset == MaxSector |
1537 | && !test_bit(Faulty, &repl->flags) | |
1538 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1539 | /* replacement has just become active */ | |
1540 | if (!rdev || | |
1541 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1542 | count++; | |
1543 | if (rdev) { | |
1544 | /* Replaced device not technically | |
1545 | * faulty, but we need to be sure | |
1546 | * it gets removed and never re-added | |
1547 | */ | |
1548 | set_bit(Faulty, &rdev->flags); | |
1549 | sysfs_notify_dirent_safe( | |
1550 | rdev->sysfs_state); | |
1551 | } | |
1552 | } | |
ddac7c7e | 1553 | if (rdev |
61e4947c | 1554 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1555 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1556 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1557 | count++; |
654e8b5a | 1558 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1559 | } |
1560 | } | |
6b965620 N |
1561 | mddev->degraded -= count; |
1562 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1563 | |
1564 | print_conf(conf); | |
6b965620 | 1565 | return count; |
1da177e4 LT |
1566 | } |
1567 | ||
fd01b88c | 1568 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1569 | { |
e8096360 | 1570 | struct r1conf *conf = mddev->private; |
199050ea | 1571 | int err = -EEXIST; |
41158c7e | 1572 | int mirror = 0; |
0eaf822c | 1573 | struct raid1_info *p; |
6c2fce2e | 1574 | int first = 0; |
30194636 | 1575 | int last = conf->raid_disks - 1; |
1da177e4 | 1576 | |
5389042f N |
1577 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1578 | return -EBUSY; | |
1579 | ||
1501efad DW |
1580 | if (md_integrity_add_rdev(rdev, mddev)) |
1581 | return -ENXIO; | |
1582 | ||
6c2fce2e NB |
1583 | if (rdev->raid_disk >= 0) |
1584 | first = last = rdev->raid_disk; | |
1585 | ||
70bcecdb GR |
1586 | /* |
1587 | * find the disk ... but prefer rdev->saved_raid_disk | |
1588 | * if possible. | |
1589 | */ | |
1590 | if (rdev->saved_raid_disk >= 0 && | |
1591 | rdev->saved_raid_disk >= first && | |
1592 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) | |
1593 | first = last = rdev->saved_raid_disk; | |
1594 | ||
7ef449d1 N |
1595 | for (mirror = first; mirror <= last; mirror++) { |
1596 | p = conf->mirrors+mirror; | |
1597 | if (!p->rdev) { | |
1da177e4 | 1598 | |
9092c02d JB |
1599 | if (mddev->gendisk) |
1600 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1601 | rdev->data_offset << 9); | |
1da177e4 LT |
1602 | |
1603 | p->head_position = 0; | |
1604 | rdev->raid_disk = mirror; | |
199050ea | 1605 | err = 0; |
6aea114a N |
1606 | /* As all devices are equivalent, we don't need a full recovery |
1607 | * if this was recently any drive of the array | |
1608 | */ | |
1609 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1610 | conf->fullsync = 1; |
d6065f7b | 1611 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1612 | break; |
1613 | } | |
7ef449d1 N |
1614 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1615 | p[conf->raid_disks].rdev == NULL) { | |
1616 | /* Add this device as a replacement */ | |
1617 | clear_bit(In_sync, &rdev->flags); | |
1618 | set_bit(Replacement, &rdev->flags); | |
1619 | rdev->raid_disk = mirror; | |
1620 | err = 0; | |
1621 | conf->fullsync = 1; | |
1622 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1623 | break; | |
1624 | } | |
1625 | } | |
9092c02d | 1626 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1627 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1628 | print_conf(conf); |
199050ea | 1629 | return err; |
1da177e4 LT |
1630 | } |
1631 | ||
b8321b68 | 1632 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1633 | { |
e8096360 | 1634 | struct r1conf *conf = mddev->private; |
1da177e4 | 1635 | int err = 0; |
b8321b68 | 1636 | int number = rdev->raid_disk; |
0eaf822c | 1637 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1638 | |
b014f14c N |
1639 | if (rdev != p->rdev) |
1640 | p = conf->mirrors + conf->raid_disks + number; | |
1641 | ||
1da177e4 | 1642 | print_conf(conf); |
b8321b68 | 1643 | if (rdev == p->rdev) { |
b2d444d7 | 1644 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1645 | atomic_read(&rdev->nr_pending)) { |
1646 | err = -EBUSY; | |
1647 | goto abort; | |
1648 | } | |
046abeed | 1649 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1650 | * is not possible. |
1651 | */ | |
1652 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1653 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1654 | mddev->degraded < conf->raid_disks) { |
1655 | err = -EBUSY; | |
1656 | goto abort; | |
1657 | } | |
1da177e4 | 1658 | p->rdev = NULL; |
fbd568a3 | 1659 | synchronize_rcu(); |
1da177e4 LT |
1660 | if (atomic_read(&rdev->nr_pending)) { |
1661 | /* lost the race, try later */ | |
1662 | err = -EBUSY; | |
1663 | p->rdev = rdev; | |
ac5e7113 | 1664 | goto abort; |
8c7a2c2b N |
1665 | } else if (conf->mirrors[conf->raid_disks + number].rdev) { |
1666 | /* We just removed a device that is being replaced. | |
1667 | * Move down the replacement. We drain all IO before | |
1668 | * doing this to avoid confusion. | |
1669 | */ | |
1670 | struct md_rdev *repl = | |
1671 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1672 | freeze_array(conf, 0); |
8c7a2c2b N |
1673 | clear_bit(Replacement, &repl->flags); |
1674 | p->rdev = repl; | |
1675 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1676 | unfreeze_array(conf); |
8c7a2c2b N |
1677 | clear_bit(WantReplacement, &rdev->flags); |
1678 | } else | |
b014f14c | 1679 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1680 | err = md_integrity_register(mddev); |
1da177e4 LT |
1681 | } |
1682 | abort: | |
1683 | ||
1684 | print_conf(conf); | |
1685 | return err; | |
1686 | } | |
1687 | ||
4246a0b6 | 1688 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1689 | { |
9f2c9d12 | 1690 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1691 | |
0fc280f6 | 1692 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1693 | |
1da177e4 LT |
1694 | /* |
1695 | * we have read a block, now it needs to be re-written, | |
1696 | * or re-read if the read failed. | |
1697 | * We don't do much here, just schedule handling by raid1d | |
1698 | */ | |
4246a0b6 | 1699 | if (!bio->bi_error) |
1da177e4 | 1700 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1701 | |
1702 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1703 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1704 | } |
1705 | ||
4246a0b6 | 1706 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1707 | { |
4246a0b6 | 1708 | int uptodate = !bio->bi_error; |
9f2c9d12 | 1709 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1710 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1711 | struct r1conf *conf = mddev->private; |
4367af55 N |
1712 | sector_t first_bad; |
1713 | int bad_sectors; | |
854abd75 | 1714 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1715 | |
6b1117d5 | 1716 | if (!uptodate) { |
57dab0bd | 1717 | sector_t sync_blocks = 0; |
6b1117d5 N |
1718 | sector_t s = r1_bio->sector; |
1719 | long sectors_to_go = r1_bio->sectors; | |
1720 | /* make sure these bits doesn't get cleared. */ | |
1721 | do { | |
5e3db645 | 1722 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1723 | &sync_blocks, 1); |
1724 | s += sync_blocks; | |
1725 | sectors_to_go -= sync_blocks; | |
1726 | } while (sectors_to_go > 0); | |
854abd75 N |
1727 | set_bit(WriteErrorSeen, &rdev->flags); |
1728 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1729 | set_bit(MD_RECOVERY_NEEDED, & |
1730 | mddev->recovery); | |
d8f05d29 | 1731 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1732 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1733 | &first_bad, &bad_sectors) && |
1734 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1735 | r1_bio->sector, | |
1736 | r1_bio->sectors, | |
1737 | &first_bad, &bad_sectors) | |
1738 | ) | |
4367af55 | 1739 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1740 | |
1da177e4 | 1741 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1742 | int s = r1_bio->sectors; |
d8f05d29 N |
1743 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1744 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1745 | reschedule_retry(r1_bio); |
1746 | else { | |
1747 | put_buf(r1_bio); | |
1748 | md_done_sync(mddev, s, uptodate); | |
1749 | } | |
1da177e4 | 1750 | } |
1da177e4 LT |
1751 | } |
1752 | ||
3cb03002 | 1753 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1754 | int sectors, struct page *page, int rw) |
1755 | { | |
1756 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) | |
1757 | /* success */ | |
1758 | return 1; | |
19d67169 | 1759 | if (rw == WRITE) { |
d8f05d29 | 1760 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1761 | if (!test_and_set_bit(WantReplacement, |
1762 | &rdev->flags)) | |
1763 | set_bit(MD_RECOVERY_NEEDED, & | |
1764 | rdev->mddev->recovery); | |
1765 | } | |
d8f05d29 N |
1766 | /* need to record an error - either for the block or the device */ |
1767 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1768 | md_error(rdev->mddev, rdev); | |
1769 | return 0; | |
1770 | } | |
1771 | ||
9f2c9d12 | 1772 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1773 | { |
a68e5870 N |
1774 | /* Try some synchronous reads of other devices to get |
1775 | * good data, much like with normal read errors. Only | |
1776 | * read into the pages we already have so we don't | |
1777 | * need to re-issue the read request. | |
1778 | * We don't need to freeze the array, because being in an | |
1779 | * active sync request, there is no normal IO, and | |
1780 | * no overlapping syncs. | |
06f60385 N |
1781 | * We don't need to check is_badblock() again as we |
1782 | * made sure that anything with a bad block in range | |
1783 | * will have bi_end_io clear. | |
a68e5870 | 1784 | */ |
fd01b88c | 1785 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1786 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1787 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1788 | sector_t sect = r1_bio->sector; | |
1789 | int sectors = r1_bio->sectors; | |
1790 | int idx = 0; | |
1791 | ||
1792 | while(sectors) { | |
1793 | int s = sectors; | |
1794 | int d = r1_bio->read_disk; | |
1795 | int success = 0; | |
3cb03002 | 1796 | struct md_rdev *rdev; |
78d7f5f7 | 1797 | int start; |
a68e5870 N |
1798 | |
1799 | if (s > (PAGE_SIZE>>9)) | |
1800 | s = PAGE_SIZE >> 9; | |
1801 | do { | |
1802 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1803 | /* No rcu protection needed here devices | |
1804 | * can only be removed when no resync is | |
1805 | * active, and resync is currently active | |
1806 | */ | |
1807 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1808 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 N |
1809 | bio->bi_io_vec[idx].bv_page, |
1810 | READ, false)) { | |
1811 | success = 1; | |
1812 | break; | |
1813 | } | |
1814 | } | |
1815 | d++; | |
8f19ccb2 | 1816 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1817 | d = 0; |
1818 | } while (!success && d != r1_bio->read_disk); | |
1819 | ||
78d7f5f7 | 1820 | if (!success) { |
a68e5870 | 1821 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1822 | int abort = 0; |
1823 | /* Cannot read from anywhere, this block is lost. | |
1824 | * Record a bad block on each device. If that doesn't | |
1825 | * work just disable and interrupt the recovery. | |
1826 | * Don't fail devices as that won't really help. | |
1827 | */ | |
a68e5870 N |
1828 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error" |
1829 | " for block %llu\n", | |
1830 | mdname(mddev), | |
1831 | bdevname(bio->bi_bdev, b), | |
1832 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1833 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1834 | rdev = conf->mirrors[d].rdev; |
1835 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1836 | continue; | |
1837 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1838 | abort = 1; | |
1839 | } | |
1840 | if (abort) { | |
d890fa2b N |
1841 | conf->recovery_disabled = |
1842 | mddev->recovery_disabled; | |
3a9f28a5 N |
1843 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1844 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1845 | put_buf(r1_bio); | |
1846 | return 0; | |
1847 | } | |
1848 | /* Try next page */ | |
1849 | sectors -= s; | |
1850 | sect += s; | |
1851 | idx++; | |
1852 | continue; | |
d11c171e | 1853 | } |
78d7f5f7 N |
1854 | |
1855 | start = d; | |
1856 | /* write it back and re-read */ | |
1857 | while (d != r1_bio->read_disk) { | |
1858 | if (d == 0) | |
8f19ccb2 | 1859 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1860 | d--; |
1861 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1862 | continue; | |
1863 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1864 | if (r1_sync_page_io(rdev, sect, s, |
1865 | bio->bi_io_vec[idx].bv_page, | |
1866 | WRITE) == 0) { | |
78d7f5f7 N |
1867 | r1_bio->bios[d]->bi_end_io = NULL; |
1868 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1869 | } |
78d7f5f7 N |
1870 | } |
1871 | d = start; | |
1872 | while (d != r1_bio->read_disk) { | |
1873 | if (d == 0) | |
8f19ccb2 | 1874 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1875 | d--; |
1876 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1877 | continue; | |
1878 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1879 | if (r1_sync_page_io(rdev, sect, s, |
1880 | bio->bi_io_vec[idx].bv_page, | |
1881 | READ) != 0) | |
9d3d8011 | 1882 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1883 | } |
a68e5870 N |
1884 | sectors -= s; |
1885 | sect += s; | |
1886 | idx ++; | |
1887 | } | |
78d7f5f7 | 1888 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4246a0b6 | 1889 | bio->bi_error = 0; |
a68e5870 N |
1890 | return 1; |
1891 | } | |
1892 | ||
c95e6385 | 1893 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1894 | { |
1895 | /* We have read all readable devices. If we haven't | |
1896 | * got the block, then there is no hope left. | |
1897 | * If we have, then we want to do a comparison | |
1898 | * and skip the write if everything is the same. | |
1899 | * If any blocks failed to read, then we need to | |
1900 | * attempt an over-write | |
1901 | */ | |
fd01b88c | 1902 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1903 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1904 | int primary; |
1905 | int i; | |
f4380a91 | 1906 | int vcnt; |
a68e5870 | 1907 | |
30bc9b53 N |
1908 | /* Fix variable parts of all bios */ |
1909 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
1910 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
1911 | int j; | |
1912 | int size; | |
4246a0b6 | 1913 | int error; |
30bc9b53 N |
1914 | struct bio *b = r1_bio->bios[i]; |
1915 | if (b->bi_end_io != end_sync_read) | |
1916 | continue; | |
4246a0b6 CH |
1917 | /* fixup the bio for reuse, but preserve errno */ |
1918 | error = b->bi_error; | |
30bc9b53 | 1919 | bio_reset(b); |
4246a0b6 | 1920 | b->bi_error = error; |
30bc9b53 | 1921 | b->bi_vcnt = vcnt; |
4f024f37 KO |
1922 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
1923 | b->bi_iter.bi_sector = r1_bio->sector + | |
30bc9b53 N |
1924 | conf->mirrors[i].rdev->data_offset; |
1925 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1926 | b->bi_end_io = end_sync_read; | |
1927 | b->bi_private = r1_bio; | |
1928 | ||
4f024f37 | 1929 | size = b->bi_iter.bi_size; |
30bc9b53 N |
1930 | for (j = 0; j < vcnt ; j++) { |
1931 | struct bio_vec *bi; | |
1932 | bi = &b->bi_io_vec[j]; | |
1933 | bi->bv_offset = 0; | |
1934 | if (size > PAGE_SIZE) | |
1935 | bi->bv_len = PAGE_SIZE; | |
1936 | else | |
1937 | bi->bv_len = size; | |
1938 | size -= PAGE_SIZE; | |
1939 | } | |
1940 | } | |
8f19ccb2 | 1941 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 1942 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4246a0b6 | 1943 | !r1_bio->bios[primary]->bi_error) { |
a68e5870 N |
1944 | r1_bio->bios[primary]->bi_end_io = NULL; |
1945 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
1946 | break; | |
1947 | } | |
1948 | r1_bio->read_disk = primary; | |
8f19ccb2 | 1949 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 1950 | int j; |
78d7f5f7 N |
1951 | struct bio *pbio = r1_bio->bios[primary]; |
1952 | struct bio *sbio = r1_bio->bios[i]; | |
4246a0b6 | 1953 | int error = sbio->bi_error; |
a68e5870 | 1954 | |
2aabaa65 | 1955 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 1956 | continue; |
4246a0b6 CH |
1957 | /* Now we can 'fixup' the error value */ |
1958 | sbio->bi_error = 0; | |
78d7f5f7 | 1959 | |
4246a0b6 | 1960 | if (!error) { |
78d7f5f7 N |
1961 | for (j = vcnt; j-- ; ) { |
1962 | struct page *p, *s; | |
1963 | p = pbio->bi_io_vec[j].bv_page; | |
1964 | s = sbio->bi_io_vec[j].bv_page; | |
1965 | if (memcmp(page_address(p), | |
1966 | page_address(s), | |
5020ad7d | 1967 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 1968 | break; |
69382e85 | 1969 | } |
78d7f5f7 N |
1970 | } else |
1971 | j = 0; | |
1972 | if (j >= 0) | |
7f7583d4 | 1973 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 1974 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4246a0b6 | 1975 | && !error)) { |
78d7f5f7 N |
1976 | /* No need to write to this device. */ |
1977 | sbio->bi_end_io = NULL; | |
1978 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
1979 | continue; | |
1980 | } | |
d3b45c2a KO |
1981 | |
1982 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 1983 | } |
a68e5870 N |
1984 | } |
1985 | ||
9f2c9d12 | 1986 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 1987 | { |
e8096360 | 1988 | struct r1conf *conf = mddev->private; |
a68e5870 | 1989 | int i; |
8f19ccb2 | 1990 | int disks = conf->raid_disks * 2; |
a68e5870 N |
1991 | struct bio *bio, *wbio; |
1992 | ||
1993 | bio = r1_bio->bios[r1_bio->read_disk]; | |
1994 | ||
a68e5870 N |
1995 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
1996 | /* ouch - failed to read all of that. */ | |
1997 | if (!fix_sync_read_error(r1_bio)) | |
1998 | return; | |
7ca78d57 N |
1999 | |
2000 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2001 | process_checks(r1_bio); |
2002 | ||
d11c171e N |
2003 | /* |
2004 | * schedule writes | |
2005 | */ | |
1da177e4 LT |
2006 | atomic_set(&r1_bio->remaining, 1); |
2007 | for (i = 0; i < disks ; i++) { | |
2008 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2009 | if (wbio->bi_end_io == NULL || |
2010 | (wbio->bi_end_io == end_sync_read && | |
2011 | (i == r1_bio->read_disk || | |
2012 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
2013 | continue; |
2014 | ||
3e198f78 N |
2015 | wbio->bi_rw = WRITE; |
2016 | wbio->bi_end_io = end_sync_write; | |
1da177e4 | 2017 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2018 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2019 | |
1da177e4 LT |
2020 | generic_make_request(wbio); |
2021 | } | |
2022 | ||
2023 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2024 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2025 | int s = r1_bio->sectors; |
2026 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2027 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2028 | reschedule_retry(r1_bio); | |
2029 | else { | |
2030 | put_buf(r1_bio); | |
2031 | md_done_sync(mddev, s, 1); | |
2032 | } | |
1da177e4 LT |
2033 | } |
2034 | } | |
2035 | ||
2036 | /* | |
2037 | * This is a kernel thread which: | |
2038 | * | |
2039 | * 1. Retries failed read operations on working mirrors. | |
2040 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2041 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2042 | */ |
2043 | ||
e8096360 | 2044 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2045 | sector_t sect, int sectors) |
2046 | { | |
fd01b88c | 2047 | struct mddev *mddev = conf->mddev; |
867868fb N |
2048 | while(sectors) { |
2049 | int s = sectors; | |
2050 | int d = read_disk; | |
2051 | int success = 0; | |
2052 | int start; | |
3cb03002 | 2053 | struct md_rdev *rdev; |
867868fb N |
2054 | |
2055 | if (s > (PAGE_SIZE>>9)) | |
2056 | s = PAGE_SIZE >> 9; | |
2057 | ||
2058 | do { | |
d2eb35ac N |
2059 | sector_t first_bad; |
2060 | int bad_sectors; | |
2061 | ||
707a6a42 N |
2062 | rcu_read_lock(); |
2063 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2064 | if (rdev && |
da8840a7 | 2065 | (test_bit(In_sync, &rdev->flags) || |
2066 | (!test_bit(Faulty, &rdev->flags) && | |
2067 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2068 | is_badblock(rdev, sect, s, |
707a6a42 N |
2069 | &first_bad, &bad_sectors) == 0) { |
2070 | atomic_inc(&rdev->nr_pending); | |
2071 | rcu_read_unlock(); | |
2072 | if (sync_page_io(rdev, sect, s<<9, | |
2073 | conf->tmppage, READ, false)) | |
2074 | success = 1; | |
2075 | rdev_dec_pending(rdev, mddev); | |
2076 | if (success) | |
2077 | break; | |
2078 | } else | |
2079 | rcu_read_unlock(); | |
2080 | d++; | |
2081 | if (d == conf->raid_disks * 2) | |
2082 | d = 0; | |
867868fb N |
2083 | } while (!success && d != read_disk); |
2084 | ||
2085 | if (!success) { | |
d8f05d29 | 2086 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2087 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2088 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2089 | md_error(mddev, rdev); | |
867868fb N |
2090 | break; |
2091 | } | |
2092 | /* write it back and re-read */ | |
2093 | start = d; | |
2094 | while (d != read_disk) { | |
2095 | if (d==0) | |
8f19ccb2 | 2096 | d = conf->raid_disks * 2; |
867868fb | 2097 | d--; |
707a6a42 N |
2098 | rcu_read_lock(); |
2099 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2100 | if (rdev && |
707a6a42 N |
2101 | !test_bit(Faulty, &rdev->flags)) { |
2102 | atomic_inc(&rdev->nr_pending); | |
2103 | rcu_read_unlock(); | |
d8f05d29 N |
2104 | r1_sync_page_io(rdev, sect, s, |
2105 | conf->tmppage, WRITE); | |
707a6a42 N |
2106 | rdev_dec_pending(rdev, mddev); |
2107 | } else | |
2108 | rcu_read_unlock(); | |
867868fb N |
2109 | } |
2110 | d = start; | |
2111 | while (d != read_disk) { | |
2112 | char b[BDEVNAME_SIZE]; | |
2113 | if (d==0) | |
8f19ccb2 | 2114 | d = conf->raid_disks * 2; |
867868fb | 2115 | d--; |
707a6a42 N |
2116 | rcu_read_lock(); |
2117 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2118 | if (rdev && |
b8cb6b4c | 2119 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2120 | atomic_inc(&rdev->nr_pending); |
2121 | rcu_read_unlock(); | |
d8f05d29 N |
2122 | if (r1_sync_page_io(rdev, sect, s, |
2123 | conf->tmppage, READ)) { | |
867868fb N |
2124 | atomic_add(s, &rdev->corrected_errors); |
2125 | printk(KERN_INFO | |
9dd1e2fa | 2126 | "md/raid1:%s: read error corrected " |
867868fb N |
2127 | "(%d sectors at %llu on %s)\n", |
2128 | mdname(mddev), s, | |
969b755a | 2129 | (unsigned long long)(sect + |
707a6a42 | 2130 | rdev->data_offset), |
867868fb N |
2131 | bdevname(rdev->bdev, b)); |
2132 | } | |
707a6a42 N |
2133 | rdev_dec_pending(rdev, mddev); |
2134 | } else | |
2135 | rcu_read_unlock(); | |
867868fb N |
2136 | } |
2137 | sectors -= s; | |
2138 | sect += s; | |
2139 | } | |
2140 | } | |
2141 | ||
9f2c9d12 | 2142 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2143 | { |
fd01b88c | 2144 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2145 | struct r1conf *conf = mddev->private; |
3cb03002 | 2146 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2147 | |
2148 | /* bio has the data to be written to device 'i' where | |
2149 | * we just recently had a write error. | |
2150 | * We repeatedly clone the bio and trim down to one block, | |
2151 | * then try the write. Where the write fails we record | |
2152 | * a bad block. | |
2153 | * It is conceivable that the bio doesn't exactly align with | |
2154 | * blocks. We must handle this somehow. | |
2155 | * | |
2156 | * We currently own a reference on the rdev. | |
2157 | */ | |
2158 | ||
2159 | int block_sectors; | |
2160 | sector_t sector; | |
2161 | int sectors; | |
2162 | int sect_to_write = r1_bio->sectors; | |
2163 | int ok = 1; | |
2164 | ||
2165 | if (rdev->badblocks.shift < 0) | |
2166 | return 0; | |
2167 | ||
ab713cdc ND |
2168 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2169 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2170 | sector = r1_bio->sector; |
2171 | sectors = ((sector + block_sectors) | |
2172 | & ~(sector_t)(block_sectors - 1)) | |
2173 | - sector; | |
2174 | ||
cd5ff9a1 N |
2175 | while (sect_to_write) { |
2176 | struct bio *wbio; | |
2177 | if (sectors > sect_to_write) | |
2178 | sectors = sect_to_write; | |
2179 | /* Write at 'sector' for 'sectors'*/ | |
2180 | ||
b783863f KO |
2181 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2182 | unsigned vcnt = r1_bio->behind_page_count; | |
2183 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2184 | ||
2185 | while (!vec->bv_page) { | |
2186 | vec++; | |
2187 | vcnt--; | |
2188 | } | |
2189 | ||
2190 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2191 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2192 | ||
2193 | wbio->bi_vcnt = vcnt; | |
2194 | } else { | |
2195 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2196 | } | |
2197 | ||
cd5ff9a1 | 2198 | wbio->bi_rw = WRITE; |
4f024f37 KO |
2199 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2200 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2201 | |
6678d83f | 2202 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2203 | wbio->bi_iter.bi_sector += rdev->data_offset; |
cd5ff9a1 | 2204 | wbio->bi_bdev = rdev->bdev; |
203d27b0 | 2205 | if (submit_bio_wait(WRITE, wbio) < 0) |
cd5ff9a1 N |
2206 | /* failure! */ |
2207 | ok = rdev_set_badblocks(rdev, sector, | |
2208 | sectors, 0) | |
2209 | && ok; | |
2210 | ||
2211 | bio_put(wbio); | |
2212 | sect_to_write -= sectors; | |
2213 | sector += sectors; | |
2214 | sectors = block_sectors; | |
2215 | } | |
2216 | return ok; | |
2217 | } | |
2218 | ||
e8096360 | 2219 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2220 | { |
2221 | int m; | |
2222 | int s = r1_bio->sectors; | |
8f19ccb2 | 2223 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2224 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2225 | struct bio *bio = r1_bio->bios[m]; |
2226 | if (bio->bi_end_io == NULL) | |
2227 | continue; | |
4246a0b6 | 2228 | if (!bio->bi_error && |
62096bce | 2229 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2230 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2231 | } |
4246a0b6 | 2232 | if (bio->bi_error && |
62096bce N |
2233 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2234 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2235 | md_error(conf->mddev, rdev); | |
2236 | } | |
2237 | } | |
2238 | put_buf(r1_bio); | |
2239 | md_done_sync(conf->mddev, s, 1); | |
2240 | } | |
2241 | ||
e8096360 | 2242 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2243 | { |
2244 | int m; | |
55ce74d4 | 2245 | bool fail = false; |
8f19ccb2 | 2246 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2247 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2248 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2249 | rdev_clear_badblocks(rdev, |
2250 | r1_bio->sector, | |
c6563a8c | 2251 | r1_bio->sectors, 0); |
62096bce N |
2252 | rdev_dec_pending(rdev, conf->mddev); |
2253 | } else if (r1_bio->bios[m] != NULL) { | |
2254 | /* This drive got a write error. We need to | |
2255 | * narrow down and record precise write | |
2256 | * errors. | |
2257 | */ | |
55ce74d4 | 2258 | fail = true; |
62096bce N |
2259 | if (!narrow_write_error(r1_bio, m)) { |
2260 | md_error(conf->mddev, | |
2261 | conf->mirrors[m].rdev); | |
2262 | /* an I/O failed, we can't clear the bitmap */ | |
2263 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2264 | } | |
2265 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2266 | conf->mddev); | |
2267 | } | |
55ce74d4 N |
2268 | if (fail) { |
2269 | spin_lock_irq(&conf->device_lock); | |
2270 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
ccfc7bf1 | 2271 | conf->nr_queued++; |
55ce74d4 N |
2272 | spin_unlock_irq(&conf->device_lock); |
2273 | md_wakeup_thread(conf->mddev->thread); | |
bd8688a1 N |
2274 | } else { |
2275 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2276 | close_write(r1_bio); | |
55ce74d4 | 2277 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2278 | } |
62096bce N |
2279 | } |
2280 | ||
e8096360 | 2281 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2282 | { |
2283 | int disk; | |
2284 | int max_sectors; | |
fd01b88c | 2285 | struct mddev *mddev = conf->mddev; |
62096bce N |
2286 | struct bio *bio; |
2287 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2288 | struct md_rdev *rdev; |
62096bce N |
2289 | |
2290 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2291 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2292 | * the block and we can fix it. | |
2293 | * We freeze all other IO, and try reading the block from | |
2294 | * other devices. When we find one, we re-write | |
2295 | * and check it that fixes the read error. | |
2296 | * This is all done synchronously while the array is | |
2297 | * frozen | |
2298 | */ | |
2299 | if (mddev->ro == 0) { | |
e2d59925 | 2300 | freeze_array(conf, 1); |
62096bce N |
2301 | fix_read_error(conf, r1_bio->read_disk, |
2302 | r1_bio->sector, r1_bio->sectors); | |
2303 | unfreeze_array(conf); | |
2304 | } else | |
2305 | md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev); | |
7ad4d4a6 | 2306 | rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); |
62096bce N |
2307 | |
2308 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2309 | bdevname(bio->bi_bdev, b); | |
2310 | read_more: | |
2311 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2312 | if (disk == -1) { | |
2313 | printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O" | |
2314 | " read error for block %llu\n", | |
2315 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
2316 | raid_end_bio_io(r1_bio); | |
2317 | } else { | |
2318 | const unsigned long do_sync | |
2319 | = r1_bio->master_bio->bi_rw & REQ_SYNC; | |
2320 | if (bio) { | |
2321 | r1_bio->bios[r1_bio->read_disk] = | |
2322 | mddev->ro ? IO_BLOCKED : NULL; | |
2323 | bio_put(bio); | |
2324 | } | |
2325 | r1_bio->read_disk = disk; | |
2326 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
4f024f37 KO |
2327 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
2328 | max_sectors); | |
62096bce N |
2329 | r1_bio->bios[r1_bio->read_disk] = bio; |
2330 | rdev = conf->mirrors[disk].rdev; | |
2331 | printk_ratelimited(KERN_ERR | |
2332 | "md/raid1:%s: redirecting sector %llu" | |
2333 | " to other mirror: %s\n", | |
2334 | mdname(mddev), | |
2335 | (unsigned long long)r1_bio->sector, | |
2336 | bdevname(rdev->bdev, b)); | |
4f024f37 | 2337 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
62096bce N |
2338 | bio->bi_bdev = rdev->bdev; |
2339 | bio->bi_end_io = raid1_end_read_request; | |
2340 | bio->bi_rw = READ | do_sync; | |
2341 | bio->bi_private = r1_bio; | |
2342 | if (max_sectors < r1_bio->sectors) { | |
2343 | /* Drat - have to split this up more */ | |
2344 | struct bio *mbio = r1_bio->master_bio; | |
2345 | int sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 2346 | - mbio->bi_iter.bi_sector); |
62096bce N |
2347 | r1_bio->sectors = max_sectors; |
2348 | spin_lock_irq(&conf->device_lock); | |
2349 | if (mbio->bi_phys_segments == 0) | |
2350 | mbio->bi_phys_segments = 2; | |
2351 | else | |
2352 | mbio->bi_phys_segments++; | |
2353 | spin_unlock_irq(&conf->device_lock); | |
2354 | generic_make_request(bio); | |
2355 | bio = NULL; | |
2356 | ||
2357 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2358 | ||
2359 | r1_bio->master_bio = mbio; | |
aa8b57aa | 2360 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
62096bce N |
2361 | r1_bio->state = 0; |
2362 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2363 | r1_bio->mddev = mddev; | |
4f024f37 KO |
2364 | r1_bio->sector = mbio->bi_iter.bi_sector + |
2365 | sectors_handled; | |
62096bce N |
2366 | |
2367 | goto read_more; | |
2368 | } else | |
2369 | generic_make_request(bio); | |
2370 | } | |
2371 | } | |
2372 | ||
4ed8731d | 2373 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2374 | { |
4ed8731d | 2375 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2376 | struct r1bio *r1_bio; |
1da177e4 | 2377 | unsigned long flags; |
e8096360 | 2378 | struct r1conf *conf = mddev->private; |
1da177e4 | 2379 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2380 | struct blk_plug plug; |
1da177e4 LT |
2381 | |
2382 | md_check_recovery(mddev); | |
e1dfa0a2 | 2383 | |
55ce74d4 N |
2384 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2385 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
2386 | LIST_HEAD(tmp); | |
2387 | spin_lock_irqsave(&conf->device_lock, flags); | |
2388 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
ccfc7bf1 ND |
2389 | while (!list_empty(&conf->bio_end_io_list)) { |
2390 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2391 | conf->nr_queued--; | |
2392 | } | |
55ce74d4 N |
2393 | } |
2394 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2395 | while (!list_empty(&tmp)) { | |
a452744b MP |
2396 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2397 | retry_list); | |
55ce74d4 | 2398 | list_del(&r1_bio->retry_list); |
bd8688a1 N |
2399 | if (mddev->degraded) |
2400 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2401 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2402 | close_write(r1_bio); | |
55ce74d4 N |
2403 | raid_end_bio_io(r1_bio); |
2404 | } | |
2405 | } | |
2406 | ||
e1dfa0a2 | 2407 | blk_start_plug(&plug); |
1da177e4 | 2408 | for (;;) { |
191ea9b2 | 2409 | |
0021b7bc | 2410 | flush_pending_writes(conf); |
191ea9b2 | 2411 | |
a35e63ef N |
2412 | spin_lock_irqsave(&conf->device_lock, flags); |
2413 | if (list_empty(head)) { | |
2414 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2415 | break; |
a35e63ef | 2416 | } |
9f2c9d12 | 2417 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2418 | list_del(head->prev); |
ddaf22ab | 2419 | conf->nr_queued--; |
1da177e4 LT |
2420 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2421 | ||
2422 | mddev = r1_bio->mddev; | |
070ec55d | 2423 | conf = mddev->private; |
4367af55 | 2424 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2425 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2426 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2427 | handle_sync_write_finished(conf, r1_bio); | |
2428 | else | |
4367af55 | 2429 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2430 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2431 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2432 | handle_write_finished(conf, r1_bio); | |
2433 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2434 | handle_read_error(conf, r1_bio); | |
2435 | else | |
d2eb35ac N |
2436 | /* just a partial read to be scheduled from separate |
2437 | * context | |
2438 | */ | |
2439 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2440 | |
1d9d5241 | 2441 | cond_resched(); |
de393cde N |
2442 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2443 | md_check_recovery(mddev); | |
1da177e4 | 2444 | } |
e1dfa0a2 | 2445 | blk_finish_plug(&plug); |
1da177e4 LT |
2446 | } |
2447 | ||
e8096360 | 2448 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2449 | { |
2450 | int buffs; | |
2451 | ||
2452 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2453 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2454 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2455 | conf->poolinfo); | |
2456 | if (!conf->r1buf_pool) | |
2457 | return -ENOMEM; | |
2458 | conf->next_resync = 0; | |
2459 | return 0; | |
2460 | } | |
2461 | ||
2462 | /* | |
2463 | * perform a "sync" on one "block" | |
2464 | * | |
2465 | * We need to make sure that no normal I/O request - particularly write | |
2466 | * requests - conflict with active sync requests. | |
2467 | * | |
2468 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2469 | * that can be installed to exclude normal IO requests. | |
2470 | */ | |
2471 | ||
849674e4 SL |
2472 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2473 | int *skipped) | |
1da177e4 | 2474 | { |
e8096360 | 2475 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2476 | struct r1bio *r1_bio; |
1da177e4 LT |
2477 | struct bio *bio; |
2478 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2479 | int disk = -1; |
1da177e4 | 2480 | int i; |
3e198f78 N |
2481 | int wonly = -1; |
2482 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2483 | sector_t sync_blocks; |
e3b9703e | 2484 | int still_degraded = 0; |
06f60385 N |
2485 | int good_sectors = RESYNC_SECTORS; |
2486 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2487 | |
2488 | if (!conf->r1buf_pool) | |
2489 | if (init_resync(conf)) | |
57afd89f | 2490 | return 0; |
1da177e4 | 2491 | |
58c0fed4 | 2492 | max_sector = mddev->dev_sectors; |
1da177e4 | 2493 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2494 | /* If we aborted, we need to abort the |
2495 | * sync on the 'current' bitmap chunk (there will | |
2496 | * only be one in raid1 resync. | |
2497 | * We can find the current addess in mddev->curr_resync | |
2498 | */ | |
6a806c51 N |
2499 | if (mddev->curr_resync < max_sector) /* aborted */ |
2500 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2501 | &sync_blocks, 1); |
6a806c51 | 2502 | else /* completed sync */ |
191ea9b2 | 2503 | conf->fullsync = 0; |
6a806c51 N |
2504 | |
2505 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2506 | close_sync(conf); |
c40f341f GR |
2507 | |
2508 | if (mddev_is_clustered(mddev)) { | |
2509 | conf->cluster_sync_low = 0; | |
2510 | conf->cluster_sync_high = 0; | |
c40f341f | 2511 | } |
1da177e4 LT |
2512 | return 0; |
2513 | } | |
2514 | ||
07d84d10 N |
2515 | if (mddev->bitmap == NULL && |
2516 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2517 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2518 | conf->fullsync == 0) { |
2519 | *skipped = 1; | |
2520 | return max_sector - sector_nr; | |
2521 | } | |
6394cca5 N |
2522 | /* before building a request, check if we can skip these blocks.. |
2523 | * This call the bitmap_start_sync doesn't actually record anything | |
2524 | */ | |
e3b9703e | 2525 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2526 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2527 | /* We can skip this block, and probably several more */ |
2528 | *skipped = 1; | |
2529 | return sync_blocks; | |
2530 | } | |
17999be4 | 2531 | |
7ac50447 TM |
2532 | /* |
2533 | * If there is non-resync activity waiting for a turn, then let it | |
2534 | * though before starting on this new sync request. | |
2535 | */ | |
2536 | if (conf->nr_waiting) | |
2537 | schedule_timeout_uninterruptible(1); | |
2538 | ||
c40f341f GR |
2539 | /* we are incrementing sector_nr below. To be safe, we check against |
2540 | * sector_nr + two times RESYNC_SECTORS | |
2541 | */ | |
2542 | ||
2543 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
2544 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
1c4588e9 | 2545 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 | 2546 | |
c2fd4c94 | 2547 | raise_barrier(conf, sector_nr); |
1da177e4 | 2548 | |
3e198f78 | 2549 | rcu_read_lock(); |
1da177e4 | 2550 | /* |
3e198f78 N |
2551 | * If we get a correctably read error during resync or recovery, |
2552 | * we might want to read from a different device. So we | |
2553 | * flag all drives that could conceivably be read from for READ, | |
2554 | * and any others (which will be non-In_sync devices) for WRITE. | |
2555 | * If a read fails, we try reading from something else for which READ | |
2556 | * is OK. | |
1da177e4 | 2557 | */ |
1da177e4 | 2558 | |
1da177e4 LT |
2559 | r1_bio->mddev = mddev; |
2560 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2561 | r1_bio->state = 0; |
1da177e4 | 2562 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2563 | |
8f19ccb2 | 2564 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2565 | struct md_rdev *rdev; |
1da177e4 | 2566 | bio = r1_bio->bios[i]; |
2aabaa65 | 2567 | bio_reset(bio); |
1da177e4 | 2568 | |
3e198f78 N |
2569 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2570 | if (rdev == NULL || | |
06f60385 | 2571 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2572 | if (i < conf->raid_disks) |
2573 | still_degraded = 1; | |
3e198f78 | 2574 | } else if (!test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
2575 | bio->bi_rw = WRITE; |
2576 | bio->bi_end_io = end_sync_write; | |
2577 | write_targets ++; | |
3e198f78 N |
2578 | } else { |
2579 | /* may need to read from here */ | |
06f60385 N |
2580 | sector_t first_bad = MaxSector; |
2581 | int bad_sectors; | |
2582 | ||
2583 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2584 | &first_bad, &bad_sectors)) { | |
2585 | if (first_bad > sector_nr) | |
2586 | good_sectors = first_bad - sector_nr; | |
2587 | else { | |
2588 | bad_sectors -= (sector_nr - first_bad); | |
2589 | if (min_bad == 0 || | |
2590 | min_bad > bad_sectors) | |
2591 | min_bad = bad_sectors; | |
2592 | } | |
2593 | } | |
2594 | if (sector_nr < first_bad) { | |
2595 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2596 | if (wonly < 0) | |
2597 | wonly = i; | |
2598 | } else { | |
2599 | if (disk < 0) | |
2600 | disk = i; | |
2601 | } | |
2602 | bio->bi_rw = READ; | |
2603 | bio->bi_end_io = end_sync_read; | |
2604 | read_targets++; | |
d57368af AL |
2605 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2606 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2607 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2608 | /* | |
2609 | * The device is suitable for reading (InSync), | |
2610 | * but has bad block(s) here. Let's try to correct them, | |
2611 | * if we are doing resync or repair. Otherwise, leave | |
2612 | * this device alone for this sync request. | |
2613 | */ | |
2614 | bio->bi_rw = WRITE; | |
2615 | bio->bi_end_io = end_sync_write; | |
2616 | write_targets++; | |
3e198f78 | 2617 | } |
3e198f78 | 2618 | } |
06f60385 N |
2619 | if (bio->bi_end_io) { |
2620 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2621 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
06f60385 N |
2622 | bio->bi_bdev = rdev->bdev; |
2623 | bio->bi_private = r1_bio; | |
2624 | } | |
1da177e4 | 2625 | } |
3e198f78 N |
2626 | rcu_read_unlock(); |
2627 | if (disk < 0) | |
2628 | disk = wonly; | |
2629 | r1_bio->read_disk = disk; | |
191ea9b2 | 2630 | |
06f60385 N |
2631 | if (read_targets == 0 && min_bad > 0) { |
2632 | /* These sectors are bad on all InSync devices, so we | |
2633 | * need to mark them bad on all write targets | |
2634 | */ | |
2635 | int ok = 1; | |
8f19ccb2 | 2636 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2637 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2638 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2639 | ok = rdev_set_badblocks(rdev, sector_nr, |
2640 | min_bad, 0 | |
2641 | ) && ok; | |
2642 | } | |
2643 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2644 | *skipped = 1; | |
2645 | put_buf(r1_bio); | |
2646 | ||
2647 | if (!ok) { | |
2648 | /* Cannot record the badblocks, so need to | |
2649 | * abort the resync. | |
2650 | * If there are multiple read targets, could just | |
2651 | * fail the really bad ones ??? | |
2652 | */ | |
2653 | conf->recovery_disabled = mddev->recovery_disabled; | |
2654 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2655 | return 0; | |
2656 | } else | |
2657 | return min_bad; | |
2658 | ||
2659 | } | |
2660 | if (min_bad > 0 && min_bad < good_sectors) { | |
2661 | /* only resync enough to reach the next bad->good | |
2662 | * transition */ | |
2663 | good_sectors = min_bad; | |
2664 | } | |
2665 | ||
3e198f78 N |
2666 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2667 | /* extra read targets are also write targets */ | |
2668 | write_targets += read_targets-1; | |
2669 | ||
2670 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2671 | /* There is nowhere to write, so all non-sync |
2672 | * drives must be failed - so we are finished | |
2673 | */ | |
b7219ccb N |
2674 | sector_t rv; |
2675 | if (min_bad > 0) | |
2676 | max_sector = sector_nr + min_bad; | |
2677 | rv = max_sector - sector_nr; | |
57afd89f | 2678 | *skipped = 1; |
1da177e4 | 2679 | put_buf(r1_bio); |
1da177e4 LT |
2680 | return rv; |
2681 | } | |
2682 | ||
c6207277 N |
2683 | if (max_sector > mddev->resync_max) |
2684 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2685 | if (max_sector > sector_nr + good_sectors) |
2686 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2687 | nr_sectors = 0; |
289e99e8 | 2688 | sync_blocks = 0; |
1da177e4 LT |
2689 | do { |
2690 | struct page *page; | |
2691 | int len = PAGE_SIZE; | |
2692 | if (sector_nr + (len>>9) > max_sector) | |
2693 | len = (max_sector - sector_nr) << 9; | |
2694 | if (len == 0) | |
2695 | break; | |
6a806c51 N |
2696 | if (sync_blocks == 0) { |
2697 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2698 | &sync_blocks, still_degraded) && |
2699 | !conf->fullsync && | |
2700 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2701 | break; |
7571ae88 | 2702 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2703 | len = sync_blocks<<9; |
ab7a30c7 | 2704 | } |
191ea9b2 | 2705 | |
8f19ccb2 | 2706 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2707 | bio = r1_bio->bios[i]; |
2708 | if (bio->bi_end_io) { | |
d11c171e | 2709 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2710 | if (bio_add_page(bio, page, len, 0) == 0) { |
2711 | /* stop here */ | |
d11c171e | 2712 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2713 | while (i > 0) { |
2714 | i--; | |
2715 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2716 | if (bio->bi_end_io==NULL) |
2717 | continue; | |
1da177e4 LT |
2718 | /* remove last page from this bio */ |
2719 | bio->bi_vcnt--; | |
4f024f37 | 2720 | bio->bi_iter.bi_size -= len; |
b7c44ed9 | 2721 | bio_clear_flag(bio, BIO_SEG_VALID); |
1da177e4 LT |
2722 | } |
2723 | goto bio_full; | |
2724 | } | |
2725 | } | |
2726 | } | |
2727 | nr_sectors += len>>9; | |
2728 | sector_nr += len>>9; | |
191ea9b2 | 2729 | sync_blocks -= (len>>9); |
1da177e4 LT |
2730 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2731 | bio_full: | |
1da177e4 LT |
2732 | r1_bio->sectors = nr_sectors; |
2733 | ||
c40f341f GR |
2734 | if (mddev_is_clustered(mddev) && |
2735 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2736 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2737 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2738 | /* Send resync message */ | |
2739 | md_cluster_ops->resync_info_update(mddev, | |
2740 | conf->cluster_sync_low, | |
2741 | conf->cluster_sync_high); | |
2742 | } | |
2743 | ||
d11c171e N |
2744 | /* For a user-requested sync, we read all readable devices and do a |
2745 | * compare | |
2746 | */ | |
2747 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2748 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2749 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2750 | bio = r1_bio->bios[i]; |
2751 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2752 | read_targets--; |
ddac7c7e | 2753 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
2754 | generic_make_request(bio); |
2755 | } | |
2756 | } | |
2757 | } else { | |
2758 | atomic_set(&r1_bio->remaining, 1); | |
2759 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2760 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 2761 | generic_make_request(bio); |
1da177e4 | 2762 | |
d11c171e | 2763 | } |
1da177e4 LT |
2764 | return nr_sectors; |
2765 | } | |
2766 | ||
fd01b88c | 2767 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2768 | { |
2769 | if (sectors) | |
2770 | return sectors; | |
2771 | ||
2772 | return mddev->dev_sectors; | |
2773 | } | |
2774 | ||
e8096360 | 2775 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2776 | { |
e8096360 | 2777 | struct r1conf *conf; |
709ae487 | 2778 | int i; |
0eaf822c | 2779 | struct raid1_info *disk; |
3cb03002 | 2780 | struct md_rdev *rdev; |
709ae487 | 2781 | int err = -ENOMEM; |
1da177e4 | 2782 | |
e8096360 | 2783 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2784 | if (!conf) |
709ae487 | 2785 | goto abort; |
1da177e4 | 2786 | |
0eaf822c | 2787 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2788 | * mddev->raid_disks * 2, |
1da177e4 LT |
2789 | GFP_KERNEL); |
2790 | if (!conf->mirrors) | |
709ae487 | 2791 | goto abort; |
1da177e4 | 2792 | |
ddaf22ab N |
2793 | conf->tmppage = alloc_page(GFP_KERNEL); |
2794 | if (!conf->tmppage) | |
709ae487 | 2795 | goto abort; |
ddaf22ab | 2796 | |
709ae487 | 2797 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2798 | if (!conf->poolinfo) |
709ae487 | 2799 | goto abort; |
8f19ccb2 | 2800 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2801 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2802 | r1bio_pool_free, | |
2803 | conf->poolinfo); | |
2804 | if (!conf->r1bio_pool) | |
709ae487 N |
2805 | goto abort; |
2806 | ||
ed9bfdf1 | 2807 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2808 | |
c19d5798 | 2809 | err = -EINVAL; |
e7e72bf6 | 2810 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2811 | rdev_for_each(rdev, mddev) { |
aba336bd | 2812 | struct request_queue *q; |
709ae487 | 2813 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2814 | if (disk_idx >= mddev->raid_disks |
2815 | || disk_idx < 0) | |
2816 | continue; | |
c19d5798 | 2817 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2818 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2819 | else |
2820 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2821 | |
c19d5798 N |
2822 | if (disk->rdev) |
2823 | goto abort; | |
1da177e4 | 2824 | disk->rdev = rdev; |
aba336bd | 2825 | q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
2826 | |
2827 | disk->head_position = 0; | |
12cee5a8 | 2828 | disk->seq_start = MaxSector; |
1da177e4 LT |
2829 | } |
2830 | conf->raid_disks = mddev->raid_disks; | |
2831 | conf->mddev = mddev; | |
1da177e4 | 2832 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 2833 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
2834 | |
2835 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2836 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2837 | |
191ea9b2 | 2838 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2839 | conf->pending_count = 0; |
d890fa2b | 2840 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2841 | |
79ef3a8a | 2842 | conf->start_next_window = MaxSector; |
2843 | conf->current_window_requests = conf->next_window_requests = 0; | |
2844 | ||
c19d5798 | 2845 | err = -EIO; |
8f19ccb2 | 2846 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2847 | |
2848 | disk = conf->mirrors + i; | |
2849 | ||
c19d5798 N |
2850 | if (i < conf->raid_disks && |
2851 | disk[conf->raid_disks].rdev) { | |
2852 | /* This slot has a replacement. */ | |
2853 | if (!disk->rdev) { | |
2854 | /* No original, just make the replacement | |
2855 | * a recovering spare | |
2856 | */ | |
2857 | disk->rdev = | |
2858 | disk[conf->raid_disks].rdev; | |
2859 | disk[conf->raid_disks].rdev = NULL; | |
2860 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2861 | /* Original is not in_sync - bad */ | |
2862 | goto abort; | |
2863 | } | |
2864 | ||
5fd6c1dc N |
2865 | if (!disk->rdev || |
2866 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2867 | disk->head_position = 0; |
4f0a5e01 JB |
2868 | if (disk->rdev && |
2869 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2870 | conf->fullsync = 1; |
be4d3280 | 2871 | } |
1da177e4 | 2872 | } |
709ae487 | 2873 | |
709ae487 | 2874 | err = -ENOMEM; |
0232605d | 2875 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
709ae487 N |
2876 | if (!conf->thread) { |
2877 | printk(KERN_ERR | |
9dd1e2fa | 2878 | "md/raid1:%s: couldn't allocate thread\n", |
709ae487 N |
2879 | mdname(mddev)); |
2880 | goto abort; | |
11ce99e6 | 2881 | } |
1da177e4 | 2882 | |
709ae487 N |
2883 | return conf; |
2884 | ||
2885 | abort: | |
2886 | if (conf) { | |
644df1a8 | 2887 | mempool_destroy(conf->r1bio_pool); |
709ae487 N |
2888 | kfree(conf->mirrors); |
2889 | safe_put_page(conf->tmppage); | |
2890 | kfree(conf->poolinfo); | |
2891 | kfree(conf); | |
2892 | } | |
2893 | return ERR_PTR(err); | |
2894 | } | |
2895 | ||
afa0f557 | 2896 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 2897 | static int raid1_run(struct mddev *mddev) |
709ae487 | 2898 | { |
e8096360 | 2899 | struct r1conf *conf; |
709ae487 | 2900 | int i; |
3cb03002 | 2901 | struct md_rdev *rdev; |
5220ea1e | 2902 | int ret; |
2ff8cc2c | 2903 | bool discard_supported = false; |
709ae487 N |
2904 | |
2905 | if (mddev->level != 1) { | |
9dd1e2fa | 2906 | printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n", |
709ae487 N |
2907 | mdname(mddev), mddev->level); |
2908 | return -EIO; | |
2909 | } | |
2910 | if (mddev->reshape_position != MaxSector) { | |
9dd1e2fa | 2911 | printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n", |
709ae487 N |
2912 | mdname(mddev)); |
2913 | return -EIO; | |
2914 | } | |
1da177e4 | 2915 | /* |
709ae487 N |
2916 | * copy the already verified devices into our private RAID1 |
2917 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 2918 | * should be freed in raid1_free()] |
1da177e4 | 2919 | */ |
709ae487 N |
2920 | if (mddev->private == NULL) |
2921 | conf = setup_conf(mddev); | |
2922 | else | |
2923 | conf = mddev->private; | |
1da177e4 | 2924 | |
709ae487 N |
2925 | if (IS_ERR(conf)) |
2926 | return PTR_ERR(conf); | |
1da177e4 | 2927 | |
c8dc9c65 | 2928 | if (mddev->queue) |
5026d7a9 PA |
2929 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
2930 | ||
dafb20fa | 2931 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
2932 | if (!mddev->gendisk) |
2933 | continue; | |
709ae487 N |
2934 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2935 | rdev->data_offset << 9); | |
2ff8cc2c SL |
2936 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2937 | discard_supported = true; | |
1da177e4 | 2938 | } |
191ea9b2 | 2939 | |
709ae487 N |
2940 | mddev->degraded = 0; |
2941 | for (i=0; i < conf->raid_disks; i++) | |
2942 | if (conf->mirrors[i].rdev == NULL || | |
2943 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
2944 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
2945 | mddev->degraded++; | |
2946 | ||
2947 | if (conf->raid_disks - mddev->degraded == 1) | |
2948 | mddev->recovery_cp = MaxSector; | |
2949 | ||
8c6ac868 | 2950 | if (mddev->recovery_cp != MaxSector) |
9dd1e2fa | 2951 | printk(KERN_NOTICE "md/raid1:%s: not clean" |
8c6ac868 AN |
2952 | " -- starting background reconstruction\n", |
2953 | mdname(mddev)); | |
f72ffdd6 | 2954 | printk(KERN_INFO |
9dd1e2fa | 2955 | "md/raid1:%s: active with %d out of %d mirrors\n", |
f72ffdd6 | 2956 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 2957 | mddev->raid_disks); |
709ae487 | 2958 | |
1da177e4 LT |
2959 | /* |
2960 | * Ok, everything is just fine now | |
2961 | */ | |
709ae487 N |
2962 | mddev->thread = conf->thread; |
2963 | conf->thread = NULL; | |
2964 | mddev->private = conf; | |
2965 | ||
1f403624 | 2966 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2967 | |
1ed7242e | 2968 | if (mddev->queue) { |
2ff8cc2c SL |
2969 | if (discard_supported) |
2970 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
2971 | mddev->queue); | |
2972 | else | |
2973 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
2974 | mddev->queue); | |
1ed7242e | 2975 | } |
5220ea1e | 2976 | |
2977 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
2978 | if (ret) { |
2979 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 2980 | raid1_free(mddev, conf); |
5aa61f42 | 2981 | } |
5220ea1e | 2982 | return ret; |
1da177e4 LT |
2983 | } |
2984 | ||
afa0f557 | 2985 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 2986 | { |
afa0f557 | 2987 | struct r1conf *conf = priv; |
409c57f3 | 2988 | |
644df1a8 | 2989 | mempool_destroy(conf->r1bio_pool); |
990a8baf | 2990 | kfree(conf->mirrors); |
0fea7ed8 | 2991 | safe_put_page(conf->tmppage); |
990a8baf | 2992 | kfree(conf->poolinfo); |
1da177e4 | 2993 | kfree(conf); |
1da177e4 LT |
2994 | } |
2995 | ||
fd01b88c | 2996 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
2997 | { |
2998 | /* no resync is happening, and there is enough space | |
2999 | * on all devices, so we can resize. | |
3000 | * We need to make sure resync covers any new space. | |
3001 | * If the array is shrinking we should possibly wait until | |
3002 | * any io in the removed space completes, but it hardly seems | |
3003 | * worth it. | |
3004 | */ | |
a4a6125a N |
3005 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3006 | if (mddev->external_size && | |
3007 | mddev->array_sectors > newsize) | |
b522adcd | 3008 | return -EINVAL; |
a4a6125a N |
3009 | if (mddev->bitmap) { |
3010 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3011 | if (ret) | |
3012 | return ret; | |
3013 | } | |
3014 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 3015 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 3016 | revalidate_disk(mddev->gendisk); |
b522adcd | 3017 | if (sectors > mddev->dev_sectors && |
b098636c | 3018 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3019 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3020 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3021 | } | |
b522adcd | 3022 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3023 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3024 | return 0; |
3025 | } | |
3026 | ||
fd01b88c | 3027 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3028 | { |
3029 | /* We need to: | |
3030 | * 1/ resize the r1bio_pool | |
3031 | * 2/ resize conf->mirrors | |
3032 | * | |
3033 | * We allocate a new r1bio_pool if we can. | |
3034 | * Then raise a device barrier and wait until all IO stops. | |
3035 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3036 | * |
3037 | * At the same time, we "pack" the devices so that all the missing | |
3038 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3039 | */ |
3040 | mempool_t *newpool, *oldpool; | |
3041 | struct pool_info *newpoolinfo; | |
0eaf822c | 3042 | struct raid1_info *newmirrors; |
e8096360 | 3043 | struct r1conf *conf = mddev->private; |
63c70c4f | 3044 | int cnt, raid_disks; |
c04be0aa | 3045 | unsigned long flags; |
b5470dc5 | 3046 | int d, d2, err; |
1da177e4 | 3047 | |
63c70c4f | 3048 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3049 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3050 | mddev->layout != mddev->new_layout || |
3051 | mddev->level != mddev->new_level) { | |
664e7c41 | 3052 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3053 | mddev->new_layout = mddev->layout; |
3054 | mddev->new_level = mddev->level; | |
3055 | return -EINVAL; | |
3056 | } | |
3057 | ||
28c1b9fd GR |
3058 | if (!mddev_is_clustered(mddev)) { |
3059 | err = md_allow_write(mddev); | |
3060 | if (err) | |
3061 | return err; | |
3062 | } | |
2a2275d6 | 3063 | |
63c70c4f N |
3064 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3065 | ||
6ea9c07c N |
3066 | if (raid_disks < conf->raid_disks) { |
3067 | cnt=0; | |
3068 | for (d= 0; d < conf->raid_disks; d++) | |
3069 | if (conf->mirrors[d].rdev) | |
3070 | cnt++; | |
3071 | if (cnt > raid_disks) | |
1da177e4 | 3072 | return -EBUSY; |
6ea9c07c | 3073 | } |
1da177e4 LT |
3074 | |
3075 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3076 | if (!newpoolinfo) | |
3077 | return -ENOMEM; | |
3078 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3079 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3080 | |
3081 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3082 | r1bio_pool_free, newpoolinfo); | |
3083 | if (!newpool) { | |
3084 | kfree(newpoolinfo); | |
3085 | return -ENOMEM; | |
3086 | } | |
0eaf822c | 3087 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3088 | GFP_KERNEL); |
1da177e4 LT |
3089 | if (!newmirrors) { |
3090 | kfree(newpoolinfo); | |
3091 | mempool_destroy(newpool); | |
3092 | return -ENOMEM; | |
3093 | } | |
1da177e4 | 3094 | |
e2d59925 | 3095 | freeze_array(conf, 0); |
1da177e4 LT |
3096 | |
3097 | /* ok, everything is stopped */ | |
3098 | oldpool = conf->r1bio_pool; | |
3099 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3100 | |
a88aa786 | 3101 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3102 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3103 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3104 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3105 | rdev->raid_disk = d2; |
36fad858 NK |
3106 | sysfs_unlink_rdev(mddev, rdev); |
3107 | if (sysfs_link_rdev(mddev, rdev)) | |
a88aa786 | 3108 | printk(KERN_WARNING |
36fad858 NK |
3109 | "md/raid1:%s: cannot register rd%d\n", |
3110 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3111 | } |
a88aa786 N |
3112 | if (rdev) |
3113 | newmirrors[d2++].rdev = rdev; | |
3114 | } | |
1da177e4 LT |
3115 | kfree(conf->mirrors); |
3116 | conf->mirrors = newmirrors; | |
3117 | kfree(conf->poolinfo); | |
3118 | conf->poolinfo = newpoolinfo; | |
3119 | ||
c04be0aa | 3120 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3121 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3122 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3123 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3124 | mddev->delta_disks = 0; |
1da177e4 | 3125 | |
e2d59925 | 3126 | unfreeze_array(conf); |
1da177e4 | 3127 | |
985ca973 | 3128 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3129 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3130 | md_wakeup_thread(mddev->thread); | |
3131 | ||
3132 | mempool_destroy(oldpool); | |
3133 | return 0; | |
3134 | } | |
3135 | ||
fd01b88c | 3136 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3137 | { |
e8096360 | 3138 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3139 | |
3140 | switch(state) { | |
6eef4b21 N |
3141 | case 2: /* wake for suspend */ |
3142 | wake_up(&conf->wait_barrier); | |
3143 | break; | |
9e6603da | 3144 | case 1: |
07169fd4 | 3145 | freeze_array(conf, 0); |
36fa3063 | 3146 | break; |
9e6603da | 3147 | case 0: |
07169fd4 | 3148 | unfreeze_array(conf); |
36fa3063 N |
3149 | break; |
3150 | } | |
36fa3063 N |
3151 | } |
3152 | ||
fd01b88c | 3153 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3154 | { |
3155 | /* raid1 can take over: | |
3156 | * raid5 with 2 devices, any layout or chunk size | |
3157 | */ | |
3158 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3159 | struct r1conf *conf; |
709ae487 N |
3160 | mddev->new_level = 1; |
3161 | mddev->new_layout = 0; | |
3162 | mddev->new_chunk_sectors = 0; | |
3163 | conf = setup_conf(mddev); | |
3164 | if (!IS_ERR(conf)) | |
07169fd4 | 3165 | /* Array must appear to be quiesced */ |
3166 | conf->array_frozen = 1; | |
709ae487 N |
3167 | return conf; |
3168 | } | |
3169 | return ERR_PTR(-EINVAL); | |
3170 | } | |
1da177e4 | 3171 | |
84fc4b56 | 3172 | static struct md_personality raid1_personality = |
1da177e4 LT |
3173 | { |
3174 | .name = "raid1", | |
2604b703 | 3175 | .level = 1, |
1da177e4 | 3176 | .owner = THIS_MODULE, |
849674e4 SL |
3177 | .make_request = raid1_make_request, |
3178 | .run = raid1_run, | |
afa0f557 | 3179 | .free = raid1_free, |
849674e4 SL |
3180 | .status = raid1_status, |
3181 | .error_handler = raid1_error, | |
1da177e4 LT |
3182 | .hot_add_disk = raid1_add_disk, |
3183 | .hot_remove_disk= raid1_remove_disk, | |
3184 | .spare_active = raid1_spare_active, | |
849674e4 | 3185 | .sync_request = raid1_sync_request, |
1da177e4 | 3186 | .resize = raid1_resize, |
80c3a6ce | 3187 | .size = raid1_size, |
63c70c4f | 3188 | .check_reshape = raid1_reshape, |
36fa3063 | 3189 | .quiesce = raid1_quiesce, |
709ae487 | 3190 | .takeover = raid1_takeover, |
5c675f83 | 3191 | .congested = raid1_congested, |
1da177e4 LT |
3192 | }; |
3193 | ||
3194 | static int __init raid_init(void) | |
3195 | { | |
2604b703 | 3196 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3197 | } |
3198 | ||
3199 | static void raid_exit(void) | |
3200 | { | |
2604b703 | 3201 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3202 | } |
3203 | ||
3204 | module_init(raid_init); | |
3205 | module_exit(raid_exit); | |
3206 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3207 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3208 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3209 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3210 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3211 | |
3212 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |