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> |
109e3765 | 40 | #include <trace/events/block.h> |
43b2e5d8 | 41 | #include "md.h" |
ef740c37 CH |
42 | #include "raid1.h" |
43 | #include "bitmap.h" | |
191ea9b2 | 44 | |
1da177e4 LT |
45 | /* |
46 | * Number of guaranteed r1bios in case of extreme VM load: | |
47 | */ | |
48 | #define NR_RAID1_BIOS 256 | |
49 | ||
473e87ce JB |
50 | /* when we get a read error on a read-only array, we redirect to another |
51 | * device without failing the first device, or trying to over-write to | |
52 | * correct the read error. To keep track of bad blocks on a per-bio | |
53 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
54 | */ | |
55 | #define IO_BLOCKED ((struct bio *)1) | |
56 | /* When we successfully write to a known bad-block, we need to remove the | |
57 | * bad-block marking which must be done from process context. So we record | |
58 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
59 | */ | |
60 | #define IO_MADE_GOOD ((struct bio *)2) | |
61 | ||
62 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
63 | ||
34db0cd6 N |
64 | /* When there are this many requests queue to be written by |
65 | * the raid1 thread, we become 'congested' to provide back-pressure | |
66 | * for writeback. | |
67 | */ | |
68 | static int max_queued_requests = 1024; | |
1da177e4 | 69 | |
79ef3a8a | 70 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
71 | sector_t bi_sector); | |
e8096360 | 72 | static void lower_barrier(struct r1conf *conf); |
1da177e4 | 73 | |
578b54ad N |
74 | #define raid1_log(md, fmt, args...) \ |
75 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) | |
76 | ||
dd0fc66f | 77 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
78 | { |
79 | struct pool_info *pi = data; | |
9f2c9d12 | 80 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
81 | |
82 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 83 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
84 | } |
85 | ||
86 | static void r1bio_pool_free(void *r1_bio, void *data) | |
87 | { | |
88 | kfree(r1_bio); | |
89 | } | |
90 | ||
91 | #define RESYNC_BLOCK_SIZE (64*1024) | |
8e005f7c | 92 | #define RESYNC_DEPTH 32 |
1da177e4 LT |
93 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
94 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
8e005f7c | 95 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
96 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
97 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
98 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
8e005f7c | 99 | #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) |
1da177e4 | 100 | |
dd0fc66f | 101 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
102 | { |
103 | struct pool_info *pi = data; | |
9f2c9d12 | 104 | struct r1bio *r1_bio; |
1da177e4 | 105 | struct bio *bio; |
da1aab3d | 106 | int need_pages; |
1da177e4 LT |
107 | int i, j; |
108 | ||
109 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 110 | if (!r1_bio) |
1da177e4 | 111 | return NULL; |
1da177e4 LT |
112 | |
113 | /* | |
114 | * Allocate bios : 1 for reading, n-1 for writing | |
115 | */ | |
116 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 117 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
118 | if (!bio) |
119 | goto out_free_bio; | |
120 | r1_bio->bios[j] = bio; | |
121 | } | |
122 | /* | |
123 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
124 | * the first bio. |
125 | * If this is a user-requested check/repair, allocate | |
126 | * RESYNC_PAGES for each bio. | |
1da177e4 | 127 | */ |
d11c171e | 128 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 129 | need_pages = pi->raid_disks; |
d11c171e | 130 | else |
da1aab3d N |
131 | need_pages = 1; |
132 | for (j = 0; j < need_pages; j++) { | |
d11c171e | 133 | bio = r1_bio->bios[j]; |
a0787606 | 134 | bio->bi_vcnt = RESYNC_PAGES; |
d11c171e | 135 | |
a0787606 | 136 | if (bio_alloc_pages(bio, gfp_flags)) |
da1aab3d | 137 | goto out_free_pages; |
d11c171e N |
138 | } |
139 | /* If not user-requests, copy the page pointers to all bios */ | |
140 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
141 | for (i=0; i<RESYNC_PAGES ; i++) | |
142 | for (j=1; j<pi->raid_disks; j++) | |
143 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
144 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
145 | } |
146 | ||
147 | r1_bio->master_bio = NULL; | |
148 | ||
149 | return r1_bio; | |
150 | ||
da1aab3d | 151 | out_free_pages: |
491221f8 GJ |
152 | while (--j >= 0) |
153 | bio_free_pages(r1_bio->bios[j]); | |
da1aab3d | 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]; | |
1d41c216 N |
354 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
355 | mdname(conf->mddev), | |
356 | bdevname(rdev->bdev, b), | |
357 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 358 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 359 | reschedule_retry(r1_bio); |
7ad4d4a6 | 360 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 361 | } |
1da177e4 LT |
362 | } |
363 | ||
9f2c9d12 | 364 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
365 | { |
366 | /* it really is the end of this request */ | |
367 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
368 | /* free extra copy of the data pages */ | |
369 | int i = r1_bio->behind_page_count; | |
370 | while (i--) | |
371 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
372 | kfree(r1_bio->behind_bvecs); | |
373 | r1_bio->behind_bvecs = NULL; | |
374 | } | |
375 | /* clear the bitmap if all writes complete successfully */ | |
376 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
377 | r1_bio->sectors, | |
378 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
379 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
380 | md_write_end(r1_bio->mddev); | |
381 | } | |
382 | ||
9f2c9d12 | 383 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 384 | { |
cd5ff9a1 N |
385 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
386 | return; | |
387 | ||
388 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
389 | reschedule_retry(r1_bio); | |
390 | else { | |
391 | close_write(r1_bio); | |
4367af55 N |
392 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
393 | reschedule_retry(r1_bio); | |
394 | else | |
395 | raid_end_bio_io(r1_bio); | |
4e78064f N |
396 | } |
397 | } | |
398 | ||
4246a0b6 | 399 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 400 | { |
9f2c9d12 | 401 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 402 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 403 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 404 | struct bio *to_put = NULL; |
e5872d58 N |
405 | int mirror = find_bio_disk(r1_bio, bio); |
406 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd SL |
407 | bool discard_error; |
408 | ||
409 | discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; | |
1da177e4 | 410 | |
e9c7469b TH |
411 | /* |
412 | * 'one mirror IO has finished' event handler: | |
413 | */ | |
e3f948cd | 414 | if (bio->bi_error && !discard_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, |
e3f948cd | 451 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
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; | |
5e2c7a36 | 749 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 750 | bio->bi_next = NULL; |
5e2c7a36 N |
751 | bio->bi_bdev = rdev->bdev; |
752 | if (test_bit(Faulty, &rdev->flags)) { | |
753 | bio->bi_error = -EIO; | |
754 | bio_endio(bio); | |
755 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
756 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
2ff8cc2c | 757 | /* Just ignore it */ |
4246a0b6 | 758 | bio_endio(bio); |
2ff8cc2c SL |
759 | else |
760 | generic_make_request(bio); | |
a35e63ef N |
761 | bio = next; |
762 | } | |
a35e63ef N |
763 | } else |
764 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
765 | } |
766 | ||
17999be4 N |
767 | /* Barriers.... |
768 | * Sometimes we need to suspend IO while we do something else, | |
769 | * either some resync/recovery, or reconfigure the array. | |
770 | * To do this we raise a 'barrier'. | |
771 | * The 'barrier' is a counter that can be raised multiple times | |
772 | * to count how many activities are happening which preclude | |
773 | * normal IO. | |
774 | * We can only raise the barrier if there is no pending IO. | |
775 | * i.e. if nr_pending == 0. | |
776 | * We choose only to raise the barrier if no-one is waiting for the | |
777 | * barrier to go down. This means that as soon as an IO request | |
778 | * is ready, no other operations which require a barrier will start | |
779 | * until the IO request has had a chance. | |
780 | * | |
781 | * So: regular IO calls 'wait_barrier'. When that returns there | |
782 | * is no backgroup IO happening, It must arrange to call | |
783 | * allow_barrier when it has finished its IO. | |
784 | * backgroup IO calls must call raise_barrier. Once that returns | |
785 | * there is no normal IO happeing. It must arrange to call | |
786 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 787 | */ |
c2fd4c94 | 788 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 LT |
789 | { |
790 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
791 | |
792 | /* Wait until no block IO is waiting */ | |
793 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
eed8c02e | 794 | conf->resync_lock); |
17999be4 N |
795 | |
796 | /* block any new IO from starting */ | |
797 | conf->barrier++; | |
c2fd4c94 | 798 | conf->next_resync = sector_nr; |
17999be4 | 799 | |
79ef3a8a | 800 | /* For these conditions we must wait: |
801 | * A: while the array is in frozen state | |
802 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach | |
803 | * the max count which allowed. | |
804 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning | |
805 | * next resync will reach to the window which normal bios are | |
806 | * handling. | |
2f73d3c5 | 807 | * D: while there are any active requests in the current window. |
79ef3a8a | 808 | */ |
17999be4 | 809 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 810 | !conf->array_frozen && |
79ef3a8a | 811 | conf->barrier < RESYNC_DEPTH && |
2f73d3c5 | 812 | conf->current_window_requests == 0 && |
79ef3a8a | 813 | (conf->start_next_window >= |
814 | conf->next_resync + RESYNC_SECTORS), | |
eed8c02e | 815 | conf->resync_lock); |
17999be4 | 816 | |
34e97f17 | 817 | conf->nr_pending++; |
17999be4 N |
818 | spin_unlock_irq(&conf->resync_lock); |
819 | } | |
820 | ||
e8096360 | 821 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
822 | { |
823 | unsigned long flags; | |
709ae487 | 824 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
825 | spin_lock_irqsave(&conf->resync_lock, flags); |
826 | conf->barrier--; | |
34e97f17 | 827 | conf->nr_pending--; |
17999be4 N |
828 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
829 | wake_up(&conf->wait_barrier); | |
830 | } | |
831 | ||
79ef3a8a | 832 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
17999be4 | 833 | { |
79ef3a8a | 834 | bool wait = false; |
835 | ||
836 | if (conf->array_frozen || !bio) | |
837 | wait = true; | |
838 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { | |
23554960 N |
839 | if ((conf->mddev->curr_resync_completed |
840 | >= bio_end_sector(bio)) || | |
f2c771a6 | 841 | (conf->start_next_window + NEXT_NORMALIO_DISTANCE |
23554960 | 842 | <= bio->bi_iter.bi_sector)) |
79ef3a8a | 843 | wait = false; |
844 | else | |
845 | wait = true; | |
846 | } | |
847 | ||
848 | return wait; | |
849 | } | |
850 | ||
851 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) | |
852 | { | |
853 | sector_t sector = 0; | |
854 | ||
17999be4 | 855 | spin_lock_irq(&conf->resync_lock); |
79ef3a8a | 856 | if (need_to_wait_for_sync(conf, bio)) { |
17999be4 | 857 | conf->nr_waiting++; |
d6b42dcb N |
858 | /* Wait for the barrier to drop. |
859 | * However if there are already pending | |
860 | * requests (preventing the barrier from | |
861 | * rising completely), and the | |
5965b642 | 862 | * per-process bio queue isn't empty, |
d6b42dcb | 863 | * then don't wait, as we need to empty |
5965b642 N |
864 | * that queue to allow conf->start_next_window |
865 | * to increase. | |
d6b42dcb | 866 | */ |
578b54ad | 867 | raid1_log(conf->mddev, "wait barrier"); |
d6b42dcb | 868 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 869 | !conf->array_frozen && |
870 | (!conf->barrier || | |
5965b642 N |
871 | ((conf->start_next_window < |
872 | conf->next_resync + RESYNC_SECTORS) && | |
873 | current->bio_list && | |
874 | !bio_list_empty(current->bio_list))), | |
eed8c02e | 875 | conf->resync_lock); |
17999be4 | 876 | conf->nr_waiting--; |
1da177e4 | 877 | } |
79ef3a8a | 878 | |
879 | if (bio && bio_data_dir(bio) == WRITE) { | |
e8ff8bf0 | 880 | if (bio->bi_iter.bi_sector >= conf->next_resync) { |
79ef3a8a | 881 | if (conf->start_next_window == MaxSector) |
882 | conf->start_next_window = | |
883 | conf->next_resync + | |
884 | NEXT_NORMALIO_DISTANCE; | |
885 | ||
886 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) | |
4f024f37 | 887 | <= bio->bi_iter.bi_sector) |
79ef3a8a | 888 | conf->next_window_requests++; |
889 | else | |
890 | conf->current_window_requests++; | |
79ef3a8a | 891 | sector = conf->start_next_window; |
41a336e0 | 892 | } |
79ef3a8a | 893 | } |
894 | ||
17999be4 | 895 | conf->nr_pending++; |
1da177e4 | 896 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 897 | return sector; |
1da177e4 LT |
898 | } |
899 | ||
79ef3a8a | 900 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
901 | sector_t bi_sector) | |
17999be4 N |
902 | { |
903 | unsigned long flags; | |
79ef3a8a | 904 | |
17999be4 N |
905 | spin_lock_irqsave(&conf->resync_lock, flags); |
906 | conf->nr_pending--; | |
79ef3a8a | 907 | if (start_next_window) { |
908 | if (start_next_window == conf->start_next_window) { | |
909 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE | |
910 | <= bi_sector) | |
911 | conf->next_window_requests--; | |
912 | else | |
913 | conf->current_window_requests--; | |
914 | } else | |
915 | conf->current_window_requests--; | |
916 | ||
917 | if (!conf->current_window_requests) { | |
918 | if (conf->next_window_requests) { | |
919 | conf->current_window_requests = | |
920 | conf->next_window_requests; | |
921 | conf->next_window_requests = 0; | |
922 | conf->start_next_window += | |
923 | NEXT_NORMALIO_DISTANCE; | |
924 | } else | |
925 | conf->start_next_window = MaxSector; | |
926 | } | |
927 | } | |
17999be4 N |
928 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
929 | wake_up(&conf->wait_barrier); | |
930 | } | |
931 | ||
e2d59925 | 932 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab N |
933 | { |
934 | /* stop syncio and normal IO and wait for everything to | |
935 | * go quite. | |
b364e3d0 | 936 | * We wait until nr_pending match nr_queued+extra |
1c830532 N |
937 | * This is called in the context of one normal IO request |
938 | * that has failed. Thus any sync request that might be pending | |
939 | * will be blocked by nr_pending, and we need to wait for | |
940 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 941 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
942 | * must match the number of pending IOs (nr_pending) before |
943 | * we continue. | |
ddaf22ab N |
944 | */ |
945 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 946 | conf->array_frozen = 1; |
578b54ad | 947 | raid1_log(conf->mddev, "wait freeze"); |
eed8c02e | 948 | wait_event_lock_irq_cmd(conf->wait_barrier, |
e2d59925 | 949 | conf->nr_pending == conf->nr_queued+extra, |
eed8c02e LC |
950 | conf->resync_lock, |
951 | flush_pending_writes(conf)); | |
ddaf22ab N |
952 | spin_unlock_irq(&conf->resync_lock); |
953 | } | |
e8096360 | 954 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
955 | { |
956 | /* reverse the effect of the freeze */ | |
957 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 958 | conf->array_frozen = 0; |
ddaf22ab N |
959 | wake_up(&conf->wait_barrier); |
960 | spin_unlock_irq(&conf->resync_lock); | |
961 | } | |
962 | ||
f72ffdd6 | 963 | /* duplicate the data pages for behind I/O |
4e78064f | 964 | */ |
9f2c9d12 | 965 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
966 | { |
967 | int i; | |
968 | struct bio_vec *bvec; | |
2ca68f5e | 969 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 970 | GFP_NOIO); |
2ca68f5e | 971 | if (unlikely(!bvecs)) |
af6d7b76 | 972 | return; |
4b6d287f | 973 | |
cb34e057 | 974 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
975 | bvecs[i] = *bvec; |
976 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
977 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 978 | goto do_sync_io; |
2ca68f5e N |
979 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
980 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
981 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
982 | kunmap(bvec->bv_page); |
983 | } | |
2ca68f5e | 984 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
985 | r1_bio->behind_page_count = bio->bi_vcnt; |
986 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
987 | return; | |
4b6d287f N |
988 | |
989 | do_sync_io: | |
af6d7b76 | 990 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
991 | if (bvecs[i].bv_page) |
992 | put_page(bvecs[i].bv_page); | |
993 | kfree(bvecs); | |
4f024f37 KO |
994 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
995 | bio->bi_iter.bi_size); | |
4b6d287f N |
996 | } |
997 | ||
f54a9d0e N |
998 | struct raid1_plug_cb { |
999 | struct blk_plug_cb cb; | |
1000 | struct bio_list pending; | |
1001 | int pending_cnt; | |
1002 | }; | |
1003 | ||
1004 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1005 | { | |
1006 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1007 | cb); | |
1008 | struct mddev *mddev = plug->cb.data; | |
1009 | struct r1conf *conf = mddev->private; | |
1010 | struct bio *bio; | |
1011 | ||
874807a8 | 1012 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1013 | spin_lock_irq(&conf->device_lock); |
1014 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1015 | conf->pending_count += plug->pending_cnt; | |
1016 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1017 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1018 | md_wakeup_thread(mddev->thread); |
1019 | kfree(plug); | |
1020 | return; | |
1021 | } | |
1022 | ||
1023 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1024 | bio = bio_list_get(&plug->pending); | |
1025 | bitmap_unplug(mddev->bitmap); | |
1026 | wake_up(&conf->wait_barrier); | |
1027 | ||
1028 | while (bio) { /* submit pending writes */ | |
1029 | struct bio *next = bio->bi_next; | |
5e2c7a36 | 1030 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
f54a9d0e | 1031 | bio->bi_next = NULL; |
5e2c7a36 N |
1032 | bio->bi_bdev = rdev->bdev; |
1033 | if (test_bit(Faulty, &rdev->flags)) { | |
1034 | bio->bi_error = -EIO; | |
1035 | bio_endio(bio); | |
1036 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
1037 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
32f9f570 | 1038 | /* Just ignore it */ |
4246a0b6 | 1039 | bio_endio(bio); |
32f9f570 SL |
1040 | else |
1041 | generic_make_request(bio); | |
f54a9d0e N |
1042 | bio = next; |
1043 | } | |
1044 | kfree(plug); | |
1045 | } | |
1046 | ||
849674e4 | 1047 | static void raid1_make_request(struct mddev *mddev, struct bio * bio) |
1da177e4 | 1048 | { |
e8096360 | 1049 | struct r1conf *conf = mddev->private; |
0eaf822c | 1050 | struct raid1_info *mirror; |
9f2c9d12 | 1051 | struct r1bio *r1_bio; |
1da177e4 | 1052 | struct bio *read_bio; |
1f68f0c4 | 1053 | int i, disks; |
84255d10 | 1054 | struct bitmap *bitmap; |
191ea9b2 | 1055 | unsigned long flags; |
796a5cf0 | 1056 | const int op = bio_op(bio); |
a362357b | 1057 | const int rw = bio_data_dir(bio); |
1eff9d32 JA |
1058 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1059 | const unsigned long do_flush_fua = (bio->bi_opf & | |
28a8f0d3 | 1060 | (REQ_PREFLUSH | REQ_FUA)); |
3cb03002 | 1061 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1062 | struct blk_plug_cb *cb; |
1063 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
1064 | int first_clone; |
1065 | int sectors_handled; | |
1066 | int max_sectors; | |
79ef3a8a | 1067 | sector_t start_next_window; |
191ea9b2 | 1068 | |
1da177e4 LT |
1069 | /* |
1070 | * Register the new request and wait if the reconstruction | |
1071 | * thread has put up a bar for new requests. | |
1072 | * Continue immediately if no resync is active currently. | |
1073 | */ | |
62de608d | 1074 | |
3d310eb7 N |
1075 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1076 | ||
6eef4b21 | 1077 | if (bio_data_dir(bio) == WRITE && |
589a1c49 GR |
1078 | ((bio_end_sector(bio) > mddev->suspend_lo && |
1079 | bio->bi_iter.bi_sector < mddev->suspend_hi) || | |
1080 | (mddev_is_clustered(mddev) && | |
90382ed9 GR |
1081 | md_cluster_ops->area_resyncing(mddev, WRITE, |
1082 | bio->bi_iter.bi_sector, bio_end_sector(bio))))) { | |
6eef4b21 N |
1083 | /* As the suspend_* range is controlled by |
1084 | * userspace, we want an interruptible | |
1085 | * wait. | |
1086 | */ | |
1087 | DEFINE_WAIT(w); | |
1088 | for (;;) { | |
1089 | flush_signals(current); | |
1090 | prepare_to_wait(&conf->wait_barrier, | |
1091 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1092 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
589a1c49 GR |
1093 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
1094 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1095 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
589a1c49 | 1096 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) |
6eef4b21 N |
1097 | break; |
1098 | schedule(); | |
1099 | } | |
1100 | finish_wait(&conf->wait_barrier, &w); | |
1101 | } | |
62de608d | 1102 | |
79ef3a8a | 1103 | start_next_window = wait_barrier(conf, bio); |
1da177e4 | 1104 | |
84255d10 N |
1105 | bitmap = mddev->bitmap; |
1106 | ||
1da177e4 | 1107 | /* |
70246286 | 1108 | * make_request() can abort the operation when read-ahead is being |
1da177e4 LT |
1109 | * used and no empty request is available. |
1110 | * | |
1111 | */ | |
1112 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1113 | ||
1114 | r1_bio->master_bio = bio; | |
aa8b57aa | 1115 | r1_bio->sectors = bio_sectors(bio); |
191ea9b2 | 1116 | r1_bio->state = 0; |
1da177e4 | 1117 | r1_bio->mddev = mddev; |
4f024f37 | 1118 | r1_bio->sector = bio->bi_iter.bi_sector; |
1da177e4 | 1119 | |
d2eb35ac N |
1120 | /* We might need to issue multiple reads to different |
1121 | * devices if there are bad blocks around, so we keep | |
1122 | * track of the number of reads in bio->bi_phys_segments. | |
1123 | * If this is 0, there is only one r1_bio and no locking | |
1124 | * will be needed when requests complete. If it is | |
1125 | * non-zero, then it is the number of not-completed requests. | |
1126 | */ | |
1127 | bio->bi_phys_segments = 0; | |
b7c44ed9 | 1128 | bio_clear_flag(bio, BIO_SEG_VALID); |
d2eb35ac | 1129 | |
a362357b | 1130 | if (rw == READ) { |
1da177e4 LT |
1131 | /* |
1132 | * read balancing logic: | |
1133 | */ | |
d2eb35ac N |
1134 | int rdisk; |
1135 | ||
1136 | read_again: | |
1137 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1da177e4 LT |
1138 | |
1139 | if (rdisk < 0) { | |
1140 | /* couldn't find anywhere to read from */ | |
1141 | raid_end_bio_io(r1_bio); | |
5a7bbad2 | 1142 | return; |
1da177e4 LT |
1143 | } |
1144 | mirror = conf->mirrors + rdisk; | |
1145 | ||
e555190d N |
1146 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1147 | bitmap) { | |
1148 | /* Reading from a write-mostly device must | |
1149 | * take care not to over-take any writes | |
1150 | * that are 'behind' | |
1151 | */ | |
578b54ad | 1152 | raid1_log(mddev, "wait behind writes"); |
e555190d N |
1153 | wait_event(bitmap->behind_wait, |
1154 | atomic_read(&bitmap->behind_writes) == 0); | |
1155 | } | |
1da177e4 | 1156 | r1_bio->read_disk = rdisk; |
f0cc9a05 | 1157 | r1_bio->start_next_window = 0; |
1da177e4 | 1158 | |
a167f663 | 1159 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1160 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1161 | max_sectors); |
1da177e4 LT |
1162 | |
1163 | r1_bio->bios[rdisk] = read_bio; | |
1164 | ||
4f024f37 KO |
1165 | read_bio->bi_iter.bi_sector = r1_bio->sector + |
1166 | mirror->rdev->data_offset; | |
1da177e4 LT |
1167 | read_bio->bi_bdev = mirror->rdev->bdev; |
1168 | read_bio->bi_end_io = raid1_end_read_request; | |
796a5cf0 | 1169 | bio_set_op_attrs(read_bio, op, do_sync); |
1da177e4 LT |
1170 | read_bio->bi_private = r1_bio; |
1171 | ||
109e3765 N |
1172 | if (mddev->gendisk) |
1173 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1174 | read_bio, disk_devt(mddev->gendisk), | |
1175 | r1_bio->sector); | |
1176 | ||
d2eb35ac N |
1177 | if (max_sectors < r1_bio->sectors) { |
1178 | /* could not read all from this device, so we will | |
1179 | * need another r1_bio. | |
1180 | */ | |
d2eb35ac N |
1181 | |
1182 | sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 1183 | - bio->bi_iter.bi_sector); |
d2eb35ac N |
1184 | r1_bio->sectors = max_sectors; |
1185 | spin_lock_irq(&conf->device_lock); | |
1186 | if (bio->bi_phys_segments == 0) | |
1187 | bio->bi_phys_segments = 2; | |
1188 | else | |
1189 | bio->bi_phys_segments++; | |
1190 | spin_unlock_irq(&conf->device_lock); | |
1191 | /* Cannot call generic_make_request directly | |
1192 | * as that will be queued in __make_request | |
1193 | * and subsequent mempool_alloc might block waiting | |
1194 | * for it. So hand bio over to raid1d. | |
1195 | */ | |
1196 | reschedule_retry(r1_bio); | |
1197 | ||
1198 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1199 | ||
1200 | r1_bio->master_bio = bio; | |
aa8b57aa | 1201 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
d2eb35ac N |
1202 | r1_bio->state = 0; |
1203 | r1_bio->mddev = mddev; | |
4f024f37 KO |
1204 | r1_bio->sector = bio->bi_iter.bi_sector + |
1205 | sectors_handled; | |
d2eb35ac N |
1206 | goto read_again; |
1207 | } else | |
1208 | generic_make_request(read_bio); | |
5a7bbad2 | 1209 | return; |
1da177e4 LT |
1210 | } |
1211 | ||
1212 | /* | |
1213 | * WRITE: | |
1214 | */ | |
34db0cd6 N |
1215 | if (conf->pending_count >= max_queued_requests) { |
1216 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1217 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1218 | wait_event(conf->wait_barrier, |
1219 | conf->pending_count < max_queued_requests); | |
1220 | } | |
1f68f0c4 | 1221 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1222 | * inc refcount on their rdev. Record them by setting |
1223 | * bios[x] to bio | |
1f68f0c4 N |
1224 | * If there are known/acknowledged bad blocks on any device on |
1225 | * which we have seen a write error, we want to avoid writing those | |
1226 | * blocks. | |
1227 | * This potentially requires several writes to write around | |
1228 | * the bad blocks. Each set of writes gets it's own r1bio | |
1229 | * with a set of bios attached. | |
1da177e4 | 1230 | */ |
c3b328ac | 1231 | |
8f19ccb2 | 1232 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1233 | retry_write: |
79ef3a8a | 1234 | r1_bio->start_next_window = start_next_window; |
6bfe0b49 | 1235 | blocked_rdev = NULL; |
1da177e4 | 1236 | rcu_read_lock(); |
1f68f0c4 | 1237 | max_sectors = r1_bio->sectors; |
1da177e4 | 1238 | for (i = 0; i < disks; i++) { |
3cb03002 | 1239 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1240 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1241 | atomic_inc(&rdev->nr_pending); | |
1242 | blocked_rdev = rdev; | |
1243 | break; | |
1244 | } | |
1f68f0c4 | 1245 | r1_bio->bios[i] = NULL; |
8ae12666 | 1246 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1247 | if (i < conf->raid_disks) |
1248 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1249 | continue; |
1250 | } | |
1251 | ||
1252 | atomic_inc(&rdev->nr_pending); | |
1253 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1254 | sector_t first_bad; | |
1255 | int bad_sectors; | |
1256 | int is_bad; | |
1257 | ||
1258 | is_bad = is_badblock(rdev, r1_bio->sector, | |
1259 | max_sectors, | |
1260 | &first_bad, &bad_sectors); | |
1261 | if (is_bad < 0) { | |
1262 | /* mustn't write here until the bad block is | |
1263 | * acknowledged*/ | |
1264 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1265 | blocked_rdev = rdev; | |
1266 | break; | |
1267 | } | |
1268 | if (is_bad && first_bad <= r1_bio->sector) { | |
1269 | /* Cannot write here at all */ | |
1270 | bad_sectors -= (r1_bio->sector - first_bad); | |
1271 | if (bad_sectors < max_sectors) | |
1272 | /* mustn't write more than bad_sectors | |
1273 | * to other devices yet | |
1274 | */ | |
1275 | max_sectors = bad_sectors; | |
03c902e1 | 1276 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1277 | /* We don't set R1BIO_Degraded as that |
1278 | * only applies if the disk is | |
1279 | * missing, so it might be re-added, | |
1280 | * and we want to know to recover this | |
1281 | * chunk. | |
1282 | * In this case the device is here, | |
1283 | * and the fact that this chunk is not | |
1284 | * in-sync is recorded in the bad | |
1285 | * block log | |
1286 | */ | |
1287 | continue; | |
964147d5 | 1288 | } |
1f68f0c4 N |
1289 | if (is_bad) { |
1290 | int good_sectors = first_bad - r1_bio->sector; | |
1291 | if (good_sectors < max_sectors) | |
1292 | max_sectors = good_sectors; | |
1293 | } | |
1294 | } | |
1295 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1296 | } |
1297 | rcu_read_unlock(); | |
1298 | ||
6bfe0b49 DW |
1299 | if (unlikely(blocked_rdev)) { |
1300 | /* Wait for this device to become unblocked */ | |
1301 | int j; | |
79ef3a8a | 1302 | sector_t old = start_next_window; |
6bfe0b49 DW |
1303 | |
1304 | for (j = 0; j < i; j++) | |
1305 | if (r1_bio->bios[j]) | |
1306 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1307 | r1_bio->state = 0; |
4f024f37 | 1308 | allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); |
578b54ad | 1309 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1310 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
79ef3a8a | 1311 | start_next_window = wait_barrier(conf, bio); |
1312 | /* | |
1313 | * We must make sure the multi r1bios of bio have | |
1314 | * the same value of bi_phys_segments | |
1315 | */ | |
1316 | if (bio->bi_phys_segments && old && | |
1317 | old != start_next_window) | |
1318 | /* Wait for the former r1bio(s) to complete */ | |
1319 | wait_event(conf->wait_barrier, | |
1320 | bio->bi_phys_segments == 1); | |
6bfe0b49 DW |
1321 | goto retry_write; |
1322 | } | |
1323 | ||
1f68f0c4 N |
1324 | if (max_sectors < r1_bio->sectors) { |
1325 | /* We are splitting this write into multiple parts, so | |
1326 | * we need to prepare for allocating another r1_bio. | |
1327 | */ | |
1328 | r1_bio->sectors = max_sectors; | |
1329 | spin_lock_irq(&conf->device_lock); | |
1330 | if (bio->bi_phys_segments == 0) | |
1331 | bio->bi_phys_segments = 2; | |
1332 | else | |
1333 | bio->bi_phys_segments++; | |
1334 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1335 | } |
4f024f37 | 1336 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
4b6d287f | 1337 | |
4e78064f | 1338 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1339 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1340 | |
1f68f0c4 | 1341 | first_clone = 1; |
1da177e4 LT |
1342 | for (i = 0; i < disks; i++) { |
1343 | struct bio *mbio; | |
1344 | if (!r1_bio->bios[i]) | |
1345 | continue; | |
1346 | ||
a167f663 | 1347 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
4f024f37 | 1348 | bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
1f68f0c4 N |
1349 | |
1350 | if (first_clone) { | |
1351 | /* do behind I/O ? | |
1352 | * Not if there are too many, or cannot | |
1353 | * allocate memory, or a reader on WriteMostly | |
1354 | * is waiting for behind writes to flush */ | |
1355 | if (bitmap && | |
1356 | (atomic_read(&bitmap->behind_writes) | |
1357 | < mddev->bitmap_info.max_write_behind) && | |
1358 | !waitqueue_active(&bitmap->behind_wait)) | |
1359 | alloc_behind_pages(mbio, r1_bio); | |
1360 | ||
1361 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1362 | r1_bio->sectors, | |
1363 | test_bit(R1BIO_BehindIO, | |
1364 | &r1_bio->state)); | |
1365 | first_clone = 0; | |
1366 | } | |
2ca68f5e | 1367 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1368 | struct bio_vec *bvec; |
1369 | int j; | |
1370 | ||
cb34e057 KO |
1371 | /* |
1372 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1373 | */ |
d74c6d51 | 1374 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1375 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1376 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1377 | atomic_inc(&r1_bio->behind_remaining); | |
1378 | } | |
1379 | ||
1f68f0c4 N |
1380 | r1_bio->bios[i] = mbio; |
1381 | ||
4f024f37 | 1382 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1383 | conf->mirrors[i].rdev->data_offset); |
109e3765 | 1384 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
1f68f0c4 | 1385 | mbio->bi_end_io = raid1_end_write_request; |
288dab8a | 1386 | bio_set_op_attrs(mbio, op, do_flush_fua | do_sync); |
1f68f0c4 N |
1387 | mbio->bi_private = r1_bio; |
1388 | ||
1da177e4 | 1389 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1390 | |
109e3765 N |
1391 | if (mddev->gendisk) |
1392 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1393 | mbio, disk_devt(mddev->gendisk), | |
1394 | r1_bio->sector); | |
1395 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1396 | mbio->bi_bdev = (void*)conf->mirrors[i].rdev; | |
1397 | ||
f54a9d0e N |
1398 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1399 | if (cb) | |
1400 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1401 | else | |
1402 | plug = NULL; | |
4e78064f | 1403 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1404 | if (plug) { |
1405 | bio_list_add(&plug->pending, mbio); | |
1406 | plug->pending_cnt++; | |
1407 | } else { | |
1408 | bio_list_add(&conf->pending_bio_list, mbio); | |
1409 | conf->pending_count++; | |
1410 | } | |
4e78064f | 1411 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1412 | if (!plug) |
b357f04a | 1413 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1414 | } |
079fa166 N |
1415 | /* Mustn't call r1_bio_write_done before this next test, |
1416 | * as it could result in the bio being freed. | |
1417 | */ | |
aa8b57aa | 1418 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1419 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1420 | /* We need another r1_bio. It has already been counted |
1421 | * in bio->bi_phys_segments | |
1422 | */ | |
1423 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1424 | r1_bio->master_bio = bio; | |
aa8b57aa | 1425 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
1f68f0c4 N |
1426 | r1_bio->state = 0; |
1427 | r1_bio->mddev = mddev; | |
4f024f37 | 1428 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
1f68f0c4 N |
1429 | goto retry_write; |
1430 | } | |
1431 | ||
079fa166 N |
1432 | r1_bio_write_done(r1_bio); |
1433 | ||
1434 | /* In case raid1d snuck in to freeze_array */ | |
1435 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1436 | } |
1437 | ||
849674e4 | 1438 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1439 | { |
e8096360 | 1440 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1441 | int i; |
1442 | ||
1443 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1444 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1445 | rcu_read_lock(); |
1446 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1447 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1448 | seq_printf(seq, "%s", |
ddac7c7e N |
1449 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1450 | } | |
1451 | rcu_read_unlock(); | |
1da177e4 LT |
1452 | seq_printf(seq, "]"); |
1453 | } | |
1454 | ||
849674e4 | 1455 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1456 | { |
1457 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1458 | struct r1conf *conf = mddev->private; |
423f04d6 | 1459 | unsigned long flags; |
1da177e4 LT |
1460 | |
1461 | /* | |
1462 | * If it is not operational, then we have already marked it as dead | |
1463 | * else if it is the last working disks, ignore the error, let the | |
1464 | * next level up know. | |
1465 | * else mark the drive as failed | |
1466 | */ | |
b2d444d7 | 1467 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1468 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1469 | /* |
1470 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1471 | * normal single drive. |
1472 | * However don't try a recovery from this drive as | |
1473 | * it is very likely to fail. | |
1da177e4 | 1474 | */ |
5389042f | 1475 | conf->recovery_disabled = mddev->recovery_disabled; |
1da177e4 | 1476 | return; |
4044ba58 | 1477 | } |
de393cde | 1478 | set_bit(Blocked, &rdev->flags); |
423f04d6 | 1479 | spin_lock_irqsave(&conf->device_lock, flags); |
c04be0aa | 1480 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1481 | mddev->degraded++; |
dd00a99e | 1482 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1483 | } else |
1484 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1485 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1486 | /* |
1487 | * if recovery is running, make sure it aborts. | |
1488 | */ | |
1489 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
85ad1d13 GJ |
1490 | set_mask_bits(&mddev->flags, 0, |
1491 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
1d41c216 N |
1492 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1493 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1494 | mdname(mddev), bdevname(rdev->bdev, b), | |
1495 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1496 | } |
1497 | ||
e8096360 | 1498 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1499 | { |
1500 | int i; | |
1da177e4 | 1501 | |
1d41c216 | 1502 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1503 | if (!conf) { |
1d41c216 | 1504 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1505 | return; |
1506 | } | |
1d41c216 N |
1507 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1508 | conf->raid_disks); | |
1da177e4 | 1509 | |
ddac7c7e | 1510 | rcu_read_lock(); |
1da177e4 LT |
1511 | for (i = 0; i < conf->raid_disks; i++) { |
1512 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1513 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1514 | if (rdev) |
1d41c216 N |
1515 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1516 | i, !test_bit(In_sync, &rdev->flags), | |
1517 | !test_bit(Faulty, &rdev->flags), | |
1518 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1519 | } |
ddac7c7e | 1520 | rcu_read_unlock(); |
1da177e4 LT |
1521 | } |
1522 | ||
e8096360 | 1523 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1524 | { |
79ef3a8a | 1525 | wait_barrier(conf, NULL); |
1526 | allow_barrier(conf, 0, 0); | |
1da177e4 LT |
1527 | |
1528 | mempool_destroy(conf->r1buf_pool); | |
1529 | conf->r1buf_pool = NULL; | |
79ef3a8a | 1530 | |
669cc7ba | 1531 | spin_lock_irq(&conf->resync_lock); |
e8ff8bf0 | 1532 | conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE; |
79ef3a8a | 1533 | conf->start_next_window = MaxSector; |
669cc7ba N |
1534 | conf->current_window_requests += |
1535 | conf->next_window_requests; | |
1536 | conf->next_window_requests = 0; | |
1537 | spin_unlock_irq(&conf->resync_lock); | |
1da177e4 LT |
1538 | } |
1539 | ||
fd01b88c | 1540 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1541 | { |
1542 | int i; | |
e8096360 | 1543 | struct r1conf *conf = mddev->private; |
6b965620 N |
1544 | int count = 0; |
1545 | unsigned long flags; | |
1da177e4 LT |
1546 | |
1547 | /* | |
f72ffdd6 | 1548 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1549 | * and mark them readable. |
1550 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1551 | * device_lock used to avoid races with raid1_end_read_request |
1552 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1553 | */ |
423f04d6 | 1554 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1555 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1556 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1557 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1558 | if (repl | |
1aee41f6 | 1559 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1560 | && repl->recovery_offset == MaxSector |
1561 | && !test_bit(Faulty, &repl->flags) | |
1562 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1563 | /* replacement has just become active */ | |
1564 | if (!rdev || | |
1565 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1566 | count++; | |
1567 | if (rdev) { | |
1568 | /* Replaced device not technically | |
1569 | * faulty, but we need to be sure | |
1570 | * it gets removed and never re-added | |
1571 | */ | |
1572 | set_bit(Faulty, &rdev->flags); | |
1573 | sysfs_notify_dirent_safe( | |
1574 | rdev->sysfs_state); | |
1575 | } | |
1576 | } | |
ddac7c7e | 1577 | if (rdev |
61e4947c | 1578 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1579 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1580 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1581 | count++; |
654e8b5a | 1582 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1583 | } |
1584 | } | |
6b965620 N |
1585 | mddev->degraded -= count; |
1586 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1587 | |
1588 | print_conf(conf); | |
6b965620 | 1589 | return count; |
1da177e4 LT |
1590 | } |
1591 | ||
fd01b88c | 1592 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1593 | { |
e8096360 | 1594 | struct r1conf *conf = mddev->private; |
199050ea | 1595 | int err = -EEXIST; |
41158c7e | 1596 | int mirror = 0; |
0eaf822c | 1597 | struct raid1_info *p; |
6c2fce2e | 1598 | int first = 0; |
30194636 | 1599 | int last = conf->raid_disks - 1; |
1da177e4 | 1600 | |
5389042f N |
1601 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1602 | return -EBUSY; | |
1603 | ||
1501efad DW |
1604 | if (md_integrity_add_rdev(rdev, mddev)) |
1605 | return -ENXIO; | |
1606 | ||
6c2fce2e NB |
1607 | if (rdev->raid_disk >= 0) |
1608 | first = last = rdev->raid_disk; | |
1609 | ||
70bcecdb GR |
1610 | /* |
1611 | * find the disk ... but prefer rdev->saved_raid_disk | |
1612 | * if possible. | |
1613 | */ | |
1614 | if (rdev->saved_raid_disk >= 0 && | |
1615 | rdev->saved_raid_disk >= first && | |
1616 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) | |
1617 | first = last = rdev->saved_raid_disk; | |
1618 | ||
7ef449d1 N |
1619 | for (mirror = first; mirror <= last; mirror++) { |
1620 | p = conf->mirrors+mirror; | |
1621 | if (!p->rdev) { | |
1da177e4 | 1622 | |
9092c02d JB |
1623 | if (mddev->gendisk) |
1624 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1625 | rdev->data_offset << 9); | |
1da177e4 LT |
1626 | |
1627 | p->head_position = 0; | |
1628 | rdev->raid_disk = mirror; | |
199050ea | 1629 | err = 0; |
6aea114a N |
1630 | /* As all devices are equivalent, we don't need a full recovery |
1631 | * if this was recently any drive of the array | |
1632 | */ | |
1633 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1634 | conf->fullsync = 1; |
d6065f7b | 1635 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1636 | break; |
1637 | } | |
7ef449d1 N |
1638 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1639 | p[conf->raid_disks].rdev == NULL) { | |
1640 | /* Add this device as a replacement */ | |
1641 | clear_bit(In_sync, &rdev->flags); | |
1642 | set_bit(Replacement, &rdev->flags); | |
1643 | rdev->raid_disk = mirror; | |
1644 | err = 0; | |
1645 | conf->fullsync = 1; | |
1646 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1647 | break; | |
1648 | } | |
1649 | } | |
9092c02d | 1650 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1651 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1652 | print_conf(conf); |
199050ea | 1653 | return err; |
1da177e4 LT |
1654 | } |
1655 | ||
b8321b68 | 1656 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1657 | { |
e8096360 | 1658 | struct r1conf *conf = mddev->private; |
1da177e4 | 1659 | int err = 0; |
b8321b68 | 1660 | int number = rdev->raid_disk; |
0eaf822c | 1661 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1662 | |
b014f14c N |
1663 | if (rdev != p->rdev) |
1664 | p = conf->mirrors + conf->raid_disks + number; | |
1665 | ||
1da177e4 | 1666 | print_conf(conf); |
b8321b68 | 1667 | if (rdev == p->rdev) { |
b2d444d7 | 1668 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1669 | atomic_read(&rdev->nr_pending)) { |
1670 | err = -EBUSY; | |
1671 | goto abort; | |
1672 | } | |
046abeed | 1673 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1674 | * is not possible. |
1675 | */ | |
1676 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1677 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1678 | mddev->degraded < conf->raid_disks) { |
1679 | err = -EBUSY; | |
1680 | goto abort; | |
1681 | } | |
1da177e4 | 1682 | p->rdev = NULL; |
d787be40 N |
1683 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1684 | synchronize_rcu(); | |
1685 | if (atomic_read(&rdev->nr_pending)) { | |
1686 | /* lost the race, try later */ | |
1687 | err = -EBUSY; | |
1688 | p->rdev = rdev; | |
1689 | goto abort; | |
1690 | } | |
1691 | } | |
1692 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1693 | /* We just removed a device that is being replaced. |
1694 | * Move down the replacement. We drain all IO before | |
1695 | * doing this to avoid confusion. | |
1696 | */ | |
1697 | struct md_rdev *repl = | |
1698 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1699 | freeze_array(conf, 0); |
8c7a2c2b N |
1700 | clear_bit(Replacement, &repl->flags); |
1701 | p->rdev = repl; | |
1702 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1703 | unfreeze_array(conf); |
8c7a2c2b N |
1704 | clear_bit(WantReplacement, &rdev->flags); |
1705 | } else | |
b014f14c | 1706 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1707 | err = md_integrity_register(mddev); |
1da177e4 LT |
1708 | } |
1709 | abort: | |
1710 | ||
1711 | print_conf(conf); | |
1712 | return err; | |
1713 | } | |
1714 | ||
4246a0b6 | 1715 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1716 | { |
9f2c9d12 | 1717 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1718 | |
0fc280f6 | 1719 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1720 | |
1da177e4 LT |
1721 | /* |
1722 | * we have read a block, now it needs to be re-written, | |
1723 | * or re-read if the read failed. | |
1724 | * We don't do much here, just schedule handling by raid1d | |
1725 | */ | |
4246a0b6 | 1726 | if (!bio->bi_error) |
1da177e4 | 1727 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1728 | |
1729 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1730 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1731 | } |
1732 | ||
4246a0b6 | 1733 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1734 | { |
4246a0b6 | 1735 | int uptodate = !bio->bi_error; |
9f2c9d12 | 1736 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1737 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1738 | struct r1conf *conf = mddev->private; |
4367af55 N |
1739 | sector_t first_bad; |
1740 | int bad_sectors; | |
854abd75 | 1741 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1742 | |
6b1117d5 | 1743 | if (!uptodate) { |
57dab0bd | 1744 | sector_t sync_blocks = 0; |
6b1117d5 N |
1745 | sector_t s = r1_bio->sector; |
1746 | long sectors_to_go = r1_bio->sectors; | |
1747 | /* make sure these bits doesn't get cleared. */ | |
1748 | do { | |
5e3db645 | 1749 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1750 | &sync_blocks, 1); |
1751 | s += sync_blocks; | |
1752 | sectors_to_go -= sync_blocks; | |
1753 | } while (sectors_to_go > 0); | |
854abd75 N |
1754 | set_bit(WriteErrorSeen, &rdev->flags); |
1755 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1756 | set_bit(MD_RECOVERY_NEEDED, & |
1757 | mddev->recovery); | |
d8f05d29 | 1758 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1759 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1760 | &first_bad, &bad_sectors) && |
1761 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1762 | r1_bio->sector, | |
1763 | r1_bio->sectors, | |
1764 | &first_bad, &bad_sectors) | |
1765 | ) | |
4367af55 | 1766 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1767 | |
1da177e4 | 1768 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1769 | int s = r1_bio->sectors; |
d8f05d29 N |
1770 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1771 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1772 | reschedule_retry(r1_bio); |
1773 | else { | |
1774 | put_buf(r1_bio); | |
1775 | md_done_sync(mddev, s, uptodate); | |
1776 | } | |
1da177e4 | 1777 | } |
1da177e4 LT |
1778 | } |
1779 | ||
3cb03002 | 1780 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1781 | int sectors, struct page *page, int rw) |
1782 | { | |
796a5cf0 | 1783 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1784 | /* success */ |
1785 | return 1; | |
19d67169 | 1786 | if (rw == WRITE) { |
d8f05d29 | 1787 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1788 | if (!test_and_set_bit(WantReplacement, |
1789 | &rdev->flags)) | |
1790 | set_bit(MD_RECOVERY_NEEDED, & | |
1791 | rdev->mddev->recovery); | |
1792 | } | |
d8f05d29 N |
1793 | /* need to record an error - either for the block or the device */ |
1794 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1795 | md_error(rdev->mddev, rdev); | |
1796 | return 0; | |
1797 | } | |
1798 | ||
9f2c9d12 | 1799 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1800 | { |
a68e5870 N |
1801 | /* Try some synchronous reads of other devices to get |
1802 | * good data, much like with normal read errors. Only | |
1803 | * read into the pages we already have so we don't | |
1804 | * need to re-issue the read request. | |
1805 | * We don't need to freeze the array, because being in an | |
1806 | * active sync request, there is no normal IO, and | |
1807 | * no overlapping syncs. | |
06f60385 N |
1808 | * We don't need to check is_badblock() again as we |
1809 | * made sure that anything with a bad block in range | |
1810 | * will have bi_end_io clear. | |
a68e5870 | 1811 | */ |
fd01b88c | 1812 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1813 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1814 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1815 | sector_t sect = r1_bio->sector; | |
1816 | int sectors = r1_bio->sectors; | |
1817 | int idx = 0; | |
1818 | ||
1819 | while(sectors) { | |
1820 | int s = sectors; | |
1821 | int d = r1_bio->read_disk; | |
1822 | int success = 0; | |
3cb03002 | 1823 | struct md_rdev *rdev; |
78d7f5f7 | 1824 | int start; |
a68e5870 N |
1825 | |
1826 | if (s > (PAGE_SIZE>>9)) | |
1827 | s = PAGE_SIZE >> 9; | |
1828 | do { | |
1829 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1830 | /* No rcu protection needed here devices | |
1831 | * can only be removed when no resync is | |
1832 | * active, and resync is currently active | |
1833 | */ | |
1834 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1835 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 | 1836 | bio->bi_io_vec[idx].bv_page, |
796a5cf0 | 1837 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
1838 | success = 1; |
1839 | break; | |
1840 | } | |
1841 | } | |
1842 | d++; | |
8f19ccb2 | 1843 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1844 | d = 0; |
1845 | } while (!success && d != r1_bio->read_disk); | |
1846 | ||
78d7f5f7 | 1847 | if (!success) { |
a68e5870 | 1848 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1849 | int abort = 0; |
1850 | /* Cannot read from anywhere, this block is lost. | |
1851 | * Record a bad block on each device. If that doesn't | |
1852 | * work just disable and interrupt the recovery. | |
1853 | * Don't fail devices as that won't really help. | |
1854 | */ | |
1d41c216 N |
1855 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
1856 | mdname(mddev), | |
1857 | bdevname(bio->bi_bdev, b), | |
1858 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1859 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1860 | rdev = conf->mirrors[d].rdev; |
1861 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1862 | continue; | |
1863 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1864 | abort = 1; | |
1865 | } | |
1866 | if (abort) { | |
d890fa2b N |
1867 | conf->recovery_disabled = |
1868 | mddev->recovery_disabled; | |
3a9f28a5 N |
1869 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1870 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1871 | put_buf(r1_bio); | |
1872 | return 0; | |
1873 | } | |
1874 | /* Try next page */ | |
1875 | sectors -= s; | |
1876 | sect += s; | |
1877 | idx++; | |
1878 | continue; | |
d11c171e | 1879 | } |
78d7f5f7 N |
1880 | |
1881 | start = d; | |
1882 | /* write it back and re-read */ | |
1883 | while (d != r1_bio->read_disk) { | |
1884 | if (d == 0) | |
8f19ccb2 | 1885 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1886 | d--; |
1887 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1888 | continue; | |
1889 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1890 | if (r1_sync_page_io(rdev, sect, s, |
1891 | bio->bi_io_vec[idx].bv_page, | |
1892 | WRITE) == 0) { | |
78d7f5f7 N |
1893 | r1_bio->bios[d]->bi_end_io = NULL; |
1894 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1895 | } |
78d7f5f7 N |
1896 | } |
1897 | d = start; | |
1898 | while (d != r1_bio->read_disk) { | |
1899 | if (d == 0) | |
8f19ccb2 | 1900 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1901 | d--; |
1902 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1903 | continue; | |
1904 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1905 | if (r1_sync_page_io(rdev, sect, s, |
1906 | bio->bi_io_vec[idx].bv_page, | |
1907 | READ) != 0) | |
9d3d8011 | 1908 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1909 | } |
a68e5870 N |
1910 | sectors -= s; |
1911 | sect += s; | |
1912 | idx ++; | |
1913 | } | |
78d7f5f7 | 1914 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4246a0b6 | 1915 | bio->bi_error = 0; |
a68e5870 N |
1916 | return 1; |
1917 | } | |
1918 | ||
c95e6385 | 1919 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1920 | { |
1921 | /* We have read all readable devices. If we haven't | |
1922 | * got the block, then there is no hope left. | |
1923 | * If we have, then we want to do a comparison | |
1924 | * and skip the write if everything is the same. | |
1925 | * If any blocks failed to read, then we need to | |
1926 | * attempt an over-write | |
1927 | */ | |
fd01b88c | 1928 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1929 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1930 | int primary; |
1931 | int i; | |
f4380a91 | 1932 | int vcnt; |
a68e5870 | 1933 | |
30bc9b53 N |
1934 | /* Fix variable parts of all bios */ |
1935 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
1936 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
1937 | int j; | |
1938 | int size; | |
4246a0b6 | 1939 | int error; |
30bc9b53 N |
1940 | struct bio *b = r1_bio->bios[i]; |
1941 | if (b->bi_end_io != end_sync_read) | |
1942 | continue; | |
4246a0b6 CH |
1943 | /* fixup the bio for reuse, but preserve errno */ |
1944 | error = b->bi_error; | |
30bc9b53 | 1945 | bio_reset(b); |
4246a0b6 | 1946 | b->bi_error = error; |
30bc9b53 | 1947 | b->bi_vcnt = vcnt; |
4f024f37 KO |
1948 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
1949 | b->bi_iter.bi_sector = r1_bio->sector + | |
30bc9b53 N |
1950 | conf->mirrors[i].rdev->data_offset; |
1951 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
1952 | b->bi_end_io = end_sync_read; | |
1953 | b->bi_private = r1_bio; | |
1954 | ||
4f024f37 | 1955 | size = b->bi_iter.bi_size; |
30bc9b53 N |
1956 | for (j = 0; j < vcnt ; j++) { |
1957 | struct bio_vec *bi; | |
1958 | bi = &b->bi_io_vec[j]; | |
1959 | bi->bv_offset = 0; | |
1960 | if (size > PAGE_SIZE) | |
1961 | bi->bv_len = PAGE_SIZE; | |
1962 | else | |
1963 | bi->bv_len = size; | |
1964 | size -= PAGE_SIZE; | |
1965 | } | |
1966 | } | |
8f19ccb2 | 1967 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 1968 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4246a0b6 | 1969 | !r1_bio->bios[primary]->bi_error) { |
a68e5870 N |
1970 | r1_bio->bios[primary]->bi_end_io = NULL; |
1971 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
1972 | break; | |
1973 | } | |
1974 | r1_bio->read_disk = primary; | |
8f19ccb2 | 1975 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 1976 | int j; |
78d7f5f7 N |
1977 | struct bio *pbio = r1_bio->bios[primary]; |
1978 | struct bio *sbio = r1_bio->bios[i]; | |
4246a0b6 | 1979 | int error = sbio->bi_error; |
a68e5870 | 1980 | |
2aabaa65 | 1981 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 1982 | continue; |
4246a0b6 CH |
1983 | /* Now we can 'fixup' the error value */ |
1984 | sbio->bi_error = 0; | |
78d7f5f7 | 1985 | |
4246a0b6 | 1986 | if (!error) { |
78d7f5f7 N |
1987 | for (j = vcnt; j-- ; ) { |
1988 | struct page *p, *s; | |
1989 | p = pbio->bi_io_vec[j].bv_page; | |
1990 | s = sbio->bi_io_vec[j].bv_page; | |
1991 | if (memcmp(page_address(p), | |
1992 | page_address(s), | |
5020ad7d | 1993 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 1994 | break; |
69382e85 | 1995 | } |
78d7f5f7 N |
1996 | } else |
1997 | j = 0; | |
1998 | if (j >= 0) | |
7f7583d4 | 1999 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2000 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4246a0b6 | 2001 | && !error)) { |
78d7f5f7 N |
2002 | /* No need to write to this device. */ |
2003 | sbio->bi_end_io = NULL; | |
2004 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2005 | continue; | |
2006 | } | |
d3b45c2a KO |
2007 | |
2008 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2009 | } |
a68e5870 N |
2010 | } |
2011 | ||
9f2c9d12 | 2012 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2013 | { |
e8096360 | 2014 | struct r1conf *conf = mddev->private; |
a68e5870 | 2015 | int i; |
8f19ccb2 | 2016 | int disks = conf->raid_disks * 2; |
a68e5870 N |
2017 | struct bio *bio, *wbio; |
2018 | ||
2019 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2020 | ||
a68e5870 N |
2021 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2022 | /* ouch - failed to read all of that. */ | |
2023 | if (!fix_sync_read_error(r1_bio)) | |
2024 | return; | |
7ca78d57 N |
2025 | |
2026 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2027 | process_checks(r1_bio); |
2028 | ||
d11c171e N |
2029 | /* |
2030 | * schedule writes | |
2031 | */ | |
1da177e4 LT |
2032 | atomic_set(&r1_bio->remaining, 1); |
2033 | for (i = 0; i < disks ; i++) { | |
2034 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2035 | if (wbio->bi_end_io == NULL || |
2036 | (wbio->bi_end_io == end_sync_read && | |
2037 | (i == r1_bio->read_disk || | |
2038 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 LT |
2039 | continue; |
2040 | ||
796a5cf0 | 2041 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
3e198f78 | 2042 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2043 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2044 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2045 | |
1da177e4 LT |
2046 | generic_make_request(wbio); |
2047 | } | |
2048 | ||
2049 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2050 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2051 | int s = r1_bio->sectors; |
2052 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2053 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2054 | reschedule_retry(r1_bio); | |
2055 | else { | |
2056 | put_buf(r1_bio); | |
2057 | md_done_sync(mddev, s, 1); | |
2058 | } | |
1da177e4 LT |
2059 | } |
2060 | } | |
2061 | ||
2062 | /* | |
2063 | * This is a kernel thread which: | |
2064 | * | |
2065 | * 1. Retries failed read operations on working mirrors. | |
2066 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2067 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2068 | */ |
2069 | ||
e8096360 | 2070 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2071 | sector_t sect, int sectors) |
2072 | { | |
fd01b88c | 2073 | struct mddev *mddev = conf->mddev; |
867868fb N |
2074 | while(sectors) { |
2075 | int s = sectors; | |
2076 | int d = read_disk; | |
2077 | int success = 0; | |
2078 | int start; | |
3cb03002 | 2079 | struct md_rdev *rdev; |
867868fb N |
2080 | |
2081 | if (s > (PAGE_SIZE>>9)) | |
2082 | s = PAGE_SIZE >> 9; | |
2083 | ||
2084 | do { | |
d2eb35ac N |
2085 | sector_t first_bad; |
2086 | int bad_sectors; | |
2087 | ||
707a6a42 N |
2088 | rcu_read_lock(); |
2089 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2090 | if (rdev && |
da8840a7 | 2091 | (test_bit(In_sync, &rdev->flags) || |
2092 | (!test_bit(Faulty, &rdev->flags) && | |
2093 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2094 | is_badblock(rdev, sect, s, |
707a6a42 N |
2095 | &first_bad, &bad_sectors) == 0) { |
2096 | atomic_inc(&rdev->nr_pending); | |
2097 | rcu_read_unlock(); | |
2098 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2099 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2100 | success = 1; |
2101 | rdev_dec_pending(rdev, mddev); | |
2102 | if (success) | |
2103 | break; | |
2104 | } else | |
2105 | rcu_read_unlock(); | |
2106 | d++; | |
2107 | if (d == conf->raid_disks * 2) | |
2108 | d = 0; | |
867868fb N |
2109 | } while (!success && d != read_disk); |
2110 | ||
2111 | if (!success) { | |
d8f05d29 | 2112 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2113 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2114 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2115 | md_error(mddev, rdev); | |
867868fb N |
2116 | break; |
2117 | } | |
2118 | /* write it back and re-read */ | |
2119 | start = d; | |
2120 | while (d != read_disk) { | |
2121 | if (d==0) | |
8f19ccb2 | 2122 | d = conf->raid_disks * 2; |
867868fb | 2123 | d--; |
707a6a42 N |
2124 | rcu_read_lock(); |
2125 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2126 | if (rdev && |
707a6a42 N |
2127 | !test_bit(Faulty, &rdev->flags)) { |
2128 | atomic_inc(&rdev->nr_pending); | |
2129 | rcu_read_unlock(); | |
d8f05d29 N |
2130 | r1_sync_page_io(rdev, sect, s, |
2131 | conf->tmppage, WRITE); | |
707a6a42 N |
2132 | rdev_dec_pending(rdev, mddev); |
2133 | } else | |
2134 | rcu_read_unlock(); | |
867868fb N |
2135 | } |
2136 | d = start; | |
2137 | while (d != read_disk) { | |
2138 | char b[BDEVNAME_SIZE]; | |
2139 | if (d==0) | |
8f19ccb2 | 2140 | d = conf->raid_disks * 2; |
867868fb | 2141 | d--; |
707a6a42 N |
2142 | rcu_read_lock(); |
2143 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2144 | if (rdev && |
b8cb6b4c | 2145 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2146 | atomic_inc(&rdev->nr_pending); |
2147 | rcu_read_unlock(); | |
d8f05d29 N |
2148 | if (r1_sync_page_io(rdev, sect, s, |
2149 | conf->tmppage, READ)) { | |
867868fb | 2150 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2151 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2152 | mdname(mddev), s, | |
2153 | (unsigned long long)(sect + | |
2154 | rdev->data_offset), | |
2155 | bdevname(rdev->bdev, b)); | |
867868fb | 2156 | } |
707a6a42 N |
2157 | rdev_dec_pending(rdev, mddev); |
2158 | } else | |
2159 | rcu_read_unlock(); | |
867868fb N |
2160 | } |
2161 | sectors -= s; | |
2162 | sect += s; | |
2163 | } | |
2164 | } | |
2165 | ||
9f2c9d12 | 2166 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2167 | { |
fd01b88c | 2168 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2169 | struct r1conf *conf = mddev->private; |
3cb03002 | 2170 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2171 | |
2172 | /* bio has the data to be written to device 'i' where | |
2173 | * we just recently had a write error. | |
2174 | * We repeatedly clone the bio and trim down to one block, | |
2175 | * then try the write. Where the write fails we record | |
2176 | * a bad block. | |
2177 | * It is conceivable that the bio doesn't exactly align with | |
2178 | * blocks. We must handle this somehow. | |
2179 | * | |
2180 | * We currently own a reference on the rdev. | |
2181 | */ | |
2182 | ||
2183 | int block_sectors; | |
2184 | sector_t sector; | |
2185 | int sectors; | |
2186 | int sect_to_write = r1_bio->sectors; | |
2187 | int ok = 1; | |
2188 | ||
2189 | if (rdev->badblocks.shift < 0) | |
2190 | return 0; | |
2191 | ||
ab713cdc ND |
2192 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2193 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2194 | sector = r1_bio->sector; |
2195 | sectors = ((sector + block_sectors) | |
2196 | & ~(sector_t)(block_sectors - 1)) | |
2197 | - sector; | |
2198 | ||
cd5ff9a1 N |
2199 | while (sect_to_write) { |
2200 | struct bio *wbio; | |
2201 | if (sectors > sect_to_write) | |
2202 | sectors = sect_to_write; | |
2203 | /* Write at 'sector' for 'sectors'*/ | |
2204 | ||
b783863f KO |
2205 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2206 | unsigned vcnt = r1_bio->behind_page_count; | |
2207 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2208 | ||
2209 | while (!vec->bv_page) { | |
2210 | vec++; | |
2211 | vcnt--; | |
2212 | } | |
2213 | ||
2214 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2215 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2216 | ||
2217 | wbio->bi_vcnt = vcnt; | |
2218 | } else { | |
2219 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2220 | } | |
2221 | ||
796a5cf0 | 2222 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2223 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2224 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2225 | |
6678d83f | 2226 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2227 | wbio->bi_iter.bi_sector += rdev->data_offset; |
cd5ff9a1 | 2228 | wbio->bi_bdev = rdev->bdev; |
4e49ea4a MC |
2229 | |
2230 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2231 | /* failure! */ |
2232 | ok = rdev_set_badblocks(rdev, sector, | |
2233 | sectors, 0) | |
2234 | && ok; | |
2235 | ||
2236 | bio_put(wbio); | |
2237 | sect_to_write -= sectors; | |
2238 | sector += sectors; | |
2239 | sectors = block_sectors; | |
2240 | } | |
2241 | return ok; | |
2242 | } | |
2243 | ||
e8096360 | 2244 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2245 | { |
2246 | int m; | |
2247 | int s = r1_bio->sectors; | |
8f19ccb2 | 2248 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2249 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2250 | struct bio *bio = r1_bio->bios[m]; |
2251 | if (bio->bi_end_io == NULL) | |
2252 | continue; | |
4246a0b6 | 2253 | if (!bio->bi_error && |
62096bce | 2254 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2255 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2256 | } |
4246a0b6 | 2257 | if (bio->bi_error && |
62096bce N |
2258 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2259 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2260 | md_error(conf->mddev, rdev); | |
2261 | } | |
2262 | } | |
2263 | put_buf(r1_bio); | |
2264 | md_done_sync(conf->mddev, s, 1); | |
2265 | } | |
2266 | ||
e8096360 | 2267 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2268 | { |
2269 | int m; | |
55ce74d4 | 2270 | bool fail = false; |
8f19ccb2 | 2271 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2272 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2273 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2274 | rdev_clear_badblocks(rdev, |
2275 | r1_bio->sector, | |
c6563a8c | 2276 | r1_bio->sectors, 0); |
62096bce N |
2277 | rdev_dec_pending(rdev, conf->mddev); |
2278 | } else if (r1_bio->bios[m] != NULL) { | |
2279 | /* This drive got a write error. We need to | |
2280 | * narrow down and record precise write | |
2281 | * errors. | |
2282 | */ | |
55ce74d4 | 2283 | fail = true; |
62096bce N |
2284 | if (!narrow_write_error(r1_bio, m)) { |
2285 | md_error(conf->mddev, | |
2286 | conf->mirrors[m].rdev); | |
2287 | /* an I/O failed, we can't clear the bitmap */ | |
2288 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2289 | } | |
2290 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2291 | conf->mddev); | |
2292 | } | |
55ce74d4 N |
2293 | if (fail) { |
2294 | spin_lock_irq(&conf->device_lock); | |
2295 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
ccfc7bf1 | 2296 | conf->nr_queued++; |
55ce74d4 N |
2297 | spin_unlock_irq(&conf->device_lock); |
2298 | md_wakeup_thread(conf->mddev->thread); | |
bd8688a1 N |
2299 | } else { |
2300 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2301 | close_write(r1_bio); | |
55ce74d4 | 2302 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2303 | } |
62096bce N |
2304 | } |
2305 | ||
e8096360 | 2306 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2307 | { |
2308 | int disk; | |
2309 | int max_sectors; | |
fd01b88c | 2310 | struct mddev *mddev = conf->mddev; |
62096bce N |
2311 | struct bio *bio; |
2312 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2313 | struct md_rdev *rdev; |
109e3765 N |
2314 | dev_t bio_dev; |
2315 | sector_t bio_sector; | |
62096bce N |
2316 | |
2317 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2318 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2319 | * the block and we can fix it. | |
2320 | * We freeze all other IO, and try reading the block from | |
2321 | * other devices. When we find one, we re-write | |
2322 | * and check it that fixes the read error. | |
2323 | * This is all done synchronously while the array is | |
2324 | * frozen | |
2325 | */ | |
7449f699 TM |
2326 | |
2327 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2328 | bdevname(bio->bi_bdev, b); | |
109e3765 N |
2329 | bio_dev = bio->bi_bdev->bd_dev; |
2330 | bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector; | |
7449f699 TM |
2331 | bio_put(bio); |
2332 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2333 | ||
62096bce | 2334 | if (mddev->ro == 0) { |
e2d59925 | 2335 | freeze_array(conf, 1); |
62096bce N |
2336 | fix_read_error(conf, r1_bio->read_disk, |
2337 | r1_bio->sector, r1_bio->sectors); | |
2338 | unfreeze_array(conf); | |
7449f699 TM |
2339 | } else { |
2340 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2341 | } | |
2342 | ||
7ad4d4a6 | 2343 | rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); |
62096bce | 2344 | |
62096bce N |
2345 | read_more: |
2346 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2347 | if (disk == -1) { | |
1d41c216 N |
2348 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
2349 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
62096bce N |
2350 | raid_end_bio_io(r1_bio); |
2351 | } else { | |
2352 | const unsigned long do_sync | |
1eff9d32 | 2353 | = r1_bio->master_bio->bi_opf & REQ_SYNC; |
62096bce N |
2354 | r1_bio->read_disk = disk; |
2355 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
4f024f37 KO |
2356 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
2357 | max_sectors); | |
62096bce N |
2358 | r1_bio->bios[r1_bio->read_disk] = bio; |
2359 | rdev = conf->mirrors[disk].rdev; | |
1d41c216 N |
2360 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", |
2361 | mdname(mddev), | |
2362 | (unsigned long long)r1_bio->sector, | |
2363 | bdevname(rdev->bdev, b)); | |
4f024f37 | 2364 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
62096bce N |
2365 | bio->bi_bdev = rdev->bdev; |
2366 | bio->bi_end_io = raid1_end_read_request; | |
796a5cf0 | 2367 | bio_set_op_attrs(bio, REQ_OP_READ, do_sync); |
62096bce N |
2368 | bio->bi_private = r1_bio; |
2369 | if (max_sectors < r1_bio->sectors) { | |
2370 | /* Drat - have to split this up more */ | |
2371 | struct bio *mbio = r1_bio->master_bio; | |
2372 | int sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 2373 | - mbio->bi_iter.bi_sector); |
62096bce N |
2374 | r1_bio->sectors = max_sectors; |
2375 | spin_lock_irq(&conf->device_lock); | |
2376 | if (mbio->bi_phys_segments == 0) | |
2377 | mbio->bi_phys_segments = 2; | |
2378 | else | |
2379 | mbio->bi_phys_segments++; | |
2380 | spin_unlock_irq(&conf->device_lock); | |
109e3765 N |
2381 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), |
2382 | bio, bio_dev, bio_sector); | |
62096bce N |
2383 | generic_make_request(bio); |
2384 | bio = NULL; | |
2385 | ||
2386 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2387 | ||
2388 | r1_bio->master_bio = mbio; | |
aa8b57aa | 2389 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
62096bce N |
2390 | r1_bio->state = 0; |
2391 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2392 | r1_bio->mddev = mddev; | |
4f024f37 KO |
2393 | r1_bio->sector = mbio->bi_iter.bi_sector + |
2394 | sectors_handled; | |
62096bce N |
2395 | |
2396 | goto read_more; | |
109e3765 N |
2397 | } else { |
2398 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), | |
2399 | bio, bio_dev, bio_sector); | |
62096bce | 2400 | generic_make_request(bio); |
109e3765 | 2401 | } |
62096bce N |
2402 | } |
2403 | } | |
2404 | ||
4ed8731d | 2405 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2406 | { |
4ed8731d | 2407 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2408 | struct r1bio *r1_bio; |
1da177e4 | 2409 | unsigned long flags; |
e8096360 | 2410 | struct r1conf *conf = mddev->private; |
1da177e4 | 2411 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2412 | struct blk_plug plug; |
1da177e4 LT |
2413 | |
2414 | md_check_recovery(mddev); | |
e1dfa0a2 | 2415 | |
55ce74d4 N |
2416 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2417 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
2418 | LIST_HEAD(tmp); | |
2419 | spin_lock_irqsave(&conf->device_lock, flags); | |
2420 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
ccfc7bf1 ND |
2421 | while (!list_empty(&conf->bio_end_io_list)) { |
2422 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2423 | conf->nr_queued--; | |
2424 | } | |
55ce74d4 N |
2425 | } |
2426 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2427 | while (!list_empty(&tmp)) { | |
a452744b MP |
2428 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2429 | retry_list); | |
55ce74d4 | 2430 | list_del(&r1_bio->retry_list); |
bd8688a1 N |
2431 | if (mddev->degraded) |
2432 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2433 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2434 | close_write(r1_bio); | |
55ce74d4 N |
2435 | raid_end_bio_io(r1_bio); |
2436 | } | |
2437 | } | |
2438 | ||
e1dfa0a2 | 2439 | blk_start_plug(&plug); |
1da177e4 | 2440 | for (;;) { |
191ea9b2 | 2441 | |
0021b7bc | 2442 | flush_pending_writes(conf); |
191ea9b2 | 2443 | |
a35e63ef N |
2444 | spin_lock_irqsave(&conf->device_lock, flags); |
2445 | if (list_empty(head)) { | |
2446 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2447 | break; |
a35e63ef | 2448 | } |
9f2c9d12 | 2449 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2450 | list_del(head->prev); |
ddaf22ab | 2451 | conf->nr_queued--; |
1da177e4 LT |
2452 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2453 | ||
2454 | mddev = r1_bio->mddev; | |
070ec55d | 2455 | conf = mddev->private; |
4367af55 | 2456 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2457 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2458 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2459 | handle_sync_write_finished(conf, r1_bio); | |
2460 | else | |
4367af55 | 2461 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2462 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2463 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2464 | handle_write_finished(conf, r1_bio); | |
2465 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2466 | handle_read_error(conf, r1_bio); | |
2467 | else | |
d2eb35ac N |
2468 | /* just a partial read to be scheduled from separate |
2469 | * context | |
2470 | */ | |
2471 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2472 | |
1d9d5241 | 2473 | cond_resched(); |
de393cde N |
2474 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) |
2475 | md_check_recovery(mddev); | |
1da177e4 | 2476 | } |
e1dfa0a2 | 2477 | blk_finish_plug(&plug); |
1da177e4 LT |
2478 | } |
2479 | ||
e8096360 | 2480 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2481 | { |
2482 | int buffs; | |
2483 | ||
2484 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2485 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2486 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2487 | conf->poolinfo); | |
2488 | if (!conf->r1buf_pool) | |
2489 | return -ENOMEM; | |
2490 | conf->next_resync = 0; | |
2491 | return 0; | |
2492 | } | |
2493 | ||
2494 | /* | |
2495 | * perform a "sync" on one "block" | |
2496 | * | |
2497 | * We need to make sure that no normal I/O request - particularly write | |
2498 | * requests - conflict with active sync requests. | |
2499 | * | |
2500 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2501 | * that can be installed to exclude normal IO requests. | |
2502 | */ | |
2503 | ||
849674e4 SL |
2504 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2505 | int *skipped) | |
1da177e4 | 2506 | { |
e8096360 | 2507 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2508 | struct r1bio *r1_bio; |
1da177e4 LT |
2509 | struct bio *bio; |
2510 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2511 | int disk = -1; |
1da177e4 | 2512 | int i; |
3e198f78 N |
2513 | int wonly = -1; |
2514 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2515 | sector_t sync_blocks; |
e3b9703e | 2516 | int still_degraded = 0; |
06f60385 N |
2517 | int good_sectors = RESYNC_SECTORS; |
2518 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2519 | |
2520 | if (!conf->r1buf_pool) | |
2521 | if (init_resync(conf)) | |
57afd89f | 2522 | return 0; |
1da177e4 | 2523 | |
58c0fed4 | 2524 | max_sector = mddev->dev_sectors; |
1da177e4 | 2525 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2526 | /* If we aborted, we need to abort the |
2527 | * sync on the 'current' bitmap chunk (there will | |
2528 | * only be one in raid1 resync. | |
2529 | * We can find the current addess in mddev->curr_resync | |
2530 | */ | |
6a806c51 N |
2531 | if (mddev->curr_resync < max_sector) /* aborted */ |
2532 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2533 | &sync_blocks, 1); |
6a806c51 | 2534 | else /* completed sync */ |
191ea9b2 | 2535 | conf->fullsync = 0; |
6a806c51 N |
2536 | |
2537 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2538 | close_sync(conf); |
c40f341f GR |
2539 | |
2540 | if (mddev_is_clustered(mddev)) { | |
2541 | conf->cluster_sync_low = 0; | |
2542 | conf->cluster_sync_high = 0; | |
c40f341f | 2543 | } |
1da177e4 LT |
2544 | return 0; |
2545 | } | |
2546 | ||
07d84d10 N |
2547 | if (mddev->bitmap == NULL && |
2548 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2549 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2550 | conf->fullsync == 0) { |
2551 | *skipped = 1; | |
2552 | return max_sector - sector_nr; | |
2553 | } | |
6394cca5 N |
2554 | /* before building a request, check if we can skip these blocks.. |
2555 | * This call the bitmap_start_sync doesn't actually record anything | |
2556 | */ | |
e3b9703e | 2557 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2558 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2559 | /* We can skip this block, and probably several more */ |
2560 | *skipped = 1; | |
2561 | return sync_blocks; | |
2562 | } | |
17999be4 | 2563 | |
7ac50447 TM |
2564 | /* |
2565 | * If there is non-resync activity waiting for a turn, then let it | |
2566 | * though before starting on this new sync request. | |
2567 | */ | |
2568 | if (conf->nr_waiting) | |
2569 | schedule_timeout_uninterruptible(1); | |
2570 | ||
c40f341f GR |
2571 | /* we are incrementing sector_nr below. To be safe, we check against |
2572 | * sector_nr + two times RESYNC_SECTORS | |
2573 | */ | |
2574 | ||
2575 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
2576 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
1c4588e9 | 2577 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 | 2578 | |
c2fd4c94 | 2579 | raise_barrier(conf, sector_nr); |
1da177e4 | 2580 | |
3e198f78 | 2581 | rcu_read_lock(); |
1da177e4 | 2582 | /* |
3e198f78 N |
2583 | * If we get a correctably read error during resync or recovery, |
2584 | * we might want to read from a different device. So we | |
2585 | * flag all drives that could conceivably be read from for READ, | |
2586 | * and any others (which will be non-In_sync devices) for WRITE. | |
2587 | * If a read fails, we try reading from something else for which READ | |
2588 | * is OK. | |
1da177e4 | 2589 | */ |
1da177e4 | 2590 | |
1da177e4 LT |
2591 | r1_bio->mddev = mddev; |
2592 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2593 | r1_bio->state = 0; |
1da177e4 | 2594 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2595 | |
8f19ccb2 | 2596 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2597 | struct md_rdev *rdev; |
1da177e4 | 2598 | bio = r1_bio->bios[i]; |
2aabaa65 | 2599 | bio_reset(bio); |
1da177e4 | 2600 | |
3e198f78 N |
2601 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2602 | if (rdev == NULL || | |
06f60385 | 2603 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2604 | if (i < conf->raid_disks) |
2605 | still_degraded = 1; | |
3e198f78 | 2606 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2607 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2608 | bio->bi_end_io = end_sync_write; |
2609 | write_targets ++; | |
3e198f78 N |
2610 | } else { |
2611 | /* may need to read from here */ | |
06f60385 N |
2612 | sector_t first_bad = MaxSector; |
2613 | int bad_sectors; | |
2614 | ||
2615 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2616 | &first_bad, &bad_sectors)) { | |
2617 | if (first_bad > sector_nr) | |
2618 | good_sectors = first_bad - sector_nr; | |
2619 | else { | |
2620 | bad_sectors -= (sector_nr - first_bad); | |
2621 | if (min_bad == 0 || | |
2622 | min_bad > bad_sectors) | |
2623 | min_bad = bad_sectors; | |
2624 | } | |
2625 | } | |
2626 | if (sector_nr < first_bad) { | |
2627 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2628 | if (wonly < 0) | |
2629 | wonly = i; | |
2630 | } else { | |
2631 | if (disk < 0) | |
2632 | disk = i; | |
2633 | } | |
796a5cf0 | 2634 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2635 | bio->bi_end_io = end_sync_read; |
2636 | read_targets++; | |
d57368af AL |
2637 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2638 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2639 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2640 | /* | |
2641 | * The device is suitable for reading (InSync), | |
2642 | * but has bad block(s) here. Let's try to correct them, | |
2643 | * if we are doing resync or repair. Otherwise, leave | |
2644 | * this device alone for this sync request. | |
2645 | */ | |
796a5cf0 | 2646 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2647 | bio->bi_end_io = end_sync_write; |
2648 | write_targets++; | |
3e198f78 | 2649 | } |
3e198f78 | 2650 | } |
06f60385 N |
2651 | if (bio->bi_end_io) { |
2652 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2653 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
06f60385 N |
2654 | bio->bi_bdev = rdev->bdev; |
2655 | bio->bi_private = r1_bio; | |
2656 | } | |
1da177e4 | 2657 | } |
3e198f78 N |
2658 | rcu_read_unlock(); |
2659 | if (disk < 0) | |
2660 | disk = wonly; | |
2661 | r1_bio->read_disk = disk; | |
191ea9b2 | 2662 | |
06f60385 N |
2663 | if (read_targets == 0 && min_bad > 0) { |
2664 | /* These sectors are bad on all InSync devices, so we | |
2665 | * need to mark them bad on all write targets | |
2666 | */ | |
2667 | int ok = 1; | |
8f19ccb2 | 2668 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2669 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2670 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2671 | ok = rdev_set_badblocks(rdev, sector_nr, |
2672 | min_bad, 0 | |
2673 | ) && ok; | |
2674 | } | |
2675 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2676 | *skipped = 1; | |
2677 | put_buf(r1_bio); | |
2678 | ||
2679 | if (!ok) { | |
2680 | /* Cannot record the badblocks, so need to | |
2681 | * abort the resync. | |
2682 | * If there are multiple read targets, could just | |
2683 | * fail the really bad ones ??? | |
2684 | */ | |
2685 | conf->recovery_disabled = mddev->recovery_disabled; | |
2686 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2687 | return 0; | |
2688 | } else | |
2689 | return min_bad; | |
2690 | ||
2691 | } | |
2692 | if (min_bad > 0 && min_bad < good_sectors) { | |
2693 | /* only resync enough to reach the next bad->good | |
2694 | * transition */ | |
2695 | good_sectors = min_bad; | |
2696 | } | |
2697 | ||
3e198f78 N |
2698 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2699 | /* extra read targets are also write targets */ | |
2700 | write_targets += read_targets-1; | |
2701 | ||
2702 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2703 | /* There is nowhere to write, so all non-sync |
2704 | * drives must be failed - so we are finished | |
2705 | */ | |
b7219ccb N |
2706 | sector_t rv; |
2707 | if (min_bad > 0) | |
2708 | max_sector = sector_nr + min_bad; | |
2709 | rv = max_sector - sector_nr; | |
57afd89f | 2710 | *skipped = 1; |
1da177e4 | 2711 | put_buf(r1_bio); |
1da177e4 LT |
2712 | return rv; |
2713 | } | |
2714 | ||
c6207277 N |
2715 | if (max_sector > mddev->resync_max) |
2716 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2717 | if (max_sector > sector_nr + good_sectors) |
2718 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2719 | nr_sectors = 0; |
289e99e8 | 2720 | sync_blocks = 0; |
1da177e4 LT |
2721 | do { |
2722 | struct page *page; | |
2723 | int len = PAGE_SIZE; | |
2724 | if (sector_nr + (len>>9) > max_sector) | |
2725 | len = (max_sector - sector_nr) << 9; | |
2726 | if (len == 0) | |
2727 | break; | |
6a806c51 N |
2728 | if (sync_blocks == 0) { |
2729 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2730 | &sync_blocks, still_degraded) && |
2731 | !conf->fullsync && | |
2732 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2733 | break; |
7571ae88 | 2734 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2735 | len = sync_blocks<<9; |
ab7a30c7 | 2736 | } |
191ea9b2 | 2737 | |
8f19ccb2 | 2738 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2739 | bio = r1_bio->bios[i]; |
2740 | if (bio->bi_end_io) { | |
d11c171e | 2741 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2742 | if (bio_add_page(bio, page, len, 0) == 0) { |
2743 | /* stop here */ | |
d11c171e | 2744 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2745 | while (i > 0) { |
2746 | i--; | |
2747 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2748 | if (bio->bi_end_io==NULL) |
2749 | continue; | |
1da177e4 LT |
2750 | /* remove last page from this bio */ |
2751 | bio->bi_vcnt--; | |
4f024f37 | 2752 | bio->bi_iter.bi_size -= len; |
b7c44ed9 | 2753 | bio_clear_flag(bio, BIO_SEG_VALID); |
1da177e4 LT |
2754 | } |
2755 | goto bio_full; | |
2756 | } | |
2757 | } | |
2758 | } | |
2759 | nr_sectors += len>>9; | |
2760 | sector_nr += len>>9; | |
191ea9b2 | 2761 | sync_blocks -= (len>>9); |
1da177e4 LT |
2762 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2763 | bio_full: | |
1da177e4 LT |
2764 | r1_bio->sectors = nr_sectors; |
2765 | ||
c40f341f GR |
2766 | if (mddev_is_clustered(mddev) && |
2767 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2768 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2769 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2770 | /* Send resync message */ | |
2771 | md_cluster_ops->resync_info_update(mddev, | |
2772 | conf->cluster_sync_low, | |
2773 | conf->cluster_sync_high); | |
2774 | } | |
2775 | ||
d11c171e N |
2776 | /* For a user-requested sync, we read all readable devices and do a |
2777 | * compare | |
2778 | */ | |
2779 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2780 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2781 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2782 | bio = r1_bio->bios[i]; |
2783 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2784 | read_targets--; |
ddac7c7e | 2785 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e N |
2786 | generic_make_request(bio); |
2787 | } | |
2788 | } | |
2789 | } else { | |
2790 | atomic_set(&r1_bio->remaining, 1); | |
2791 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2792 | md_sync_acct(bio->bi_bdev, nr_sectors); |
d11c171e | 2793 | generic_make_request(bio); |
1da177e4 | 2794 | |
d11c171e | 2795 | } |
1da177e4 LT |
2796 | return nr_sectors; |
2797 | } | |
2798 | ||
fd01b88c | 2799 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2800 | { |
2801 | if (sectors) | |
2802 | return sectors; | |
2803 | ||
2804 | return mddev->dev_sectors; | |
2805 | } | |
2806 | ||
e8096360 | 2807 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2808 | { |
e8096360 | 2809 | struct r1conf *conf; |
709ae487 | 2810 | int i; |
0eaf822c | 2811 | struct raid1_info *disk; |
3cb03002 | 2812 | struct md_rdev *rdev; |
709ae487 | 2813 | int err = -ENOMEM; |
1da177e4 | 2814 | |
e8096360 | 2815 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2816 | if (!conf) |
709ae487 | 2817 | goto abort; |
1da177e4 | 2818 | |
0eaf822c | 2819 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2820 | * mddev->raid_disks * 2, |
1da177e4 LT |
2821 | GFP_KERNEL); |
2822 | if (!conf->mirrors) | |
709ae487 | 2823 | goto abort; |
1da177e4 | 2824 | |
ddaf22ab N |
2825 | conf->tmppage = alloc_page(GFP_KERNEL); |
2826 | if (!conf->tmppage) | |
709ae487 | 2827 | goto abort; |
ddaf22ab | 2828 | |
709ae487 | 2829 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2830 | if (!conf->poolinfo) |
709ae487 | 2831 | goto abort; |
8f19ccb2 | 2832 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2833 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2834 | r1bio_pool_free, | |
2835 | conf->poolinfo); | |
2836 | if (!conf->r1bio_pool) | |
709ae487 N |
2837 | goto abort; |
2838 | ||
ed9bfdf1 | 2839 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2840 | |
c19d5798 | 2841 | err = -EINVAL; |
e7e72bf6 | 2842 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2843 | rdev_for_each(rdev, mddev) { |
aba336bd | 2844 | struct request_queue *q; |
709ae487 | 2845 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2846 | if (disk_idx >= mddev->raid_disks |
2847 | || disk_idx < 0) | |
2848 | continue; | |
c19d5798 | 2849 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2850 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2851 | else |
2852 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2853 | |
c19d5798 N |
2854 | if (disk->rdev) |
2855 | goto abort; | |
1da177e4 | 2856 | disk->rdev = rdev; |
aba336bd | 2857 | q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
2858 | |
2859 | disk->head_position = 0; | |
12cee5a8 | 2860 | disk->seq_start = MaxSector; |
1da177e4 LT |
2861 | } |
2862 | conf->raid_disks = mddev->raid_disks; | |
2863 | conf->mddev = mddev; | |
1da177e4 | 2864 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 2865 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
2866 | |
2867 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2868 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2869 | |
191ea9b2 | 2870 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2871 | conf->pending_count = 0; |
d890fa2b | 2872 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2873 | |
79ef3a8a | 2874 | conf->start_next_window = MaxSector; |
2875 | conf->current_window_requests = conf->next_window_requests = 0; | |
2876 | ||
c19d5798 | 2877 | err = -EIO; |
8f19ccb2 | 2878 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2879 | |
2880 | disk = conf->mirrors + i; | |
2881 | ||
c19d5798 N |
2882 | if (i < conf->raid_disks && |
2883 | disk[conf->raid_disks].rdev) { | |
2884 | /* This slot has a replacement. */ | |
2885 | if (!disk->rdev) { | |
2886 | /* No original, just make the replacement | |
2887 | * a recovering spare | |
2888 | */ | |
2889 | disk->rdev = | |
2890 | disk[conf->raid_disks].rdev; | |
2891 | disk[conf->raid_disks].rdev = NULL; | |
2892 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2893 | /* Original is not in_sync - bad */ | |
2894 | goto abort; | |
2895 | } | |
2896 | ||
5fd6c1dc N |
2897 | if (!disk->rdev || |
2898 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2899 | disk->head_position = 0; |
4f0a5e01 JB |
2900 | if (disk->rdev && |
2901 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2902 | conf->fullsync = 1; |
be4d3280 | 2903 | } |
1da177e4 | 2904 | } |
709ae487 | 2905 | |
709ae487 | 2906 | err = -ENOMEM; |
0232605d | 2907 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 2908 | if (!conf->thread) |
709ae487 | 2909 | goto abort; |
1da177e4 | 2910 | |
709ae487 N |
2911 | return conf; |
2912 | ||
2913 | abort: | |
2914 | if (conf) { | |
644df1a8 | 2915 | mempool_destroy(conf->r1bio_pool); |
709ae487 N |
2916 | kfree(conf->mirrors); |
2917 | safe_put_page(conf->tmppage); | |
2918 | kfree(conf->poolinfo); | |
2919 | kfree(conf); | |
2920 | } | |
2921 | return ERR_PTR(err); | |
2922 | } | |
2923 | ||
afa0f557 | 2924 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 2925 | static int raid1_run(struct mddev *mddev) |
709ae487 | 2926 | { |
e8096360 | 2927 | struct r1conf *conf; |
709ae487 | 2928 | int i; |
3cb03002 | 2929 | struct md_rdev *rdev; |
5220ea1e | 2930 | int ret; |
2ff8cc2c | 2931 | bool discard_supported = false; |
709ae487 N |
2932 | |
2933 | if (mddev->level != 1) { | |
1d41c216 N |
2934 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
2935 | mdname(mddev), mddev->level); | |
709ae487 N |
2936 | return -EIO; |
2937 | } | |
2938 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
2939 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
2940 | mdname(mddev)); | |
709ae487 N |
2941 | return -EIO; |
2942 | } | |
1da177e4 | 2943 | /* |
709ae487 N |
2944 | * copy the already verified devices into our private RAID1 |
2945 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 2946 | * should be freed in raid1_free()] |
1da177e4 | 2947 | */ |
709ae487 N |
2948 | if (mddev->private == NULL) |
2949 | conf = setup_conf(mddev); | |
2950 | else | |
2951 | conf = mddev->private; | |
1da177e4 | 2952 | |
709ae487 N |
2953 | if (IS_ERR(conf)) |
2954 | return PTR_ERR(conf); | |
1da177e4 | 2955 | |
c8dc9c65 | 2956 | if (mddev->queue) |
5026d7a9 PA |
2957 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
2958 | ||
dafb20fa | 2959 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
2960 | if (!mddev->gendisk) |
2961 | continue; | |
709ae487 N |
2962 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
2963 | rdev->data_offset << 9); | |
2ff8cc2c SL |
2964 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2965 | discard_supported = true; | |
1da177e4 | 2966 | } |
191ea9b2 | 2967 | |
709ae487 N |
2968 | mddev->degraded = 0; |
2969 | for (i=0; i < conf->raid_disks; i++) | |
2970 | if (conf->mirrors[i].rdev == NULL || | |
2971 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
2972 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
2973 | mddev->degraded++; | |
2974 | ||
2975 | if (conf->raid_disks - mddev->degraded == 1) | |
2976 | mddev->recovery_cp = MaxSector; | |
2977 | ||
8c6ac868 | 2978 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
2979 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
2980 | mdname(mddev)); | |
2981 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 2982 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 2983 | mddev->raid_disks); |
709ae487 | 2984 | |
1da177e4 LT |
2985 | /* |
2986 | * Ok, everything is just fine now | |
2987 | */ | |
709ae487 N |
2988 | mddev->thread = conf->thread; |
2989 | conf->thread = NULL; | |
2990 | mddev->private = conf; | |
46533ff7 | 2991 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 2992 | |
1f403624 | 2993 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 2994 | |
1ed7242e | 2995 | if (mddev->queue) { |
2ff8cc2c SL |
2996 | if (discard_supported) |
2997 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
2998 | mddev->queue); | |
2999 | else | |
3000 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3001 | mddev->queue); | |
1ed7242e | 3002 | } |
5220ea1e | 3003 | |
3004 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
3005 | if (ret) { |
3006 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 3007 | raid1_free(mddev, conf); |
5aa61f42 | 3008 | } |
5220ea1e | 3009 | return ret; |
1da177e4 LT |
3010 | } |
3011 | ||
afa0f557 | 3012 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3013 | { |
afa0f557 | 3014 | struct r1conf *conf = priv; |
409c57f3 | 3015 | |
644df1a8 | 3016 | mempool_destroy(conf->r1bio_pool); |
990a8baf | 3017 | kfree(conf->mirrors); |
0fea7ed8 | 3018 | safe_put_page(conf->tmppage); |
990a8baf | 3019 | kfree(conf->poolinfo); |
1da177e4 | 3020 | kfree(conf); |
1da177e4 LT |
3021 | } |
3022 | ||
fd01b88c | 3023 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3024 | { |
3025 | /* no resync is happening, and there is enough space | |
3026 | * on all devices, so we can resize. | |
3027 | * We need to make sure resync covers any new space. | |
3028 | * If the array is shrinking we should possibly wait until | |
3029 | * any io in the removed space completes, but it hardly seems | |
3030 | * worth it. | |
3031 | */ | |
a4a6125a N |
3032 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3033 | if (mddev->external_size && | |
3034 | mddev->array_sectors > newsize) | |
b522adcd | 3035 | return -EINVAL; |
a4a6125a N |
3036 | if (mddev->bitmap) { |
3037 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3038 | if (ret) | |
3039 | return ret; | |
3040 | } | |
3041 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 3042 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 3043 | revalidate_disk(mddev->gendisk); |
b522adcd | 3044 | if (sectors > mddev->dev_sectors && |
b098636c | 3045 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3046 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3047 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3048 | } | |
b522adcd | 3049 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3050 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3051 | return 0; |
3052 | } | |
3053 | ||
fd01b88c | 3054 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3055 | { |
3056 | /* We need to: | |
3057 | * 1/ resize the r1bio_pool | |
3058 | * 2/ resize conf->mirrors | |
3059 | * | |
3060 | * We allocate a new r1bio_pool if we can. | |
3061 | * Then raise a device barrier and wait until all IO stops. | |
3062 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3063 | * |
3064 | * At the same time, we "pack" the devices so that all the missing | |
3065 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3066 | */ |
3067 | mempool_t *newpool, *oldpool; | |
3068 | struct pool_info *newpoolinfo; | |
0eaf822c | 3069 | struct raid1_info *newmirrors; |
e8096360 | 3070 | struct r1conf *conf = mddev->private; |
63c70c4f | 3071 | int cnt, raid_disks; |
c04be0aa | 3072 | unsigned long flags; |
b5470dc5 | 3073 | int d, d2, err; |
1da177e4 | 3074 | |
63c70c4f | 3075 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3076 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3077 | mddev->layout != mddev->new_layout || |
3078 | mddev->level != mddev->new_level) { | |
664e7c41 | 3079 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3080 | mddev->new_layout = mddev->layout; |
3081 | mddev->new_level = mddev->level; | |
3082 | return -EINVAL; | |
3083 | } | |
3084 | ||
28c1b9fd GR |
3085 | if (!mddev_is_clustered(mddev)) { |
3086 | err = md_allow_write(mddev); | |
3087 | if (err) | |
3088 | return err; | |
3089 | } | |
2a2275d6 | 3090 | |
63c70c4f N |
3091 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3092 | ||
6ea9c07c N |
3093 | if (raid_disks < conf->raid_disks) { |
3094 | cnt=0; | |
3095 | for (d= 0; d < conf->raid_disks; d++) | |
3096 | if (conf->mirrors[d].rdev) | |
3097 | cnt++; | |
3098 | if (cnt > raid_disks) | |
1da177e4 | 3099 | return -EBUSY; |
6ea9c07c | 3100 | } |
1da177e4 LT |
3101 | |
3102 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3103 | if (!newpoolinfo) | |
3104 | return -ENOMEM; | |
3105 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3106 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3107 | |
3108 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3109 | r1bio_pool_free, newpoolinfo); | |
3110 | if (!newpool) { | |
3111 | kfree(newpoolinfo); | |
3112 | return -ENOMEM; | |
3113 | } | |
0eaf822c | 3114 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3115 | GFP_KERNEL); |
1da177e4 LT |
3116 | if (!newmirrors) { |
3117 | kfree(newpoolinfo); | |
3118 | mempool_destroy(newpool); | |
3119 | return -ENOMEM; | |
3120 | } | |
1da177e4 | 3121 | |
e2d59925 | 3122 | freeze_array(conf, 0); |
1da177e4 LT |
3123 | |
3124 | /* ok, everything is stopped */ | |
3125 | oldpool = conf->r1bio_pool; | |
3126 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3127 | |
a88aa786 | 3128 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3129 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3130 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3131 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3132 | rdev->raid_disk = d2; |
36fad858 NK |
3133 | sysfs_unlink_rdev(mddev, rdev); |
3134 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3135 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3136 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3137 | } |
a88aa786 N |
3138 | if (rdev) |
3139 | newmirrors[d2++].rdev = rdev; | |
3140 | } | |
1da177e4 LT |
3141 | kfree(conf->mirrors); |
3142 | conf->mirrors = newmirrors; | |
3143 | kfree(conf->poolinfo); | |
3144 | conf->poolinfo = newpoolinfo; | |
3145 | ||
c04be0aa | 3146 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3147 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3148 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3149 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3150 | mddev->delta_disks = 0; |
1da177e4 | 3151 | |
e2d59925 | 3152 | unfreeze_array(conf); |
1da177e4 | 3153 | |
985ca973 | 3154 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3155 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3156 | md_wakeup_thread(mddev->thread); | |
3157 | ||
3158 | mempool_destroy(oldpool); | |
3159 | return 0; | |
3160 | } | |
3161 | ||
fd01b88c | 3162 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3163 | { |
e8096360 | 3164 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3165 | |
3166 | switch(state) { | |
6eef4b21 N |
3167 | case 2: /* wake for suspend */ |
3168 | wake_up(&conf->wait_barrier); | |
3169 | break; | |
9e6603da | 3170 | case 1: |
07169fd4 | 3171 | freeze_array(conf, 0); |
36fa3063 | 3172 | break; |
9e6603da | 3173 | case 0: |
07169fd4 | 3174 | unfreeze_array(conf); |
36fa3063 N |
3175 | break; |
3176 | } | |
36fa3063 N |
3177 | } |
3178 | ||
fd01b88c | 3179 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3180 | { |
3181 | /* raid1 can take over: | |
3182 | * raid5 with 2 devices, any layout or chunk size | |
3183 | */ | |
3184 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3185 | struct r1conf *conf; |
709ae487 N |
3186 | mddev->new_level = 1; |
3187 | mddev->new_layout = 0; | |
3188 | mddev->new_chunk_sectors = 0; | |
3189 | conf = setup_conf(mddev); | |
3190 | if (!IS_ERR(conf)) | |
07169fd4 | 3191 | /* Array must appear to be quiesced */ |
3192 | conf->array_frozen = 1; | |
709ae487 N |
3193 | return conf; |
3194 | } | |
3195 | return ERR_PTR(-EINVAL); | |
3196 | } | |
1da177e4 | 3197 | |
84fc4b56 | 3198 | static struct md_personality raid1_personality = |
1da177e4 LT |
3199 | { |
3200 | .name = "raid1", | |
2604b703 | 3201 | .level = 1, |
1da177e4 | 3202 | .owner = THIS_MODULE, |
849674e4 SL |
3203 | .make_request = raid1_make_request, |
3204 | .run = raid1_run, | |
afa0f557 | 3205 | .free = raid1_free, |
849674e4 SL |
3206 | .status = raid1_status, |
3207 | .error_handler = raid1_error, | |
1da177e4 LT |
3208 | .hot_add_disk = raid1_add_disk, |
3209 | .hot_remove_disk= raid1_remove_disk, | |
3210 | .spare_active = raid1_spare_active, | |
849674e4 | 3211 | .sync_request = raid1_sync_request, |
1da177e4 | 3212 | .resize = raid1_resize, |
80c3a6ce | 3213 | .size = raid1_size, |
63c70c4f | 3214 | .check_reshape = raid1_reshape, |
36fa3063 | 3215 | .quiesce = raid1_quiesce, |
709ae487 | 3216 | .takeover = raid1_takeover, |
5c675f83 | 3217 | .congested = raid1_congested, |
1da177e4 LT |
3218 | }; |
3219 | ||
3220 | static int __init raid_init(void) | |
3221 | { | |
2604b703 | 3222 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3223 | } |
3224 | ||
3225 | static void raid_exit(void) | |
3226 | { | |
2604b703 | 3227 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3228 | } |
3229 | ||
3230 | module_init(raid_init); | |
3231 | module_exit(raid_exit); | |
3232 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3233 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3234 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3235 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3236 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3237 | |
3238 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |