]>
Commit | Line | Data |
---|---|---|
af1a8899 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * raid1.c : Multiple Devices driver for Linux | |
4 | * | |
5 | * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat | |
6 | * | |
7 | * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
8 | * | |
9 | * RAID-1 management functions. | |
10 | * | |
11 | * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 | |
12 | * | |
96de0e25 | 13 | * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> |
1da177e4 LT |
14 | * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> |
15 | * | |
191ea9b2 N |
16 | * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support |
17 | * bitmapped intelligence in resync: | |
18 | * | |
19 | * - bitmap marked during normal i/o | |
20 | * - bitmap used to skip nondirty blocks during sync | |
21 | * | |
22 | * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: | |
23 | * - persistent bitmap code | |
1da177e4 LT |
24 | */ |
25 | ||
5a0e3ad6 | 26 | #include <linux/slab.h> |
25570727 | 27 | #include <linux/delay.h> |
bff61975 | 28 | #include <linux/blkdev.h> |
056075c7 | 29 | #include <linux/module.h> |
bff61975 | 30 | #include <linux/seq_file.h> |
8bda470e | 31 | #include <linux/ratelimit.h> |
3f07c014 | 32 | |
109e3765 | 33 | #include <trace/events/block.h> |
3f07c014 | 34 | |
43b2e5d8 | 35 | #include "md.h" |
ef740c37 | 36 | #include "raid1.h" |
935fe098 | 37 | #include "md-bitmap.h" |
191ea9b2 | 38 | |
394ed8e4 SL |
39 | #define UNSUPPORTED_MDDEV_FLAGS \ |
40 | ((1L << MD_HAS_JOURNAL) | \ | |
ea0213e0 | 41 | (1L << MD_JOURNAL_CLEAN) | \ |
ddc08823 PB |
42 | (1L << MD_HAS_PPL) | \ |
43 | (1L << MD_HAS_MULTIPLE_PPLS)) | |
394ed8e4 | 44 | |
fd76863e | 45 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr); |
46 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr); | |
1da177e4 | 47 | |
578b54ad N |
48 | #define raid1_log(md, fmt, args...) \ |
49 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) | |
50 | ||
fb0eb5df ML |
51 | #include "raid1-10.c" |
52 | ||
3e148a32 GJ |
53 | static int check_and_add_wb(struct md_rdev *rdev, sector_t lo, sector_t hi) |
54 | { | |
55 | struct wb_info *wi, *temp_wi; | |
56 | unsigned long flags; | |
57 | int ret = 0; | |
58 | struct mddev *mddev = rdev->mddev; | |
59 | ||
60 | wi = mempool_alloc(mddev->wb_info_pool, GFP_NOIO); | |
61 | ||
62 | spin_lock_irqsave(&rdev->wb_list_lock, flags); | |
63 | list_for_each_entry(temp_wi, &rdev->wb_list, list) { | |
64 | /* collision happened */ | |
65 | if (hi > temp_wi->lo && lo < temp_wi->hi) { | |
66 | ret = -EBUSY; | |
67 | break; | |
68 | } | |
69 | } | |
70 | ||
71 | if (!ret) { | |
72 | wi->lo = lo; | |
73 | wi->hi = hi; | |
74 | list_add(&wi->list, &rdev->wb_list); | |
75 | } else | |
76 | mempool_free(wi, mddev->wb_info_pool); | |
77 | spin_unlock_irqrestore(&rdev->wb_list_lock, flags); | |
78 | ||
79 | return ret; | |
80 | } | |
81 | ||
82 | static void remove_wb(struct md_rdev *rdev, sector_t lo, sector_t hi) | |
83 | { | |
84 | struct wb_info *wi; | |
85 | unsigned long flags; | |
86 | int found = 0; | |
87 | struct mddev *mddev = rdev->mddev; | |
88 | ||
89 | spin_lock_irqsave(&rdev->wb_list_lock, flags); | |
90 | list_for_each_entry(wi, &rdev->wb_list, list) | |
91 | if (hi == wi->hi && lo == wi->lo) { | |
92 | list_del(&wi->list); | |
93 | mempool_free(wi, mddev->wb_info_pool); | |
94 | found = 1; | |
95 | break; | |
96 | } | |
97 | ||
98 | if (!found) | |
16d4b746 | 99 | WARN(1, "The write behind IO is not recorded\n"); |
3e148a32 GJ |
100 | spin_unlock_irqrestore(&rdev->wb_list_lock, flags); |
101 | wake_up(&rdev->wb_io_wait); | |
102 | } | |
103 | ||
98d30c58 ML |
104 | /* |
105 | * for resync bio, r1bio pointer can be retrieved from the per-bio | |
106 | * 'struct resync_pages'. | |
107 | */ | |
108 | static inline struct r1bio *get_resync_r1bio(struct bio *bio) | |
109 | { | |
110 | return get_resync_pages(bio)->raid_bio; | |
111 | } | |
112 | ||
dd0fc66f | 113 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
114 | { |
115 | struct pool_info *pi = data; | |
9f2c9d12 | 116 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
117 | |
118 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 119 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
120 | } |
121 | ||
8e005f7c | 122 | #define RESYNC_DEPTH 32 |
1da177e4 | 123 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
8e005f7c | 124 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
125 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
126 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
127 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
1da177e4 | 128 | |
dd0fc66f | 129 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
130 | { |
131 | struct pool_info *pi = data; | |
9f2c9d12 | 132 | struct r1bio *r1_bio; |
1da177e4 | 133 | struct bio *bio; |
da1aab3d | 134 | int need_pages; |
98d30c58 ML |
135 | int j; |
136 | struct resync_pages *rps; | |
1da177e4 LT |
137 | |
138 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 139 | if (!r1_bio) |
1da177e4 | 140 | return NULL; |
1da177e4 | 141 | |
6da2ec56 KC |
142 | rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages), |
143 | gfp_flags); | |
98d30c58 ML |
144 | if (!rps) |
145 | goto out_free_r1bio; | |
146 | ||
1da177e4 LT |
147 | /* |
148 | * Allocate bios : 1 for reading, n-1 for writing | |
149 | */ | |
150 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 151 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
152 | if (!bio) |
153 | goto out_free_bio; | |
154 | r1_bio->bios[j] = bio; | |
155 | } | |
156 | /* | |
157 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
158 | * the first bio. |
159 | * If this is a user-requested check/repair, allocate | |
160 | * RESYNC_PAGES for each bio. | |
1da177e4 | 161 | */ |
d11c171e | 162 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 163 | need_pages = pi->raid_disks; |
d11c171e | 164 | else |
da1aab3d | 165 | need_pages = 1; |
98d30c58 ML |
166 | for (j = 0; j < pi->raid_disks; j++) { |
167 | struct resync_pages *rp = &rps[j]; | |
168 | ||
d11c171e | 169 | bio = r1_bio->bios[j]; |
d11c171e | 170 | |
98d30c58 ML |
171 | if (j < need_pages) { |
172 | if (resync_alloc_pages(rp, gfp_flags)) | |
173 | goto out_free_pages; | |
174 | } else { | |
175 | memcpy(rp, &rps[0], sizeof(*rp)); | |
176 | resync_get_all_pages(rp); | |
177 | } | |
178 | ||
98d30c58 ML |
179 | rp->raid_bio = r1_bio; |
180 | bio->bi_private = rp; | |
1da177e4 LT |
181 | } |
182 | ||
183 | r1_bio->master_bio = NULL; | |
184 | ||
185 | return r1_bio; | |
186 | ||
da1aab3d | 187 | out_free_pages: |
491221f8 | 188 | while (--j >= 0) |
98d30c58 | 189 | resync_free_pages(&rps[j]); |
da1aab3d | 190 | |
1da177e4 | 191 | out_free_bio: |
8f19ccb2 | 192 | while (++j < pi->raid_disks) |
1da177e4 | 193 | bio_put(r1_bio->bios[j]); |
98d30c58 ML |
194 | kfree(rps); |
195 | ||
196 | out_free_r1bio: | |
c7afa803 | 197 | rbio_pool_free(r1_bio, data); |
1da177e4 LT |
198 | return NULL; |
199 | } | |
200 | ||
201 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
202 | { | |
203 | struct pool_info *pi = data; | |
98d30c58 | 204 | int i; |
9f2c9d12 | 205 | struct r1bio *r1bio = __r1_bio; |
98d30c58 | 206 | struct resync_pages *rp = NULL; |
1da177e4 | 207 | |
98d30c58 ML |
208 | for (i = pi->raid_disks; i--; ) { |
209 | rp = get_resync_pages(r1bio->bios[i]); | |
210 | resync_free_pages(rp); | |
1da177e4 | 211 | bio_put(r1bio->bios[i]); |
98d30c58 ML |
212 | } |
213 | ||
214 | /* resync pages array stored in the 1st bio's .bi_private */ | |
215 | kfree(rp); | |
1da177e4 | 216 | |
c7afa803 | 217 | rbio_pool_free(r1bio, data); |
1da177e4 LT |
218 | } |
219 | ||
e8096360 | 220 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
221 | { |
222 | int i; | |
223 | ||
8f19ccb2 | 224 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 225 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 226 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
227 | bio_put(*bio); |
228 | *bio = NULL; | |
229 | } | |
230 | } | |
231 | ||
9f2c9d12 | 232 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 233 | { |
e8096360 | 234 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 235 | |
1da177e4 | 236 | put_all_bios(conf, r1_bio); |
afeee514 | 237 | mempool_free(r1_bio, &conf->r1bio_pool); |
1da177e4 LT |
238 | } |
239 | ||
9f2c9d12 | 240 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 241 | { |
e8096360 | 242 | struct r1conf *conf = r1_bio->mddev->private; |
af5f42a7 | 243 | sector_t sect = r1_bio->sector; |
3e198f78 N |
244 | int i; |
245 | ||
8f19ccb2 | 246 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
247 | struct bio *bio = r1_bio->bios[i]; |
248 | if (bio->bi_end_io) | |
249 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
250 | } | |
1da177e4 | 251 | |
afeee514 | 252 | mempool_free(r1_bio, &conf->r1buf_pool); |
1da177e4 | 253 | |
af5f42a7 | 254 | lower_barrier(conf, sect); |
1da177e4 LT |
255 | } |
256 | ||
9f2c9d12 | 257 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
258 | { |
259 | unsigned long flags; | |
fd01b88c | 260 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 261 | struct r1conf *conf = mddev->private; |
fd76863e | 262 | int idx; |
1da177e4 | 263 | |
fd76863e | 264 | idx = sector_to_idx(r1_bio->sector); |
1da177e4 LT |
265 | spin_lock_irqsave(&conf->device_lock, flags); |
266 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
824e47da | 267 | atomic_inc(&conf->nr_queued[idx]); |
1da177e4 LT |
268 | spin_unlock_irqrestore(&conf->device_lock, flags); |
269 | ||
17999be4 | 270 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
271 | md_wakeup_thread(mddev->thread); |
272 | } | |
273 | ||
274 | /* | |
275 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
276 | * operation and are ready to return a success/failure code to the buffer | |
277 | * cache layer. | |
278 | */ | |
9f2c9d12 | 279 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
280 | { |
281 | struct bio *bio = r1_bio->master_bio; | |
e8096360 | 282 | struct r1conf *conf = r1_bio->mddev->private; |
d2eb35ac N |
283 | |
284 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4e4cbee9 | 285 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 286 | |
37011e3a N |
287 | bio_endio(bio); |
288 | /* | |
289 | * Wake up any possible resync thread that waits for the device | |
290 | * to go idle. | |
291 | */ | |
292 | allow_barrier(conf, r1_bio->sector); | |
d2eb35ac N |
293 | } |
294 | ||
9f2c9d12 | 295 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
296 | { |
297 | struct bio *bio = r1_bio->master_bio; | |
298 | ||
4b6d287f N |
299 | /* if nobody has done the final endio yet, do it now */ |
300 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
301 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
302 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
303 | (unsigned long long) bio->bi_iter.bi_sector, |
304 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 305 | |
d2eb35ac | 306 | call_bio_endio(r1_bio); |
4b6d287f | 307 | } |
1da177e4 LT |
308 | free_r1bio(r1_bio); |
309 | } | |
310 | ||
311 | /* | |
312 | * Update disk head position estimator based on IRQ completion info. | |
313 | */ | |
9f2c9d12 | 314 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 315 | { |
e8096360 | 316 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
317 | |
318 | conf->mirrors[disk].head_position = | |
319 | r1_bio->sector + (r1_bio->sectors); | |
320 | } | |
321 | ||
ba3ae3be NK |
322 | /* |
323 | * Find the disk number which triggered given bio | |
324 | */ | |
9f2c9d12 | 325 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
326 | { |
327 | int mirror; | |
30194636 N |
328 | struct r1conf *conf = r1_bio->mddev->private; |
329 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 330 | |
8f19ccb2 | 331 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
332 | if (r1_bio->bios[mirror] == bio) |
333 | break; | |
334 | ||
8f19ccb2 | 335 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
336 | update_head_pos(mirror, r1_bio); |
337 | ||
338 | return mirror; | |
339 | } | |
340 | ||
4246a0b6 | 341 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 342 | { |
4e4cbee9 | 343 | int uptodate = !bio->bi_status; |
9f2c9d12 | 344 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 345 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 346 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 347 | |
1da177e4 LT |
348 | /* |
349 | * this branch is our 'one mirror IO has finished' event handler: | |
350 | */ | |
e5872d58 | 351 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 352 | |
dd00a99e N |
353 | if (uptodate) |
354 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
355 | else if (test_bit(FailFast, &rdev->flags) && |
356 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
357 | /* This was a fail-fast read so we definitely | |
358 | * want to retry */ | |
359 | ; | |
dd00a99e N |
360 | else { |
361 | /* If all other devices have failed, we want to return | |
362 | * the error upwards rather than fail the last device. | |
363 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 364 | */ |
dd00a99e N |
365 | unsigned long flags; |
366 | spin_lock_irqsave(&conf->device_lock, flags); | |
367 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
368 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 369 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
370 | uptodate = 1; |
371 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
372 | } | |
1da177e4 | 373 | |
7ad4d4a6 | 374 | if (uptodate) { |
1da177e4 | 375 | raid_end_bio_io(r1_bio); |
e5872d58 | 376 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 377 | } else { |
1da177e4 LT |
378 | /* |
379 | * oops, read error: | |
380 | */ | |
381 | char b[BDEVNAME_SIZE]; | |
1d41c216 N |
382 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
383 | mdname(conf->mddev), | |
384 | bdevname(rdev->bdev, b), | |
385 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 386 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 387 | reschedule_retry(r1_bio); |
7ad4d4a6 | 388 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 389 | } |
1da177e4 LT |
390 | } |
391 | ||
9f2c9d12 | 392 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
393 | { |
394 | /* it really is the end of this request */ | |
395 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
841c1316 ML |
396 | bio_free_pages(r1_bio->behind_master_bio); |
397 | bio_put(r1_bio->behind_master_bio); | |
398 | r1_bio->behind_master_bio = NULL; | |
cd5ff9a1 N |
399 | } |
400 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
401 | md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
402 | r1_bio->sectors, | |
403 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
404 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
cd5ff9a1 N |
405 | md_write_end(r1_bio->mddev); |
406 | } | |
407 | ||
9f2c9d12 | 408 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 409 | { |
cd5ff9a1 N |
410 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
411 | return; | |
412 | ||
413 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
414 | reschedule_retry(r1_bio); | |
415 | else { | |
416 | close_write(r1_bio); | |
4367af55 N |
417 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
418 | reschedule_retry(r1_bio); | |
419 | else | |
420 | raid_end_bio_io(r1_bio); | |
4e78064f N |
421 | } |
422 | } | |
423 | ||
4246a0b6 | 424 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 425 | { |
9f2c9d12 | 426 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 427 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 428 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 429 | struct bio *to_put = NULL; |
e5872d58 N |
430 | int mirror = find_bio_disk(r1_bio, bio); |
431 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd SL |
432 | bool discard_error; |
433 | ||
4e4cbee9 | 434 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 435 | |
e9c7469b TH |
436 | /* |
437 | * 'one mirror IO has finished' event handler: | |
438 | */ | |
4e4cbee9 | 439 | if (bio->bi_status && !discard_error) { |
e5872d58 N |
440 | set_bit(WriteErrorSeen, &rdev->flags); |
441 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
442 | set_bit(MD_RECOVERY_NEEDED, & |
443 | conf->mddev->recovery); | |
444 | ||
212e7eb7 N |
445 | if (test_bit(FailFast, &rdev->flags) && |
446 | (bio->bi_opf & MD_FAILFAST) && | |
447 | /* We never try FailFast to WriteMostly devices */ | |
448 | !test_bit(WriteMostly, &rdev->flags)) { | |
449 | md_error(r1_bio->mddev, rdev); | |
0f7be3f2 YY |
450 | } |
451 | ||
452 | /* | |
453 | * When the device is faulty, it is not necessary to | |
454 | * handle write error. | |
455 | * For failfast, this is the only remaining device, | |
456 | * We need to retry the write without FailFast. | |
457 | */ | |
458 | if (!test_bit(Faulty, &rdev->flags)) | |
212e7eb7 | 459 | set_bit(R1BIO_WriteError, &r1_bio->state); |
0f7be3f2 YY |
460 | else { |
461 | /* Finished with this branch */ | |
462 | r1_bio->bios[mirror] = NULL; | |
463 | to_put = bio; | |
464 | } | |
4367af55 | 465 | } else { |
1da177e4 | 466 | /* |
e9c7469b TH |
467 | * Set R1BIO_Uptodate in our master bio, so that we |
468 | * will return a good error code for to the higher | |
469 | * levels even if IO on some other mirrored buffer | |
470 | * fails. | |
471 | * | |
472 | * The 'master' represents the composite IO operation | |
473 | * to user-side. So if something waits for IO, then it | |
474 | * will wait for the 'master' bio. | |
1da177e4 | 475 | */ |
4367af55 N |
476 | sector_t first_bad; |
477 | int bad_sectors; | |
478 | ||
cd5ff9a1 N |
479 | r1_bio->bios[mirror] = NULL; |
480 | to_put = bio; | |
3056e3ae AL |
481 | /* |
482 | * Do not set R1BIO_Uptodate if the current device is | |
483 | * rebuilding or Faulty. This is because we cannot use | |
484 | * such device for properly reading the data back (we could | |
485 | * potentially use it, if the current write would have felt | |
486 | * before rdev->recovery_offset, but for simplicity we don't | |
487 | * check this here. | |
488 | */ | |
e5872d58 N |
489 | if (test_bit(In_sync, &rdev->flags) && |
490 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 491 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 492 | |
4367af55 | 493 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 494 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 495 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
496 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
497 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
498 | } | |
499 | } | |
500 | ||
e9c7469b | 501 | if (behind) { |
3e148a32 GJ |
502 | if (test_bit(WBCollisionCheck, &rdev->flags)) { |
503 | sector_t lo = r1_bio->sector; | |
504 | sector_t hi = r1_bio->sector + r1_bio->sectors; | |
505 | ||
506 | remove_wb(rdev, lo, hi); | |
507 | } | |
e5872d58 | 508 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
509 | atomic_dec(&r1_bio->behind_remaining); |
510 | ||
511 | /* | |
512 | * In behind mode, we ACK the master bio once the I/O | |
513 | * has safely reached all non-writemostly | |
514 | * disks. Setting the Returned bit ensures that this | |
515 | * gets done only once -- we don't ever want to return | |
516 | * -EIO here, instead we'll wait | |
517 | */ | |
518 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
519 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
520 | /* Maybe we can return now */ | |
521 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
522 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
523 | pr_debug("raid1: behind end write sectors" |
524 | " %llu-%llu\n", | |
4f024f37 KO |
525 | (unsigned long long) mbio->bi_iter.bi_sector, |
526 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 527 | call_bio_endio(r1_bio); |
4b6d287f N |
528 | } |
529 | } | |
530 | } | |
4367af55 | 531 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 532 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 533 | |
1da177e4 | 534 | /* |
1da177e4 LT |
535 | * Let's see if all mirrored write operations have finished |
536 | * already. | |
537 | */ | |
af6d7b76 | 538 | r1_bio_write_done(r1_bio); |
c70810b3 | 539 | |
04b857f7 N |
540 | if (to_put) |
541 | bio_put(to_put); | |
1da177e4 LT |
542 | } |
543 | ||
fd76863e | 544 | static sector_t align_to_barrier_unit_end(sector_t start_sector, |
545 | sector_t sectors) | |
546 | { | |
547 | sector_t len; | |
548 | ||
549 | WARN_ON(sectors == 0); | |
550 | /* | |
551 | * len is the number of sectors from start_sector to end of the | |
552 | * barrier unit which start_sector belongs to. | |
553 | */ | |
554 | len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - | |
555 | start_sector; | |
556 | ||
557 | if (len > sectors) | |
558 | len = sectors; | |
559 | ||
560 | return len; | |
561 | } | |
562 | ||
1da177e4 LT |
563 | /* |
564 | * This routine returns the disk from which the requested read should | |
565 | * be done. There is a per-array 'next expected sequential IO' sector | |
566 | * number - if this matches on the next IO then we use the last disk. | |
567 | * There is also a per-disk 'last know head position' sector that is | |
568 | * maintained from IRQ contexts, both the normal and the resync IO | |
569 | * completion handlers update this position correctly. If there is no | |
570 | * perfect sequential match then we pick the disk whose head is closest. | |
571 | * | |
572 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
573 | * because position is mirror, not device based. | |
574 | * | |
575 | * The rdev for the device selected will have nr_pending incremented. | |
576 | */ | |
e8096360 | 577 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 578 | { |
af3a2cd6 | 579 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
580 | int sectors; |
581 | int best_good_sectors; | |
9dedf603 SL |
582 | int best_disk, best_dist_disk, best_pending_disk; |
583 | int has_nonrot_disk; | |
be4d3280 | 584 | int disk; |
76073054 | 585 | sector_t best_dist; |
9dedf603 | 586 | unsigned int min_pending; |
3cb03002 | 587 | struct md_rdev *rdev; |
f3ac8bf7 | 588 | int choose_first; |
12cee5a8 | 589 | int choose_next_idle; |
1da177e4 LT |
590 | |
591 | rcu_read_lock(); | |
592 | /* | |
8ddf9efe | 593 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
594 | * device if no resync is going on, or below the resync window. |
595 | * We take the first readable disk when above the resync window. | |
596 | */ | |
597 | retry: | |
d2eb35ac | 598 | sectors = r1_bio->sectors; |
76073054 | 599 | best_disk = -1; |
9dedf603 | 600 | best_dist_disk = -1; |
76073054 | 601 | best_dist = MaxSector; |
9dedf603 SL |
602 | best_pending_disk = -1; |
603 | min_pending = UINT_MAX; | |
d2eb35ac | 604 | best_good_sectors = 0; |
9dedf603 | 605 | has_nonrot_disk = 0; |
12cee5a8 | 606 | choose_next_idle = 0; |
2e52d449 | 607 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 608 | |
7d49ffcf GR |
609 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
610 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 611 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
612 | this_sector + sectors))) |
613 | choose_first = 1; | |
614 | else | |
615 | choose_first = 0; | |
1da177e4 | 616 | |
be4d3280 | 617 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 618 | sector_t dist; |
d2eb35ac N |
619 | sector_t first_bad; |
620 | int bad_sectors; | |
9dedf603 | 621 | unsigned int pending; |
12cee5a8 | 622 | bool nonrot; |
d2eb35ac | 623 | |
f3ac8bf7 N |
624 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
625 | if (r1_bio->bios[disk] == IO_BLOCKED | |
626 | || rdev == NULL | |
76073054 | 627 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 628 | continue; |
76073054 N |
629 | if (!test_bit(In_sync, &rdev->flags) && |
630 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 631 | continue; |
76073054 N |
632 | if (test_bit(WriteMostly, &rdev->flags)) { |
633 | /* Don't balance among write-mostly, just | |
634 | * use the first as a last resort */ | |
d1901ef0 | 635 | if (best_dist_disk < 0) { |
307729c8 N |
636 | if (is_badblock(rdev, this_sector, sectors, |
637 | &first_bad, &bad_sectors)) { | |
816b0acf | 638 | if (first_bad <= this_sector) |
307729c8 N |
639 | /* Cannot use this */ |
640 | continue; | |
641 | best_good_sectors = first_bad - this_sector; | |
642 | } else | |
643 | best_good_sectors = sectors; | |
d1901ef0 TH |
644 | best_dist_disk = disk; |
645 | best_pending_disk = disk; | |
307729c8 | 646 | } |
76073054 N |
647 | continue; |
648 | } | |
649 | /* This is a reasonable device to use. It might | |
650 | * even be best. | |
651 | */ | |
d2eb35ac N |
652 | if (is_badblock(rdev, this_sector, sectors, |
653 | &first_bad, &bad_sectors)) { | |
654 | if (best_dist < MaxSector) | |
655 | /* already have a better device */ | |
656 | continue; | |
657 | if (first_bad <= this_sector) { | |
658 | /* cannot read here. If this is the 'primary' | |
659 | * device, then we must not read beyond | |
660 | * bad_sectors from another device.. | |
661 | */ | |
662 | bad_sectors -= (this_sector - first_bad); | |
663 | if (choose_first && sectors > bad_sectors) | |
664 | sectors = bad_sectors; | |
665 | if (best_good_sectors > sectors) | |
666 | best_good_sectors = sectors; | |
667 | ||
668 | } else { | |
669 | sector_t good_sectors = first_bad - this_sector; | |
670 | if (good_sectors > best_good_sectors) { | |
671 | best_good_sectors = good_sectors; | |
672 | best_disk = disk; | |
673 | } | |
674 | if (choose_first) | |
675 | break; | |
676 | } | |
677 | continue; | |
d82dd0e3 TM |
678 | } else { |
679 | if ((sectors > best_good_sectors) && (best_disk >= 0)) | |
680 | best_disk = -1; | |
d2eb35ac | 681 | best_good_sectors = sectors; |
d82dd0e3 | 682 | } |
d2eb35ac | 683 | |
2e52d449 N |
684 | if (best_disk >= 0) |
685 | /* At least two disks to choose from so failfast is OK */ | |
686 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
687 | ||
12cee5a8 SL |
688 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
689 | has_nonrot_disk |= nonrot; | |
9dedf603 | 690 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 691 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 692 | if (choose_first) { |
76073054 | 693 | best_disk = disk; |
1da177e4 LT |
694 | break; |
695 | } | |
12cee5a8 SL |
696 | /* Don't change to another disk for sequential reads */ |
697 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
698 | || dist == 0) { | |
699 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
700 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
701 | ||
702 | best_disk = disk; | |
703 | /* | |
704 | * If buffered sequential IO size exceeds optimal | |
705 | * iosize, check if there is idle disk. If yes, choose | |
706 | * the idle disk. read_balance could already choose an | |
707 | * idle disk before noticing it's a sequential IO in | |
708 | * this disk. This doesn't matter because this disk | |
709 | * will idle, next time it will be utilized after the | |
710 | * first disk has IO size exceeds optimal iosize. In | |
711 | * this way, iosize of the first disk will be optimal | |
712 | * iosize at least. iosize of the second disk might be | |
713 | * small, but not a big deal since when the second disk | |
714 | * starts IO, the first disk is likely still busy. | |
715 | */ | |
716 | if (nonrot && opt_iosize > 0 && | |
717 | mirror->seq_start != MaxSector && | |
718 | mirror->next_seq_sect > opt_iosize && | |
719 | mirror->next_seq_sect - opt_iosize >= | |
720 | mirror->seq_start) { | |
721 | choose_next_idle = 1; | |
722 | continue; | |
723 | } | |
724 | break; | |
725 | } | |
12cee5a8 SL |
726 | |
727 | if (choose_next_idle) | |
728 | continue; | |
9dedf603 SL |
729 | |
730 | if (min_pending > pending) { | |
731 | min_pending = pending; | |
732 | best_pending_disk = disk; | |
733 | } | |
734 | ||
76073054 N |
735 | if (dist < best_dist) { |
736 | best_dist = dist; | |
9dedf603 | 737 | best_dist_disk = disk; |
1da177e4 | 738 | } |
f3ac8bf7 | 739 | } |
1da177e4 | 740 | |
9dedf603 SL |
741 | /* |
742 | * If all disks are rotational, choose the closest disk. If any disk is | |
743 | * non-rotational, choose the disk with less pending request even the | |
744 | * disk is rotational, which might/might not be optimal for raids with | |
745 | * mixed ratation/non-rotational disks depending on workload. | |
746 | */ | |
747 | if (best_disk == -1) { | |
2e52d449 | 748 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
749 | best_disk = best_pending_disk; |
750 | else | |
751 | best_disk = best_dist_disk; | |
752 | } | |
753 | ||
76073054 N |
754 | if (best_disk >= 0) { |
755 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
756 | if (!rdev) |
757 | goto retry; | |
758 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 759 | sectors = best_good_sectors; |
12cee5a8 SL |
760 | |
761 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
762 | conf->mirrors[best_disk].seq_start = this_sector; | |
763 | ||
be4d3280 | 764 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
765 | } |
766 | rcu_read_unlock(); | |
d2eb35ac | 767 | *max_sectors = sectors; |
1da177e4 | 768 | |
76073054 | 769 | return best_disk; |
1da177e4 LT |
770 | } |
771 | ||
5c675f83 | 772 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 773 | { |
e8096360 | 774 | struct r1conf *conf = mddev->private; |
0d129228 N |
775 | int i, ret = 0; |
776 | ||
4452226e | 777 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
778 | conf->pending_count >= max_queued_requests) |
779 | return 1; | |
780 | ||
0d129228 | 781 | rcu_read_lock(); |
f53e29fc | 782 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 783 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 784 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 785 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 786 | |
1ed7242e JB |
787 | BUG_ON(!q); |
788 | ||
0d129228 N |
789 | /* Note the '|| 1' - when read_balance prefers |
790 | * non-congested targets, it can be removed | |
791 | */ | |
4452226e | 792 | if ((bits & (1 << WB_async_congested)) || 1) |
dc3b17cc | 793 | ret |= bdi_congested(q->backing_dev_info, bits); |
0d129228 | 794 | else |
dc3b17cc | 795 | ret &= bdi_congested(q->backing_dev_info, bits); |
0d129228 N |
796 | } |
797 | } | |
798 | rcu_read_unlock(); | |
799 | return ret; | |
800 | } | |
0d129228 | 801 | |
673ca68d N |
802 | static void flush_bio_list(struct r1conf *conf, struct bio *bio) |
803 | { | |
804 | /* flush any pending bitmap writes to disk before proceeding w/ I/O */ | |
e64e4018 | 805 | md_bitmap_unplug(conf->mddev->bitmap); |
673ca68d N |
806 | wake_up(&conf->wait_barrier); |
807 | ||
808 | while (bio) { /* submit pending writes */ | |
809 | struct bio *next = bio->bi_next; | |
74d46992 | 810 | struct md_rdev *rdev = (void *)bio->bi_disk; |
673ca68d | 811 | bio->bi_next = NULL; |
74d46992 | 812 | bio_set_dev(bio, rdev->bdev); |
673ca68d | 813 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 814 | bio_io_error(bio); |
673ca68d | 815 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
74d46992 | 816 | !blk_queue_discard(bio->bi_disk->queue))) |
673ca68d N |
817 | /* Just ignore it */ |
818 | bio_endio(bio); | |
819 | else | |
820 | generic_make_request(bio); | |
821 | bio = next; | |
822 | } | |
823 | } | |
824 | ||
e8096360 | 825 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
826 | { |
827 | /* Any writes that have been queued but are awaiting | |
828 | * bitmap updates get flushed here. | |
a35e63ef | 829 | */ |
a35e63ef N |
830 | spin_lock_irq(&conf->device_lock); |
831 | ||
832 | if (conf->pending_bio_list.head) { | |
18022a1b | 833 | struct blk_plug plug; |
a35e63ef | 834 | struct bio *bio; |
18022a1b | 835 | |
a35e63ef | 836 | bio = bio_list_get(&conf->pending_bio_list); |
34db0cd6 | 837 | conf->pending_count = 0; |
a35e63ef | 838 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
839 | |
840 | /* | |
841 | * As this is called in a wait_event() loop (see freeze_array), | |
842 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
843 | * cause a warning when we prepare to wait again. As it is | |
844 | * rare that this path is taken, it is perfectly safe to force | |
845 | * us to go around the wait_event() loop again, so the warning | |
846 | * is a false-positive. Silence the warning by resetting | |
847 | * thread state | |
848 | */ | |
849 | __set_current_state(TASK_RUNNING); | |
18022a1b | 850 | blk_start_plug(&plug); |
673ca68d | 851 | flush_bio_list(conf, bio); |
18022a1b | 852 | blk_finish_plug(&plug); |
a35e63ef N |
853 | } else |
854 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
855 | } |
856 | ||
17999be4 N |
857 | /* Barriers.... |
858 | * Sometimes we need to suspend IO while we do something else, | |
859 | * either some resync/recovery, or reconfigure the array. | |
860 | * To do this we raise a 'barrier'. | |
861 | * The 'barrier' is a counter that can be raised multiple times | |
862 | * to count how many activities are happening which preclude | |
863 | * normal IO. | |
864 | * We can only raise the barrier if there is no pending IO. | |
865 | * i.e. if nr_pending == 0. | |
866 | * We choose only to raise the barrier if no-one is waiting for the | |
867 | * barrier to go down. This means that as soon as an IO request | |
868 | * is ready, no other operations which require a barrier will start | |
869 | * until the IO request has had a chance. | |
870 | * | |
871 | * So: regular IO calls 'wait_barrier'. When that returns there | |
872 | * is no backgroup IO happening, It must arrange to call | |
873 | * allow_barrier when it has finished its IO. | |
874 | * backgroup IO calls must call raise_barrier. Once that returns | |
875 | * there is no normal IO happeing. It must arrange to call | |
876 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 877 | */ |
8c242593 | 878 | static sector_t raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 | 879 | { |
fd76863e | 880 | int idx = sector_to_idx(sector_nr); |
881 | ||
1da177e4 | 882 | spin_lock_irq(&conf->resync_lock); |
17999be4 N |
883 | |
884 | /* Wait until no block IO is waiting */ | |
824e47da | 885 | wait_event_lock_irq(conf->wait_barrier, |
886 | !atomic_read(&conf->nr_waiting[idx]), | |
eed8c02e | 887 | conf->resync_lock); |
17999be4 N |
888 | |
889 | /* block any new IO from starting */ | |
824e47da | 890 | atomic_inc(&conf->barrier[idx]); |
891 | /* | |
892 | * In raise_barrier() we firstly increase conf->barrier[idx] then | |
893 | * check conf->nr_pending[idx]. In _wait_barrier() we firstly | |
894 | * increase conf->nr_pending[idx] then check conf->barrier[idx]. | |
895 | * A memory barrier here to make sure conf->nr_pending[idx] won't | |
896 | * be fetched before conf->barrier[idx] is increased. Otherwise | |
897 | * there will be a race between raise_barrier() and _wait_barrier(). | |
898 | */ | |
899 | smp_mb__after_atomic(); | |
17999be4 | 900 | |
79ef3a8a | 901 | /* For these conditions we must wait: |
902 | * A: while the array is in frozen state | |
fd76863e | 903 | * B: while conf->nr_pending[idx] is not 0, meaning regular I/O |
904 | * existing in corresponding I/O barrier bucket. | |
905 | * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches | |
906 | * max resync count which allowed on current I/O barrier bucket. | |
79ef3a8a | 907 | */ |
17999be4 | 908 | wait_event_lock_irq(conf->wait_barrier, |
8c242593 | 909 | (!conf->array_frozen && |
824e47da | 910 | !atomic_read(&conf->nr_pending[idx]) && |
8c242593 YY |
911 | atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || |
912 | test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), | |
eed8c02e | 913 | conf->resync_lock); |
17999be4 | 914 | |
8c242593 YY |
915 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
916 | atomic_dec(&conf->barrier[idx]); | |
917 | spin_unlock_irq(&conf->resync_lock); | |
918 | wake_up(&conf->wait_barrier); | |
919 | return -EINTR; | |
920 | } | |
921 | ||
43ac9b84 | 922 | atomic_inc(&conf->nr_sync_pending); |
17999be4 | 923 | spin_unlock_irq(&conf->resync_lock); |
8c242593 YY |
924 | |
925 | return 0; | |
17999be4 N |
926 | } |
927 | ||
fd76863e | 928 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 929 | { |
fd76863e | 930 | int idx = sector_to_idx(sector_nr); |
931 | ||
824e47da | 932 | BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); |
fd76863e | 933 | |
824e47da | 934 | atomic_dec(&conf->barrier[idx]); |
43ac9b84 | 935 | atomic_dec(&conf->nr_sync_pending); |
17999be4 N |
936 | wake_up(&conf->wait_barrier); |
937 | } | |
938 | ||
fd76863e | 939 | static void _wait_barrier(struct r1conf *conf, int idx) |
17999be4 | 940 | { |
824e47da | 941 | /* |
942 | * We need to increase conf->nr_pending[idx] very early here, | |
943 | * then raise_barrier() can be blocked when it waits for | |
944 | * conf->nr_pending[idx] to be 0. Then we can avoid holding | |
945 | * conf->resync_lock when there is no barrier raised in same | |
946 | * barrier unit bucket. Also if the array is frozen, I/O | |
947 | * should be blocked until array is unfrozen. | |
948 | */ | |
949 | atomic_inc(&conf->nr_pending[idx]); | |
950 | /* | |
951 | * In _wait_barrier() we firstly increase conf->nr_pending[idx], then | |
952 | * check conf->barrier[idx]. In raise_barrier() we firstly increase | |
953 | * conf->barrier[idx], then check conf->nr_pending[idx]. A memory | |
954 | * barrier is necessary here to make sure conf->barrier[idx] won't be | |
955 | * fetched before conf->nr_pending[idx] is increased. Otherwise there | |
956 | * will be a race between _wait_barrier() and raise_barrier(). | |
957 | */ | |
958 | smp_mb__after_atomic(); | |
79ef3a8a | 959 | |
824e47da | 960 | /* |
961 | * Don't worry about checking two atomic_t variables at same time | |
962 | * here. If during we check conf->barrier[idx], the array is | |
963 | * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is | |
964 | * 0, it is safe to return and make the I/O continue. Because the | |
965 | * array is frozen, all I/O returned here will eventually complete | |
966 | * or be queued, no race will happen. See code comment in | |
967 | * frozen_array(). | |
968 | */ | |
969 | if (!READ_ONCE(conf->array_frozen) && | |
970 | !atomic_read(&conf->barrier[idx])) | |
971 | return; | |
79ef3a8a | 972 | |
824e47da | 973 | /* |
974 | * After holding conf->resync_lock, conf->nr_pending[idx] | |
975 | * should be decreased before waiting for barrier to drop. | |
976 | * Otherwise, we may encounter a race condition because | |
977 | * raise_barrer() might be waiting for conf->nr_pending[idx] | |
978 | * to be 0 at same time. | |
979 | */ | |
980 | spin_lock_irq(&conf->resync_lock); | |
981 | atomic_inc(&conf->nr_waiting[idx]); | |
982 | atomic_dec(&conf->nr_pending[idx]); | |
983 | /* | |
984 | * In case freeze_array() is waiting for | |
985 | * get_unqueued_pending() == extra | |
986 | */ | |
987 | wake_up(&conf->wait_barrier); | |
988 | /* Wait for the barrier in same barrier unit bucket to drop. */ | |
989 | wait_event_lock_irq(conf->wait_barrier, | |
990 | !conf->array_frozen && | |
991 | !atomic_read(&conf->barrier[idx]), | |
992 | conf->resync_lock); | |
993 | atomic_inc(&conf->nr_pending[idx]); | |
994 | atomic_dec(&conf->nr_waiting[idx]); | |
fd76863e | 995 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 996 | } |
997 | ||
fd76863e | 998 | static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr) |
79ef3a8a | 999 | { |
fd76863e | 1000 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 1001 | |
824e47da | 1002 | /* |
1003 | * Very similar to _wait_barrier(). The difference is, for read | |
1004 | * I/O we don't need wait for sync I/O, but if the whole array | |
1005 | * is frozen, the read I/O still has to wait until the array is | |
1006 | * unfrozen. Since there is no ordering requirement with | |
1007 | * conf->barrier[idx] here, memory barrier is unnecessary as well. | |
1008 | */ | |
1009 | atomic_inc(&conf->nr_pending[idx]); | |
79ef3a8a | 1010 | |
824e47da | 1011 | if (!READ_ONCE(conf->array_frozen)) |
1012 | return; | |
1013 | ||
1014 | spin_lock_irq(&conf->resync_lock); | |
1015 | atomic_inc(&conf->nr_waiting[idx]); | |
1016 | atomic_dec(&conf->nr_pending[idx]); | |
1017 | /* | |
1018 | * In case freeze_array() is waiting for | |
1019 | * get_unqueued_pending() == extra | |
1020 | */ | |
1021 | wake_up(&conf->wait_barrier); | |
1022 | /* Wait for array to be unfrozen */ | |
1023 | wait_event_lock_irq(conf->wait_barrier, | |
1024 | !conf->array_frozen, | |
1025 | conf->resync_lock); | |
1026 | atomic_inc(&conf->nr_pending[idx]); | |
1027 | atomic_dec(&conf->nr_waiting[idx]); | |
1da177e4 LT |
1028 | spin_unlock_irq(&conf->resync_lock); |
1029 | } | |
1030 | ||
fd76863e | 1031 | static void wait_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 1032 | { |
fd76863e | 1033 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 1034 | |
fd76863e | 1035 | _wait_barrier(conf, idx); |
1036 | } | |
1037 | ||
fd76863e | 1038 | static void _allow_barrier(struct r1conf *conf, int idx) |
17999be4 | 1039 | { |
824e47da | 1040 | atomic_dec(&conf->nr_pending[idx]); |
17999be4 N |
1041 | wake_up(&conf->wait_barrier); |
1042 | } | |
1043 | ||
fd76863e | 1044 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr) |
1045 | { | |
1046 | int idx = sector_to_idx(sector_nr); | |
1047 | ||
1048 | _allow_barrier(conf, idx); | |
1049 | } | |
1050 | ||
fd76863e | 1051 | /* conf->resync_lock should be held */ |
1052 | static int get_unqueued_pending(struct r1conf *conf) | |
1053 | { | |
1054 | int idx, ret; | |
1055 | ||
43ac9b84 XN |
1056 | ret = atomic_read(&conf->nr_sync_pending); |
1057 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
824e47da | 1058 | ret += atomic_read(&conf->nr_pending[idx]) - |
1059 | atomic_read(&conf->nr_queued[idx]); | |
fd76863e | 1060 | |
1061 | return ret; | |
1062 | } | |
1063 | ||
e2d59925 | 1064 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab | 1065 | { |
fd76863e | 1066 | /* Stop sync I/O and normal I/O and wait for everything to |
11353b9d | 1067 | * go quiet. |
fd76863e | 1068 | * This is called in two situations: |
1069 | * 1) management command handlers (reshape, remove disk, quiesce). | |
1070 | * 2) one normal I/O request failed. | |
1071 | ||
1072 | * After array_frozen is set to 1, new sync IO will be blocked at | |
1073 | * raise_barrier(), and new normal I/O will blocked at _wait_barrier() | |
1074 | * or wait_read_barrier(). The flying I/Os will either complete or be | |
1075 | * queued. When everything goes quite, there are only queued I/Os left. | |
1076 | ||
1077 | * Every flying I/O contributes to a conf->nr_pending[idx], idx is the | |
1078 | * barrier bucket index which this I/O request hits. When all sync and | |
1079 | * normal I/O are queued, sum of all conf->nr_pending[] will match sum | |
1080 | * of all conf->nr_queued[]. But normal I/O failure is an exception, | |
1081 | * in handle_read_error(), we may call freeze_array() before trying to | |
1082 | * fix the read error. In this case, the error read I/O is not queued, | |
1083 | * so get_unqueued_pending() == 1. | |
1084 | * | |
1085 | * Therefore before this function returns, we need to wait until | |
1086 | * get_unqueued_pendings(conf) gets equal to extra. For | |
1087 | * normal I/O context, extra is 1, in rested situations extra is 0. | |
ddaf22ab N |
1088 | */ |
1089 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1090 | conf->array_frozen = 1; |
578b54ad | 1091 | raid1_log(conf->mddev, "wait freeze"); |
fd76863e | 1092 | wait_event_lock_irq_cmd( |
1093 | conf->wait_barrier, | |
1094 | get_unqueued_pending(conf) == extra, | |
1095 | conf->resync_lock, | |
1096 | flush_pending_writes(conf)); | |
ddaf22ab N |
1097 | spin_unlock_irq(&conf->resync_lock); |
1098 | } | |
e8096360 | 1099 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
1100 | { |
1101 | /* reverse the effect of the freeze */ | |
1102 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1103 | conf->array_frozen = 0; |
ddaf22ab | 1104 | spin_unlock_irq(&conf->resync_lock); |
824e47da | 1105 | wake_up(&conf->wait_barrier); |
ddaf22ab N |
1106 | } |
1107 | ||
16d56e2f | 1108 | static void alloc_behind_master_bio(struct r1bio *r1_bio, |
cb83efcf | 1109 | struct bio *bio) |
4b6d287f | 1110 | { |
cb83efcf | 1111 | int size = bio->bi_iter.bi_size; |
841c1316 ML |
1112 | unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1113 | int i = 0; | |
1114 | struct bio *behind_bio = NULL; | |
1115 | ||
1116 | behind_bio = bio_alloc_mddev(GFP_NOIO, vcnt, r1_bio->mddev); | |
1117 | if (!behind_bio) | |
16d56e2f | 1118 | return; |
4b6d287f | 1119 | |
41743c1f | 1120 | /* discard op, we don't support writezero/writesame yet */ |
16d56e2f SL |
1121 | if (!bio_has_data(bio)) { |
1122 | behind_bio->bi_iter.bi_size = size; | |
41743c1f | 1123 | goto skip_copy; |
16d56e2f | 1124 | } |
41743c1f | 1125 | |
dba40d46 MD |
1126 | behind_bio->bi_write_hint = bio->bi_write_hint; |
1127 | ||
841c1316 ML |
1128 | while (i < vcnt && size) { |
1129 | struct page *page; | |
1130 | int len = min_t(int, PAGE_SIZE, size); | |
1131 | ||
1132 | page = alloc_page(GFP_NOIO); | |
1133 | if (unlikely(!page)) | |
1134 | goto free_pages; | |
1135 | ||
1136 | bio_add_page(behind_bio, page, len, 0); | |
1137 | ||
1138 | size -= len; | |
1139 | i++; | |
4b6d287f | 1140 | } |
841c1316 | 1141 | |
cb83efcf | 1142 | bio_copy_data(behind_bio, bio); |
41743c1f | 1143 | skip_copy: |
56a64c17 | 1144 | r1_bio->behind_master_bio = behind_bio; |
af6d7b76 | 1145 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
4b6d287f | 1146 | |
16d56e2f | 1147 | return; |
841c1316 ML |
1148 | |
1149 | free_pages: | |
4f024f37 KO |
1150 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1151 | bio->bi_iter.bi_size); | |
841c1316 | 1152 | bio_free_pages(behind_bio); |
16d56e2f | 1153 | bio_put(behind_bio); |
4b6d287f N |
1154 | } |
1155 | ||
f54a9d0e N |
1156 | struct raid1_plug_cb { |
1157 | struct blk_plug_cb cb; | |
1158 | struct bio_list pending; | |
1159 | int pending_cnt; | |
1160 | }; | |
1161 | ||
1162 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1163 | { | |
1164 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1165 | cb); | |
1166 | struct mddev *mddev = plug->cb.data; | |
1167 | struct r1conf *conf = mddev->private; | |
1168 | struct bio *bio; | |
1169 | ||
874807a8 | 1170 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1171 | spin_lock_irq(&conf->device_lock); |
1172 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1173 | conf->pending_count += plug->pending_cnt; | |
1174 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1175 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1176 | md_wakeup_thread(mddev->thread); |
1177 | kfree(plug); | |
1178 | return; | |
1179 | } | |
1180 | ||
1181 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1182 | bio = bio_list_get(&plug->pending); | |
673ca68d | 1183 | flush_bio_list(conf, bio); |
f54a9d0e N |
1184 | kfree(plug); |
1185 | } | |
1186 | ||
689389a0 N |
1187 | static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio) |
1188 | { | |
1189 | r1_bio->master_bio = bio; | |
1190 | r1_bio->sectors = bio_sectors(bio); | |
1191 | r1_bio->state = 0; | |
1192 | r1_bio->mddev = mddev; | |
1193 | r1_bio->sector = bio->bi_iter.bi_sector; | |
1194 | } | |
1195 | ||
fd76863e | 1196 | static inline struct r1bio * |
689389a0 | 1197 | alloc_r1bio(struct mddev *mddev, struct bio *bio) |
fd76863e | 1198 | { |
1199 | struct r1conf *conf = mddev->private; | |
1200 | struct r1bio *r1_bio; | |
1201 | ||
afeee514 | 1202 | r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); |
689389a0 N |
1203 | /* Ensure no bio records IO_BLOCKED */ |
1204 | memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); | |
1205 | init_r1bio(r1_bio, mddev, bio); | |
fd76863e | 1206 | return r1_bio; |
1207 | } | |
1208 | ||
c230e7e5 | 1209 | static void raid1_read_request(struct mddev *mddev, struct bio *bio, |
689389a0 | 1210 | int max_read_sectors, struct r1bio *r1_bio) |
1da177e4 | 1211 | { |
e8096360 | 1212 | struct r1conf *conf = mddev->private; |
0eaf822c | 1213 | struct raid1_info *mirror; |
1da177e4 | 1214 | struct bio *read_bio; |
3b046a97 RL |
1215 | struct bitmap *bitmap = mddev->bitmap; |
1216 | const int op = bio_op(bio); | |
1217 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
3b046a97 RL |
1218 | int max_sectors; |
1219 | int rdisk; | |
689389a0 N |
1220 | bool print_msg = !!r1_bio; |
1221 | char b[BDEVNAME_SIZE]; | |
3b046a97 | 1222 | |
fd76863e | 1223 | /* |
689389a0 N |
1224 | * If r1_bio is set, we are blocking the raid1d thread |
1225 | * so there is a tiny risk of deadlock. So ask for | |
1226 | * emergency memory if needed. | |
fd76863e | 1227 | */ |
689389a0 | 1228 | gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO; |
fd76863e | 1229 | |
689389a0 N |
1230 | if (print_msg) { |
1231 | /* Need to get the block device name carefully */ | |
1232 | struct md_rdev *rdev; | |
1233 | rcu_read_lock(); | |
1234 | rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); | |
1235 | if (rdev) | |
1236 | bdevname(rdev->bdev, b); | |
1237 | else | |
1238 | strcpy(b, "???"); | |
1239 | rcu_read_unlock(); | |
1240 | } | |
3b046a97 | 1241 | |
fd76863e | 1242 | /* |
fd76863e | 1243 | * Still need barrier for READ in case that whole |
1244 | * array is frozen. | |
fd76863e | 1245 | */ |
fd76863e | 1246 | wait_read_barrier(conf, bio->bi_iter.bi_sector); |
1247 | ||
689389a0 N |
1248 | if (!r1_bio) |
1249 | r1_bio = alloc_r1bio(mddev, bio); | |
1250 | else | |
1251 | init_r1bio(r1_bio, mddev, bio); | |
c230e7e5 | 1252 | r1_bio->sectors = max_read_sectors; |
fd76863e | 1253 | |
1254 | /* | |
1255 | * make_request() can abort the operation when read-ahead is being | |
1256 | * used and no empty request is available. | |
1257 | */ | |
3b046a97 RL |
1258 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
1259 | ||
1260 | if (rdisk < 0) { | |
1261 | /* couldn't find anywhere to read from */ | |
689389a0 N |
1262 | if (print_msg) { |
1263 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1264 | mdname(mddev), | |
1265 | b, | |
1266 | (unsigned long long)r1_bio->sector); | |
1267 | } | |
3b046a97 RL |
1268 | raid_end_bio_io(r1_bio); |
1269 | return; | |
1270 | } | |
1271 | mirror = conf->mirrors + rdisk; | |
1272 | ||
689389a0 N |
1273 | if (print_msg) |
1274 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", | |
1275 | mdname(mddev), | |
1276 | (unsigned long long)r1_bio->sector, | |
1277 | bdevname(mirror->rdev->bdev, b)); | |
1278 | ||
3b046a97 RL |
1279 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1280 | bitmap) { | |
1281 | /* | |
1282 | * Reading from a write-mostly device must take care not to | |
1283 | * over-take any writes that are 'behind' | |
1284 | */ | |
1285 | raid1_log(mddev, "wait behind writes"); | |
1286 | wait_event(bitmap->behind_wait, | |
1287 | atomic_read(&bitmap->behind_writes) == 0); | |
1288 | } | |
c230e7e5 N |
1289 | |
1290 | if (max_sectors < bio_sectors(bio)) { | |
1291 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1292 | gfp, &conf->bio_split); |
c230e7e5 N |
1293 | bio_chain(split, bio); |
1294 | generic_make_request(bio); | |
1295 | bio = split; | |
1296 | r1_bio->master_bio = bio; | |
1297 | r1_bio->sectors = max_sectors; | |
1298 | } | |
1299 | ||
3b046a97 | 1300 | r1_bio->read_disk = rdisk; |
3b046a97 | 1301 | |
afeee514 | 1302 | read_bio = bio_clone_fast(bio, gfp, &mddev->bio_set); |
3b046a97 RL |
1303 | |
1304 | r1_bio->bios[rdisk] = read_bio; | |
1305 | ||
1306 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1307 | mirror->rdev->data_offset; | |
74d46992 | 1308 | bio_set_dev(read_bio, mirror->rdev->bdev); |
3b046a97 RL |
1309 | read_bio->bi_end_io = raid1_end_read_request; |
1310 | bio_set_op_attrs(read_bio, op, do_sync); | |
1311 | if (test_bit(FailFast, &mirror->rdev->flags) && | |
1312 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1313 | read_bio->bi_opf |= MD_FAILFAST; | |
1314 | read_bio->bi_private = r1_bio; | |
1315 | ||
1316 | if (mddev->gendisk) | |
74d46992 CH |
1317 | trace_block_bio_remap(read_bio->bi_disk->queue, read_bio, |
1318 | disk_devt(mddev->gendisk), r1_bio->sector); | |
3b046a97 | 1319 | |
c230e7e5 | 1320 | generic_make_request(read_bio); |
3b046a97 RL |
1321 | } |
1322 | ||
c230e7e5 N |
1323 | static void raid1_write_request(struct mddev *mddev, struct bio *bio, |
1324 | int max_write_sectors) | |
3b046a97 RL |
1325 | { |
1326 | struct r1conf *conf = mddev->private; | |
fd76863e | 1327 | struct r1bio *r1_bio; |
1f68f0c4 | 1328 | int i, disks; |
3b046a97 | 1329 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1330 | unsigned long flags; |
3cb03002 | 1331 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1332 | struct blk_plug_cb *cb; |
1333 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 | 1334 | int first_clone; |
1f68f0c4 | 1335 | int max_sectors; |
191ea9b2 | 1336 | |
b3143b9a | 1337 | if (mddev_is_clustered(mddev) && |
90382ed9 | 1338 | md_cluster_ops->area_resyncing(mddev, WRITE, |
b3143b9a | 1339 | bio->bi_iter.bi_sector, bio_end_sector(bio))) { |
3b046a97 | 1340 | |
6eef4b21 N |
1341 | DEFINE_WAIT(w); |
1342 | for (;;) { | |
6eef4b21 | 1343 | prepare_to_wait(&conf->wait_barrier, |
ae89fd3d | 1344 | &w, TASK_IDLE); |
f81f7302 | 1345 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, |
385f4d7f | 1346 | bio->bi_iter.bi_sector, |
b3143b9a | 1347 | bio_end_sector(bio))) |
6eef4b21 N |
1348 | break; |
1349 | schedule(); | |
1350 | } | |
1351 | finish_wait(&conf->wait_barrier, &w); | |
1352 | } | |
f81f7302 GJ |
1353 | |
1354 | /* | |
1355 | * Register the new request and wait if the reconstruction | |
1356 | * thread has put up a bar for new requests. | |
1357 | * Continue immediately if no resync is active currently. | |
1358 | */ | |
fd76863e | 1359 | wait_barrier(conf, bio->bi_iter.bi_sector); |
1360 | ||
689389a0 | 1361 | r1_bio = alloc_r1bio(mddev, bio); |
c230e7e5 | 1362 | r1_bio->sectors = max_write_sectors; |
1da177e4 | 1363 | |
34db0cd6 N |
1364 | if (conf->pending_count >= max_queued_requests) { |
1365 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1366 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1367 | wait_event(conf->wait_barrier, |
1368 | conf->pending_count < max_queued_requests); | |
1369 | } | |
1f68f0c4 | 1370 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1371 | * inc refcount on their rdev. Record them by setting |
1372 | * bios[x] to bio | |
1f68f0c4 N |
1373 | * If there are known/acknowledged bad blocks on any device on |
1374 | * which we have seen a write error, we want to avoid writing those | |
1375 | * blocks. | |
1376 | * This potentially requires several writes to write around | |
1377 | * the bad blocks. Each set of writes gets it's own r1bio | |
1378 | * with a set of bios attached. | |
1da177e4 | 1379 | */ |
c3b328ac | 1380 | |
8f19ccb2 | 1381 | disks = conf->raid_disks * 2; |
6bfe0b49 DW |
1382 | retry_write: |
1383 | blocked_rdev = NULL; | |
1da177e4 | 1384 | rcu_read_lock(); |
1f68f0c4 | 1385 | max_sectors = r1_bio->sectors; |
1da177e4 | 1386 | for (i = 0; i < disks; i++) { |
3cb03002 | 1387 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1388 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1389 | atomic_inc(&rdev->nr_pending); | |
1390 | blocked_rdev = rdev; | |
1391 | break; | |
1392 | } | |
1f68f0c4 | 1393 | r1_bio->bios[i] = NULL; |
8ae12666 | 1394 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1395 | if (i < conf->raid_disks) |
1396 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1397 | continue; |
1398 | } | |
1399 | ||
1400 | atomic_inc(&rdev->nr_pending); | |
1401 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1402 | sector_t first_bad; | |
1403 | int bad_sectors; | |
1404 | int is_bad; | |
1405 | ||
3b046a97 | 1406 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1407 | &first_bad, &bad_sectors); |
1408 | if (is_bad < 0) { | |
1409 | /* mustn't write here until the bad block is | |
1410 | * acknowledged*/ | |
1411 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1412 | blocked_rdev = rdev; | |
1413 | break; | |
1414 | } | |
1415 | if (is_bad && first_bad <= r1_bio->sector) { | |
1416 | /* Cannot write here at all */ | |
1417 | bad_sectors -= (r1_bio->sector - first_bad); | |
1418 | if (bad_sectors < max_sectors) | |
1419 | /* mustn't write more than bad_sectors | |
1420 | * to other devices yet | |
1421 | */ | |
1422 | max_sectors = bad_sectors; | |
03c902e1 | 1423 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1424 | /* We don't set R1BIO_Degraded as that |
1425 | * only applies if the disk is | |
1426 | * missing, so it might be re-added, | |
1427 | * and we want to know to recover this | |
1428 | * chunk. | |
1429 | * In this case the device is here, | |
1430 | * and the fact that this chunk is not | |
1431 | * in-sync is recorded in the bad | |
1432 | * block log | |
1433 | */ | |
1434 | continue; | |
964147d5 | 1435 | } |
1f68f0c4 N |
1436 | if (is_bad) { |
1437 | int good_sectors = first_bad - r1_bio->sector; | |
1438 | if (good_sectors < max_sectors) | |
1439 | max_sectors = good_sectors; | |
1440 | } | |
1441 | } | |
1442 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1443 | } |
1444 | rcu_read_unlock(); | |
1445 | ||
6bfe0b49 DW |
1446 | if (unlikely(blocked_rdev)) { |
1447 | /* Wait for this device to become unblocked */ | |
1448 | int j; | |
1449 | ||
1450 | for (j = 0; j < i; j++) | |
1451 | if (r1_bio->bios[j]) | |
1452 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1453 | r1_bio->state = 0; |
fd76863e | 1454 | allow_barrier(conf, bio->bi_iter.bi_sector); |
578b54ad | 1455 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1456 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
fd76863e | 1457 | wait_barrier(conf, bio->bi_iter.bi_sector); |
6bfe0b49 DW |
1458 | goto retry_write; |
1459 | } | |
1460 | ||
c230e7e5 N |
1461 | if (max_sectors < bio_sectors(bio)) { |
1462 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1463 | GFP_NOIO, &conf->bio_split); |
c230e7e5 N |
1464 | bio_chain(split, bio); |
1465 | generic_make_request(bio); | |
1466 | bio = split; | |
1467 | r1_bio->master_bio = bio; | |
1f68f0c4 | 1468 | r1_bio->sectors = max_sectors; |
191ea9b2 | 1469 | } |
4b6d287f | 1470 | |
4e78064f | 1471 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1472 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1473 | |
1f68f0c4 | 1474 | first_clone = 1; |
d8c84c4f | 1475 | |
1da177e4 | 1476 | for (i = 0; i < disks; i++) { |
8e58e327 | 1477 | struct bio *mbio = NULL; |
1da177e4 LT |
1478 | if (!r1_bio->bios[i]) |
1479 | continue; | |
1480 | ||
1f68f0c4 N |
1481 | if (first_clone) { |
1482 | /* do behind I/O ? | |
1483 | * Not if there are too many, or cannot | |
1484 | * allocate memory, or a reader on WriteMostly | |
1485 | * is waiting for behind writes to flush */ | |
1486 | if (bitmap && | |
1487 | (atomic_read(&bitmap->behind_writes) | |
1488 | < mddev->bitmap_info.max_write_behind) && | |
8e58e327 | 1489 | !waitqueue_active(&bitmap->behind_wait)) { |
16d56e2f | 1490 | alloc_behind_master_bio(r1_bio, bio); |
8e58e327 | 1491 | } |
1f68f0c4 | 1492 | |
e64e4018 AS |
1493 | md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, |
1494 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
1f68f0c4 N |
1495 | first_clone = 0; |
1496 | } | |
8e58e327 | 1497 | |
16d56e2f SL |
1498 | if (r1_bio->behind_master_bio) |
1499 | mbio = bio_clone_fast(r1_bio->behind_master_bio, | |
afeee514 | 1500 | GFP_NOIO, &mddev->bio_set); |
16d56e2f | 1501 | else |
afeee514 | 1502 | mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
8e58e327 | 1503 | |
841c1316 | 1504 | if (r1_bio->behind_master_bio) { |
3e148a32 GJ |
1505 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
1506 | ||
1507 | if (test_bit(WBCollisionCheck, &rdev->flags)) { | |
1508 | sector_t lo = r1_bio->sector; | |
1509 | sector_t hi = r1_bio->sector + r1_bio->sectors; | |
1510 | ||
1511 | wait_event(rdev->wb_io_wait, | |
1512 | check_and_add_wb(rdev, lo, hi) == 0); | |
1513 | } | |
1514 | if (test_bit(WriteMostly, &rdev->flags)) | |
4b6d287f N |
1515 | atomic_inc(&r1_bio->behind_remaining); |
1516 | } | |
1517 | ||
1f68f0c4 N |
1518 | r1_bio->bios[i] = mbio; |
1519 | ||
4f024f37 | 1520 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1521 | conf->mirrors[i].rdev->data_offset); |
74d46992 | 1522 | bio_set_dev(mbio, conf->mirrors[i].rdev->bdev); |
1f68f0c4 | 1523 | mbio->bi_end_io = raid1_end_write_request; |
a682e003 | 1524 | mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); |
212e7eb7 N |
1525 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && |
1526 | !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && | |
1527 | conf->raid_disks - mddev->degraded > 1) | |
1528 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1529 | mbio->bi_private = r1_bio; |
1530 | ||
1da177e4 | 1531 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1532 | |
109e3765 | 1533 | if (mddev->gendisk) |
74d46992 | 1534 | trace_block_bio_remap(mbio->bi_disk->queue, |
109e3765 N |
1535 | mbio, disk_devt(mddev->gendisk), |
1536 | r1_bio->sector); | |
1537 | /* flush_pending_writes() needs access to the rdev so...*/ | |
74d46992 | 1538 | mbio->bi_disk = (void *)conf->mirrors[i].rdev; |
109e3765 | 1539 | |
f54a9d0e N |
1540 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1541 | if (cb) | |
1542 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1543 | else | |
1544 | plug = NULL; | |
f54a9d0e N |
1545 | if (plug) { |
1546 | bio_list_add(&plug->pending, mbio); | |
1547 | plug->pending_cnt++; | |
1548 | } else { | |
23b245c0 | 1549 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1550 | bio_list_add(&conf->pending_bio_list, mbio); |
1551 | conf->pending_count++; | |
23b245c0 | 1552 | spin_unlock_irqrestore(&conf->device_lock, flags); |
b357f04a | 1553 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1554 | } |
1da177e4 | 1555 | } |
1f68f0c4 | 1556 | |
079fa166 N |
1557 | r1_bio_write_done(r1_bio); |
1558 | ||
1559 | /* In case raid1d snuck in to freeze_array */ | |
1560 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1561 | } |
1562 | ||
cc27b0c7 | 1563 | static bool raid1_make_request(struct mddev *mddev, struct bio *bio) |
3b046a97 | 1564 | { |
fd76863e | 1565 | sector_t sectors; |
3b046a97 | 1566 | |
aff8da09 SL |
1567 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
1568 | md_flush_request(mddev, bio); | |
cc27b0c7 | 1569 | return true; |
aff8da09 | 1570 | } |
3b046a97 | 1571 | |
c230e7e5 N |
1572 | /* |
1573 | * There is a limit to the maximum size, but | |
1574 | * the read/write handler might find a lower limit | |
1575 | * due to bad blocks. To avoid multiple splits, | |
1576 | * we pass the maximum number of sectors down | |
1577 | * and let the lower level perform the split. | |
1578 | */ | |
1579 | sectors = align_to_barrier_unit_end( | |
1580 | bio->bi_iter.bi_sector, bio_sectors(bio)); | |
61eb2b43 | 1581 | |
c230e7e5 | 1582 | if (bio_data_dir(bio) == READ) |
689389a0 | 1583 | raid1_read_request(mddev, bio, sectors, NULL); |
cc27b0c7 N |
1584 | else { |
1585 | if (!md_write_start(mddev,bio)) | |
1586 | return false; | |
c230e7e5 | 1587 | raid1_write_request(mddev, bio, sectors); |
cc27b0c7 N |
1588 | } |
1589 | return true; | |
3b046a97 RL |
1590 | } |
1591 | ||
849674e4 | 1592 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1593 | { |
e8096360 | 1594 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1595 | int i; |
1596 | ||
1597 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1598 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1599 | rcu_read_lock(); |
1600 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1601 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1602 | seq_printf(seq, "%s", |
ddac7c7e N |
1603 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1604 | } | |
1605 | rcu_read_unlock(); | |
1da177e4 LT |
1606 | seq_printf(seq, "]"); |
1607 | } | |
1608 | ||
849674e4 | 1609 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1610 | { |
1611 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1612 | struct r1conf *conf = mddev->private; |
423f04d6 | 1613 | unsigned long flags; |
1da177e4 LT |
1614 | |
1615 | /* | |
1616 | * If it is not operational, then we have already marked it as dead | |
1617 | * else if it is the last working disks, ignore the error, let the | |
1618 | * next level up know. | |
1619 | * else mark the drive as failed | |
1620 | */ | |
2e52d449 | 1621 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1622 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1623 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1624 | /* |
1625 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1626 | * normal single drive. |
1627 | * However don't try a recovery from this drive as | |
1628 | * it is very likely to fail. | |
1da177e4 | 1629 | */ |
5389042f | 1630 | conf->recovery_disabled = mddev->recovery_disabled; |
2e52d449 | 1631 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1632 | return; |
4044ba58 | 1633 | } |
de393cde | 1634 | set_bit(Blocked, &rdev->flags); |
ebda52fa | 1635 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1636 | mddev->degraded++; |
ebda52fa | 1637 | set_bit(Faulty, &rdev->flags); |
423f04d6 | 1638 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1639 | /* |
1640 | * if recovery is running, make sure it aborts. | |
1641 | */ | |
1642 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1643 | set_mask_bits(&mddev->sb_flags, 0, |
1644 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
1d41c216 N |
1645 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1646 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1647 | mdname(mddev), bdevname(rdev->bdev, b), | |
1648 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1649 | } |
1650 | ||
e8096360 | 1651 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1652 | { |
1653 | int i; | |
1da177e4 | 1654 | |
1d41c216 | 1655 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1656 | if (!conf) { |
1d41c216 | 1657 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1658 | return; |
1659 | } | |
1d41c216 N |
1660 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1661 | conf->raid_disks); | |
1da177e4 | 1662 | |
ddac7c7e | 1663 | rcu_read_lock(); |
1da177e4 LT |
1664 | for (i = 0; i < conf->raid_disks; i++) { |
1665 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1666 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1667 | if (rdev) |
1d41c216 N |
1668 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1669 | i, !test_bit(In_sync, &rdev->flags), | |
1670 | !test_bit(Faulty, &rdev->flags), | |
1671 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1672 | } |
ddac7c7e | 1673 | rcu_read_unlock(); |
1da177e4 LT |
1674 | } |
1675 | ||
e8096360 | 1676 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1677 | { |
f6eca2d4 ND |
1678 | int idx; |
1679 | ||
1680 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) { | |
1681 | _wait_barrier(conf, idx); | |
1682 | _allow_barrier(conf, idx); | |
1683 | } | |
1da177e4 | 1684 | |
afeee514 | 1685 | mempool_exit(&conf->r1buf_pool); |
1da177e4 LT |
1686 | } |
1687 | ||
fd01b88c | 1688 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1689 | { |
1690 | int i; | |
e8096360 | 1691 | struct r1conf *conf = mddev->private; |
6b965620 N |
1692 | int count = 0; |
1693 | unsigned long flags; | |
1da177e4 LT |
1694 | |
1695 | /* | |
f72ffdd6 | 1696 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1697 | * and mark them readable. |
1698 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1699 | * device_lock used to avoid races with raid1_end_read_request |
1700 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1701 | */ |
423f04d6 | 1702 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1703 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1704 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1705 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1706 | if (repl | |
1aee41f6 | 1707 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1708 | && repl->recovery_offset == MaxSector |
1709 | && !test_bit(Faulty, &repl->flags) | |
1710 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1711 | /* replacement has just become active */ | |
1712 | if (!rdev || | |
1713 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1714 | count++; | |
1715 | if (rdev) { | |
1716 | /* Replaced device not technically | |
1717 | * faulty, but we need to be sure | |
1718 | * it gets removed and never re-added | |
1719 | */ | |
1720 | set_bit(Faulty, &rdev->flags); | |
1721 | sysfs_notify_dirent_safe( | |
1722 | rdev->sysfs_state); | |
1723 | } | |
1724 | } | |
ddac7c7e | 1725 | if (rdev |
61e4947c | 1726 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1727 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1728 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1729 | count++; |
654e8b5a | 1730 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1731 | } |
1732 | } | |
6b965620 N |
1733 | mddev->degraded -= count; |
1734 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1735 | |
1736 | print_conf(conf); | |
6b965620 | 1737 | return count; |
1da177e4 LT |
1738 | } |
1739 | ||
fd01b88c | 1740 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1741 | { |
e8096360 | 1742 | struct r1conf *conf = mddev->private; |
199050ea | 1743 | int err = -EEXIST; |
41158c7e | 1744 | int mirror = 0; |
0eaf822c | 1745 | struct raid1_info *p; |
6c2fce2e | 1746 | int first = 0; |
30194636 | 1747 | int last = conf->raid_disks - 1; |
1da177e4 | 1748 | |
5389042f N |
1749 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1750 | return -EBUSY; | |
1751 | ||
1501efad DW |
1752 | if (md_integrity_add_rdev(rdev, mddev)) |
1753 | return -ENXIO; | |
1754 | ||
6c2fce2e NB |
1755 | if (rdev->raid_disk >= 0) |
1756 | first = last = rdev->raid_disk; | |
1757 | ||
70bcecdb GR |
1758 | /* |
1759 | * find the disk ... but prefer rdev->saved_raid_disk | |
1760 | * if possible. | |
1761 | */ | |
1762 | if (rdev->saved_raid_disk >= 0 && | |
1763 | rdev->saved_raid_disk >= first && | |
9e753ba9 | 1764 | rdev->saved_raid_disk < conf->raid_disks && |
70bcecdb GR |
1765 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1766 | first = last = rdev->saved_raid_disk; | |
1767 | ||
7ef449d1 | 1768 | for (mirror = first; mirror <= last; mirror++) { |
ebfeb444 | 1769 | p = conf->mirrors + mirror; |
7ef449d1 | 1770 | if (!p->rdev) { |
9092c02d JB |
1771 | if (mddev->gendisk) |
1772 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1773 | rdev->data_offset << 9); | |
1da177e4 LT |
1774 | |
1775 | p->head_position = 0; | |
1776 | rdev->raid_disk = mirror; | |
199050ea | 1777 | err = 0; |
6aea114a N |
1778 | /* As all devices are equivalent, we don't need a full recovery |
1779 | * if this was recently any drive of the array | |
1780 | */ | |
1781 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1782 | conf->fullsync = 1; |
d6065f7b | 1783 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1784 | break; |
1785 | } | |
7ef449d1 N |
1786 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1787 | p[conf->raid_disks].rdev == NULL) { | |
1788 | /* Add this device as a replacement */ | |
1789 | clear_bit(In_sync, &rdev->flags); | |
1790 | set_bit(Replacement, &rdev->flags); | |
1791 | rdev->raid_disk = mirror; | |
1792 | err = 0; | |
1793 | conf->fullsync = 1; | |
1794 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1795 | break; | |
1796 | } | |
1797 | } | |
9092c02d | 1798 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
8b904b5b | 1799 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1800 | print_conf(conf); |
199050ea | 1801 | return err; |
1da177e4 LT |
1802 | } |
1803 | ||
b8321b68 | 1804 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1805 | { |
e8096360 | 1806 | struct r1conf *conf = mddev->private; |
1da177e4 | 1807 | int err = 0; |
b8321b68 | 1808 | int number = rdev->raid_disk; |
0eaf822c | 1809 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1810 | |
b014f14c N |
1811 | if (rdev != p->rdev) |
1812 | p = conf->mirrors + conf->raid_disks + number; | |
1813 | ||
1da177e4 | 1814 | print_conf(conf); |
b8321b68 | 1815 | if (rdev == p->rdev) { |
b2d444d7 | 1816 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1817 | atomic_read(&rdev->nr_pending)) { |
1818 | err = -EBUSY; | |
1819 | goto abort; | |
1820 | } | |
046abeed | 1821 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1822 | * is not possible. |
1823 | */ | |
1824 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1825 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1826 | mddev->degraded < conf->raid_disks) { |
1827 | err = -EBUSY; | |
1828 | goto abort; | |
1829 | } | |
1da177e4 | 1830 | p->rdev = NULL; |
d787be40 N |
1831 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1832 | synchronize_rcu(); | |
1833 | if (atomic_read(&rdev->nr_pending)) { | |
1834 | /* lost the race, try later */ | |
1835 | err = -EBUSY; | |
1836 | p->rdev = rdev; | |
1837 | goto abort; | |
1838 | } | |
1839 | } | |
1840 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1841 | /* We just removed a device that is being replaced. |
1842 | * Move down the replacement. We drain all IO before | |
1843 | * doing this to avoid confusion. | |
1844 | */ | |
1845 | struct md_rdev *repl = | |
1846 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1847 | freeze_array(conf, 0); |
3de59bb9 YY |
1848 | if (atomic_read(&repl->nr_pending)) { |
1849 | /* It means that some queued IO of retry_list | |
1850 | * hold repl. Thus, we cannot set replacement | |
1851 | * as NULL, avoiding rdev NULL pointer | |
1852 | * dereference in sync_request_write and | |
1853 | * handle_write_finished. | |
1854 | */ | |
1855 | err = -EBUSY; | |
1856 | unfreeze_array(conf); | |
1857 | goto abort; | |
1858 | } | |
8c7a2c2b N |
1859 | clear_bit(Replacement, &repl->flags); |
1860 | p->rdev = repl; | |
1861 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1862 | unfreeze_array(conf); |
e5bc9c3c GJ |
1863 | } |
1864 | ||
1865 | clear_bit(WantReplacement, &rdev->flags); | |
a91a2785 | 1866 | err = md_integrity_register(mddev); |
1da177e4 LT |
1867 | } |
1868 | abort: | |
1869 | ||
1870 | print_conf(conf); | |
1871 | return err; | |
1872 | } | |
1873 | ||
4246a0b6 | 1874 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1875 | { |
98d30c58 | 1876 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
1da177e4 | 1877 | |
0fc280f6 | 1878 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1879 | |
1da177e4 LT |
1880 | /* |
1881 | * we have read a block, now it needs to be re-written, | |
1882 | * or re-read if the read failed. | |
1883 | * We don't do much here, just schedule handling by raid1d | |
1884 | */ | |
4e4cbee9 | 1885 | if (!bio->bi_status) |
1da177e4 | 1886 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1887 | |
1888 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1889 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1890 | } |
1891 | ||
dfcc34c9 ND |
1892 | static void abort_sync_write(struct mddev *mddev, struct r1bio *r1_bio) |
1893 | { | |
1894 | sector_t sync_blocks = 0; | |
1895 | sector_t s = r1_bio->sector; | |
1896 | long sectors_to_go = r1_bio->sectors; | |
1897 | ||
1898 | /* make sure these bits don't get cleared. */ | |
1899 | do { | |
1900 | md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); | |
1901 | s += sync_blocks; | |
1902 | sectors_to_go -= sync_blocks; | |
1903 | } while (sectors_to_go > 0); | |
1904 | } | |
1905 | ||
4246a0b6 | 1906 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1907 | { |
4e4cbee9 | 1908 | int uptodate = !bio->bi_status; |
98d30c58 | 1909 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
fd01b88c | 1910 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1911 | struct r1conf *conf = mddev->private; |
4367af55 N |
1912 | sector_t first_bad; |
1913 | int bad_sectors; | |
854abd75 | 1914 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1915 | |
6b1117d5 | 1916 | if (!uptodate) { |
dfcc34c9 | 1917 | abort_sync_write(mddev, r1_bio); |
854abd75 N |
1918 | set_bit(WriteErrorSeen, &rdev->flags); |
1919 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1920 | set_bit(MD_RECOVERY_NEEDED, & |
1921 | mddev->recovery); | |
d8f05d29 | 1922 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1923 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1924 | &first_bad, &bad_sectors) && |
1925 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1926 | r1_bio->sector, | |
1927 | r1_bio->sectors, | |
1928 | &first_bad, &bad_sectors) | |
1929 | ) | |
4367af55 | 1930 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1931 | |
1da177e4 | 1932 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1933 | int s = r1_bio->sectors; |
d8f05d29 N |
1934 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1935 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1936 | reschedule_retry(r1_bio); |
1937 | else { | |
1938 | put_buf(r1_bio); | |
1939 | md_done_sync(mddev, s, uptodate); | |
1940 | } | |
1da177e4 | 1941 | } |
1da177e4 LT |
1942 | } |
1943 | ||
3cb03002 | 1944 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1945 | int sectors, struct page *page, int rw) |
1946 | { | |
796a5cf0 | 1947 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1948 | /* success */ |
1949 | return 1; | |
19d67169 | 1950 | if (rw == WRITE) { |
d8f05d29 | 1951 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1952 | if (!test_and_set_bit(WantReplacement, |
1953 | &rdev->flags)) | |
1954 | set_bit(MD_RECOVERY_NEEDED, & | |
1955 | rdev->mddev->recovery); | |
1956 | } | |
d8f05d29 N |
1957 | /* need to record an error - either for the block or the device */ |
1958 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1959 | md_error(rdev->mddev, rdev); | |
1960 | return 0; | |
1961 | } | |
1962 | ||
9f2c9d12 | 1963 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1964 | { |
a68e5870 N |
1965 | /* Try some synchronous reads of other devices to get |
1966 | * good data, much like with normal read errors. Only | |
1967 | * read into the pages we already have so we don't | |
1968 | * need to re-issue the read request. | |
1969 | * We don't need to freeze the array, because being in an | |
1970 | * active sync request, there is no normal IO, and | |
1971 | * no overlapping syncs. | |
06f60385 N |
1972 | * We don't need to check is_badblock() again as we |
1973 | * made sure that anything with a bad block in range | |
1974 | * will have bi_end_io clear. | |
a68e5870 | 1975 | */ |
fd01b88c | 1976 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1977 | struct r1conf *conf = mddev->private; |
a68e5870 | 1978 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
44cf0f4d | 1979 | struct page **pages = get_resync_pages(bio)->pages; |
a68e5870 N |
1980 | sector_t sect = r1_bio->sector; |
1981 | int sectors = r1_bio->sectors; | |
1982 | int idx = 0; | |
2e52d449 N |
1983 | struct md_rdev *rdev; |
1984 | ||
1985 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
1986 | if (test_bit(FailFast, &rdev->flags)) { | |
1987 | /* Don't try recovering from here - just fail it | |
1988 | * ... unless it is the last working device of course */ | |
1989 | md_error(mddev, rdev); | |
1990 | if (test_bit(Faulty, &rdev->flags)) | |
1991 | /* Don't try to read from here, but make sure | |
1992 | * put_buf does it's thing | |
1993 | */ | |
1994 | bio->bi_end_io = end_sync_write; | |
1995 | } | |
a68e5870 N |
1996 | |
1997 | while(sectors) { | |
1998 | int s = sectors; | |
1999 | int d = r1_bio->read_disk; | |
2000 | int success = 0; | |
78d7f5f7 | 2001 | int start; |
a68e5870 N |
2002 | |
2003 | if (s > (PAGE_SIZE>>9)) | |
2004 | s = PAGE_SIZE >> 9; | |
2005 | do { | |
2006 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
2007 | /* No rcu protection needed here devices | |
2008 | * can only be removed when no resync is | |
2009 | * active, and resync is currently active | |
2010 | */ | |
2011 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 2012 | if (sync_page_io(rdev, sect, s<<9, |
44cf0f4d | 2013 | pages[idx], |
796a5cf0 | 2014 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
2015 | success = 1; |
2016 | break; | |
2017 | } | |
2018 | } | |
2019 | d++; | |
8f19ccb2 | 2020 | if (d == conf->raid_disks * 2) |
a68e5870 N |
2021 | d = 0; |
2022 | } while (!success && d != r1_bio->read_disk); | |
2023 | ||
78d7f5f7 | 2024 | if (!success) { |
a68e5870 | 2025 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
2026 | int abort = 0; |
2027 | /* Cannot read from anywhere, this block is lost. | |
2028 | * Record a bad block on each device. If that doesn't | |
2029 | * work just disable and interrupt the recovery. | |
2030 | * Don't fail devices as that won't really help. | |
2031 | */ | |
1d41c216 | 2032 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
74d46992 | 2033 | mdname(mddev), bio_devname(bio, b), |
1d41c216 | 2034 | (unsigned long long)r1_bio->sector); |
8f19ccb2 | 2035 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
2036 | rdev = conf->mirrors[d].rdev; |
2037 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
2038 | continue; | |
2039 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
2040 | abort = 1; | |
2041 | } | |
2042 | if (abort) { | |
d890fa2b N |
2043 | conf->recovery_disabled = |
2044 | mddev->recovery_disabled; | |
3a9f28a5 N |
2045 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
2046 | md_done_sync(mddev, r1_bio->sectors, 0); | |
2047 | put_buf(r1_bio); | |
2048 | return 0; | |
2049 | } | |
2050 | /* Try next page */ | |
2051 | sectors -= s; | |
2052 | sect += s; | |
2053 | idx++; | |
2054 | continue; | |
d11c171e | 2055 | } |
78d7f5f7 N |
2056 | |
2057 | start = d; | |
2058 | /* write it back and re-read */ | |
2059 | while (d != r1_bio->read_disk) { | |
2060 | if (d == 0) | |
8f19ccb2 | 2061 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2062 | d--; |
2063 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2064 | continue; | |
2065 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2066 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2067 | pages[idx], |
d8f05d29 | 2068 | WRITE) == 0) { |
78d7f5f7 N |
2069 | r1_bio->bios[d]->bi_end_io = NULL; |
2070 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 2071 | } |
78d7f5f7 N |
2072 | } |
2073 | d = start; | |
2074 | while (d != r1_bio->read_disk) { | |
2075 | if (d == 0) | |
8f19ccb2 | 2076 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2077 | d--; |
2078 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2079 | continue; | |
2080 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2081 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2082 | pages[idx], |
d8f05d29 | 2083 | READ) != 0) |
9d3d8011 | 2084 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 2085 | } |
a68e5870 N |
2086 | sectors -= s; |
2087 | sect += s; | |
2088 | idx ++; | |
2089 | } | |
78d7f5f7 | 2090 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4e4cbee9 | 2091 | bio->bi_status = 0; |
a68e5870 N |
2092 | return 1; |
2093 | } | |
2094 | ||
c95e6385 | 2095 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
2096 | { |
2097 | /* We have read all readable devices. If we haven't | |
2098 | * got the block, then there is no hope left. | |
2099 | * If we have, then we want to do a comparison | |
2100 | * and skip the write if everything is the same. | |
2101 | * If any blocks failed to read, then we need to | |
2102 | * attempt an over-write | |
2103 | */ | |
fd01b88c | 2104 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2105 | struct r1conf *conf = mddev->private; |
a68e5870 N |
2106 | int primary; |
2107 | int i; | |
f4380a91 | 2108 | int vcnt; |
a68e5870 | 2109 | |
30bc9b53 N |
2110 | /* Fix variable parts of all bios */ |
2111 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
2112 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
4e4cbee9 | 2113 | blk_status_t status; |
30bc9b53 | 2114 | struct bio *b = r1_bio->bios[i]; |
98d30c58 | 2115 | struct resync_pages *rp = get_resync_pages(b); |
30bc9b53 N |
2116 | if (b->bi_end_io != end_sync_read) |
2117 | continue; | |
4246a0b6 | 2118 | /* fixup the bio for reuse, but preserve errno */ |
4e4cbee9 | 2119 | status = b->bi_status; |
30bc9b53 | 2120 | bio_reset(b); |
4e4cbee9 | 2121 | b->bi_status = status; |
4f024f37 | 2122 | b->bi_iter.bi_sector = r1_bio->sector + |
30bc9b53 | 2123 | conf->mirrors[i].rdev->data_offset; |
74d46992 | 2124 | bio_set_dev(b, conf->mirrors[i].rdev->bdev); |
30bc9b53 | 2125 | b->bi_end_io = end_sync_read; |
98d30c58 ML |
2126 | rp->raid_bio = r1_bio; |
2127 | b->bi_private = rp; | |
30bc9b53 | 2128 | |
fb0eb5df ML |
2129 | /* initialize bvec table again */ |
2130 | md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); | |
30bc9b53 | 2131 | } |
8f19ccb2 | 2132 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2133 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4e4cbee9 | 2134 | !r1_bio->bios[primary]->bi_status) { |
a68e5870 N |
2135 | r1_bio->bios[primary]->bi_end_io = NULL; |
2136 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2137 | break; | |
2138 | } | |
2139 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2140 | for (i = 0; i < conf->raid_disks * 2; i++) { |
2b070cfe | 2141 | int j = 0; |
78d7f5f7 N |
2142 | struct bio *pbio = r1_bio->bios[primary]; |
2143 | struct bio *sbio = r1_bio->bios[i]; | |
4e4cbee9 | 2144 | blk_status_t status = sbio->bi_status; |
44cf0f4d ML |
2145 | struct page **ppages = get_resync_pages(pbio)->pages; |
2146 | struct page **spages = get_resync_pages(sbio)->pages; | |
60928a91 | 2147 | struct bio_vec *bi; |
8fc04e6e | 2148 | int page_len[RESYNC_PAGES] = { 0 }; |
6dc4f100 | 2149 | struct bvec_iter_all iter_all; |
a68e5870 | 2150 | |
2aabaa65 | 2151 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2152 | continue; |
4246a0b6 | 2153 | /* Now we can 'fixup' the error value */ |
4e4cbee9 | 2154 | sbio->bi_status = 0; |
78d7f5f7 | 2155 | |
2b070cfe CH |
2156 | bio_for_each_segment_all(bi, sbio, iter_all) |
2157 | page_len[j++] = bi->bv_len; | |
60928a91 | 2158 | |
4e4cbee9 | 2159 | if (!status) { |
78d7f5f7 | 2160 | for (j = vcnt; j-- ; ) { |
44cf0f4d ML |
2161 | if (memcmp(page_address(ppages[j]), |
2162 | page_address(spages[j]), | |
60928a91 | 2163 | page_len[j])) |
78d7f5f7 | 2164 | break; |
69382e85 | 2165 | } |
78d7f5f7 N |
2166 | } else |
2167 | j = 0; | |
2168 | if (j >= 0) | |
7f7583d4 | 2169 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2170 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4e4cbee9 | 2171 | && !status)) { |
78d7f5f7 N |
2172 | /* No need to write to this device. */ |
2173 | sbio->bi_end_io = NULL; | |
2174 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2175 | continue; | |
2176 | } | |
d3b45c2a KO |
2177 | |
2178 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2179 | } |
a68e5870 N |
2180 | } |
2181 | ||
9f2c9d12 | 2182 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2183 | { |
e8096360 | 2184 | struct r1conf *conf = mddev->private; |
a68e5870 | 2185 | int i; |
8f19ccb2 | 2186 | int disks = conf->raid_disks * 2; |
037d2ff6 | 2187 | struct bio *wbio; |
a68e5870 | 2188 | |
a68e5870 N |
2189 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2190 | /* ouch - failed to read all of that. */ | |
2191 | if (!fix_sync_read_error(r1_bio)) | |
2192 | return; | |
7ca78d57 N |
2193 | |
2194 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2195 | process_checks(r1_bio); |
2196 | ||
d11c171e N |
2197 | /* |
2198 | * schedule writes | |
2199 | */ | |
1da177e4 LT |
2200 | atomic_set(&r1_bio->remaining, 1); |
2201 | for (i = 0; i < disks ; i++) { | |
2202 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2203 | if (wbio->bi_end_io == NULL || |
2204 | (wbio->bi_end_io == end_sync_read && | |
2205 | (i == r1_bio->read_disk || | |
2206 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 | 2207 | continue; |
dfcc34c9 ND |
2208 | if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { |
2209 | abort_sync_write(mddev, r1_bio); | |
0c9d5b12 | 2210 | continue; |
dfcc34c9 | 2211 | } |
1da177e4 | 2212 | |
796a5cf0 | 2213 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
212e7eb7 N |
2214 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2215 | wbio->bi_opf |= MD_FAILFAST; | |
2216 | ||
3e198f78 | 2217 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2218 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2219 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2220 | |
1da177e4 LT |
2221 | generic_make_request(wbio); |
2222 | } | |
2223 | ||
2224 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2225 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2226 | int s = r1_bio->sectors; |
2227 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2228 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2229 | reschedule_retry(r1_bio); | |
2230 | else { | |
2231 | put_buf(r1_bio); | |
2232 | md_done_sync(mddev, s, 1); | |
2233 | } | |
1da177e4 LT |
2234 | } |
2235 | } | |
2236 | ||
2237 | /* | |
2238 | * This is a kernel thread which: | |
2239 | * | |
2240 | * 1. Retries failed read operations on working mirrors. | |
2241 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2242 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2243 | */ |
2244 | ||
e8096360 | 2245 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2246 | sector_t sect, int sectors) |
2247 | { | |
fd01b88c | 2248 | struct mddev *mddev = conf->mddev; |
867868fb N |
2249 | while(sectors) { |
2250 | int s = sectors; | |
2251 | int d = read_disk; | |
2252 | int success = 0; | |
2253 | int start; | |
3cb03002 | 2254 | struct md_rdev *rdev; |
867868fb N |
2255 | |
2256 | if (s > (PAGE_SIZE>>9)) | |
2257 | s = PAGE_SIZE >> 9; | |
2258 | ||
2259 | do { | |
d2eb35ac N |
2260 | sector_t first_bad; |
2261 | int bad_sectors; | |
2262 | ||
707a6a42 N |
2263 | rcu_read_lock(); |
2264 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2265 | if (rdev && |
da8840a7 | 2266 | (test_bit(In_sync, &rdev->flags) || |
2267 | (!test_bit(Faulty, &rdev->flags) && | |
2268 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2269 | is_badblock(rdev, sect, s, |
707a6a42 N |
2270 | &first_bad, &bad_sectors) == 0) { |
2271 | atomic_inc(&rdev->nr_pending); | |
2272 | rcu_read_unlock(); | |
2273 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2274 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2275 | success = 1; |
2276 | rdev_dec_pending(rdev, mddev); | |
2277 | if (success) | |
2278 | break; | |
2279 | } else | |
2280 | rcu_read_unlock(); | |
2281 | d++; | |
2282 | if (d == conf->raid_disks * 2) | |
2283 | d = 0; | |
867868fb N |
2284 | } while (!success && d != read_disk); |
2285 | ||
2286 | if (!success) { | |
d8f05d29 | 2287 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2288 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2289 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2290 | md_error(mddev, rdev); | |
867868fb N |
2291 | break; |
2292 | } | |
2293 | /* write it back and re-read */ | |
2294 | start = d; | |
2295 | while (d != read_disk) { | |
2296 | if (d==0) | |
8f19ccb2 | 2297 | d = conf->raid_disks * 2; |
867868fb | 2298 | d--; |
707a6a42 N |
2299 | rcu_read_lock(); |
2300 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2301 | if (rdev && |
707a6a42 N |
2302 | !test_bit(Faulty, &rdev->flags)) { |
2303 | atomic_inc(&rdev->nr_pending); | |
2304 | rcu_read_unlock(); | |
d8f05d29 N |
2305 | r1_sync_page_io(rdev, sect, s, |
2306 | conf->tmppage, WRITE); | |
707a6a42 N |
2307 | rdev_dec_pending(rdev, mddev); |
2308 | } else | |
2309 | rcu_read_unlock(); | |
867868fb N |
2310 | } |
2311 | d = start; | |
2312 | while (d != read_disk) { | |
2313 | char b[BDEVNAME_SIZE]; | |
2314 | if (d==0) | |
8f19ccb2 | 2315 | d = conf->raid_disks * 2; |
867868fb | 2316 | d--; |
707a6a42 N |
2317 | rcu_read_lock(); |
2318 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2319 | if (rdev && |
b8cb6b4c | 2320 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2321 | atomic_inc(&rdev->nr_pending); |
2322 | rcu_read_unlock(); | |
d8f05d29 N |
2323 | if (r1_sync_page_io(rdev, sect, s, |
2324 | conf->tmppage, READ)) { | |
867868fb | 2325 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2326 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2327 | mdname(mddev), s, | |
2328 | (unsigned long long)(sect + | |
2329 | rdev->data_offset), | |
2330 | bdevname(rdev->bdev, b)); | |
867868fb | 2331 | } |
707a6a42 N |
2332 | rdev_dec_pending(rdev, mddev); |
2333 | } else | |
2334 | rcu_read_unlock(); | |
867868fb N |
2335 | } |
2336 | sectors -= s; | |
2337 | sect += s; | |
2338 | } | |
2339 | } | |
2340 | ||
9f2c9d12 | 2341 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2342 | { |
fd01b88c | 2343 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2344 | struct r1conf *conf = mddev->private; |
3cb03002 | 2345 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2346 | |
2347 | /* bio has the data to be written to device 'i' where | |
2348 | * we just recently had a write error. | |
2349 | * We repeatedly clone the bio and trim down to one block, | |
2350 | * then try the write. Where the write fails we record | |
2351 | * a bad block. | |
2352 | * It is conceivable that the bio doesn't exactly align with | |
2353 | * blocks. We must handle this somehow. | |
2354 | * | |
2355 | * We currently own a reference on the rdev. | |
2356 | */ | |
2357 | ||
2358 | int block_sectors; | |
2359 | sector_t sector; | |
2360 | int sectors; | |
2361 | int sect_to_write = r1_bio->sectors; | |
2362 | int ok = 1; | |
2363 | ||
2364 | if (rdev->badblocks.shift < 0) | |
2365 | return 0; | |
2366 | ||
ab713cdc ND |
2367 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2368 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2369 | sector = r1_bio->sector; |
2370 | sectors = ((sector + block_sectors) | |
2371 | & ~(sector_t)(block_sectors - 1)) | |
2372 | - sector; | |
2373 | ||
cd5ff9a1 N |
2374 | while (sect_to_write) { |
2375 | struct bio *wbio; | |
2376 | if (sectors > sect_to_write) | |
2377 | sectors = sect_to_write; | |
2378 | /* Write at 'sector' for 'sectors'*/ | |
2379 | ||
b783863f | 2380 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
841c1316 ML |
2381 | wbio = bio_clone_fast(r1_bio->behind_master_bio, |
2382 | GFP_NOIO, | |
afeee514 | 2383 | &mddev->bio_set); |
b783863f | 2384 | } else { |
d7a10308 | 2385 | wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, |
afeee514 | 2386 | &mddev->bio_set); |
b783863f KO |
2387 | } |
2388 | ||
796a5cf0 | 2389 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2390 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2391 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2392 | |
6678d83f | 2393 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2394 | wbio->bi_iter.bi_sector += rdev->data_offset; |
74d46992 | 2395 | bio_set_dev(wbio, rdev->bdev); |
4e49ea4a MC |
2396 | |
2397 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2398 | /* failure! */ |
2399 | ok = rdev_set_badblocks(rdev, sector, | |
2400 | sectors, 0) | |
2401 | && ok; | |
2402 | ||
2403 | bio_put(wbio); | |
2404 | sect_to_write -= sectors; | |
2405 | sector += sectors; | |
2406 | sectors = block_sectors; | |
2407 | } | |
2408 | return ok; | |
2409 | } | |
2410 | ||
e8096360 | 2411 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2412 | { |
2413 | int m; | |
2414 | int s = r1_bio->sectors; | |
8f19ccb2 | 2415 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2416 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2417 | struct bio *bio = r1_bio->bios[m]; |
2418 | if (bio->bi_end_io == NULL) | |
2419 | continue; | |
4e4cbee9 | 2420 | if (!bio->bi_status && |
62096bce | 2421 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2422 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2423 | } |
4e4cbee9 | 2424 | if (bio->bi_status && |
62096bce N |
2425 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2426 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2427 | md_error(conf->mddev, rdev); | |
2428 | } | |
2429 | } | |
2430 | put_buf(r1_bio); | |
2431 | md_done_sync(conf->mddev, s, 1); | |
2432 | } | |
2433 | ||
e8096360 | 2434 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2435 | { |
fd76863e | 2436 | int m, idx; |
55ce74d4 | 2437 | bool fail = false; |
fd76863e | 2438 | |
8f19ccb2 | 2439 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2440 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2441 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2442 | rdev_clear_badblocks(rdev, |
2443 | r1_bio->sector, | |
c6563a8c | 2444 | r1_bio->sectors, 0); |
62096bce N |
2445 | rdev_dec_pending(rdev, conf->mddev); |
2446 | } else if (r1_bio->bios[m] != NULL) { | |
2447 | /* This drive got a write error. We need to | |
2448 | * narrow down and record precise write | |
2449 | * errors. | |
2450 | */ | |
55ce74d4 | 2451 | fail = true; |
62096bce N |
2452 | if (!narrow_write_error(r1_bio, m)) { |
2453 | md_error(conf->mddev, | |
2454 | conf->mirrors[m].rdev); | |
2455 | /* an I/O failed, we can't clear the bitmap */ | |
2456 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2457 | } | |
2458 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2459 | conf->mddev); | |
2460 | } | |
55ce74d4 N |
2461 | if (fail) { |
2462 | spin_lock_irq(&conf->device_lock); | |
2463 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
fd76863e | 2464 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2465 | atomic_inc(&conf->nr_queued[idx]); |
55ce74d4 | 2466 | spin_unlock_irq(&conf->device_lock); |
824e47da | 2467 | /* |
2468 | * In case freeze_array() is waiting for condition | |
2469 | * get_unqueued_pending() == extra to be true. | |
2470 | */ | |
2471 | wake_up(&conf->wait_barrier); | |
55ce74d4 | 2472 | md_wakeup_thread(conf->mddev->thread); |
bd8688a1 N |
2473 | } else { |
2474 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2475 | close_write(r1_bio); | |
55ce74d4 | 2476 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2477 | } |
62096bce N |
2478 | } |
2479 | ||
e8096360 | 2480 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2481 | { |
fd01b88c | 2482 | struct mddev *mddev = conf->mddev; |
62096bce | 2483 | struct bio *bio; |
3cb03002 | 2484 | struct md_rdev *rdev; |
62096bce N |
2485 | |
2486 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2487 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2488 | * the block and we can fix it. | |
2489 | * We freeze all other IO, and try reading the block from | |
2490 | * other devices. When we find one, we re-write | |
2491 | * and check it that fixes the read error. | |
2492 | * This is all done synchronously while the array is | |
2493 | * frozen | |
2494 | */ | |
7449f699 TM |
2495 | |
2496 | bio = r1_bio->bios[r1_bio->read_disk]; | |
7449f699 TM |
2497 | bio_put(bio); |
2498 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2499 | ||
2e52d449 N |
2500 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2501 | if (mddev->ro == 0 | |
2502 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2503 | freeze_array(conf, 1); |
62096bce N |
2504 | fix_read_error(conf, r1_bio->read_disk, |
2505 | r1_bio->sector, r1_bio->sectors); | |
2506 | unfreeze_array(conf); | |
b33d1062 GK |
2507 | } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { |
2508 | md_error(mddev, rdev); | |
7449f699 TM |
2509 | } else { |
2510 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2511 | } | |
2512 | ||
2e52d449 | 2513 | rdev_dec_pending(rdev, conf->mddev); |
689389a0 N |
2514 | allow_barrier(conf, r1_bio->sector); |
2515 | bio = r1_bio->master_bio; | |
62096bce | 2516 | |
689389a0 N |
2517 | /* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */ |
2518 | r1_bio->state = 0; | |
2519 | raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); | |
62096bce N |
2520 | } |
2521 | ||
4ed8731d | 2522 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2523 | { |
4ed8731d | 2524 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2525 | struct r1bio *r1_bio; |
1da177e4 | 2526 | unsigned long flags; |
e8096360 | 2527 | struct r1conf *conf = mddev->private; |
1da177e4 | 2528 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2529 | struct blk_plug plug; |
fd76863e | 2530 | int idx; |
1da177e4 LT |
2531 | |
2532 | md_check_recovery(mddev); | |
e1dfa0a2 | 2533 | |
55ce74d4 | 2534 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2535 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2536 | LIST_HEAD(tmp); |
2537 | spin_lock_irqsave(&conf->device_lock, flags); | |
fd76863e | 2538 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
2539 | list_splice_init(&conf->bio_end_io_list, &tmp); | |
55ce74d4 N |
2540 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2541 | while (!list_empty(&tmp)) { | |
a452744b MP |
2542 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2543 | retry_list); | |
55ce74d4 | 2544 | list_del(&r1_bio->retry_list); |
fd76863e | 2545 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2546 | atomic_dec(&conf->nr_queued[idx]); |
bd8688a1 N |
2547 | if (mddev->degraded) |
2548 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2549 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2550 | close_write(r1_bio); | |
55ce74d4 N |
2551 | raid_end_bio_io(r1_bio); |
2552 | } | |
2553 | } | |
2554 | ||
e1dfa0a2 | 2555 | blk_start_plug(&plug); |
1da177e4 | 2556 | for (;;) { |
191ea9b2 | 2557 | |
0021b7bc | 2558 | flush_pending_writes(conf); |
191ea9b2 | 2559 | |
a35e63ef N |
2560 | spin_lock_irqsave(&conf->device_lock, flags); |
2561 | if (list_empty(head)) { | |
2562 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2563 | break; |
a35e63ef | 2564 | } |
9f2c9d12 | 2565 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2566 | list_del(head->prev); |
fd76863e | 2567 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2568 | atomic_dec(&conf->nr_queued[idx]); |
1da177e4 LT |
2569 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2570 | ||
2571 | mddev = r1_bio->mddev; | |
070ec55d | 2572 | conf = mddev->private; |
4367af55 | 2573 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2574 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2575 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2576 | handle_sync_write_finished(conf, r1_bio); | |
2577 | else | |
4367af55 | 2578 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2579 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2580 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2581 | handle_write_finished(conf, r1_bio); | |
2582 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2583 | handle_read_error(conf, r1_bio); | |
2584 | else | |
c230e7e5 | 2585 | WARN_ON_ONCE(1); |
62096bce | 2586 | |
1d9d5241 | 2587 | cond_resched(); |
2953079c | 2588 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2589 | md_check_recovery(mddev); |
1da177e4 | 2590 | } |
e1dfa0a2 | 2591 | blk_finish_plug(&plug); |
1da177e4 LT |
2592 | } |
2593 | ||
e8096360 | 2594 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2595 | { |
2596 | int buffs; | |
2597 | ||
2598 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 KO |
2599 | BUG_ON(mempool_initialized(&conf->r1buf_pool)); |
2600 | ||
2601 | return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, | |
2602 | r1buf_pool_free, conf->poolinfo); | |
1da177e4 LT |
2603 | } |
2604 | ||
208410b5 SL |
2605 | static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf) |
2606 | { | |
afeee514 | 2607 | struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); |
208410b5 SL |
2608 | struct resync_pages *rps; |
2609 | struct bio *bio; | |
2610 | int i; | |
2611 | ||
2612 | for (i = conf->poolinfo->raid_disks; i--; ) { | |
2613 | bio = r1bio->bios[i]; | |
2614 | rps = bio->bi_private; | |
2615 | bio_reset(bio); | |
2616 | bio->bi_private = rps; | |
2617 | } | |
2618 | r1bio->master_bio = NULL; | |
2619 | return r1bio; | |
2620 | } | |
2621 | ||
1da177e4 LT |
2622 | /* |
2623 | * perform a "sync" on one "block" | |
2624 | * | |
2625 | * We need to make sure that no normal I/O request - particularly write | |
2626 | * requests - conflict with active sync requests. | |
2627 | * | |
2628 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2629 | * that can be installed to exclude normal IO requests. | |
2630 | */ | |
2631 | ||
849674e4 SL |
2632 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2633 | int *skipped) | |
1da177e4 | 2634 | { |
e8096360 | 2635 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2636 | struct r1bio *r1_bio; |
1da177e4 LT |
2637 | struct bio *bio; |
2638 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2639 | int disk = -1; |
1da177e4 | 2640 | int i; |
3e198f78 N |
2641 | int wonly = -1; |
2642 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2643 | sector_t sync_blocks; |
e3b9703e | 2644 | int still_degraded = 0; |
06f60385 N |
2645 | int good_sectors = RESYNC_SECTORS; |
2646 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
fd76863e | 2647 | int idx = sector_to_idx(sector_nr); |
022e510f | 2648 | int page_idx = 0; |
1da177e4 | 2649 | |
afeee514 | 2650 | if (!mempool_initialized(&conf->r1buf_pool)) |
1da177e4 | 2651 | if (init_resync(conf)) |
57afd89f | 2652 | return 0; |
1da177e4 | 2653 | |
58c0fed4 | 2654 | max_sector = mddev->dev_sectors; |
1da177e4 | 2655 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2656 | /* If we aborted, we need to abort the |
2657 | * sync on the 'current' bitmap chunk (there will | |
2658 | * only be one in raid1 resync. | |
2659 | * We can find the current addess in mddev->curr_resync | |
2660 | */ | |
6a806c51 | 2661 | if (mddev->curr_resync < max_sector) /* aborted */ |
e64e4018 AS |
2662 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
2663 | &sync_blocks, 1); | |
6a806c51 | 2664 | else /* completed sync */ |
191ea9b2 | 2665 | conf->fullsync = 0; |
6a806c51 | 2666 | |
e64e4018 | 2667 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2668 | close_sync(conf); |
c40f341f GR |
2669 | |
2670 | if (mddev_is_clustered(mddev)) { | |
2671 | conf->cluster_sync_low = 0; | |
2672 | conf->cluster_sync_high = 0; | |
c40f341f | 2673 | } |
1da177e4 LT |
2674 | return 0; |
2675 | } | |
2676 | ||
07d84d10 N |
2677 | if (mddev->bitmap == NULL && |
2678 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2679 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2680 | conf->fullsync == 0) { |
2681 | *skipped = 1; | |
2682 | return max_sector - sector_nr; | |
2683 | } | |
6394cca5 N |
2684 | /* before building a request, check if we can skip these blocks.. |
2685 | * This call the bitmap_start_sync doesn't actually record anything | |
2686 | */ | |
e64e4018 | 2687 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2688 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2689 | /* We can skip this block, and probably several more */ |
2690 | *skipped = 1; | |
2691 | return sync_blocks; | |
2692 | } | |
17999be4 | 2693 | |
7ac50447 TM |
2694 | /* |
2695 | * If there is non-resync activity waiting for a turn, then let it | |
2696 | * though before starting on this new sync request. | |
2697 | */ | |
824e47da | 2698 | if (atomic_read(&conf->nr_waiting[idx])) |
7ac50447 TM |
2699 | schedule_timeout_uninterruptible(1); |
2700 | ||
c40f341f GR |
2701 | /* we are incrementing sector_nr below. To be safe, we check against |
2702 | * sector_nr + two times RESYNC_SECTORS | |
2703 | */ | |
2704 | ||
e64e4018 | 2705 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
c40f341f | 2706 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); |
17999be4 | 2707 | |
8c242593 YY |
2708 | |
2709 | if (raise_barrier(conf, sector_nr)) | |
2710 | return 0; | |
2711 | ||
2712 | r1_bio = raid1_alloc_init_r1buf(conf); | |
1da177e4 | 2713 | |
3e198f78 | 2714 | rcu_read_lock(); |
1da177e4 | 2715 | /* |
3e198f78 N |
2716 | * If we get a correctably read error during resync or recovery, |
2717 | * we might want to read from a different device. So we | |
2718 | * flag all drives that could conceivably be read from for READ, | |
2719 | * and any others (which will be non-In_sync devices) for WRITE. | |
2720 | * If a read fails, we try reading from something else for which READ | |
2721 | * is OK. | |
1da177e4 | 2722 | */ |
1da177e4 | 2723 | |
1da177e4 LT |
2724 | r1_bio->mddev = mddev; |
2725 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2726 | r1_bio->state = 0; |
1da177e4 | 2727 | set_bit(R1BIO_IsSync, &r1_bio->state); |
fd76863e | 2728 | /* make sure good_sectors won't go across barrier unit boundary */ |
2729 | good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); | |
1da177e4 | 2730 | |
8f19ccb2 | 2731 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2732 | struct md_rdev *rdev; |
1da177e4 | 2733 | bio = r1_bio->bios[i]; |
1da177e4 | 2734 | |
3e198f78 N |
2735 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2736 | if (rdev == NULL || | |
06f60385 | 2737 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2738 | if (i < conf->raid_disks) |
2739 | still_degraded = 1; | |
3e198f78 | 2740 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2741 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2742 | bio->bi_end_io = end_sync_write; |
2743 | write_targets ++; | |
3e198f78 N |
2744 | } else { |
2745 | /* may need to read from here */ | |
06f60385 N |
2746 | sector_t first_bad = MaxSector; |
2747 | int bad_sectors; | |
2748 | ||
2749 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2750 | &first_bad, &bad_sectors)) { | |
2751 | if (first_bad > sector_nr) | |
2752 | good_sectors = first_bad - sector_nr; | |
2753 | else { | |
2754 | bad_sectors -= (sector_nr - first_bad); | |
2755 | if (min_bad == 0 || | |
2756 | min_bad > bad_sectors) | |
2757 | min_bad = bad_sectors; | |
2758 | } | |
2759 | } | |
2760 | if (sector_nr < first_bad) { | |
2761 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2762 | if (wonly < 0) | |
2763 | wonly = i; | |
2764 | } else { | |
2765 | if (disk < 0) | |
2766 | disk = i; | |
2767 | } | |
796a5cf0 | 2768 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2769 | bio->bi_end_io = end_sync_read; |
2770 | read_targets++; | |
d57368af AL |
2771 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2772 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2773 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2774 | /* | |
2775 | * The device is suitable for reading (InSync), | |
2776 | * but has bad block(s) here. Let's try to correct them, | |
2777 | * if we are doing resync or repair. Otherwise, leave | |
2778 | * this device alone for this sync request. | |
2779 | */ | |
796a5cf0 | 2780 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2781 | bio->bi_end_io = end_sync_write; |
2782 | write_targets++; | |
3e198f78 | 2783 | } |
3e198f78 | 2784 | } |
06f60385 N |
2785 | if (bio->bi_end_io) { |
2786 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2787 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
74d46992 | 2788 | bio_set_dev(bio, rdev->bdev); |
2e52d449 N |
2789 | if (test_bit(FailFast, &rdev->flags)) |
2790 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2791 | } |
1da177e4 | 2792 | } |
3e198f78 N |
2793 | rcu_read_unlock(); |
2794 | if (disk < 0) | |
2795 | disk = wonly; | |
2796 | r1_bio->read_disk = disk; | |
191ea9b2 | 2797 | |
06f60385 N |
2798 | if (read_targets == 0 && min_bad > 0) { |
2799 | /* These sectors are bad on all InSync devices, so we | |
2800 | * need to mark them bad on all write targets | |
2801 | */ | |
2802 | int ok = 1; | |
8f19ccb2 | 2803 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2804 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2805 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2806 | ok = rdev_set_badblocks(rdev, sector_nr, |
2807 | min_bad, 0 | |
2808 | ) && ok; | |
2809 | } | |
2953079c | 2810 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2811 | *skipped = 1; |
2812 | put_buf(r1_bio); | |
2813 | ||
2814 | if (!ok) { | |
2815 | /* Cannot record the badblocks, so need to | |
2816 | * abort the resync. | |
2817 | * If there are multiple read targets, could just | |
2818 | * fail the really bad ones ??? | |
2819 | */ | |
2820 | conf->recovery_disabled = mddev->recovery_disabled; | |
2821 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2822 | return 0; | |
2823 | } else | |
2824 | return min_bad; | |
2825 | ||
2826 | } | |
2827 | if (min_bad > 0 && min_bad < good_sectors) { | |
2828 | /* only resync enough to reach the next bad->good | |
2829 | * transition */ | |
2830 | good_sectors = min_bad; | |
2831 | } | |
2832 | ||
3e198f78 N |
2833 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2834 | /* extra read targets are also write targets */ | |
2835 | write_targets += read_targets-1; | |
2836 | ||
2837 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2838 | /* There is nowhere to write, so all non-sync |
2839 | * drives must be failed - so we are finished | |
2840 | */ | |
b7219ccb N |
2841 | sector_t rv; |
2842 | if (min_bad > 0) | |
2843 | max_sector = sector_nr + min_bad; | |
2844 | rv = max_sector - sector_nr; | |
57afd89f | 2845 | *skipped = 1; |
1da177e4 | 2846 | put_buf(r1_bio); |
1da177e4 LT |
2847 | return rv; |
2848 | } | |
2849 | ||
c6207277 N |
2850 | if (max_sector > mddev->resync_max) |
2851 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2852 | if (max_sector > sector_nr + good_sectors) |
2853 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2854 | nr_sectors = 0; |
289e99e8 | 2855 | sync_blocks = 0; |
1da177e4 LT |
2856 | do { |
2857 | struct page *page; | |
2858 | int len = PAGE_SIZE; | |
2859 | if (sector_nr + (len>>9) > max_sector) | |
2860 | len = (max_sector - sector_nr) << 9; | |
2861 | if (len == 0) | |
2862 | break; | |
6a806c51 | 2863 | if (sync_blocks == 0) { |
e64e4018 AS |
2864 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
2865 | &sync_blocks, still_degraded) && | |
e5de485f N |
2866 | !conf->fullsync && |
2867 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2868 | break; |
7571ae88 | 2869 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2870 | len = sync_blocks<<9; |
ab7a30c7 | 2871 | } |
191ea9b2 | 2872 | |
8f19ccb2 | 2873 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
98d30c58 ML |
2874 | struct resync_pages *rp; |
2875 | ||
1da177e4 | 2876 | bio = r1_bio->bios[i]; |
98d30c58 | 2877 | rp = get_resync_pages(bio); |
1da177e4 | 2878 | if (bio->bi_end_io) { |
022e510f | 2879 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
2880 | |
2881 | /* | |
2882 | * won't fail because the vec table is big | |
2883 | * enough to hold all these pages | |
2884 | */ | |
2885 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
2886 | } |
2887 | } | |
2888 | nr_sectors += len>>9; | |
2889 | sector_nr += len>>9; | |
191ea9b2 | 2890 | sync_blocks -= (len>>9); |
022e510f | 2891 | } while (++page_idx < RESYNC_PAGES); |
98d30c58 | 2892 | |
1da177e4 LT |
2893 | r1_bio->sectors = nr_sectors; |
2894 | ||
c40f341f GR |
2895 | if (mddev_is_clustered(mddev) && |
2896 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2897 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2898 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2899 | /* Send resync message */ | |
2900 | md_cluster_ops->resync_info_update(mddev, | |
2901 | conf->cluster_sync_low, | |
2902 | conf->cluster_sync_high); | |
2903 | } | |
2904 | ||
d11c171e N |
2905 | /* For a user-requested sync, we read all readable devices and do a |
2906 | * compare | |
2907 | */ | |
2908 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2909 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2910 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2911 | bio = r1_bio->bios[i]; |
2912 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2913 | read_targets--; |
74d46992 | 2914 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2915 | if (read_targets == 1) |
2916 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e N |
2917 | generic_make_request(bio); |
2918 | } | |
2919 | } | |
2920 | } else { | |
2921 | atomic_set(&r1_bio->remaining, 1); | |
2922 | bio = r1_bio->bios[r1_bio->read_disk]; | |
74d46992 | 2923 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2924 | if (read_targets == 1) |
2925 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e | 2926 | generic_make_request(bio); |
d11c171e | 2927 | } |
1da177e4 LT |
2928 | return nr_sectors; |
2929 | } | |
2930 | ||
fd01b88c | 2931 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2932 | { |
2933 | if (sectors) | |
2934 | return sectors; | |
2935 | ||
2936 | return mddev->dev_sectors; | |
2937 | } | |
2938 | ||
e8096360 | 2939 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2940 | { |
e8096360 | 2941 | struct r1conf *conf; |
709ae487 | 2942 | int i; |
0eaf822c | 2943 | struct raid1_info *disk; |
3cb03002 | 2944 | struct md_rdev *rdev; |
709ae487 | 2945 | int err = -ENOMEM; |
1da177e4 | 2946 | |
e8096360 | 2947 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2948 | if (!conf) |
709ae487 | 2949 | goto abort; |
1da177e4 | 2950 | |
fd76863e | 2951 | conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, |
824e47da | 2952 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2953 | if (!conf->nr_pending) |
2954 | goto abort; | |
2955 | ||
2956 | conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2957 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2958 | if (!conf->nr_waiting) |
2959 | goto abort; | |
2960 | ||
2961 | conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2962 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2963 | if (!conf->nr_queued) |
2964 | goto abort; | |
2965 | ||
2966 | conf->barrier = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2967 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2968 | if (!conf->barrier) |
2969 | goto abort; | |
2970 | ||
6396bb22 KC |
2971 | conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
2972 | mddev->raid_disks, 2), | |
2973 | GFP_KERNEL); | |
1da177e4 | 2974 | if (!conf->mirrors) |
709ae487 | 2975 | goto abort; |
1da177e4 | 2976 | |
ddaf22ab N |
2977 | conf->tmppage = alloc_page(GFP_KERNEL); |
2978 | if (!conf->tmppage) | |
709ae487 | 2979 | goto abort; |
ddaf22ab | 2980 | |
709ae487 | 2981 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2982 | if (!conf->poolinfo) |
709ae487 | 2983 | goto abort; |
8f19ccb2 | 2984 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
3f677f9c | 2985 | err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 2986 | rbio_pool_free, conf->poolinfo); |
afeee514 | 2987 | if (err) |
709ae487 N |
2988 | goto abort; |
2989 | ||
afeee514 KO |
2990 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
2991 | if (err) | |
c230e7e5 N |
2992 | goto abort; |
2993 | ||
ed9bfdf1 | 2994 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2995 | |
c19d5798 | 2996 | err = -EINVAL; |
e7e72bf6 | 2997 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2998 | rdev_for_each(rdev, mddev) { |
709ae487 | 2999 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
3000 | if (disk_idx >= mddev->raid_disks |
3001 | || disk_idx < 0) | |
3002 | continue; | |
c19d5798 | 3003 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 3004 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
3005 | else |
3006 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 3007 | |
c19d5798 N |
3008 | if (disk->rdev) |
3009 | goto abort; | |
1da177e4 | 3010 | disk->rdev = rdev; |
1da177e4 | 3011 | disk->head_position = 0; |
12cee5a8 | 3012 | disk->seq_start = MaxSector; |
1da177e4 LT |
3013 | } |
3014 | conf->raid_disks = mddev->raid_disks; | |
3015 | conf->mddev = mddev; | |
1da177e4 | 3016 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 3017 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
3018 | |
3019 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 3020 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 3021 | |
191ea9b2 | 3022 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 3023 | conf->pending_count = 0; |
d890fa2b | 3024 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 3025 | |
c19d5798 | 3026 | err = -EIO; |
8f19ccb2 | 3027 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
3028 | |
3029 | disk = conf->mirrors + i; | |
3030 | ||
c19d5798 N |
3031 | if (i < conf->raid_disks && |
3032 | disk[conf->raid_disks].rdev) { | |
3033 | /* This slot has a replacement. */ | |
3034 | if (!disk->rdev) { | |
3035 | /* No original, just make the replacement | |
3036 | * a recovering spare | |
3037 | */ | |
3038 | disk->rdev = | |
3039 | disk[conf->raid_disks].rdev; | |
3040 | disk[conf->raid_disks].rdev = NULL; | |
3041 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
3042 | /* Original is not in_sync - bad */ | |
3043 | goto abort; | |
3044 | } | |
3045 | ||
5fd6c1dc N |
3046 | if (!disk->rdev || |
3047 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 3048 | disk->head_position = 0; |
4f0a5e01 JB |
3049 | if (disk->rdev && |
3050 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 3051 | conf->fullsync = 1; |
be4d3280 | 3052 | } |
1da177e4 | 3053 | } |
709ae487 | 3054 | |
709ae487 | 3055 | err = -ENOMEM; |
0232605d | 3056 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 3057 | if (!conf->thread) |
709ae487 | 3058 | goto abort; |
1da177e4 | 3059 | |
709ae487 N |
3060 | return conf; |
3061 | ||
3062 | abort: | |
3063 | if (conf) { | |
afeee514 | 3064 | mempool_exit(&conf->r1bio_pool); |
709ae487 N |
3065 | kfree(conf->mirrors); |
3066 | safe_put_page(conf->tmppage); | |
3067 | kfree(conf->poolinfo); | |
fd76863e | 3068 | kfree(conf->nr_pending); |
3069 | kfree(conf->nr_waiting); | |
3070 | kfree(conf->nr_queued); | |
3071 | kfree(conf->barrier); | |
afeee514 | 3072 | bioset_exit(&conf->bio_split); |
709ae487 N |
3073 | kfree(conf); |
3074 | } | |
3075 | return ERR_PTR(err); | |
3076 | } | |
3077 | ||
afa0f557 | 3078 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 3079 | static int raid1_run(struct mddev *mddev) |
709ae487 | 3080 | { |
e8096360 | 3081 | struct r1conf *conf; |
709ae487 | 3082 | int i; |
3cb03002 | 3083 | struct md_rdev *rdev; |
5220ea1e | 3084 | int ret; |
2ff8cc2c | 3085 | bool discard_supported = false; |
709ae487 N |
3086 | |
3087 | if (mddev->level != 1) { | |
1d41c216 N |
3088 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3089 | mdname(mddev), mddev->level); | |
709ae487 N |
3090 | return -EIO; |
3091 | } | |
3092 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3093 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3094 | mdname(mddev)); | |
709ae487 N |
3095 | return -EIO; |
3096 | } | |
a415c0f1 N |
3097 | if (mddev_init_writes_pending(mddev) < 0) |
3098 | return -ENOMEM; | |
1da177e4 | 3099 | /* |
709ae487 N |
3100 | * copy the already verified devices into our private RAID1 |
3101 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3102 | * should be freed in raid1_free()] |
1da177e4 | 3103 | */ |
709ae487 N |
3104 | if (mddev->private == NULL) |
3105 | conf = setup_conf(mddev); | |
3106 | else | |
3107 | conf = mddev->private; | |
1da177e4 | 3108 | |
709ae487 N |
3109 | if (IS_ERR(conf)) |
3110 | return PTR_ERR(conf); | |
1da177e4 | 3111 | |
3deff1a7 | 3112 | if (mddev->queue) { |
5026d7a9 | 3113 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 CH |
3114 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
3115 | } | |
5026d7a9 | 3116 | |
dafb20fa | 3117 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3118 | if (!mddev->gendisk) |
3119 | continue; | |
709ae487 N |
3120 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3121 | rdev->data_offset << 9); | |
2ff8cc2c SL |
3122 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
3123 | discard_supported = true; | |
1da177e4 | 3124 | } |
191ea9b2 | 3125 | |
709ae487 | 3126 | mddev->degraded = 0; |
ebfeb444 | 3127 | for (i = 0; i < conf->raid_disks; i++) |
709ae487 N |
3128 | if (conf->mirrors[i].rdev == NULL || |
3129 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3130 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3131 | mddev->degraded++; | |
b4e28886 YY |
3132 | /* |
3133 | * RAID1 needs at least one disk in active | |
3134 | */ | |
3135 | if (conf->raid_disks - mddev->degraded < 1) { | |
3136 | ret = -EINVAL; | |
3137 | goto abort; | |
3138 | } | |
709ae487 N |
3139 | |
3140 | if (conf->raid_disks - mddev->degraded == 1) | |
3141 | mddev->recovery_cp = MaxSector; | |
3142 | ||
8c6ac868 | 3143 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3144 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3145 | mdname(mddev)); | |
3146 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3147 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3148 | mddev->raid_disks); |
709ae487 | 3149 | |
1da177e4 LT |
3150 | /* |
3151 | * Ok, everything is just fine now | |
3152 | */ | |
709ae487 N |
3153 | mddev->thread = conf->thread; |
3154 | conf->thread = NULL; | |
3155 | mddev->private = conf; | |
46533ff7 | 3156 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3157 | |
1f403624 | 3158 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3159 | |
1ed7242e | 3160 | if (mddev->queue) { |
2ff8cc2c | 3161 | if (discard_supported) |
8b904b5b | 3162 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
2ff8cc2c SL |
3163 | mddev->queue); |
3164 | else | |
8b904b5b | 3165 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
2ff8cc2c | 3166 | mddev->queue); |
1ed7242e | 3167 | } |
5220ea1e | 3168 | |
ebfeb444 | 3169 | ret = md_integrity_register(mddev); |
5aa61f42 N |
3170 | if (ret) { |
3171 | md_unregister_thread(&mddev->thread); | |
b4e28886 | 3172 | goto abort; |
5aa61f42 | 3173 | } |
b4e28886 YY |
3174 | return 0; |
3175 | ||
3176 | abort: | |
3177 | raid1_free(mddev, conf); | |
5220ea1e | 3178 | return ret; |
1da177e4 LT |
3179 | } |
3180 | ||
afa0f557 | 3181 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3182 | { |
afa0f557 | 3183 | struct r1conf *conf = priv; |
409c57f3 | 3184 | |
afeee514 | 3185 | mempool_exit(&conf->r1bio_pool); |
990a8baf | 3186 | kfree(conf->mirrors); |
0fea7ed8 | 3187 | safe_put_page(conf->tmppage); |
990a8baf | 3188 | kfree(conf->poolinfo); |
fd76863e | 3189 | kfree(conf->nr_pending); |
3190 | kfree(conf->nr_waiting); | |
3191 | kfree(conf->nr_queued); | |
3192 | kfree(conf->barrier); | |
afeee514 | 3193 | bioset_exit(&conf->bio_split); |
1da177e4 | 3194 | kfree(conf); |
1da177e4 LT |
3195 | } |
3196 | ||
fd01b88c | 3197 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3198 | { |
3199 | /* no resync is happening, and there is enough space | |
3200 | * on all devices, so we can resize. | |
3201 | * We need to make sure resync covers any new space. | |
3202 | * If the array is shrinking we should possibly wait until | |
3203 | * any io in the removed space completes, but it hardly seems | |
3204 | * worth it. | |
3205 | */ | |
a4a6125a N |
3206 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3207 | if (mddev->external_size && | |
3208 | mddev->array_sectors > newsize) | |
b522adcd | 3209 | return -EINVAL; |
a4a6125a | 3210 | if (mddev->bitmap) { |
e64e4018 | 3211 | int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); |
a4a6125a N |
3212 | if (ret) |
3213 | return ret; | |
3214 | } | |
3215 | md_set_array_sectors(mddev, newsize); | |
b522adcd | 3216 | if (sectors > mddev->dev_sectors && |
b098636c | 3217 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3218 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3219 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3220 | } | |
b522adcd | 3221 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3222 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3223 | return 0; |
3224 | } | |
3225 | ||
fd01b88c | 3226 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3227 | { |
3228 | /* We need to: | |
3229 | * 1/ resize the r1bio_pool | |
3230 | * 2/ resize conf->mirrors | |
3231 | * | |
3232 | * We allocate a new r1bio_pool if we can. | |
3233 | * Then raise a device barrier and wait until all IO stops. | |
3234 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3235 | * |
3236 | * At the same time, we "pack" the devices so that all the missing | |
3237 | * devices have the higher raid_disk numbers. | |
1da177e4 | 3238 | */ |
afeee514 | 3239 | mempool_t newpool, oldpool; |
1da177e4 | 3240 | struct pool_info *newpoolinfo; |
0eaf822c | 3241 | struct raid1_info *newmirrors; |
e8096360 | 3242 | struct r1conf *conf = mddev->private; |
63c70c4f | 3243 | int cnt, raid_disks; |
c04be0aa | 3244 | unsigned long flags; |
2214c260 | 3245 | int d, d2; |
afeee514 KO |
3246 | int ret; |
3247 | ||
3248 | memset(&newpool, 0, sizeof(newpool)); | |
3249 | memset(&oldpool, 0, sizeof(oldpool)); | |
1da177e4 | 3250 | |
63c70c4f | 3251 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3252 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3253 | mddev->layout != mddev->new_layout || |
3254 | mddev->level != mddev->new_level) { | |
664e7c41 | 3255 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3256 | mddev->new_layout = mddev->layout; |
3257 | mddev->new_level = mddev->level; | |
3258 | return -EINVAL; | |
3259 | } | |
3260 | ||
2214c260 AP |
3261 | if (!mddev_is_clustered(mddev)) |
3262 | md_allow_write(mddev); | |
2a2275d6 | 3263 | |
63c70c4f N |
3264 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3265 | ||
6ea9c07c N |
3266 | if (raid_disks < conf->raid_disks) { |
3267 | cnt=0; | |
3268 | for (d= 0; d < conf->raid_disks; d++) | |
3269 | if (conf->mirrors[d].rdev) | |
3270 | cnt++; | |
3271 | if (cnt > raid_disks) | |
1da177e4 | 3272 | return -EBUSY; |
6ea9c07c | 3273 | } |
1da177e4 LT |
3274 | |
3275 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3276 | if (!newpoolinfo) | |
3277 | return -ENOMEM; | |
3278 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3279 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 | 3280 | |
3f677f9c | 3281 | ret = mempool_init(&newpool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 3282 | rbio_pool_free, newpoolinfo); |
afeee514 | 3283 | if (ret) { |
1da177e4 | 3284 | kfree(newpoolinfo); |
afeee514 | 3285 | return ret; |
1da177e4 | 3286 | } |
6396bb22 KC |
3287 | newmirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
3288 | raid_disks, 2), | |
8f19ccb2 | 3289 | GFP_KERNEL); |
1da177e4 LT |
3290 | if (!newmirrors) { |
3291 | kfree(newpoolinfo); | |
afeee514 | 3292 | mempool_exit(&newpool); |
1da177e4 LT |
3293 | return -ENOMEM; |
3294 | } | |
1da177e4 | 3295 | |
e2d59925 | 3296 | freeze_array(conf, 0); |
1da177e4 LT |
3297 | |
3298 | /* ok, everything is stopped */ | |
3299 | oldpool = conf->r1bio_pool; | |
3300 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3301 | |
a88aa786 | 3302 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3303 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3304 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3305 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3306 | rdev->raid_disk = d2; |
36fad858 NK |
3307 | sysfs_unlink_rdev(mddev, rdev); |
3308 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3309 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3310 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3311 | } |
a88aa786 N |
3312 | if (rdev) |
3313 | newmirrors[d2++].rdev = rdev; | |
3314 | } | |
1da177e4 LT |
3315 | kfree(conf->mirrors); |
3316 | conf->mirrors = newmirrors; | |
3317 | kfree(conf->poolinfo); | |
3318 | conf->poolinfo = newpoolinfo; | |
3319 | ||
c04be0aa | 3320 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3321 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3322 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3323 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3324 | mddev->delta_disks = 0; |
1da177e4 | 3325 | |
e2d59925 | 3326 | unfreeze_array(conf); |
1da177e4 | 3327 | |
985ca973 | 3328 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3329 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3330 | md_wakeup_thread(mddev->thread); | |
3331 | ||
afeee514 | 3332 | mempool_exit(&oldpool); |
1da177e4 LT |
3333 | return 0; |
3334 | } | |
3335 | ||
b03e0ccb | 3336 | static void raid1_quiesce(struct mddev *mddev, int quiesce) |
36fa3063 | 3337 | { |
e8096360 | 3338 | struct r1conf *conf = mddev->private; |
36fa3063 | 3339 | |
b03e0ccb | 3340 | if (quiesce) |
07169fd4 | 3341 | freeze_array(conf, 0); |
b03e0ccb | 3342 | else |
07169fd4 | 3343 | unfreeze_array(conf); |
36fa3063 N |
3344 | } |
3345 | ||
fd01b88c | 3346 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3347 | { |
3348 | /* raid1 can take over: | |
3349 | * raid5 with 2 devices, any layout or chunk size | |
3350 | */ | |
3351 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3352 | struct r1conf *conf; |
709ae487 N |
3353 | mddev->new_level = 1; |
3354 | mddev->new_layout = 0; | |
3355 | mddev->new_chunk_sectors = 0; | |
3356 | conf = setup_conf(mddev); | |
6995f0b2 | 3357 | if (!IS_ERR(conf)) { |
07169fd4 | 3358 | /* Array must appear to be quiesced */ |
3359 | conf->array_frozen = 1; | |
394ed8e4 SL |
3360 | mddev_clear_unsupported_flags(mddev, |
3361 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3362 | } |
709ae487 N |
3363 | return conf; |
3364 | } | |
3365 | return ERR_PTR(-EINVAL); | |
3366 | } | |
1da177e4 | 3367 | |
84fc4b56 | 3368 | static struct md_personality raid1_personality = |
1da177e4 LT |
3369 | { |
3370 | .name = "raid1", | |
2604b703 | 3371 | .level = 1, |
1da177e4 | 3372 | .owner = THIS_MODULE, |
849674e4 SL |
3373 | .make_request = raid1_make_request, |
3374 | .run = raid1_run, | |
afa0f557 | 3375 | .free = raid1_free, |
849674e4 SL |
3376 | .status = raid1_status, |
3377 | .error_handler = raid1_error, | |
1da177e4 LT |
3378 | .hot_add_disk = raid1_add_disk, |
3379 | .hot_remove_disk= raid1_remove_disk, | |
3380 | .spare_active = raid1_spare_active, | |
849674e4 | 3381 | .sync_request = raid1_sync_request, |
1da177e4 | 3382 | .resize = raid1_resize, |
80c3a6ce | 3383 | .size = raid1_size, |
63c70c4f | 3384 | .check_reshape = raid1_reshape, |
36fa3063 | 3385 | .quiesce = raid1_quiesce, |
709ae487 | 3386 | .takeover = raid1_takeover, |
5c675f83 | 3387 | .congested = raid1_congested, |
1da177e4 LT |
3388 | }; |
3389 | ||
3390 | static int __init raid_init(void) | |
3391 | { | |
2604b703 | 3392 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3393 | } |
3394 | ||
3395 | static void raid_exit(void) | |
3396 | { | |
2604b703 | 3397 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3398 | } |
3399 | ||
3400 | module_init(raid_init); | |
3401 | module_exit(raid_exit); | |
3402 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3403 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3404 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3405 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3406 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3407 | |
3408 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |