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> |
69b00b5b | 32 | #include <linux/interval_tree_generic.h> |
3f07c014 | 33 | |
109e3765 | 34 | #include <trace/events/block.h> |
3f07c014 | 35 | |
43b2e5d8 | 36 | #include "md.h" |
ef740c37 | 37 | #include "raid1.h" |
935fe098 | 38 | #include "md-bitmap.h" |
191ea9b2 | 39 | |
394ed8e4 SL |
40 | #define UNSUPPORTED_MDDEV_FLAGS \ |
41 | ((1L << MD_HAS_JOURNAL) | \ | |
ea0213e0 | 42 | (1L << MD_JOURNAL_CLEAN) | \ |
ddc08823 PB |
43 | (1L << MD_HAS_PPL) | \ |
44 | (1L << MD_HAS_MULTIPLE_PPLS)) | |
394ed8e4 | 45 | |
fd76863e | 46 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr); |
47 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr); | |
1da177e4 | 48 | |
1979dbbe | 49 | #define RAID_1_10_NAME "raid1" |
fb0eb5df ML |
50 | #include "raid1-10.c" |
51 | ||
69b00b5b GJ |
52 | #define START(node) ((node)->start) |
53 | #define LAST(node) ((node)->last) | |
54 | INTERVAL_TREE_DEFINE(struct serial_info, node, sector_t, _subtree_last, | |
55 | START, LAST, static inline, raid1_rb); | |
56 | ||
d0d2d8ba GJ |
57 | static int check_and_add_serial(struct md_rdev *rdev, struct r1bio *r1_bio, |
58 | struct serial_info *si, int idx) | |
3e148a32 | 59 | { |
3e148a32 GJ |
60 | unsigned long flags; |
61 | int ret = 0; | |
d0d2d8ba GJ |
62 | sector_t lo = r1_bio->sector; |
63 | sector_t hi = lo + r1_bio->sectors; | |
025471f9 | 64 | struct serial_in_rdev *serial = &rdev->serial[idx]; |
3e148a32 | 65 | |
69b00b5b GJ |
66 | spin_lock_irqsave(&serial->serial_lock, flags); |
67 | /* collision happened */ | |
68 | if (raid1_rb_iter_first(&serial->serial_rb, lo, hi)) | |
69 | ret = -EBUSY; | |
d0d2d8ba | 70 | else { |
69b00b5b GJ |
71 | si->start = lo; |
72 | si->last = hi; | |
73 | raid1_rb_insert(si, &serial->serial_rb); | |
d0d2d8ba | 74 | } |
69b00b5b | 75 | spin_unlock_irqrestore(&serial->serial_lock, flags); |
3e148a32 GJ |
76 | |
77 | return ret; | |
78 | } | |
79 | ||
d0d2d8ba GJ |
80 | static void wait_for_serialization(struct md_rdev *rdev, struct r1bio *r1_bio) |
81 | { | |
82 | struct mddev *mddev = rdev->mddev; | |
83 | struct serial_info *si; | |
84 | int idx = sector_to_idx(r1_bio->sector); | |
85 | struct serial_in_rdev *serial = &rdev->serial[idx]; | |
86 | ||
87 | if (WARN_ON(!mddev->serial_info_pool)) | |
88 | return; | |
89 | si = mempool_alloc(mddev->serial_info_pool, GFP_NOIO); | |
90 | wait_event(serial->serial_io_wait, | |
91 | check_and_add_serial(rdev, r1_bio, si, idx) == 0); | |
92 | } | |
93 | ||
404659cf | 94 | static void remove_serial(struct md_rdev *rdev, sector_t lo, sector_t hi) |
3e148a32 | 95 | { |
69b00b5b | 96 | struct serial_info *si; |
3e148a32 GJ |
97 | unsigned long flags; |
98 | int found = 0; | |
99 | struct mddev *mddev = rdev->mddev; | |
025471f9 GJ |
100 | int idx = sector_to_idx(lo); |
101 | struct serial_in_rdev *serial = &rdev->serial[idx]; | |
69b00b5b GJ |
102 | |
103 | spin_lock_irqsave(&serial->serial_lock, flags); | |
104 | for (si = raid1_rb_iter_first(&serial->serial_rb, lo, hi); | |
105 | si; si = raid1_rb_iter_next(si, lo, hi)) { | |
106 | if (si->start == lo && si->last == hi) { | |
107 | raid1_rb_remove(si, &serial->serial_rb); | |
108 | mempool_free(si, mddev->serial_info_pool); | |
3e148a32 GJ |
109 | found = 1; |
110 | break; | |
111 | } | |
69b00b5b | 112 | } |
3e148a32 | 113 | if (!found) |
404659cf | 114 | WARN(1, "The write IO is not recorded for serialization\n"); |
69b00b5b GJ |
115 | spin_unlock_irqrestore(&serial->serial_lock, flags); |
116 | wake_up(&serial->serial_io_wait); | |
3e148a32 GJ |
117 | } |
118 | ||
98d30c58 ML |
119 | /* |
120 | * for resync bio, r1bio pointer can be retrieved from the per-bio | |
121 | * 'struct resync_pages'. | |
122 | */ | |
123 | static inline struct r1bio *get_resync_r1bio(struct bio *bio) | |
124 | { | |
125 | return get_resync_pages(bio)->raid_bio; | |
126 | } | |
127 | ||
dd0fc66f | 128 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
129 | { |
130 | struct pool_info *pi = data; | |
9f2c9d12 | 131 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
132 | |
133 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 134 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
135 | } |
136 | ||
8e005f7c | 137 | #define RESYNC_DEPTH 32 |
1da177e4 | 138 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
8e005f7c | 139 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
140 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
141 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
142 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
1da177e4 | 143 | |
dd0fc66f | 144 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
145 | { |
146 | struct pool_info *pi = data; | |
9f2c9d12 | 147 | struct r1bio *r1_bio; |
1da177e4 | 148 | struct bio *bio; |
da1aab3d | 149 | int need_pages; |
98d30c58 ML |
150 | int j; |
151 | struct resync_pages *rps; | |
1da177e4 LT |
152 | |
153 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 154 | if (!r1_bio) |
1da177e4 | 155 | return NULL; |
1da177e4 | 156 | |
6da2ec56 KC |
157 | rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages), |
158 | gfp_flags); | |
98d30c58 ML |
159 | if (!rps) |
160 | goto out_free_r1bio; | |
161 | ||
1da177e4 LT |
162 | /* |
163 | * Allocate bios : 1 for reading, n-1 for writing | |
164 | */ | |
165 | for (j = pi->raid_disks ; j-- ; ) { | |
066ff571 | 166 | bio = bio_kmalloc(RESYNC_PAGES, gfp_flags); |
1da177e4 LT |
167 | if (!bio) |
168 | goto out_free_bio; | |
066ff571 | 169 | bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); |
1da177e4 LT |
170 | r1_bio->bios[j] = bio; |
171 | } | |
172 | /* | |
173 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
174 | * the first bio. |
175 | * If this is a user-requested check/repair, allocate | |
176 | * RESYNC_PAGES for each bio. | |
1da177e4 | 177 | */ |
d11c171e | 178 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 179 | need_pages = pi->raid_disks; |
d11c171e | 180 | else |
da1aab3d | 181 | need_pages = 1; |
98d30c58 ML |
182 | for (j = 0; j < pi->raid_disks; j++) { |
183 | struct resync_pages *rp = &rps[j]; | |
184 | ||
d11c171e | 185 | bio = r1_bio->bios[j]; |
d11c171e | 186 | |
98d30c58 ML |
187 | if (j < need_pages) { |
188 | if (resync_alloc_pages(rp, gfp_flags)) | |
189 | goto out_free_pages; | |
190 | } else { | |
191 | memcpy(rp, &rps[0], sizeof(*rp)); | |
192 | resync_get_all_pages(rp); | |
193 | } | |
194 | ||
98d30c58 ML |
195 | rp->raid_bio = r1_bio; |
196 | bio->bi_private = rp; | |
1da177e4 LT |
197 | } |
198 | ||
199 | r1_bio->master_bio = NULL; | |
200 | ||
201 | return r1_bio; | |
202 | ||
da1aab3d | 203 | out_free_pages: |
491221f8 | 204 | while (--j >= 0) |
98d30c58 | 205 | resync_free_pages(&rps[j]); |
da1aab3d | 206 | |
1da177e4 | 207 | out_free_bio: |
066ff571 CH |
208 | while (++j < pi->raid_disks) { |
209 | bio_uninit(r1_bio->bios[j]); | |
210 | kfree(r1_bio->bios[j]); | |
211 | } | |
98d30c58 ML |
212 | kfree(rps); |
213 | ||
214 | out_free_r1bio: | |
c7afa803 | 215 | rbio_pool_free(r1_bio, data); |
1da177e4 LT |
216 | return NULL; |
217 | } | |
218 | ||
219 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
220 | { | |
221 | struct pool_info *pi = data; | |
98d30c58 | 222 | int i; |
9f2c9d12 | 223 | struct r1bio *r1bio = __r1_bio; |
98d30c58 | 224 | struct resync_pages *rp = NULL; |
1da177e4 | 225 | |
98d30c58 ML |
226 | for (i = pi->raid_disks; i--; ) { |
227 | rp = get_resync_pages(r1bio->bios[i]); | |
228 | resync_free_pages(rp); | |
066ff571 CH |
229 | bio_uninit(r1bio->bios[i]); |
230 | kfree(r1bio->bios[i]); | |
98d30c58 ML |
231 | } |
232 | ||
233 | /* resync pages array stored in the 1st bio's .bi_private */ | |
234 | kfree(rp); | |
1da177e4 | 235 | |
c7afa803 | 236 | rbio_pool_free(r1bio, data); |
1da177e4 LT |
237 | } |
238 | ||
e8096360 | 239 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
240 | { |
241 | int i; | |
242 | ||
8f19ccb2 | 243 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 244 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 245 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
246 | bio_put(*bio); |
247 | *bio = NULL; | |
248 | } | |
249 | } | |
250 | ||
9f2c9d12 | 251 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 252 | { |
e8096360 | 253 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 254 | |
1da177e4 | 255 | put_all_bios(conf, r1_bio); |
afeee514 | 256 | mempool_free(r1_bio, &conf->r1bio_pool); |
1da177e4 LT |
257 | } |
258 | ||
9f2c9d12 | 259 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 260 | { |
e8096360 | 261 | struct r1conf *conf = r1_bio->mddev->private; |
af5f42a7 | 262 | sector_t sect = r1_bio->sector; |
3e198f78 N |
263 | int i; |
264 | ||
8f19ccb2 | 265 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
266 | struct bio *bio = r1_bio->bios[i]; |
267 | if (bio->bi_end_io) | |
268 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
269 | } | |
1da177e4 | 270 | |
afeee514 | 271 | mempool_free(r1_bio, &conf->r1buf_pool); |
1da177e4 | 272 | |
af5f42a7 | 273 | lower_barrier(conf, sect); |
1da177e4 LT |
274 | } |
275 | ||
9f2c9d12 | 276 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
277 | { |
278 | unsigned long flags; | |
fd01b88c | 279 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 280 | struct r1conf *conf = mddev->private; |
fd76863e | 281 | int idx; |
1da177e4 | 282 | |
fd76863e | 283 | idx = sector_to_idx(r1_bio->sector); |
1da177e4 LT |
284 | spin_lock_irqsave(&conf->device_lock, flags); |
285 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
824e47da | 286 | atomic_inc(&conf->nr_queued[idx]); |
1da177e4 LT |
287 | spin_unlock_irqrestore(&conf->device_lock, flags); |
288 | ||
17999be4 | 289 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
290 | md_wakeup_thread(mddev->thread); |
291 | } | |
292 | ||
293 | /* | |
294 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
295 | * operation and are ready to return a success/failure code to the buffer | |
296 | * cache layer. | |
297 | */ | |
9f2c9d12 | 298 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
299 | { |
300 | struct bio *bio = r1_bio->master_bio; | |
d2eb35ac N |
301 | |
302 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4e4cbee9 | 303 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 304 | |
37011e3a | 305 | bio_endio(bio); |
d2eb35ac N |
306 | } |
307 | ||
9f2c9d12 | 308 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
309 | { |
310 | struct bio *bio = r1_bio->master_bio; | |
c91114c2 | 311 | struct r1conf *conf = r1_bio->mddev->private; |
c5d736f5 | 312 | sector_t sector = r1_bio->sector; |
1da177e4 | 313 | |
4b6d287f N |
314 | /* if nobody has done the final endio yet, do it now */ |
315 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
316 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
317 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
318 | (unsigned long long) bio->bi_iter.bi_sector, |
319 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 320 | |
d2eb35ac | 321 | call_bio_endio(r1_bio); |
4b6d287f | 322 | } |
c5d736f5 XH |
323 | |
324 | free_r1bio(r1_bio); | |
c91114c2 DJ |
325 | /* |
326 | * Wake up any possible resync thread that waits for the device | |
327 | * to go idle. All I/Os, even write-behind writes, are done. | |
328 | */ | |
c5d736f5 | 329 | allow_barrier(conf, sector); |
1da177e4 LT |
330 | } |
331 | ||
332 | /* | |
333 | * Update disk head position estimator based on IRQ completion info. | |
334 | */ | |
9f2c9d12 | 335 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 336 | { |
e8096360 | 337 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
338 | |
339 | conf->mirrors[disk].head_position = | |
340 | r1_bio->sector + (r1_bio->sectors); | |
341 | } | |
342 | ||
ba3ae3be NK |
343 | /* |
344 | * Find the disk number which triggered given bio | |
345 | */ | |
9f2c9d12 | 346 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
347 | { |
348 | int mirror; | |
30194636 N |
349 | struct r1conf *conf = r1_bio->mddev->private; |
350 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 351 | |
8f19ccb2 | 352 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
353 | if (r1_bio->bios[mirror] == bio) |
354 | break; | |
355 | ||
8f19ccb2 | 356 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
357 | update_head_pos(mirror, r1_bio); |
358 | ||
359 | return mirror; | |
360 | } | |
361 | ||
4246a0b6 | 362 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 363 | { |
4e4cbee9 | 364 | int uptodate = !bio->bi_status; |
9f2c9d12 | 365 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 366 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 367 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 368 | |
1da177e4 LT |
369 | /* |
370 | * this branch is our 'one mirror IO has finished' event handler: | |
371 | */ | |
e5872d58 | 372 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 373 | |
dd00a99e N |
374 | if (uptodate) |
375 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
376 | else if (test_bit(FailFast, &rdev->flags) && |
377 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
378 | /* This was a fail-fast read so we definitely | |
379 | * want to retry */ | |
380 | ; | |
dd00a99e N |
381 | else { |
382 | /* If all other devices have failed, we want to return | |
383 | * the error upwards rather than fail the last device. | |
384 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 385 | */ |
dd00a99e N |
386 | unsigned long flags; |
387 | spin_lock_irqsave(&conf->device_lock, flags); | |
388 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
389 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 390 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
391 | uptodate = 1; |
392 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
393 | } | |
1da177e4 | 394 | |
7ad4d4a6 | 395 | if (uptodate) { |
1da177e4 | 396 | raid_end_bio_io(r1_bio); |
e5872d58 | 397 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 398 | } else { |
1da177e4 LT |
399 | /* |
400 | * oops, read error: | |
401 | */ | |
913cce5a | 402 | pr_err_ratelimited("md/raid1:%s: %pg: rescheduling sector %llu\n", |
1d41c216 | 403 | mdname(conf->mddev), |
913cce5a | 404 | rdev->bdev, |
1d41c216 | 405 | (unsigned long long)r1_bio->sector); |
d2eb35ac | 406 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 407 | reschedule_retry(r1_bio); |
7ad4d4a6 | 408 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 409 | } |
1da177e4 LT |
410 | } |
411 | ||
9f2c9d12 | 412 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
413 | { |
414 | /* it really is the end of this request */ | |
415 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
841c1316 ML |
416 | bio_free_pages(r1_bio->behind_master_bio); |
417 | bio_put(r1_bio->behind_master_bio); | |
418 | r1_bio->behind_master_bio = NULL; | |
cd5ff9a1 N |
419 | } |
420 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
421 | md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
422 | r1_bio->sectors, | |
423 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
424 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
cd5ff9a1 N |
425 | md_write_end(r1_bio->mddev); |
426 | } | |
427 | ||
9f2c9d12 | 428 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 429 | { |
cd5ff9a1 N |
430 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
431 | return; | |
432 | ||
433 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
434 | reschedule_retry(r1_bio); | |
435 | else { | |
436 | close_write(r1_bio); | |
4367af55 N |
437 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
438 | reschedule_retry(r1_bio); | |
439 | else | |
440 | raid_end_bio_io(r1_bio); | |
4e78064f N |
441 | } |
442 | } | |
443 | ||
4246a0b6 | 444 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 445 | { |
9f2c9d12 | 446 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 447 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 448 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 449 | struct bio *to_put = NULL; |
e5872d58 N |
450 | int mirror = find_bio_disk(r1_bio, bio); |
451 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd | 452 | bool discard_error; |
69df9cfc GJ |
453 | sector_t lo = r1_bio->sector; |
454 | sector_t hi = r1_bio->sector + r1_bio->sectors; | |
e3f948cd | 455 | |
4e4cbee9 | 456 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 457 | |
e9c7469b TH |
458 | /* |
459 | * 'one mirror IO has finished' event handler: | |
460 | */ | |
4e4cbee9 | 461 | if (bio->bi_status && !discard_error) { |
e5872d58 N |
462 | set_bit(WriteErrorSeen, &rdev->flags); |
463 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
464 | set_bit(MD_RECOVERY_NEEDED, & |
465 | conf->mddev->recovery); | |
466 | ||
212e7eb7 N |
467 | if (test_bit(FailFast, &rdev->flags) && |
468 | (bio->bi_opf & MD_FAILFAST) && | |
469 | /* We never try FailFast to WriteMostly devices */ | |
470 | !test_bit(WriteMostly, &rdev->flags)) { | |
471 | md_error(r1_bio->mddev, rdev); | |
eeba6809 YY |
472 | } |
473 | ||
474 | /* | |
475 | * When the device is faulty, it is not necessary to | |
476 | * handle write error. | |
eeba6809 YY |
477 | */ |
478 | if (!test_bit(Faulty, &rdev->flags)) | |
212e7eb7 | 479 | set_bit(R1BIO_WriteError, &r1_bio->state); |
eeba6809 | 480 | else { |
2417b986 PC |
481 | /* Fail the request */ |
482 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
eeba6809 YY |
483 | /* Finished with this branch */ |
484 | r1_bio->bios[mirror] = NULL; | |
485 | to_put = bio; | |
486 | } | |
4367af55 | 487 | } else { |
1da177e4 | 488 | /* |
e9c7469b TH |
489 | * Set R1BIO_Uptodate in our master bio, so that we |
490 | * will return a good error code for to the higher | |
491 | * levels even if IO on some other mirrored buffer | |
492 | * fails. | |
493 | * | |
494 | * The 'master' represents the composite IO operation | |
495 | * to user-side. So if something waits for IO, then it | |
496 | * will wait for the 'master' bio. | |
1da177e4 | 497 | */ |
cd5ff9a1 N |
498 | r1_bio->bios[mirror] = NULL; |
499 | to_put = bio; | |
3056e3ae AL |
500 | /* |
501 | * Do not set R1BIO_Uptodate if the current device is | |
502 | * rebuilding or Faulty. This is because we cannot use | |
503 | * such device for properly reading the data back (we could | |
504 | * potentially use it, if the current write would have felt | |
505 | * before rdev->recovery_offset, but for simplicity we don't | |
506 | * check this here. | |
507 | */ | |
e5872d58 N |
508 | if (test_bit(In_sync, &rdev->flags) && |
509 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 510 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 511 | |
4367af55 | 512 | /* Maybe we can clear some bad blocks. */ |
3a0f007b YK |
513 | if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) && |
514 | !discard_error) { | |
4367af55 N |
515 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
516 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
517 | } | |
518 | } | |
519 | ||
e9c7469b | 520 | if (behind) { |
69df9cfc | 521 | if (test_bit(CollisionCheck, &rdev->flags)) |
404659cf | 522 | remove_serial(rdev, lo, hi); |
e5872d58 | 523 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
524 | atomic_dec(&r1_bio->behind_remaining); |
525 | ||
526 | /* | |
527 | * In behind mode, we ACK the master bio once the I/O | |
528 | * has safely reached all non-writemostly | |
529 | * disks. Setting the Returned bit ensures that this | |
530 | * gets done only once -- we don't ever want to return | |
531 | * -EIO here, instead we'll wait | |
532 | */ | |
533 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
534 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
535 | /* Maybe we can return now */ | |
536 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
537 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
538 | pr_debug("raid1: behind end write sectors" |
539 | " %llu-%llu\n", | |
4f024f37 KO |
540 | (unsigned long long) mbio->bi_iter.bi_sector, |
541 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 542 | call_bio_endio(r1_bio); |
4b6d287f N |
543 | } |
544 | } | |
69df9cfc GJ |
545 | } else if (rdev->mddev->serialize_policy) |
546 | remove_serial(rdev, lo, hi); | |
4367af55 | 547 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 548 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 549 | |
1da177e4 | 550 | /* |
1da177e4 LT |
551 | * Let's see if all mirrored write operations have finished |
552 | * already. | |
553 | */ | |
af6d7b76 | 554 | r1_bio_write_done(r1_bio); |
c70810b3 | 555 | |
04b857f7 N |
556 | if (to_put) |
557 | bio_put(to_put); | |
1da177e4 LT |
558 | } |
559 | ||
fd76863e | 560 | static sector_t align_to_barrier_unit_end(sector_t start_sector, |
561 | sector_t sectors) | |
562 | { | |
563 | sector_t len; | |
564 | ||
565 | WARN_ON(sectors == 0); | |
566 | /* | |
567 | * len is the number of sectors from start_sector to end of the | |
568 | * barrier unit which start_sector belongs to. | |
569 | */ | |
570 | len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - | |
571 | start_sector; | |
572 | ||
573 | if (len > sectors) | |
574 | len = sectors; | |
575 | ||
576 | return len; | |
577 | } | |
578 | ||
31a73331 YK |
579 | static void update_read_sectors(struct r1conf *conf, int disk, |
580 | sector_t this_sector, int len) | |
581 | { | |
582 | struct raid1_info *info = &conf->mirrors[disk]; | |
583 | ||
584 | atomic_inc(&info->rdev->nr_pending); | |
585 | if (info->next_seq_sect != this_sector) | |
586 | info->seq_start = this_sector; | |
587 | info->next_seq_sect = this_sector + len; | |
588 | } | |
589 | ||
590 | static int choose_first_rdev(struct r1conf *conf, struct r1bio *r1_bio, | |
591 | int *max_sectors) | |
592 | { | |
593 | sector_t this_sector = r1_bio->sector; | |
594 | int len = r1_bio->sectors; | |
595 | int disk; | |
596 | ||
597 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { | |
598 | struct md_rdev *rdev; | |
599 | int read_len; | |
600 | ||
601 | if (r1_bio->bios[disk] == IO_BLOCKED) | |
602 | continue; | |
603 | ||
604 | rdev = conf->mirrors[disk].rdev; | |
605 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
606 | continue; | |
607 | ||
608 | /* choose the first disk even if it has some bad blocks. */ | |
609 | read_len = raid1_check_read_range(rdev, this_sector, &len); | |
610 | if (read_len > 0) { | |
611 | update_read_sectors(conf, disk, this_sector, read_len); | |
612 | *max_sectors = read_len; | |
613 | return disk; | |
614 | } | |
615 | } | |
616 | ||
617 | return -1; | |
618 | } | |
619 | ||
9f3ced79 YK |
620 | static int choose_bb_rdev(struct r1conf *conf, struct r1bio *r1_bio, |
621 | int *max_sectors) | |
622 | { | |
623 | sector_t this_sector = r1_bio->sector; | |
624 | int best_disk = -1; | |
625 | int best_len = 0; | |
626 | int disk; | |
627 | ||
628 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { | |
629 | struct md_rdev *rdev; | |
630 | int len; | |
631 | int read_len; | |
632 | ||
633 | if (r1_bio->bios[disk] == IO_BLOCKED) | |
634 | continue; | |
635 | ||
636 | rdev = conf->mirrors[disk].rdev; | |
637 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
638 | test_bit(WriteMostly, &rdev->flags)) | |
639 | continue; | |
640 | ||
641 | /* keep track of the disk with the most readable sectors. */ | |
642 | len = r1_bio->sectors; | |
643 | read_len = raid1_check_read_range(rdev, this_sector, &len); | |
644 | if (read_len > best_len) { | |
645 | best_disk = disk; | |
646 | best_len = read_len; | |
647 | } | |
648 | } | |
649 | ||
650 | if (best_disk != -1) { | |
651 | *max_sectors = best_len; | |
652 | update_read_sectors(conf, best_disk, this_sector, best_len); | |
653 | } | |
654 | ||
655 | return best_disk; | |
656 | } | |
657 | ||
dfa8ecd1 YK |
658 | static int choose_slow_rdev(struct r1conf *conf, struct r1bio *r1_bio, |
659 | int *max_sectors) | |
660 | { | |
661 | sector_t this_sector = r1_bio->sector; | |
662 | int bb_disk = -1; | |
663 | int bb_read_len = 0; | |
664 | int disk; | |
665 | ||
666 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { | |
667 | struct md_rdev *rdev; | |
668 | int len; | |
669 | int read_len; | |
670 | ||
671 | if (r1_bio->bios[disk] == IO_BLOCKED) | |
672 | continue; | |
673 | ||
674 | rdev = conf->mirrors[disk].rdev; | |
675 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
676 | !test_bit(WriteMostly, &rdev->flags)) | |
677 | continue; | |
678 | ||
679 | /* there are no bad blocks, we can use this disk */ | |
680 | len = r1_bio->sectors; | |
681 | read_len = raid1_check_read_range(rdev, this_sector, &len); | |
682 | if (read_len == r1_bio->sectors) { | |
683 | update_read_sectors(conf, disk, this_sector, read_len); | |
684 | return disk; | |
685 | } | |
686 | ||
687 | /* | |
688 | * there are partial bad blocks, choose the rdev with largest | |
689 | * read length. | |
690 | */ | |
691 | if (read_len > bb_read_len) { | |
692 | bb_disk = disk; | |
693 | bb_read_len = read_len; | |
694 | } | |
695 | } | |
696 | ||
697 | if (bb_disk != -1) { | |
698 | *max_sectors = bb_read_len; | |
699 | update_read_sectors(conf, bb_disk, this_sector, bb_read_len); | |
700 | } | |
701 | ||
702 | return bb_disk; | |
703 | } | |
704 | ||
ba58f57f YK |
705 | static bool is_sequential(struct r1conf *conf, int disk, struct r1bio *r1_bio) |
706 | { | |
707 | /* TODO: address issues with this check and concurrency. */ | |
708 | return conf->mirrors[disk].next_seq_sect == r1_bio->sector || | |
709 | conf->mirrors[disk].head_position == r1_bio->sector; | |
710 | } | |
711 | ||
712 | /* | |
713 | * If buffered sequential IO size exceeds optimal iosize, check if there is idle | |
714 | * disk. If yes, choose the idle disk. | |
715 | */ | |
716 | static bool should_choose_next(struct r1conf *conf, int disk) | |
717 | { | |
718 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
719 | int opt_iosize; | |
720 | ||
721 | if (!test_bit(Nonrot, &mirror->rdev->flags)) | |
722 | return false; | |
723 | ||
724 | opt_iosize = bdev_io_opt(mirror->rdev->bdev) >> 9; | |
725 | return opt_iosize > 0 && mirror->seq_start != MaxSector && | |
726 | mirror->next_seq_sect > opt_iosize && | |
727 | mirror->next_seq_sect - opt_iosize >= mirror->seq_start; | |
728 | } | |
729 | ||
0091c5a2 | 730 | static bool rdev_readable(struct md_rdev *rdev, struct r1bio *r1_bio) |
1da177e4 | 731 | { |
0091c5a2 YK |
732 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
733 | return false; | |
d2eb35ac | 734 | |
0091c5a2 YK |
735 | /* still in recovery */ |
736 | if (!test_bit(In_sync, &rdev->flags) && | |
737 | rdev->recovery_offset < r1_bio->sector + r1_bio->sectors) | |
738 | return false; | |
739 | ||
740 | /* don't read from slow disk unless have to */ | |
741 | if (test_bit(WriteMostly, &rdev->flags)) | |
742 | return false; | |
743 | ||
744 | /* don't split IO for bad blocks unless have to */ | |
745 | if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors)) | |
746 | return false; | |
747 | ||
748 | return true; | |
749 | } | |
750 | ||
751 | struct read_balance_ctl { | |
752 | sector_t closest_dist; | |
753 | int closest_dist_disk; | |
754 | int min_pending; | |
755 | int min_pending_disk; | |
756 | int sequential_disk; | |
757 | int readable_disks; | |
758 | }; | |
759 | ||
760 | static int choose_best_rdev(struct r1conf *conf, struct r1bio *r1_bio) | |
761 | { | |
762 | int disk; | |
763 | struct read_balance_ctl ctl = { | |
764 | .closest_dist_disk = -1, | |
765 | .closest_dist = MaxSector, | |
766 | .min_pending_disk = -1, | |
767 | .min_pending = UINT_MAX, | |
768 | .sequential_disk = -1, | |
769 | }; | |
31a73331 | 770 | |
be4d3280 | 771 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
0091c5a2 | 772 | struct md_rdev *rdev; |
76073054 | 773 | sector_t dist; |
9dedf603 | 774 | unsigned int pending; |
d2eb35ac | 775 | |
0091c5a2 | 776 | if (r1_bio->bios[disk] == IO_BLOCKED) |
76073054 | 777 | continue; |
0091c5a2 YK |
778 | |
779 | rdev = conf->mirrors[disk].rdev; | |
780 | if (!rdev_readable(rdev, r1_bio)) | |
d2eb35ac | 781 | continue; |
d2eb35ac | 782 | |
0091c5a2 YK |
783 | /* At least two disks to choose from so failfast is OK */ |
784 | if (ctl.readable_disks++ == 1) | |
2e52d449 N |
785 | set_bit(R1BIO_FailFast, &r1_bio->state); |
786 | ||
9dedf603 | 787 | pending = atomic_read(&rdev->nr_pending); |
0091c5a2 YK |
788 | dist = abs(r1_bio->sector - conf->mirrors[disk].head_position); |
789 | ||
12cee5a8 | 790 | /* Don't change to another disk for sequential reads */ |
ba58f57f | 791 | if (is_sequential(conf, disk, r1_bio)) { |
0091c5a2 YK |
792 | if (!should_choose_next(conf, disk)) |
793 | return disk; | |
794 | ||
ba58f57f YK |
795 | /* |
796 | * Add 'pending' to avoid choosing this disk if | |
797 | * there is other idle disk. | |
798 | */ | |
799 | pending++; | |
800 | /* | |
801 | * If there is no other idle disk, this disk | |
802 | * will be chosen. | |
803 | */ | |
0091c5a2 | 804 | ctl.sequential_disk = disk; |
12cee5a8 | 805 | } |
12cee5a8 | 806 | |
0091c5a2 YK |
807 | if (ctl.min_pending > pending) { |
808 | ctl.min_pending = pending; | |
809 | ctl.min_pending_disk = disk; | |
9dedf603 SL |
810 | } |
811 | ||
0091c5a2 YK |
812 | if (ctl.closest_dist > dist) { |
813 | ctl.closest_dist = dist; | |
814 | ctl.closest_dist_disk = disk; | |
1da177e4 | 815 | } |
f3ac8bf7 | 816 | } |
1da177e4 | 817 | |
257ac239 YK |
818 | /* |
819 | * sequential IO size exceeds optimal iosize, however, there is no other | |
820 | * idle disk, so choose the sequential disk. | |
821 | */ | |
0091c5a2 YK |
822 | if (ctl.sequential_disk != -1 && ctl.min_pending != 0) |
823 | return ctl.sequential_disk; | |
257ac239 | 824 | |
9dedf603 SL |
825 | /* |
826 | * If all disks are rotational, choose the closest disk. If any disk is | |
827 | * non-rotational, choose the disk with less pending request even the | |
828 | * disk is rotational, which might/might not be optimal for raids with | |
829 | * mixed ratation/non-rotational disks depending on workload. | |
830 | */ | |
0091c5a2 YK |
831 | if (ctl.min_pending_disk != -1 && |
832 | (READ_ONCE(conf->nonrot_disks) || ctl.min_pending == 0)) | |
833 | return ctl.min_pending_disk; | |
834 | else | |
835 | return ctl.closest_dist_disk; | |
836 | } | |
9dedf603 | 837 | |
0091c5a2 YK |
838 | /* |
839 | * This routine returns the disk from which the requested read should be done. | |
840 | * | |
841 | * 1) If resync is in progress, find the first usable disk and use it even if it | |
842 | * has some bad blocks. | |
843 | * | |
844 | * 2) Now that there is no resync, loop through all disks and skipping slow | |
845 | * disks and disks with bad blocks for now. Only pay attention to key disk | |
846 | * choice. | |
847 | * | |
848 | * 3) If we've made it this far, now look for disks with bad blocks and choose | |
849 | * the one with most number of sectors. | |
850 | * | |
851 | * 4) If we are all the way at the end, we have no choice but to use a disk even | |
852 | * if it is write mostly. | |
853 | * | |
854 | * The rdev for the device selected will have nr_pending incremented. | |
855 | */ | |
856 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, | |
857 | int *max_sectors) | |
858 | { | |
859 | int disk; | |
12cee5a8 | 860 | |
0091c5a2 | 861 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
1da177e4 | 862 | |
0091c5a2 YK |
863 | if (raid1_should_read_first(conf->mddev, r1_bio->sector, |
864 | r1_bio->sectors)) | |
865 | return choose_first_rdev(conf, r1_bio, max_sectors); | |
866 | ||
867 | disk = choose_best_rdev(conf, r1_bio); | |
868 | if (disk >= 0) { | |
869 | *max_sectors = r1_bio->sectors; | |
870 | update_read_sectors(conf, disk, r1_bio->sector, | |
871 | r1_bio->sectors); | |
872 | return disk; | |
873 | } | |
dfa8ecd1 | 874 | |
9f3ced79 YK |
875 | /* |
876 | * If we are here it means we didn't find a perfectly good disk so | |
877 | * now spend a bit more time trying to find one with the most good | |
878 | * sectors. | |
879 | */ | |
880 | disk = choose_bb_rdev(conf, r1_bio, max_sectors); | |
881 | if (disk >= 0) | |
882 | return disk; | |
883 | ||
dfa8ecd1 | 884 | return choose_slow_rdev(conf, r1_bio, max_sectors); |
1da177e4 LT |
885 | } |
886 | ||
21bd9a68 JW |
887 | static void wake_up_barrier(struct r1conf *conf) |
888 | { | |
889 | if (wq_has_sleeper(&conf->wait_barrier)) | |
890 | wake_up(&conf->wait_barrier); | |
891 | } | |
892 | ||
673ca68d N |
893 | static void flush_bio_list(struct r1conf *conf, struct bio *bio) |
894 | { | |
895 | /* flush any pending bitmap writes to disk before proceeding w/ I/O */ | |
9efcc2c3 | 896 | raid1_prepare_flush_writes(conf->mddev->bitmap); |
21bd9a68 | 897 | wake_up_barrier(conf); |
673ca68d N |
898 | |
899 | while (bio) { /* submit pending writes */ | |
900 | struct bio *next = bio->bi_next; | |
8295efbe YK |
901 | |
902 | raid1_submit_write(bio); | |
673ca68d | 903 | bio = next; |
5fa4f8ba | 904 | cond_resched(); |
673ca68d N |
905 | } |
906 | } | |
907 | ||
e8096360 | 908 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
909 | { |
910 | /* Any writes that have been queued but are awaiting | |
911 | * bitmap updates get flushed here. | |
a35e63ef | 912 | */ |
a35e63ef N |
913 | spin_lock_irq(&conf->device_lock); |
914 | ||
915 | if (conf->pending_bio_list.head) { | |
18022a1b | 916 | struct blk_plug plug; |
a35e63ef | 917 | struct bio *bio; |
18022a1b | 918 | |
a35e63ef | 919 | bio = bio_list_get(&conf->pending_bio_list); |
a35e63ef | 920 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
921 | |
922 | /* | |
923 | * As this is called in a wait_event() loop (see freeze_array), | |
924 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
925 | * cause a warning when we prepare to wait again. As it is | |
926 | * rare that this path is taken, it is perfectly safe to force | |
927 | * us to go around the wait_event() loop again, so the warning | |
928 | * is a false-positive. Silence the warning by resetting | |
929 | * thread state | |
930 | */ | |
931 | __set_current_state(TASK_RUNNING); | |
18022a1b | 932 | blk_start_plug(&plug); |
673ca68d | 933 | flush_bio_list(conf, bio); |
18022a1b | 934 | blk_finish_plug(&plug); |
a35e63ef N |
935 | } else |
936 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
937 | } |
938 | ||
17999be4 N |
939 | /* Barriers.... |
940 | * Sometimes we need to suspend IO while we do something else, | |
941 | * either some resync/recovery, or reconfigure the array. | |
942 | * To do this we raise a 'barrier'. | |
943 | * The 'barrier' is a counter that can be raised multiple times | |
944 | * to count how many activities are happening which preclude | |
945 | * normal IO. | |
946 | * We can only raise the barrier if there is no pending IO. | |
947 | * i.e. if nr_pending == 0. | |
948 | * We choose only to raise the barrier if no-one is waiting for the | |
949 | * barrier to go down. This means that as soon as an IO request | |
950 | * is ready, no other operations which require a barrier will start | |
951 | * until the IO request has had a chance. | |
952 | * | |
953 | * So: regular IO calls 'wait_barrier'. When that returns there | |
954 | * is no backgroup IO happening, It must arrange to call | |
955 | * allow_barrier when it has finished its IO. | |
956 | * backgroup IO calls must call raise_barrier. Once that returns | |
957 | * there is no normal IO happeing. It must arrange to call | |
958 | * lower_barrier when the particular background IO completes. | |
4675719d HT |
959 | * |
960 | * If resync/recovery is interrupted, returns -EINTR; | |
961 | * Otherwise, returns 0. | |
1da177e4 | 962 | */ |
4675719d | 963 | static int raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 | 964 | { |
fd76863e | 965 | int idx = sector_to_idx(sector_nr); |
966 | ||
1da177e4 | 967 | spin_lock_irq(&conf->resync_lock); |
17999be4 N |
968 | |
969 | /* Wait until no block IO is waiting */ | |
824e47da | 970 | wait_event_lock_irq(conf->wait_barrier, |
971 | !atomic_read(&conf->nr_waiting[idx]), | |
eed8c02e | 972 | conf->resync_lock); |
17999be4 N |
973 | |
974 | /* block any new IO from starting */ | |
824e47da | 975 | atomic_inc(&conf->barrier[idx]); |
976 | /* | |
977 | * In raise_barrier() we firstly increase conf->barrier[idx] then | |
978 | * check conf->nr_pending[idx]. In _wait_barrier() we firstly | |
979 | * increase conf->nr_pending[idx] then check conf->barrier[idx]. | |
980 | * A memory barrier here to make sure conf->nr_pending[idx] won't | |
981 | * be fetched before conf->barrier[idx] is increased. Otherwise | |
982 | * there will be a race between raise_barrier() and _wait_barrier(). | |
983 | */ | |
984 | smp_mb__after_atomic(); | |
17999be4 | 985 | |
79ef3a8a | 986 | /* For these conditions we must wait: |
987 | * A: while the array is in frozen state | |
fd76863e | 988 | * B: while conf->nr_pending[idx] is not 0, meaning regular I/O |
989 | * existing in corresponding I/O barrier bucket. | |
990 | * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches | |
991 | * max resync count which allowed on current I/O barrier bucket. | |
79ef3a8a | 992 | */ |
17999be4 | 993 | wait_event_lock_irq(conf->wait_barrier, |
8c242593 | 994 | (!conf->array_frozen && |
824e47da | 995 | !atomic_read(&conf->nr_pending[idx]) && |
8c242593 YY |
996 | atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || |
997 | test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), | |
eed8c02e | 998 | conf->resync_lock); |
17999be4 | 999 | |
8c242593 YY |
1000 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
1001 | atomic_dec(&conf->barrier[idx]); | |
1002 | spin_unlock_irq(&conf->resync_lock); | |
1003 | wake_up(&conf->wait_barrier); | |
1004 | return -EINTR; | |
1005 | } | |
1006 | ||
43ac9b84 | 1007 | atomic_inc(&conf->nr_sync_pending); |
17999be4 | 1008 | spin_unlock_irq(&conf->resync_lock); |
8c242593 YY |
1009 | |
1010 | return 0; | |
17999be4 N |
1011 | } |
1012 | ||
fd76863e | 1013 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 1014 | { |
fd76863e | 1015 | int idx = sector_to_idx(sector_nr); |
1016 | ||
824e47da | 1017 | BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); |
fd76863e | 1018 | |
824e47da | 1019 | atomic_dec(&conf->barrier[idx]); |
43ac9b84 | 1020 | atomic_dec(&conf->nr_sync_pending); |
17999be4 N |
1021 | wake_up(&conf->wait_barrier); |
1022 | } | |
1023 | ||
5aa70503 | 1024 | static bool _wait_barrier(struct r1conf *conf, int idx, bool nowait) |
17999be4 | 1025 | { |
5aa70503 VV |
1026 | bool ret = true; |
1027 | ||
824e47da | 1028 | /* |
1029 | * We need to increase conf->nr_pending[idx] very early here, | |
1030 | * then raise_barrier() can be blocked when it waits for | |
1031 | * conf->nr_pending[idx] to be 0. Then we can avoid holding | |
1032 | * conf->resync_lock when there is no barrier raised in same | |
1033 | * barrier unit bucket. Also if the array is frozen, I/O | |
1034 | * should be blocked until array is unfrozen. | |
1035 | */ | |
1036 | atomic_inc(&conf->nr_pending[idx]); | |
1037 | /* | |
1038 | * In _wait_barrier() we firstly increase conf->nr_pending[idx], then | |
1039 | * check conf->barrier[idx]. In raise_barrier() we firstly increase | |
1040 | * conf->barrier[idx], then check conf->nr_pending[idx]. A memory | |
1041 | * barrier is necessary here to make sure conf->barrier[idx] won't be | |
1042 | * fetched before conf->nr_pending[idx] is increased. Otherwise there | |
1043 | * will be a race between _wait_barrier() and raise_barrier(). | |
1044 | */ | |
1045 | smp_mb__after_atomic(); | |
79ef3a8a | 1046 | |
824e47da | 1047 | /* |
1048 | * Don't worry about checking two atomic_t variables at same time | |
1049 | * here. If during we check conf->barrier[idx], the array is | |
1050 | * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is | |
1051 | * 0, it is safe to return and make the I/O continue. Because the | |
1052 | * array is frozen, all I/O returned here will eventually complete | |
1053 | * or be queued, no race will happen. See code comment in | |
1054 | * frozen_array(). | |
1055 | */ | |
1056 | if (!READ_ONCE(conf->array_frozen) && | |
1057 | !atomic_read(&conf->barrier[idx])) | |
5aa70503 | 1058 | return ret; |
79ef3a8a | 1059 | |
824e47da | 1060 | /* |
1061 | * After holding conf->resync_lock, conf->nr_pending[idx] | |
1062 | * should be decreased before waiting for barrier to drop. | |
1063 | * Otherwise, we may encounter a race condition because | |
1064 | * raise_barrer() might be waiting for conf->nr_pending[idx] | |
1065 | * to be 0 at same time. | |
1066 | */ | |
1067 | spin_lock_irq(&conf->resync_lock); | |
1068 | atomic_inc(&conf->nr_waiting[idx]); | |
1069 | atomic_dec(&conf->nr_pending[idx]); | |
1070 | /* | |
1071 | * In case freeze_array() is waiting for | |
1072 | * get_unqueued_pending() == extra | |
1073 | */ | |
21bd9a68 | 1074 | wake_up_barrier(conf); |
824e47da | 1075 | /* Wait for the barrier in same barrier unit bucket to drop. */ |
5aa70503 VV |
1076 | |
1077 | /* Return false when nowait flag is set */ | |
1078 | if (nowait) { | |
1079 | ret = false; | |
1080 | } else { | |
1081 | wait_event_lock_irq(conf->wait_barrier, | |
1082 | !conf->array_frozen && | |
1083 | !atomic_read(&conf->barrier[idx]), | |
1084 | conf->resync_lock); | |
1085 | atomic_inc(&conf->nr_pending[idx]); | |
1086 | } | |
1087 | ||
824e47da | 1088 | atomic_dec(&conf->nr_waiting[idx]); |
fd76863e | 1089 | spin_unlock_irq(&conf->resync_lock); |
5aa70503 | 1090 | return ret; |
79ef3a8a | 1091 | } |
1092 | ||
5aa70503 | 1093 | static bool wait_read_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait) |
79ef3a8a | 1094 | { |
fd76863e | 1095 | int idx = sector_to_idx(sector_nr); |
5aa70503 | 1096 | bool ret = true; |
79ef3a8a | 1097 | |
824e47da | 1098 | /* |
1099 | * Very similar to _wait_barrier(). The difference is, for read | |
1100 | * I/O we don't need wait for sync I/O, but if the whole array | |
1101 | * is frozen, the read I/O still has to wait until the array is | |
1102 | * unfrozen. Since there is no ordering requirement with | |
1103 | * conf->barrier[idx] here, memory barrier is unnecessary as well. | |
1104 | */ | |
1105 | atomic_inc(&conf->nr_pending[idx]); | |
79ef3a8a | 1106 | |
824e47da | 1107 | if (!READ_ONCE(conf->array_frozen)) |
5aa70503 | 1108 | return ret; |
824e47da | 1109 | |
1110 | spin_lock_irq(&conf->resync_lock); | |
1111 | atomic_inc(&conf->nr_waiting[idx]); | |
1112 | atomic_dec(&conf->nr_pending[idx]); | |
1113 | /* | |
1114 | * In case freeze_array() is waiting for | |
1115 | * get_unqueued_pending() == extra | |
1116 | */ | |
21bd9a68 | 1117 | wake_up_barrier(conf); |
824e47da | 1118 | /* Wait for array to be unfrozen */ |
5aa70503 VV |
1119 | |
1120 | /* Return false when nowait flag is set */ | |
1121 | if (nowait) { | |
1122 | /* Return false when nowait flag is set */ | |
1123 | ret = false; | |
1124 | } else { | |
1125 | wait_event_lock_irq(conf->wait_barrier, | |
1126 | !conf->array_frozen, | |
1127 | conf->resync_lock); | |
1128 | atomic_inc(&conf->nr_pending[idx]); | |
1129 | } | |
1130 | ||
824e47da | 1131 | atomic_dec(&conf->nr_waiting[idx]); |
1da177e4 | 1132 | spin_unlock_irq(&conf->resync_lock); |
5aa70503 | 1133 | return ret; |
1da177e4 LT |
1134 | } |
1135 | ||
5aa70503 | 1136 | static bool wait_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait) |
17999be4 | 1137 | { |
fd76863e | 1138 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 1139 | |
5aa70503 | 1140 | return _wait_barrier(conf, idx, nowait); |
fd76863e | 1141 | } |
1142 | ||
fd76863e | 1143 | static void _allow_barrier(struct r1conf *conf, int idx) |
17999be4 | 1144 | { |
824e47da | 1145 | atomic_dec(&conf->nr_pending[idx]); |
21bd9a68 | 1146 | wake_up_barrier(conf); |
17999be4 N |
1147 | } |
1148 | ||
fd76863e | 1149 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr) |
1150 | { | |
1151 | int idx = sector_to_idx(sector_nr); | |
1152 | ||
1153 | _allow_barrier(conf, idx); | |
1154 | } | |
1155 | ||
fd76863e | 1156 | /* conf->resync_lock should be held */ |
1157 | static int get_unqueued_pending(struct r1conf *conf) | |
1158 | { | |
1159 | int idx, ret; | |
1160 | ||
43ac9b84 XN |
1161 | ret = atomic_read(&conf->nr_sync_pending); |
1162 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
824e47da | 1163 | ret += atomic_read(&conf->nr_pending[idx]) - |
1164 | atomic_read(&conf->nr_queued[idx]); | |
fd76863e | 1165 | |
1166 | return ret; | |
1167 | } | |
1168 | ||
e2d59925 | 1169 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab | 1170 | { |
fd76863e | 1171 | /* Stop sync I/O and normal I/O and wait for everything to |
11353b9d | 1172 | * go quiet. |
fd76863e | 1173 | * This is called in two situations: |
1174 | * 1) management command handlers (reshape, remove disk, quiesce). | |
1175 | * 2) one normal I/O request failed. | |
1176 | ||
1177 | * After array_frozen is set to 1, new sync IO will be blocked at | |
1178 | * raise_barrier(), and new normal I/O will blocked at _wait_barrier() | |
1179 | * or wait_read_barrier(). The flying I/Os will either complete or be | |
1180 | * queued. When everything goes quite, there are only queued I/Os left. | |
1181 | ||
1182 | * Every flying I/O contributes to a conf->nr_pending[idx], idx is the | |
1183 | * barrier bucket index which this I/O request hits. When all sync and | |
1184 | * normal I/O are queued, sum of all conf->nr_pending[] will match sum | |
1185 | * of all conf->nr_queued[]. But normal I/O failure is an exception, | |
1186 | * in handle_read_error(), we may call freeze_array() before trying to | |
1187 | * fix the read error. In this case, the error read I/O is not queued, | |
1188 | * so get_unqueued_pending() == 1. | |
1189 | * | |
1190 | * Therefore before this function returns, we need to wait until | |
1191 | * get_unqueued_pendings(conf) gets equal to extra. For | |
1192 | * normal I/O context, extra is 1, in rested situations extra is 0. | |
ddaf22ab N |
1193 | */ |
1194 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1195 | conf->array_frozen = 1; |
28be4fd3 | 1196 | mddev_add_trace_msg(conf->mddev, "raid1 wait freeze"); |
fd76863e | 1197 | wait_event_lock_irq_cmd( |
1198 | conf->wait_barrier, | |
1199 | get_unqueued_pending(conf) == extra, | |
1200 | conf->resync_lock, | |
1201 | flush_pending_writes(conf)); | |
ddaf22ab N |
1202 | spin_unlock_irq(&conf->resync_lock); |
1203 | } | |
e8096360 | 1204 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
1205 | { |
1206 | /* reverse the effect of the freeze */ | |
1207 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1208 | conf->array_frozen = 0; |
ddaf22ab | 1209 | spin_unlock_irq(&conf->resync_lock); |
824e47da | 1210 | wake_up(&conf->wait_barrier); |
ddaf22ab N |
1211 | } |
1212 | ||
16d56e2f | 1213 | static void alloc_behind_master_bio(struct r1bio *r1_bio, |
cb83efcf | 1214 | struct bio *bio) |
4b6d287f | 1215 | { |
cb83efcf | 1216 | int size = bio->bi_iter.bi_size; |
841c1316 ML |
1217 | unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1218 | int i = 0; | |
1219 | struct bio *behind_bio = NULL; | |
1220 | ||
609be106 CH |
1221 | behind_bio = bio_alloc_bioset(NULL, vcnt, 0, GFP_NOIO, |
1222 | &r1_bio->mddev->bio_set); | |
4b6d287f | 1223 | |
41743c1f | 1224 | /* discard op, we don't support writezero/writesame yet */ |
16d56e2f SL |
1225 | if (!bio_has_data(bio)) { |
1226 | behind_bio->bi_iter.bi_size = size; | |
41743c1f | 1227 | goto skip_copy; |
16d56e2f | 1228 | } |
41743c1f | 1229 | |
841c1316 ML |
1230 | while (i < vcnt && size) { |
1231 | struct page *page; | |
1232 | int len = min_t(int, PAGE_SIZE, size); | |
1233 | ||
1234 | page = alloc_page(GFP_NOIO); | |
1235 | if (unlikely(!page)) | |
1236 | goto free_pages; | |
1237 | ||
b42473cd JT |
1238 | if (!bio_add_page(behind_bio, page, len, 0)) { |
1239 | put_page(page); | |
1240 | goto free_pages; | |
1241 | } | |
841c1316 ML |
1242 | |
1243 | size -= len; | |
1244 | i++; | |
4b6d287f | 1245 | } |
841c1316 | 1246 | |
cb83efcf | 1247 | bio_copy_data(behind_bio, bio); |
41743c1f | 1248 | skip_copy: |
56a64c17 | 1249 | r1_bio->behind_master_bio = behind_bio; |
af6d7b76 | 1250 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
4b6d287f | 1251 | |
16d56e2f | 1252 | return; |
841c1316 ML |
1253 | |
1254 | free_pages: | |
4f024f37 KO |
1255 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1256 | bio->bi_iter.bi_size); | |
841c1316 | 1257 | bio_free_pages(behind_bio); |
16d56e2f | 1258 | bio_put(behind_bio); |
4b6d287f N |
1259 | } |
1260 | ||
f54a9d0e N |
1261 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) |
1262 | { | |
1263 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1264 | cb); | |
1265 | struct mddev *mddev = plug->cb.data; | |
1266 | struct r1conf *conf = mddev->private; | |
1267 | struct bio *bio; | |
1268 | ||
9efcc2c3 | 1269 | if (from_schedule) { |
f54a9d0e N |
1270 | spin_lock_irq(&conf->device_lock); |
1271 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
f54a9d0e | 1272 | spin_unlock_irq(&conf->device_lock); |
21bd9a68 | 1273 | wake_up_barrier(conf); |
f54a9d0e N |
1274 | md_wakeup_thread(mddev->thread); |
1275 | kfree(plug); | |
1276 | return; | |
1277 | } | |
1278 | ||
1279 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1280 | bio = bio_list_get(&plug->pending); | |
673ca68d | 1281 | flush_bio_list(conf, bio); |
f54a9d0e N |
1282 | kfree(plug); |
1283 | } | |
1284 | ||
689389a0 N |
1285 | static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio) |
1286 | { | |
1287 | r1_bio->master_bio = bio; | |
1288 | r1_bio->sectors = bio_sectors(bio); | |
1289 | r1_bio->state = 0; | |
1290 | r1_bio->mddev = mddev; | |
1291 | r1_bio->sector = bio->bi_iter.bi_sector; | |
1292 | } | |
1293 | ||
fd76863e | 1294 | static inline struct r1bio * |
689389a0 | 1295 | alloc_r1bio(struct mddev *mddev, struct bio *bio) |
fd76863e | 1296 | { |
1297 | struct r1conf *conf = mddev->private; | |
1298 | struct r1bio *r1_bio; | |
1299 | ||
afeee514 | 1300 | r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); |
689389a0 N |
1301 | /* Ensure no bio records IO_BLOCKED */ |
1302 | memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); | |
1303 | init_r1bio(r1_bio, mddev, bio); | |
fd76863e | 1304 | return r1_bio; |
1305 | } | |
1306 | ||
c230e7e5 | 1307 | static void raid1_read_request(struct mddev *mddev, struct bio *bio, |
689389a0 | 1308 | int max_read_sectors, struct r1bio *r1_bio) |
1da177e4 | 1309 | { |
e8096360 | 1310 | struct r1conf *conf = mddev->private; |
0eaf822c | 1311 | struct raid1_info *mirror; |
1da177e4 | 1312 | struct bio *read_bio; |
3b046a97 | 1313 | struct bitmap *bitmap = mddev->bitmap; |
3c5e514d BVA |
1314 | const enum req_op op = bio_op(bio); |
1315 | const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; | |
3b046a97 RL |
1316 | int max_sectors; |
1317 | int rdisk; | |
9b8ae7b9 | 1318 | bool r1bio_existed = !!r1_bio; |
689389a0 | 1319 | char b[BDEVNAME_SIZE]; |
3b046a97 | 1320 | |
fd76863e | 1321 | /* |
689389a0 N |
1322 | * If r1_bio is set, we are blocking the raid1d thread |
1323 | * so there is a tiny risk of deadlock. So ask for | |
1324 | * emergency memory if needed. | |
fd76863e | 1325 | */ |
689389a0 | 1326 | gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO; |
fd76863e | 1327 | |
9b8ae7b9 | 1328 | if (r1bio_existed) { |
689389a0 | 1329 | /* Need to get the block device name carefully */ |
2d32777d YK |
1330 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1331 | ||
689389a0 | 1332 | if (rdev) |
900d156b | 1333 | snprintf(b, sizeof(b), "%pg", rdev->bdev); |
689389a0 N |
1334 | else |
1335 | strcpy(b, "???"); | |
689389a0 | 1336 | } |
3b046a97 | 1337 | |
fd76863e | 1338 | /* |
fd76863e | 1339 | * Still need barrier for READ in case that whole |
1340 | * array is frozen. | |
fd76863e | 1341 | */ |
5aa70503 VV |
1342 | if (!wait_read_barrier(conf, bio->bi_iter.bi_sector, |
1343 | bio->bi_opf & REQ_NOWAIT)) { | |
1344 | bio_wouldblock_error(bio); | |
1345 | return; | |
1346 | } | |
fd76863e | 1347 | |
689389a0 N |
1348 | if (!r1_bio) |
1349 | r1_bio = alloc_r1bio(mddev, bio); | |
1350 | else | |
1351 | init_r1bio(r1_bio, mddev, bio); | |
c230e7e5 | 1352 | r1_bio->sectors = max_read_sectors; |
fd76863e | 1353 | |
1354 | /* | |
1355 | * make_request() can abort the operation when read-ahead is being | |
1356 | * used and no empty request is available. | |
1357 | */ | |
3b046a97 RL |
1358 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
1359 | ||
1360 | if (rdisk < 0) { | |
1361 | /* couldn't find anywhere to read from */ | |
9b8ae7b9 | 1362 | if (r1bio_existed) { |
689389a0 N |
1363 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
1364 | mdname(mddev), | |
1365 | b, | |
1366 | (unsigned long long)r1_bio->sector); | |
1367 | } | |
3b046a97 RL |
1368 | raid_end_bio_io(r1_bio); |
1369 | return; | |
1370 | } | |
1371 | mirror = conf->mirrors + rdisk; | |
1372 | ||
9b8ae7b9 | 1373 | if (r1bio_existed) |
913cce5a | 1374 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n", |
689389a0 N |
1375 | mdname(mddev), |
1376 | (unsigned long long)r1_bio->sector, | |
913cce5a | 1377 | mirror->rdev->bdev); |
689389a0 | 1378 | |
3b046a97 RL |
1379 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1380 | bitmap) { | |
1381 | /* | |
1382 | * Reading from a write-mostly device must take care not to | |
1383 | * over-take any writes that are 'behind' | |
1384 | */ | |
28be4fd3 | 1385 | mddev_add_trace_msg(mddev, "raid1 wait behind writes"); |
3b046a97 RL |
1386 | wait_event(bitmap->behind_wait, |
1387 | atomic_read(&bitmap->behind_writes) == 0); | |
1388 | } | |
c230e7e5 N |
1389 | |
1390 | if (max_sectors < bio_sectors(bio)) { | |
1391 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1392 | gfp, &conf->bio_split); |
c230e7e5 | 1393 | bio_chain(split, bio); |
ed00aabd | 1394 | submit_bio_noacct(bio); |
c230e7e5 N |
1395 | bio = split; |
1396 | r1_bio->master_bio = bio; | |
1397 | r1_bio->sectors = max_sectors; | |
1398 | } | |
1399 | ||
3b046a97 | 1400 | r1_bio->read_disk = rdisk; |
bb2a9ace YK |
1401 | if (!r1bio_existed) { |
1402 | md_account_bio(mddev, &bio); | |
1403 | r1_bio->master_bio = bio; | |
1404 | } | |
abfc426d CH |
1405 | read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp, |
1406 | &mddev->bio_set); | |
3b046a97 RL |
1407 | |
1408 | r1_bio->bios[rdisk] = read_bio; | |
1409 | ||
1410 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1411 | mirror->rdev->data_offset; | |
3b046a97 | 1412 | read_bio->bi_end_io = raid1_end_read_request; |
c34b7ac6 | 1413 | read_bio->bi_opf = op | do_sync; |
3b046a97 RL |
1414 | if (test_bit(FailFast, &mirror->rdev->flags) && |
1415 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1416 | read_bio->bi_opf |= MD_FAILFAST; | |
1417 | read_bio->bi_private = r1_bio; | |
c396b90e | 1418 | mddev_trace_remap(mddev, read_bio, r1_bio->sector); |
ed00aabd | 1419 | submit_bio_noacct(read_bio); |
3b046a97 RL |
1420 | } |
1421 | ||
c230e7e5 N |
1422 | static void raid1_write_request(struct mddev *mddev, struct bio *bio, |
1423 | int max_write_sectors) | |
3b046a97 RL |
1424 | { |
1425 | struct r1conf *conf = mddev->private; | |
fd76863e | 1426 | struct r1bio *r1_bio; |
1f68f0c4 | 1427 | int i, disks; |
3b046a97 | 1428 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1429 | unsigned long flags; |
3cb03002 | 1430 | struct md_rdev *blocked_rdev; |
1f68f0c4 | 1431 | int first_clone; |
1f68f0c4 | 1432 | int max_sectors; |
6607cd31 | 1433 | bool write_behind = false; |
9e55a22f | 1434 | bool is_discard = (bio_op(bio) == REQ_OP_DISCARD); |
191ea9b2 | 1435 | |
b3143b9a | 1436 | if (mddev_is_clustered(mddev) && |
90382ed9 | 1437 | md_cluster_ops->area_resyncing(mddev, WRITE, |
b3143b9a | 1438 | bio->bi_iter.bi_sector, bio_end_sector(bio))) { |
3b046a97 | 1439 | |
6eef4b21 | 1440 | DEFINE_WAIT(w); |
5aa70503 VV |
1441 | if (bio->bi_opf & REQ_NOWAIT) { |
1442 | bio_wouldblock_error(bio); | |
1443 | return; | |
1444 | } | |
6eef4b21 | 1445 | for (;;) { |
6eef4b21 | 1446 | prepare_to_wait(&conf->wait_barrier, |
ae89fd3d | 1447 | &w, TASK_IDLE); |
f81f7302 | 1448 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, |
385f4d7f | 1449 | bio->bi_iter.bi_sector, |
b3143b9a | 1450 | bio_end_sector(bio))) |
6eef4b21 N |
1451 | break; |
1452 | schedule(); | |
1453 | } | |
1454 | finish_wait(&conf->wait_barrier, &w); | |
1455 | } | |
f81f7302 GJ |
1456 | |
1457 | /* | |
1458 | * Register the new request and wait if the reconstruction | |
1459 | * thread has put up a bar for new requests. | |
1460 | * Continue immediately if no resync is active currently. | |
1461 | */ | |
5aa70503 VV |
1462 | if (!wait_barrier(conf, bio->bi_iter.bi_sector, |
1463 | bio->bi_opf & REQ_NOWAIT)) { | |
1464 | bio_wouldblock_error(bio); | |
1465 | return; | |
1466 | } | |
fd76863e | 1467 | |
992db13a | 1468 | retry_write: |
689389a0 | 1469 | r1_bio = alloc_r1bio(mddev, bio); |
c230e7e5 | 1470 | r1_bio->sectors = max_write_sectors; |
1da177e4 | 1471 | |
1f68f0c4 | 1472 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1473 | * inc refcount on their rdev. Record them by setting |
1474 | * bios[x] to bio | |
1f68f0c4 N |
1475 | * If there are known/acknowledged bad blocks on any device on |
1476 | * which we have seen a write error, we want to avoid writing those | |
1477 | * blocks. | |
1478 | * This potentially requires several writes to write around | |
1479 | * the bad blocks. Each set of writes gets it's own r1bio | |
1480 | * with a set of bios attached. | |
1da177e4 | 1481 | */ |
c3b328ac | 1482 | |
8f19ccb2 | 1483 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1484 | blocked_rdev = NULL; |
1f68f0c4 | 1485 | max_sectors = r1_bio->sectors; |
1da177e4 | 1486 | for (i = 0; i < disks; i++) { |
2d32777d | 1487 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
6607cd31 GJ |
1488 | |
1489 | /* | |
1490 | * The write-behind io is only attempted on drives marked as | |
1491 | * write-mostly, which means we could allocate write behind | |
1492 | * bio later. | |
1493 | */ | |
9e55a22f | 1494 | if (!is_discard && rdev && test_bit(WriteMostly, &rdev->flags)) |
6607cd31 GJ |
1495 | write_behind = true; |
1496 | ||
6bfe0b49 DW |
1497 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1498 | atomic_inc(&rdev->nr_pending); | |
1499 | blocked_rdev = rdev; | |
1500 | break; | |
1501 | } | |
1f68f0c4 | 1502 | r1_bio->bios[i] = NULL; |
8ae12666 | 1503 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1504 | if (i < conf->raid_disks) |
1505 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1506 | continue; |
1507 | } | |
1508 | ||
1509 | atomic_inc(&rdev->nr_pending); | |
1510 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1511 | sector_t first_bad; | |
1512 | int bad_sectors; | |
1513 | int is_bad; | |
1514 | ||
3b046a97 | 1515 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1516 | &first_bad, &bad_sectors); |
1517 | if (is_bad < 0) { | |
1518 | /* mustn't write here until the bad block is | |
1519 | * acknowledged*/ | |
1520 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1521 | blocked_rdev = rdev; | |
1522 | break; | |
1523 | } | |
1524 | if (is_bad && first_bad <= r1_bio->sector) { | |
1525 | /* Cannot write here at all */ | |
1526 | bad_sectors -= (r1_bio->sector - first_bad); | |
1527 | if (bad_sectors < max_sectors) | |
1528 | /* mustn't write more than bad_sectors | |
1529 | * to other devices yet | |
1530 | */ | |
1531 | max_sectors = bad_sectors; | |
03c902e1 | 1532 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1533 | /* We don't set R1BIO_Degraded as that |
1534 | * only applies if the disk is | |
1535 | * missing, so it might be re-added, | |
1536 | * and we want to know to recover this | |
1537 | * chunk. | |
1538 | * In this case the device is here, | |
1539 | * and the fact that this chunk is not | |
1540 | * in-sync is recorded in the bad | |
1541 | * block log | |
1542 | */ | |
1543 | continue; | |
964147d5 | 1544 | } |
1f68f0c4 N |
1545 | if (is_bad) { |
1546 | int good_sectors = first_bad - r1_bio->sector; | |
1547 | if (good_sectors < max_sectors) | |
1548 | max_sectors = good_sectors; | |
1549 | } | |
1550 | } | |
1551 | r1_bio->bios[i] = bio; | |
1da177e4 | 1552 | } |
1da177e4 | 1553 | |
6bfe0b49 DW |
1554 | if (unlikely(blocked_rdev)) { |
1555 | /* Wait for this device to become unblocked */ | |
1556 | int j; | |
1557 | ||
1558 | for (j = 0; j < i; j++) | |
1559 | if (r1_bio->bios[j]) | |
1560 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
992db13a | 1561 | free_r1bio(r1_bio); |
fd76863e | 1562 | allow_barrier(conf, bio->bi_iter.bi_sector); |
5aa70503 VV |
1563 | |
1564 | if (bio->bi_opf & REQ_NOWAIT) { | |
1565 | bio_wouldblock_error(bio); | |
1566 | return; | |
1567 | } | |
28be4fd3 CH |
1568 | mddev_add_trace_msg(mddev, "raid1 wait rdev %d blocked", |
1569 | blocked_rdev->raid_disk); | |
6bfe0b49 | 1570 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
5aa70503 | 1571 | wait_barrier(conf, bio->bi_iter.bi_sector, false); |
6bfe0b49 DW |
1572 | goto retry_write; |
1573 | } | |
1574 | ||
6607cd31 GJ |
1575 | /* |
1576 | * When using a bitmap, we may call alloc_behind_master_bio below. | |
1577 | * alloc_behind_master_bio allocates a copy of the data payload a page | |
1578 | * at a time and thus needs a new bio that can fit the whole payload | |
1579 | * this bio in page sized chunks. | |
1580 | */ | |
1581 | if (write_behind && bitmap) | |
1582 | max_sectors = min_t(int, max_sectors, | |
1583 | BIO_MAX_VECS * (PAGE_SIZE >> 9)); | |
c230e7e5 N |
1584 | if (max_sectors < bio_sectors(bio)) { |
1585 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1586 | GFP_NOIO, &conf->bio_split); |
c230e7e5 | 1587 | bio_chain(split, bio); |
ed00aabd | 1588 | submit_bio_noacct(bio); |
c230e7e5 N |
1589 | bio = split; |
1590 | r1_bio->master_bio = bio; | |
1f68f0c4 | 1591 | r1_bio->sectors = max_sectors; |
191ea9b2 | 1592 | } |
4b6d287f | 1593 | |
bb2a9ace YK |
1594 | md_account_bio(mddev, &bio); |
1595 | r1_bio->master_bio = bio; | |
4e78064f | 1596 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1597 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1598 | |
1f68f0c4 | 1599 | first_clone = 1; |
d8c84c4f | 1600 | |
1da177e4 | 1601 | for (i = 0; i < disks; i++) { |
8e58e327 | 1602 | struct bio *mbio = NULL; |
69df9cfc | 1603 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
1da177e4 LT |
1604 | if (!r1_bio->bios[i]) |
1605 | continue; | |
1606 | ||
46669e86 | 1607 | if (first_clone) { |
1f68f0c4 N |
1608 | /* do behind I/O ? |
1609 | * Not if there are too many, or cannot | |
1610 | * allocate memory, or a reader on WriteMostly | |
1611 | * is waiting for behind writes to flush */ | |
6b2460e6 | 1612 | if (bitmap && write_behind && |
1f68f0c4 N |
1613 | (atomic_read(&bitmap->behind_writes) |
1614 | < mddev->bitmap_info.max_write_behind) && | |
8e58e327 | 1615 | !waitqueue_active(&bitmap->behind_wait)) { |
16d56e2f | 1616 | alloc_behind_master_bio(r1_bio, bio); |
8e58e327 | 1617 | } |
1f68f0c4 | 1618 | |
e64e4018 AS |
1619 | md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, |
1620 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
1f68f0c4 N |
1621 | first_clone = 0; |
1622 | } | |
8e58e327 | 1623 | |
841c1316 | 1624 | if (r1_bio->behind_master_bio) { |
abfc426d CH |
1625 | mbio = bio_alloc_clone(rdev->bdev, |
1626 | r1_bio->behind_master_bio, | |
1627 | GFP_NOIO, &mddev->bio_set); | |
69df9cfc | 1628 | if (test_bit(CollisionCheck, &rdev->flags)) |
d0d2d8ba | 1629 | wait_for_serialization(rdev, r1_bio); |
3e148a32 | 1630 | if (test_bit(WriteMostly, &rdev->flags)) |
4b6d287f | 1631 | atomic_inc(&r1_bio->behind_remaining); |
abfc426d CH |
1632 | } else { |
1633 | mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, | |
1634 | &mddev->bio_set); | |
1635 | ||
1636 | if (mddev->serialize_policy) | |
1637 | wait_for_serialization(rdev, r1_bio); | |
1638 | } | |
4b6d287f | 1639 | |
1f68f0c4 N |
1640 | r1_bio->bios[i] = mbio; |
1641 | ||
2e94275e | 1642 | mbio->bi_iter.bi_sector = (r1_bio->sector + rdev->data_offset); |
1f68f0c4 | 1643 | mbio->bi_end_io = raid1_end_write_request; |
a682e003 | 1644 | mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); |
2e94275e GJ |
1645 | if (test_bit(FailFast, &rdev->flags) && |
1646 | !test_bit(WriteMostly, &rdev->flags) && | |
212e7eb7 N |
1647 | conf->raid_disks - mddev->degraded > 1) |
1648 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1649 | mbio->bi_private = r1_bio; |
1650 | ||
1da177e4 | 1651 | atomic_inc(&r1_bio->remaining); |
c396b90e | 1652 | mddev_trace_remap(mddev, mbio, r1_bio->sector); |
109e3765 | 1653 | /* flush_pending_writes() needs access to the rdev so...*/ |
2e94275e | 1654 | mbio->bi_bdev = (void *)rdev; |
460af1f9 | 1655 | if (!raid1_add_bio_to_plug(mddev, mbio, raid1_unplug, disks)) { |
23b245c0 | 1656 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e | 1657 | bio_list_add(&conf->pending_bio_list, mbio); |
23b245c0 | 1658 | spin_unlock_irqrestore(&conf->device_lock, flags); |
b357f04a | 1659 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1660 | } |
1da177e4 | 1661 | } |
1f68f0c4 | 1662 | |
079fa166 N |
1663 | r1_bio_write_done(r1_bio); |
1664 | ||
1665 | /* In case raid1d snuck in to freeze_array */ | |
21bd9a68 | 1666 | wake_up_barrier(conf); |
1da177e4 LT |
1667 | } |
1668 | ||
cc27b0c7 | 1669 | static bool raid1_make_request(struct mddev *mddev, struct bio *bio) |
3b046a97 | 1670 | { |
fd76863e | 1671 | sector_t sectors; |
3b046a97 | 1672 | |
775d7831 DJ |
1673 | if (unlikely(bio->bi_opf & REQ_PREFLUSH) |
1674 | && md_flush_request(mddev, bio)) | |
cc27b0c7 | 1675 | return true; |
3b046a97 | 1676 | |
c230e7e5 N |
1677 | /* |
1678 | * There is a limit to the maximum size, but | |
1679 | * the read/write handler might find a lower limit | |
1680 | * due to bad blocks. To avoid multiple splits, | |
1681 | * we pass the maximum number of sectors down | |
1682 | * and let the lower level perform the split. | |
1683 | */ | |
1684 | sectors = align_to_barrier_unit_end( | |
1685 | bio->bi_iter.bi_sector, bio_sectors(bio)); | |
61eb2b43 | 1686 | |
c230e7e5 | 1687 | if (bio_data_dir(bio) == READ) |
689389a0 | 1688 | raid1_read_request(mddev, bio, sectors, NULL); |
cc27b0c7 N |
1689 | else { |
1690 | if (!md_write_start(mddev,bio)) | |
1691 | return false; | |
c230e7e5 | 1692 | raid1_write_request(mddev, bio, sectors); |
cc27b0c7 N |
1693 | } |
1694 | return true; | |
3b046a97 RL |
1695 | } |
1696 | ||
849674e4 | 1697 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1698 | { |
e8096360 | 1699 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1700 | int i; |
1701 | ||
2d32777d YK |
1702 | lockdep_assert_held(&mddev->lock); |
1703 | ||
1da177e4 | 1704 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, |
11ce99e6 | 1705 | conf->raid_disks - mddev->degraded); |
ddac7c7e | 1706 | for (i = 0; i < conf->raid_disks; i++) { |
2d32777d YK |
1707 | struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev); |
1708 | ||
1da177e4 | 1709 | seq_printf(seq, "%s", |
ddac7c7e N |
1710 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1711 | } | |
1da177e4 LT |
1712 | seq_printf(seq, "]"); |
1713 | } | |
1714 | ||
9631abdb MT |
1715 | /** |
1716 | * raid1_error() - RAID1 error handler. | |
1717 | * @mddev: affected md device. | |
1718 | * @rdev: member device to fail. | |
1719 | * | |
1720 | * The routine acknowledges &rdev failure and determines new @mddev state. | |
1721 | * If it failed, then: | |
1722 | * - &MD_BROKEN flag is set in &mddev->flags. | |
1723 | * - recovery is disabled. | |
1724 | * Otherwise, it must be degraded: | |
1725 | * - recovery is interrupted. | |
1726 | * - &mddev->degraded is bumped. | |
1727 | * | |
1728 | * @rdev is marked as &Faulty excluding case when array is failed and | |
1729 | * &mddev->fail_last_dev is off. | |
1730 | */ | |
849674e4 | 1731 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1732 | { |
e8096360 | 1733 | struct r1conf *conf = mddev->private; |
423f04d6 | 1734 | unsigned long flags; |
1da177e4 | 1735 | |
2e52d449 | 1736 | spin_lock_irqsave(&conf->device_lock, flags); |
9631abdb MT |
1737 | |
1738 | if (test_bit(In_sync, &rdev->flags) && | |
1739 | (conf->raid_disks - mddev->degraded) == 1) { | |
1740 | set_bit(MD_BROKEN, &mddev->flags); | |
1741 | ||
1742 | if (!mddev->fail_last_dev) { | |
1743 | conf->recovery_disabled = mddev->recovery_disabled; | |
1744 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1745 | return; | |
1746 | } | |
4044ba58 | 1747 | } |
de393cde | 1748 | set_bit(Blocked, &rdev->flags); |
ebda52fa | 1749 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1750 | mddev->degraded++; |
ebda52fa | 1751 | set_bit(Faulty, &rdev->flags); |
423f04d6 | 1752 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1753 | /* |
1754 | * if recovery is running, make sure it aborts. | |
1755 | */ | |
1756 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1757 | set_mask_bits(&mddev->sb_flags, 0, |
1758 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
913cce5a | 1759 | pr_crit("md/raid1:%s: Disk failure on %pg, disabling device.\n" |
1d41c216 | 1760 | "md/raid1:%s: Operation continuing on %d devices.\n", |
913cce5a | 1761 | mdname(mddev), rdev->bdev, |
1d41c216 | 1762 | mdname(mddev), conf->raid_disks - mddev->degraded); |
1da177e4 LT |
1763 | } |
1764 | ||
e8096360 | 1765 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1766 | { |
1767 | int i; | |
1da177e4 | 1768 | |
1d41c216 | 1769 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1770 | if (!conf) { |
1d41c216 | 1771 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1772 | return; |
1773 | } | |
1d41c216 N |
1774 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1775 | conf->raid_disks); | |
1da177e4 | 1776 | |
2d32777d | 1777 | lockdep_assert_held(&conf->mddev->reconfig_mutex); |
1da177e4 | 1778 | for (i = 0; i < conf->raid_disks; i++) { |
2d32777d | 1779 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
ddac7c7e | 1780 | if (rdev) |
913cce5a | 1781 | pr_debug(" disk %d, wo:%d, o:%d, dev:%pg\n", |
1d41c216 N |
1782 | i, !test_bit(In_sync, &rdev->flags), |
1783 | !test_bit(Faulty, &rdev->flags), | |
913cce5a | 1784 | rdev->bdev); |
1da177e4 LT |
1785 | } |
1786 | } | |
1787 | ||
e8096360 | 1788 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1789 | { |
f6eca2d4 ND |
1790 | int idx; |
1791 | ||
1792 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) { | |
5aa70503 | 1793 | _wait_barrier(conf, idx, false); |
f6eca2d4 ND |
1794 | _allow_barrier(conf, idx); |
1795 | } | |
1da177e4 | 1796 | |
afeee514 | 1797 | mempool_exit(&conf->r1buf_pool); |
1da177e4 LT |
1798 | } |
1799 | ||
fd01b88c | 1800 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1801 | { |
1802 | int i; | |
e8096360 | 1803 | struct r1conf *conf = mddev->private; |
6b965620 N |
1804 | int count = 0; |
1805 | unsigned long flags; | |
1da177e4 LT |
1806 | |
1807 | /* | |
f72ffdd6 | 1808 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1809 | * and mark them readable. |
1810 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1811 | * device_lock used to avoid races with raid1_end_read_request |
1812 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1813 | */ |
423f04d6 | 1814 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1815 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1816 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1817 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1818 | if (repl | |
1aee41f6 | 1819 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1820 | && repl->recovery_offset == MaxSector |
1821 | && !test_bit(Faulty, &repl->flags) | |
1822 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1823 | /* replacement has just become active */ | |
1824 | if (!rdev || | |
1825 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1826 | count++; | |
1827 | if (rdev) { | |
1828 | /* Replaced device not technically | |
1829 | * faulty, but we need to be sure | |
1830 | * it gets removed and never re-added | |
1831 | */ | |
1832 | set_bit(Faulty, &rdev->flags); | |
1833 | sysfs_notify_dirent_safe( | |
1834 | rdev->sysfs_state); | |
1835 | } | |
1836 | } | |
ddac7c7e | 1837 | if (rdev |
61e4947c | 1838 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1839 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1840 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1841 | count++; |
654e8b5a | 1842 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1843 | } |
1844 | } | |
6b965620 N |
1845 | mddev->degraded -= count; |
1846 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1847 | |
1848 | print_conf(conf); | |
6b965620 | 1849 | return count; |
1da177e4 LT |
1850 | } |
1851 | ||
969d6589 YK |
1852 | static bool raid1_add_conf(struct r1conf *conf, struct md_rdev *rdev, int disk, |
1853 | bool replacement) | |
1854 | { | |
1855 | struct raid1_info *info = conf->mirrors + disk; | |
1856 | ||
1857 | if (replacement) | |
1858 | info += conf->raid_disks; | |
1859 | ||
1860 | if (info->rdev) | |
1861 | return false; | |
1862 | ||
2c27d09d YK |
1863 | if (bdev_nonrot(rdev->bdev)) { |
1864 | set_bit(Nonrot, &rdev->flags); | |
1865 | WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks + 1); | |
1866 | } | |
1867 | ||
969d6589 YK |
1868 | rdev->raid_disk = disk; |
1869 | info->head_position = 0; | |
1870 | info->seq_start = MaxSector; | |
1871 | WRITE_ONCE(info->rdev, rdev); | |
1872 | ||
1873 | return true; | |
1874 | } | |
1875 | ||
1876 | static bool raid1_remove_conf(struct r1conf *conf, int disk) | |
1877 | { | |
1878 | struct raid1_info *info = conf->mirrors + disk; | |
1879 | struct md_rdev *rdev = info->rdev; | |
1880 | ||
1881 | if (!rdev || test_bit(In_sync, &rdev->flags) || | |
1882 | atomic_read(&rdev->nr_pending)) | |
1883 | return false; | |
1884 | ||
1885 | /* Only remove non-faulty devices if recovery is not possible. */ | |
1886 | if (!test_bit(Faulty, &rdev->flags) && | |
1887 | rdev->mddev->recovery_disabled != conf->recovery_disabled && | |
1888 | rdev->mddev->degraded < conf->raid_disks) | |
1889 | return false; | |
1890 | ||
2c27d09d YK |
1891 | if (test_and_clear_bit(Nonrot, &rdev->flags)) |
1892 | WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks - 1); | |
1893 | ||
969d6589 YK |
1894 | WRITE_ONCE(info->rdev, NULL); |
1895 | return true; | |
1896 | } | |
1897 | ||
fd01b88c | 1898 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1899 | { |
e8096360 | 1900 | struct r1conf *conf = mddev->private; |
199050ea | 1901 | int err = -EEXIST; |
ffb1e7a0 | 1902 | int mirror = 0, repl_slot = -1; |
0eaf822c | 1903 | struct raid1_info *p; |
6c2fce2e | 1904 | int first = 0; |
30194636 | 1905 | int last = conf->raid_disks - 1; |
1da177e4 | 1906 | |
5389042f N |
1907 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1908 | return -EBUSY; | |
1909 | ||
1501efad DW |
1910 | if (md_integrity_add_rdev(rdev, mddev)) |
1911 | return -ENXIO; | |
1912 | ||
6c2fce2e NB |
1913 | if (rdev->raid_disk >= 0) |
1914 | first = last = rdev->raid_disk; | |
1915 | ||
70bcecdb GR |
1916 | /* |
1917 | * find the disk ... but prefer rdev->saved_raid_disk | |
1918 | * if possible. | |
1919 | */ | |
1920 | if (rdev->saved_raid_disk >= 0 && | |
1921 | rdev->saved_raid_disk >= first && | |
9e753ba9 | 1922 | rdev->saved_raid_disk < conf->raid_disks && |
70bcecdb GR |
1923 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1924 | first = last = rdev->saved_raid_disk; | |
1925 | ||
7ef449d1 | 1926 | for (mirror = first; mirror <= last; mirror++) { |
ebfeb444 | 1927 | p = conf->mirrors + mirror; |
7ef449d1 | 1928 | if (!p->rdev) { |
97894f7d CH |
1929 | err = mddev_stack_new_rdev(mddev, rdev); |
1930 | if (err) | |
1931 | return err; | |
1da177e4 | 1932 | |
969d6589 | 1933 | raid1_add_conf(conf, rdev, mirror, false); |
6aea114a N |
1934 | /* As all devices are equivalent, we don't need a full recovery |
1935 | * if this was recently any drive of the array | |
1936 | */ | |
1937 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1938 | conf->fullsync = 1; |
1da177e4 LT |
1939 | break; |
1940 | } | |
7ef449d1 | 1941 | if (test_bit(WantReplacement, &p->rdev->flags) && |
ffb1e7a0 LN |
1942 | p[conf->raid_disks].rdev == NULL && repl_slot < 0) |
1943 | repl_slot = mirror; | |
7ef449d1 | 1944 | } |
ffb1e7a0 LN |
1945 | |
1946 | if (err && repl_slot >= 0) { | |
1947 | /* Add this device as a replacement */ | |
ffb1e7a0 LN |
1948 | clear_bit(In_sync, &rdev->flags); |
1949 | set_bit(Replacement, &rdev->flags); | |
969d6589 | 1950 | raid1_add_conf(conf, rdev, repl_slot, true); |
ffb1e7a0 LN |
1951 | err = 0; |
1952 | conf->fullsync = 1; | |
ffb1e7a0 LN |
1953 | } |
1954 | ||
1da177e4 | 1955 | print_conf(conf); |
199050ea | 1956 | return err; |
1da177e4 LT |
1957 | } |
1958 | ||
b8321b68 | 1959 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1960 | { |
e8096360 | 1961 | struct r1conf *conf = mddev->private; |
1da177e4 | 1962 | int err = 0; |
b8321b68 | 1963 | int number = rdev->raid_disk; |
df203da4 | 1964 | struct raid1_info *p = conf->mirrors + number; |
8b0472b5 ZS |
1965 | |
1966 | if (unlikely(number >= conf->raid_disks)) | |
1967 | goto abort; | |
1968 | ||
969d6589 YK |
1969 | if (rdev != p->rdev) { |
1970 | number += conf->raid_disks; | |
1971 | p = conf->mirrors + number; | |
1972 | } | |
b014f14c | 1973 | |
1da177e4 | 1974 | print_conf(conf); |
b8321b68 | 1975 | if (rdev == p->rdev) { |
969d6589 | 1976 | if (!raid1_remove_conf(conf, number)) { |
dfc70645 N |
1977 | err = -EBUSY; |
1978 | goto abort; | |
1979 | } | |
969d6589 YK |
1980 | |
1981 | if (number < conf->raid_disks && | |
1982 | conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1983 | /* We just removed a device that is being replaced. |
1984 | * Move down the replacement. We drain all IO before | |
1985 | * doing this to avoid confusion. | |
1986 | */ | |
1987 | struct md_rdev *repl = | |
1988 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1989 | freeze_array(conf, 0); |
3de59bb9 YY |
1990 | if (atomic_read(&repl->nr_pending)) { |
1991 | /* It means that some queued IO of retry_list | |
1992 | * hold repl. Thus, we cannot set replacement | |
1993 | * as NULL, avoiding rdev NULL pointer | |
1994 | * dereference in sync_request_write and | |
1995 | * handle_write_finished. | |
1996 | */ | |
1997 | err = -EBUSY; | |
1998 | unfreeze_array(conf); | |
1999 | goto abort; | |
2000 | } | |
8c7a2c2b | 2001 | clear_bit(Replacement, &repl->flags); |
2d32777d | 2002 | WRITE_ONCE(p->rdev, repl); |
8c7a2c2b | 2003 | conf->mirrors[conf->raid_disks + number].rdev = NULL; |
e2d59925 | 2004 | unfreeze_array(conf); |
e5bc9c3c GJ |
2005 | } |
2006 | ||
2007 | clear_bit(WantReplacement, &rdev->flags); | |
a91a2785 | 2008 | err = md_integrity_register(mddev); |
1da177e4 LT |
2009 | } |
2010 | abort: | |
2011 | ||
2012 | print_conf(conf); | |
2013 | return err; | |
2014 | } | |
2015 | ||
4246a0b6 | 2016 | static void end_sync_read(struct bio *bio) |
1da177e4 | 2017 | { |
98d30c58 | 2018 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
1da177e4 | 2019 | |
0fc280f6 | 2020 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 2021 | |
1da177e4 LT |
2022 | /* |
2023 | * we have read a block, now it needs to be re-written, | |
2024 | * or re-read if the read failed. | |
2025 | * We don't do much here, just schedule handling by raid1d | |
2026 | */ | |
4e4cbee9 | 2027 | if (!bio->bi_status) |
1da177e4 | 2028 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
2029 | |
2030 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
2031 | reschedule_retry(r1_bio); | |
1da177e4 LT |
2032 | } |
2033 | ||
dfcc34c9 ND |
2034 | static void abort_sync_write(struct mddev *mddev, struct r1bio *r1_bio) |
2035 | { | |
2036 | sector_t sync_blocks = 0; | |
2037 | sector_t s = r1_bio->sector; | |
2038 | long sectors_to_go = r1_bio->sectors; | |
2039 | ||
2040 | /* make sure these bits don't get cleared. */ | |
2041 | do { | |
2042 | md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); | |
2043 | s += sync_blocks; | |
2044 | sectors_to_go -= sync_blocks; | |
2045 | } while (sectors_to_go > 0); | |
2046 | } | |
2047 | ||
449808a2 HT |
2048 | static void put_sync_write_buf(struct r1bio *r1_bio, int uptodate) |
2049 | { | |
2050 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
2051 | struct mddev *mddev = r1_bio->mddev; | |
2052 | int s = r1_bio->sectors; | |
2053 | ||
2054 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2055 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2056 | reschedule_retry(r1_bio); | |
2057 | else { | |
2058 | put_buf(r1_bio); | |
2059 | md_done_sync(mddev, s, uptodate); | |
2060 | } | |
2061 | } | |
2062 | } | |
2063 | ||
4246a0b6 | 2064 | static void end_sync_write(struct bio *bio) |
1da177e4 | 2065 | { |
4e4cbee9 | 2066 | int uptodate = !bio->bi_status; |
98d30c58 | 2067 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
fd01b88c | 2068 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2069 | struct r1conf *conf = mddev->private; |
854abd75 | 2070 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 2071 | |
6b1117d5 | 2072 | if (!uptodate) { |
dfcc34c9 | 2073 | abort_sync_write(mddev, r1_bio); |
854abd75 N |
2074 | set_bit(WriteErrorSeen, &rdev->flags); |
2075 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
2076 | set_bit(MD_RECOVERY_NEEDED, & |
2077 | mddev->recovery); | |
d8f05d29 | 2078 | set_bit(R1BIO_WriteError, &r1_bio->state); |
3a0f007b YK |
2079 | } else if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) && |
2080 | !rdev_has_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
2081 | r1_bio->sector, r1_bio->sectors)) { | |
4367af55 | 2082 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
3a0f007b | 2083 | } |
e3b9703e | 2084 | |
449808a2 | 2085 | put_sync_write_buf(r1_bio, uptodate); |
1da177e4 LT |
2086 | } |
2087 | ||
3cb03002 | 2088 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
7dab2455 | 2089 | int sectors, struct page *page, blk_opf_t rw) |
d8f05d29 | 2090 | { |
4ce4c73f | 2091 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) |
d8f05d29 N |
2092 | /* success */ |
2093 | return 1; | |
7dab2455 | 2094 | if (rw == REQ_OP_WRITE) { |
d8f05d29 | 2095 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
2096 | if (!test_and_set_bit(WantReplacement, |
2097 | &rdev->flags)) | |
2098 | set_bit(MD_RECOVERY_NEEDED, & | |
2099 | rdev->mddev->recovery); | |
2100 | } | |
d8f05d29 N |
2101 | /* need to record an error - either for the block or the device */ |
2102 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2103 | md_error(rdev->mddev, rdev); | |
2104 | return 0; | |
2105 | } | |
2106 | ||
9f2c9d12 | 2107 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 2108 | { |
a68e5870 N |
2109 | /* Try some synchronous reads of other devices to get |
2110 | * good data, much like with normal read errors. Only | |
2111 | * read into the pages we already have so we don't | |
2112 | * need to re-issue the read request. | |
2113 | * We don't need to freeze the array, because being in an | |
2114 | * active sync request, there is no normal IO, and | |
2115 | * no overlapping syncs. | |
06f60385 N |
2116 | * We don't need to check is_badblock() again as we |
2117 | * made sure that anything with a bad block in range | |
2118 | * will have bi_end_io clear. | |
a68e5870 | 2119 | */ |
fd01b88c | 2120 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2121 | struct r1conf *conf = mddev->private; |
a68e5870 | 2122 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
44cf0f4d | 2123 | struct page **pages = get_resync_pages(bio)->pages; |
a68e5870 N |
2124 | sector_t sect = r1_bio->sector; |
2125 | int sectors = r1_bio->sectors; | |
2126 | int idx = 0; | |
2e52d449 N |
2127 | struct md_rdev *rdev; |
2128 | ||
2129 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
2130 | if (test_bit(FailFast, &rdev->flags)) { | |
2131 | /* Don't try recovering from here - just fail it | |
2132 | * ... unless it is the last working device of course */ | |
2133 | md_error(mddev, rdev); | |
2134 | if (test_bit(Faulty, &rdev->flags)) | |
2135 | /* Don't try to read from here, but make sure | |
2136 | * put_buf does it's thing | |
2137 | */ | |
2138 | bio->bi_end_io = end_sync_write; | |
2139 | } | |
a68e5870 N |
2140 | |
2141 | while(sectors) { | |
2142 | int s = sectors; | |
2143 | int d = r1_bio->read_disk; | |
2144 | int success = 0; | |
78d7f5f7 | 2145 | int start; |
a68e5870 N |
2146 | |
2147 | if (s > (PAGE_SIZE>>9)) | |
2148 | s = PAGE_SIZE >> 9; | |
2149 | do { | |
2150 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
2151 | /* No rcu protection needed here devices | |
2152 | * can only be removed when no resync is | |
2153 | * active, and resync is currently active | |
2154 | */ | |
2155 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 2156 | if (sync_page_io(rdev, sect, s<<9, |
44cf0f4d | 2157 | pages[idx], |
4ce4c73f | 2158 | REQ_OP_READ, false)) { |
a68e5870 N |
2159 | success = 1; |
2160 | break; | |
2161 | } | |
2162 | } | |
2163 | d++; | |
8f19ccb2 | 2164 | if (d == conf->raid_disks * 2) |
a68e5870 N |
2165 | d = 0; |
2166 | } while (!success && d != r1_bio->read_disk); | |
2167 | ||
78d7f5f7 | 2168 | if (!success) { |
3a9f28a5 N |
2169 | int abort = 0; |
2170 | /* Cannot read from anywhere, this block is lost. | |
2171 | * Record a bad block on each device. If that doesn't | |
2172 | * work just disable and interrupt the recovery. | |
2173 | * Don't fail devices as that won't really help. | |
2174 | */ | |
ac483eb3 CH |
2175 | pr_crit_ratelimited("md/raid1:%s: %pg: unrecoverable I/O read error for block %llu\n", |
2176 | mdname(mddev), bio->bi_bdev, | |
1d41c216 | 2177 | (unsigned long long)r1_bio->sector); |
8f19ccb2 | 2178 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
2179 | rdev = conf->mirrors[d].rdev; |
2180 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
2181 | continue; | |
2182 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
2183 | abort = 1; | |
2184 | } | |
2185 | if (abort) { | |
d890fa2b N |
2186 | conf->recovery_disabled = |
2187 | mddev->recovery_disabled; | |
3a9f28a5 N |
2188 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
2189 | md_done_sync(mddev, r1_bio->sectors, 0); | |
2190 | put_buf(r1_bio); | |
2191 | return 0; | |
2192 | } | |
2193 | /* Try next page */ | |
2194 | sectors -= s; | |
2195 | sect += s; | |
2196 | idx++; | |
2197 | continue; | |
d11c171e | 2198 | } |
78d7f5f7 N |
2199 | |
2200 | start = d; | |
2201 | /* write it back and re-read */ | |
2202 | while (d != r1_bio->read_disk) { | |
2203 | if (d == 0) | |
8f19ccb2 | 2204 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2205 | d--; |
2206 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2207 | continue; | |
2208 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2209 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2210 | pages[idx], |
7dab2455 | 2211 | REQ_OP_WRITE) == 0) { |
78d7f5f7 N |
2212 | r1_bio->bios[d]->bi_end_io = NULL; |
2213 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 2214 | } |
78d7f5f7 N |
2215 | } |
2216 | d = start; | |
2217 | while (d != r1_bio->read_disk) { | |
2218 | if (d == 0) | |
8f19ccb2 | 2219 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2220 | d--; |
2221 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2222 | continue; | |
2223 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2224 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2225 | pages[idx], |
7dab2455 | 2226 | REQ_OP_READ) != 0) |
9d3d8011 | 2227 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 2228 | } |
a68e5870 N |
2229 | sectors -= s; |
2230 | sect += s; | |
2231 | idx ++; | |
2232 | } | |
78d7f5f7 | 2233 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4e4cbee9 | 2234 | bio->bi_status = 0; |
a68e5870 N |
2235 | return 1; |
2236 | } | |
2237 | ||
c95e6385 | 2238 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
2239 | { |
2240 | /* We have read all readable devices. If we haven't | |
2241 | * got the block, then there is no hope left. | |
2242 | * If we have, then we want to do a comparison | |
2243 | * and skip the write if everything is the same. | |
2244 | * If any blocks failed to read, then we need to | |
2245 | * attempt an over-write | |
2246 | */ | |
fd01b88c | 2247 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2248 | struct r1conf *conf = mddev->private; |
a68e5870 N |
2249 | int primary; |
2250 | int i; | |
f4380a91 | 2251 | int vcnt; |
a68e5870 | 2252 | |
30bc9b53 N |
2253 | /* Fix variable parts of all bios */ |
2254 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
2255 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
4e4cbee9 | 2256 | blk_status_t status; |
30bc9b53 | 2257 | struct bio *b = r1_bio->bios[i]; |
98d30c58 | 2258 | struct resync_pages *rp = get_resync_pages(b); |
30bc9b53 N |
2259 | if (b->bi_end_io != end_sync_read) |
2260 | continue; | |
4246a0b6 | 2261 | /* fixup the bio for reuse, but preserve errno */ |
4e4cbee9 | 2262 | status = b->bi_status; |
a7c50c94 | 2263 | bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ); |
4e4cbee9 | 2264 | b->bi_status = status; |
4f024f37 | 2265 | b->bi_iter.bi_sector = r1_bio->sector + |
30bc9b53 | 2266 | conf->mirrors[i].rdev->data_offset; |
30bc9b53 | 2267 | b->bi_end_io = end_sync_read; |
98d30c58 ML |
2268 | rp->raid_bio = r1_bio; |
2269 | b->bi_private = rp; | |
30bc9b53 | 2270 | |
fb0eb5df ML |
2271 | /* initialize bvec table again */ |
2272 | md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); | |
30bc9b53 | 2273 | } |
8f19ccb2 | 2274 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2275 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4e4cbee9 | 2276 | !r1_bio->bios[primary]->bi_status) { |
a68e5870 N |
2277 | r1_bio->bios[primary]->bi_end_io = NULL; |
2278 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2279 | break; | |
2280 | } | |
2281 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2282 | for (i = 0; i < conf->raid_disks * 2; i++) { |
2b070cfe | 2283 | int j = 0; |
78d7f5f7 N |
2284 | struct bio *pbio = r1_bio->bios[primary]; |
2285 | struct bio *sbio = r1_bio->bios[i]; | |
4e4cbee9 | 2286 | blk_status_t status = sbio->bi_status; |
44cf0f4d ML |
2287 | struct page **ppages = get_resync_pages(pbio)->pages; |
2288 | struct page **spages = get_resync_pages(sbio)->pages; | |
60928a91 | 2289 | struct bio_vec *bi; |
8fc04e6e | 2290 | int page_len[RESYNC_PAGES] = { 0 }; |
6dc4f100 | 2291 | struct bvec_iter_all iter_all; |
a68e5870 | 2292 | |
2aabaa65 | 2293 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2294 | continue; |
4246a0b6 | 2295 | /* Now we can 'fixup' the error value */ |
4e4cbee9 | 2296 | sbio->bi_status = 0; |
78d7f5f7 | 2297 | |
2b070cfe CH |
2298 | bio_for_each_segment_all(bi, sbio, iter_all) |
2299 | page_len[j++] = bi->bv_len; | |
60928a91 | 2300 | |
4e4cbee9 | 2301 | if (!status) { |
78d7f5f7 | 2302 | for (j = vcnt; j-- ; ) { |
44cf0f4d ML |
2303 | if (memcmp(page_address(ppages[j]), |
2304 | page_address(spages[j]), | |
60928a91 | 2305 | page_len[j])) |
78d7f5f7 | 2306 | break; |
69382e85 | 2307 | } |
78d7f5f7 N |
2308 | } else |
2309 | j = 0; | |
2310 | if (j >= 0) | |
7f7583d4 | 2311 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2312 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4e4cbee9 | 2313 | && !status)) { |
78d7f5f7 N |
2314 | /* No need to write to this device. */ |
2315 | sbio->bi_end_io = NULL; | |
2316 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2317 | continue; | |
2318 | } | |
d3b45c2a KO |
2319 | |
2320 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2321 | } |
a68e5870 N |
2322 | } |
2323 | ||
9f2c9d12 | 2324 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2325 | { |
e8096360 | 2326 | struct r1conf *conf = mddev->private; |
a68e5870 | 2327 | int i; |
8f19ccb2 | 2328 | int disks = conf->raid_disks * 2; |
037d2ff6 | 2329 | struct bio *wbio; |
a68e5870 | 2330 | |
a68e5870 N |
2331 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2332 | /* ouch - failed to read all of that. */ | |
2333 | if (!fix_sync_read_error(r1_bio)) | |
2334 | return; | |
7ca78d57 N |
2335 | |
2336 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2337 | process_checks(r1_bio); |
2338 | ||
d11c171e N |
2339 | /* |
2340 | * schedule writes | |
2341 | */ | |
1da177e4 LT |
2342 | atomic_set(&r1_bio->remaining, 1); |
2343 | for (i = 0; i < disks ; i++) { | |
2344 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2345 | if (wbio->bi_end_io == NULL || |
2346 | (wbio->bi_end_io == end_sync_read && | |
2347 | (i == r1_bio->read_disk || | |
2348 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 | 2349 | continue; |
dfcc34c9 ND |
2350 | if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { |
2351 | abort_sync_write(mddev, r1_bio); | |
0c9d5b12 | 2352 | continue; |
dfcc34c9 | 2353 | } |
1da177e4 | 2354 | |
c34b7ac6 | 2355 | wbio->bi_opf = REQ_OP_WRITE; |
212e7eb7 N |
2356 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2357 | wbio->bi_opf |= MD_FAILFAST; | |
2358 | ||
3e198f78 | 2359 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2360 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2361 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2362 | |
ed00aabd | 2363 | submit_bio_noacct(wbio); |
1da177e4 LT |
2364 | } |
2365 | ||
449808a2 | 2366 | put_sync_write_buf(r1_bio, 1); |
1da177e4 LT |
2367 | } |
2368 | ||
2369 | /* | |
2370 | * This is a kernel thread which: | |
2371 | * | |
2372 | * 1. Retries failed read operations on working mirrors. | |
2373 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2374 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2375 | */ |
2376 | ||
ca294b34 | 2377 | static void fix_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
867868fb | 2378 | { |
ca294b34 LN |
2379 | sector_t sect = r1_bio->sector; |
2380 | int sectors = r1_bio->sectors; | |
2381 | int read_disk = r1_bio->read_disk; | |
fd01b88c | 2382 | struct mddev *mddev = conf->mddev; |
9f3fe29d | 2383 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
ca294b34 LN |
2384 | |
2385 | if (exceed_read_errors(mddev, rdev)) { | |
2386 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2387 | return; | |
2388 | } | |
2389 | ||
867868fb N |
2390 | while(sectors) { |
2391 | int s = sectors; | |
2392 | int d = read_disk; | |
2393 | int success = 0; | |
2394 | int start; | |
867868fb N |
2395 | |
2396 | if (s > (PAGE_SIZE>>9)) | |
2397 | s = PAGE_SIZE >> 9; | |
2398 | ||
2399 | do { | |
2d32777d | 2400 | rdev = conf->mirrors[d].rdev; |
867868fb | 2401 | if (rdev && |
da8840a7 | 2402 | (test_bit(In_sync, &rdev->flags) || |
2403 | (!test_bit(Faulty, &rdev->flags) && | |
2404 | rdev->recovery_offset >= sect + s)) && | |
3a0f007b | 2405 | rdev_has_badblock(rdev, sect, s) == 0) { |
707a6a42 | 2406 | atomic_inc(&rdev->nr_pending); |
707a6a42 | 2407 | if (sync_page_io(rdev, sect, s<<9, |
4ce4c73f | 2408 | conf->tmppage, REQ_OP_READ, false)) |
707a6a42 N |
2409 | success = 1; |
2410 | rdev_dec_pending(rdev, mddev); | |
2411 | if (success) | |
2412 | break; | |
2d32777d YK |
2413 | } |
2414 | ||
707a6a42 N |
2415 | d++; |
2416 | if (d == conf->raid_disks * 2) | |
2417 | d = 0; | |
02c67a3b | 2418 | } while (d != read_disk); |
867868fb N |
2419 | |
2420 | if (!success) { | |
d8f05d29 | 2421 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2422 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2423 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2424 | md_error(mddev, rdev); | |
867868fb N |
2425 | break; |
2426 | } | |
2427 | /* write it back and re-read */ | |
2428 | start = d; | |
2429 | while (d != read_disk) { | |
2430 | if (d==0) | |
8f19ccb2 | 2431 | d = conf->raid_disks * 2; |
867868fb | 2432 | d--; |
2d32777d | 2433 | rdev = conf->mirrors[d].rdev; |
867868fb | 2434 | if (rdev && |
707a6a42 N |
2435 | !test_bit(Faulty, &rdev->flags)) { |
2436 | atomic_inc(&rdev->nr_pending); | |
d8f05d29 | 2437 | r1_sync_page_io(rdev, sect, s, |
7dab2455 | 2438 | conf->tmppage, REQ_OP_WRITE); |
707a6a42 | 2439 | rdev_dec_pending(rdev, mddev); |
2d32777d | 2440 | } |
867868fb N |
2441 | } |
2442 | d = start; | |
2443 | while (d != read_disk) { | |
867868fb | 2444 | if (d==0) |
8f19ccb2 | 2445 | d = conf->raid_disks * 2; |
867868fb | 2446 | d--; |
2d32777d | 2447 | rdev = conf->mirrors[d].rdev; |
867868fb | 2448 | if (rdev && |
b8cb6b4c | 2449 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 | 2450 | atomic_inc(&rdev->nr_pending); |
d8f05d29 | 2451 | if (r1_sync_page_io(rdev, sect, s, |
7dab2455 | 2452 | conf->tmppage, REQ_OP_READ)) { |
867868fb | 2453 | atomic_add(s, &rdev->corrected_errors); |
913cce5a | 2454 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %pg)\n", |
1d41c216 N |
2455 | mdname(mddev), s, |
2456 | (unsigned long long)(sect + | |
2457 | rdev->data_offset), | |
913cce5a | 2458 | rdev->bdev); |
867868fb | 2459 | } |
707a6a42 | 2460 | rdev_dec_pending(rdev, mddev); |
2d32777d | 2461 | } |
867868fb N |
2462 | } |
2463 | sectors -= s; | |
2464 | sect += s; | |
2465 | } | |
2466 | } | |
2467 | ||
9f2c9d12 | 2468 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2469 | { |
fd01b88c | 2470 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2471 | struct r1conf *conf = mddev->private; |
3cb03002 | 2472 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2473 | |
2474 | /* bio has the data to be written to device 'i' where | |
2475 | * we just recently had a write error. | |
2476 | * We repeatedly clone the bio and trim down to one block, | |
2477 | * then try the write. Where the write fails we record | |
2478 | * a bad block. | |
2479 | * It is conceivable that the bio doesn't exactly align with | |
2480 | * blocks. We must handle this somehow. | |
2481 | * | |
2482 | * We currently own a reference on the rdev. | |
2483 | */ | |
2484 | ||
2485 | int block_sectors; | |
2486 | sector_t sector; | |
2487 | int sectors; | |
2488 | int sect_to_write = r1_bio->sectors; | |
2489 | int ok = 1; | |
2490 | ||
2491 | if (rdev->badblocks.shift < 0) | |
2492 | return 0; | |
2493 | ||
ab713cdc ND |
2494 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2495 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2496 | sector = r1_bio->sector; |
2497 | sectors = ((sector + block_sectors) | |
2498 | & ~(sector_t)(block_sectors - 1)) | |
2499 | - sector; | |
2500 | ||
cd5ff9a1 N |
2501 | while (sect_to_write) { |
2502 | struct bio *wbio; | |
2503 | if (sectors > sect_to_write) | |
2504 | sectors = sect_to_write; | |
2505 | /* Write at 'sector' for 'sectors'*/ | |
2506 | ||
b783863f | 2507 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
abfc426d CH |
2508 | wbio = bio_alloc_clone(rdev->bdev, |
2509 | r1_bio->behind_master_bio, | |
2510 | GFP_NOIO, &mddev->bio_set); | |
b783863f | 2511 | } else { |
abfc426d CH |
2512 | wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio, |
2513 | GFP_NOIO, &mddev->bio_set); | |
b783863f KO |
2514 | } |
2515 | ||
c34b7ac6 | 2516 | wbio->bi_opf = REQ_OP_WRITE; |
4f024f37 KO |
2517 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2518 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2519 | |
6678d83f | 2520 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2521 | wbio->bi_iter.bi_sector += rdev->data_offset; |
4e49ea4a MC |
2522 | |
2523 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2524 | /* failure! */ |
2525 | ok = rdev_set_badblocks(rdev, sector, | |
2526 | sectors, 0) | |
2527 | && ok; | |
2528 | ||
2529 | bio_put(wbio); | |
2530 | sect_to_write -= sectors; | |
2531 | sector += sectors; | |
2532 | sectors = block_sectors; | |
2533 | } | |
2534 | return ok; | |
2535 | } | |
2536 | ||
e8096360 | 2537 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2538 | { |
2539 | int m; | |
2540 | int s = r1_bio->sectors; | |
8f19ccb2 | 2541 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2542 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2543 | struct bio *bio = r1_bio->bios[m]; |
2544 | if (bio->bi_end_io == NULL) | |
2545 | continue; | |
4e4cbee9 | 2546 | if (!bio->bi_status && |
62096bce | 2547 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2548 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2549 | } |
4e4cbee9 | 2550 | if (bio->bi_status && |
62096bce N |
2551 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2552 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2553 | md_error(conf->mddev, rdev); | |
2554 | } | |
2555 | } | |
2556 | put_buf(r1_bio); | |
2557 | md_done_sync(conf->mddev, s, 1); | |
2558 | } | |
2559 | ||
e8096360 | 2560 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2561 | { |
fd76863e | 2562 | int m, idx; |
55ce74d4 | 2563 | bool fail = false; |
fd76863e | 2564 | |
8f19ccb2 | 2565 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2566 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2567 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2568 | rdev_clear_badblocks(rdev, |
2569 | r1_bio->sector, | |
c6563a8c | 2570 | r1_bio->sectors, 0); |
62096bce N |
2571 | rdev_dec_pending(rdev, conf->mddev); |
2572 | } else if (r1_bio->bios[m] != NULL) { | |
2573 | /* This drive got a write error. We need to | |
2574 | * narrow down and record precise write | |
2575 | * errors. | |
2576 | */ | |
55ce74d4 | 2577 | fail = true; |
62096bce N |
2578 | if (!narrow_write_error(r1_bio, m)) { |
2579 | md_error(conf->mddev, | |
2580 | conf->mirrors[m].rdev); | |
2581 | /* an I/O failed, we can't clear the bitmap */ | |
2582 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2583 | } | |
2584 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2585 | conf->mddev); | |
2586 | } | |
55ce74d4 N |
2587 | if (fail) { |
2588 | spin_lock_irq(&conf->device_lock); | |
2589 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
fd76863e | 2590 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2591 | atomic_inc(&conf->nr_queued[idx]); |
55ce74d4 | 2592 | spin_unlock_irq(&conf->device_lock); |
824e47da | 2593 | /* |
2594 | * In case freeze_array() is waiting for condition | |
2595 | * get_unqueued_pending() == extra to be true. | |
2596 | */ | |
2597 | wake_up(&conf->wait_barrier); | |
55ce74d4 | 2598 | md_wakeup_thread(conf->mddev->thread); |
bd8688a1 N |
2599 | } else { |
2600 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2601 | close_write(r1_bio); | |
55ce74d4 | 2602 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2603 | } |
62096bce N |
2604 | } |
2605 | ||
e8096360 | 2606 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2607 | { |
fd01b88c | 2608 | struct mddev *mddev = conf->mddev; |
62096bce | 2609 | struct bio *bio; |
3cb03002 | 2610 | struct md_rdev *rdev; |
c069da44 | 2611 | sector_t sector; |
62096bce N |
2612 | |
2613 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2614 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2615 | * the block and we can fix it. | |
2616 | * We freeze all other IO, and try reading the block from | |
2617 | * other devices. When we find one, we re-write | |
2618 | * and check it that fixes the read error. | |
2619 | * This is all done synchronously while the array is | |
2620 | * frozen | |
2621 | */ | |
7449f699 TM |
2622 | |
2623 | bio = r1_bio->bios[r1_bio->read_disk]; | |
7449f699 TM |
2624 | bio_put(bio); |
2625 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2626 | ||
2e52d449 N |
2627 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2628 | if (mddev->ro == 0 | |
2629 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2630 | freeze_array(conf, 1); |
ca294b34 | 2631 | fix_read_error(conf, r1_bio); |
62096bce | 2632 | unfreeze_array(conf); |
b33d1062 GK |
2633 | } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { |
2634 | md_error(mddev, rdev); | |
7449f699 TM |
2635 | } else { |
2636 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2637 | } | |
2638 | ||
2e52d449 | 2639 | rdev_dec_pending(rdev, conf->mddev); |
c069da44 | 2640 | sector = r1_bio->sector; |
689389a0 | 2641 | bio = r1_bio->master_bio; |
62096bce | 2642 | |
689389a0 N |
2643 | /* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */ |
2644 | r1_bio->state = 0; | |
2645 | raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); | |
c069da44 | 2646 | allow_barrier(conf, sector); |
62096bce N |
2647 | } |
2648 | ||
4ed8731d | 2649 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2650 | { |
4ed8731d | 2651 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2652 | struct r1bio *r1_bio; |
1da177e4 | 2653 | unsigned long flags; |
e8096360 | 2654 | struct r1conf *conf = mddev->private; |
1da177e4 | 2655 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2656 | struct blk_plug plug; |
fd76863e | 2657 | int idx; |
1da177e4 LT |
2658 | |
2659 | md_check_recovery(mddev); | |
e1dfa0a2 | 2660 | |
55ce74d4 | 2661 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2662 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2663 | LIST_HEAD(tmp); |
2664 | spin_lock_irqsave(&conf->device_lock, flags); | |
fd76863e | 2665 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
2666 | list_splice_init(&conf->bio_end_io_list, &tmp); | |
55ce74d4 N |
2667 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2668 | while (!list_empty(&tmp)) { | |
a452744b MP |
2669 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2670 | retry_list); | |
55ce74d4 | 2671 | list_del(&r1_bio->retry_list); |
fd76863e | 2672 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2673 | atomic_dec(&conf->nr_queued[idx]); |
bd8688a1 N |
2674 | if (mddev->degraded) |
2675 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2676 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2677 | close_write(r1_bio); | |
55ce74d4 N |
2678 | raid_end_bio_io(r1_bio); |
2679 | } | |
2680 | } | |
2681 | ||
e1dfa0a2 | 2682 | blk_start_plug(&plug); |
1da177e4 | 2683 | for (;;) { |
191ea9b2 | 2684 | |
0021b7bc | 2685 | flush_pending_writes(conf); |
191ea9b2 | 2686 | |
a35e63ef N |
2687 | spin_lock_irqsave(&conf->device_lock, flags); |
2688 | if (list_empty(head)) { | |
2689 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2690 | break; |
a35e63ef | 2691 | } |
9f2c9d12 | 2692 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2693 | list_del(head->prev); |
fd76863e | 2694 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2695 | atomic_dec(&conf->nr_queued[idx]); |
1da177e4 LT |
2696 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2697 | ||
2698 | mddev = r1_bio->mddev; | |
070ec55d | 2699 | conf = mddev->private; |
4367af55 | 2700 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2701 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2702 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2703 | handle_sync_write_finished(conf, r1_bio); | |
2704 | else | |
4367af55 | 2705 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2706 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2707 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2708 | handle_write_finished(conf, r1_bio); | |
2709 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2710 | handle_read_error(conf, r1_bio); | |
2711 | else | |
c230e7e5 | 2712 | WARN_ON_ONCE(1); |
62096bce | 2713 | |
1d9d5241 | 2714 | cond_resched(); |
2953079c | 2715 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2716 | md_check_recovery(mddev); |
1da177e4 | 2717 | } |
e1dfa0a2 | 2718 | blk_finish_plug(&plug); |
1da177e4 LT |
2719 | } |
2720 | ||
e8096360 | 2721 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2722 | { |
2723 | int buffs; | |
2724 | ||
2725 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 KO |
2726 | BUG_ON(mempool_initialized(&conf->r1buf_pool)); |
2727 | ||
2728 | return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, | |
2729 | r1buf_pool_free, conf->poolinfo); | |
1da177e4 LT |
2730 | } |
2731 | ||
208410b5 SL |
2732 | static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf) |
2733 | { | |
afeee514 | 2734 | struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); |
208410b5 SL |
2735 | struct resync_pages *rps; |
2736 | struct bio *bio; | |
2737 | int i; | |
2738 | ||
2739 | for (i = conf->poolinfo->raid_disks; i--; ) { | |
2740 | bio = r1bio->bios[i]; | |
2741 | rps = bio->bi_private; | |
a7c50c94 | 2742 | bio_reset(bio, NULL, 0); |
208410b5 SL |
2743 | bio->bi_private = rps; |
2744 | } | |
2745 | r1bio->master_bio = NULL; | |
2746 | return r1bio; | |
2747 | } | |
2748 | ||
1da177e4 LT |
2749 | /* |
2750 | * perform a "sync" on one "block" | |
2751 | * | |
2752 | * We need to make sure that no normal I/O request - particularly write | |
2753 | * requests - conflict with active sync requests. | |
2754 | * | |
2755 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2756 | * that can be installed to exclude normal IO requests. | |
2757 | */ | |
2758 | ||
849674e4 SL |
2759 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2760 | int *skipped) | |
1da177e4 | 2761 | { |
e8096360 | 2762 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2763 | struct r1bio *r1_bio; |
1da177e4 LT |
2764 | struct bio *bio; |
2765 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2766 | int disk = -1; |
1da177e4 | 2767 | int i; |
3e198f78 N |
2768 | int wonly = -1; |
2769 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2770 | sector_t sync_blocks; |
e3b9703e | 2771 | int still_degraded = 0; |
06f60385 N |
2772 | int good_sectors = RESYNC_SECTORS; |
2773 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
fd76863e | 2774 | int idx = sector_to_idx(sector_nr); |
022e510f | 2775 | int page_idx = 0; |
1da177e4 | 2776 | |
afeee514 | 2777 | if (!mempool_initialized(&conf->r1buf_pool)) |
1da177e4 | 2778 | if (init_resync(conf)) |
57afd89f | 2779 | return 0; |
1da177e4 | 2780 | |
58c0fed4 | 2781 | max_sector = mddev->dev_sectors; |
1da177e4 | 2782 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2783 | /* If we aborted, we need to abort the |
2784 | * sync on the 'current' bitmap chunk (there will | |
2785 | * only be one in raid1 resync. | |
2786 | * We can find the current addess in mddev->curr_resync | |
2787 | */ | |
6a806c51 | 2788 | if (mddev->curr_resync < max_sector) /* aborted */ |
e64e4018 AS |
2789 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
2790 | &sync_blocks, 1); | |
6a806c51 | 2791 | else /* completed sync */ |
191ea9b2 | 2792 | conf->fullsync = 0; |
6a806c51 | 2793 | |
e64e4018 | 2794 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2795 | close_sync(conf); |
c40f341f GR |
2796 | |
2797 | if (mddev_is_clustered(mddev)) { | |
2798 | conf->cluster_sync_low = 0; | |
2799 | conf->cluster_sync_high = 0; | |
c40f341f | 2800 | } |
1da177e4 LT |
2801 | return 0; |
2802 | } | |
2803 | ||
07d84d10 N |
2804 | if (mddev->bitmap == NULL && |
2805 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2806 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2807 | conf->fullsync == 0) { |
2808 | *skipped = 1; | |
2809 | return max_sector - sector_nr; | |
2810 | } | |
6394cca5 N |
2811 | /* before building a request, check if we can skip these blocks.. |
2812 | * This call the bitmap_start_sync doesn't actually record anything | |
2813 | */ | |
e64e4018 | 2814 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2815 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2816 | /* We can skip this block, and probably several more */ |
2817 | *skipped = 1; | |
2818 | return sync_blocks; | |
2819 | } | |
17999be4 | 2820 | |
7ac50447 TM |
2821 | /* |
2822 | * If there is non-resync activity waiting for a turn, then let it | |
2823 | * though before starting on this new sync request. | |
2824 | */ | |
824e47da | 2825 | if (atomic_read(&conf->nr_waiting[idx])) |
7ac50447 TM |
2826 | schedule_timeout_uninterruptible(1); |
2827 | ||
c40f341f GR |
2828 | /* we are incrementing sector_nr below. To be safe, we check against |
2829 | * sector_nr + two times RESYNC_SECTORS | |
2830 | */ | |
2831 | ||
e64e4018 | 2832 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
c40f341f | 2833 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); |
17999be4 | 2834 | |
8c242593 YY |
2835 | |
2836 | if (raise_barrier(conf, sector_nr)) | |
2837 | return 0; | |
2838 | ||
2839 | r1_bio = raid1_alloc_init_r1buf(conf); | |
1da177e4 LT |
2840 | |
2841 | /* | |
3e198f78 N |
2842 | * If we get a correctably read error during resync or recovery, |
2843 | * we might want to read from a different device. So we | |
2844 | * flag all drives that could conceivably be read from for READ, | |
2845 | * and any others (which will be non-In_sync devices) for WRITE. | |
2846 | * If a read fails, we try reading from something else for which READ | |
2847 | * is OK. | |
1da177e4 | 2848 | */ |
1da177e4 | 2849 | |
1da177e4 LT |
2850 | r1_bio->mddev = mddev; |
2851 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2852 | r1_bio->state = 0; |
1da177e4 | 2853 | set_bit(R1BIO_IsSync, &r1_bio->state); |
fd76863e | 2854 | /* make sure good_sectors won't go across barrier unit boundary */ |
2855 | good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); | |
1da177e4 | 2856 | |
8f19ccb2 | 2857 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2858 | struct md_rdev *rdev; |
1da177e4 | 2859 | bio = r1_bio->bios[i]; |
1da177e4 | 2860 | |
2d32777d | 2861 | rdev = conf->mirrors[i].rdev; |
3e198f78 | 2862 | if (rdev == NULL || |
06f60385 | 2863 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2864 | if (i < conf->raid_disks) |
2865 | still_degraded = 1; | |
3e198f78 | 2866 | } else if (!test_bit(In_sync, &rdev->flags)) { |
c34b7ac6 | 2867 | bio->bi_opf = REQ_OP_WRITE; |
1da177e4 LT |
2868 | bio->bi_end_io = end_sync_write; |
2869 | write_targets ++; | |
3e198f78 N |
2870 | } else { |
2871 | /* may need to read from here */ | |
06f60385 N |
2872 | sector_t first_bad = MaxSector; |
2873 | int bad_sectors; | |
2874 | ||
2875 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2876 | &first_bad, &bad_sectors)) { | |
2877 | if (first_bad > sector_nr) | |
2878 | good_sectors = first_bad - sector_nr; | |
2879 | else { | |
2880 | bad_sectors -= (sector_nr - first_bad); | |
2881 | if (min_bad == 0 || | |
2882 | min_bad > bad_sectors) | |
2883 | min_bad = bad_sectors; | |
2884 | } | |
2885 | } | |
2886 | if (sector_nr < first_bad) { | |
2887 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2888 | if (wonly < 0) | |
2889 | wonly = i; | |
2890 | } else { | |
2891 | if (disk < 0) | |
2892 | disk = i; | |
2893 | } | |
c34b7ac6 | 2894 | bio->bi_opf = REQ_OP_READ; |
06f60385 N |
2895 | bio->bi_end_io = end_sync_read; |
2896 | read_targets++; | |
d57368af AL |
2897 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2898 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2899 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2900 | /* | |
2901 | * The device is suitable for reading (InSync), | |
2902 | * but has bad block(s) here. Let's try to correct them, | |
2903 | * if we are doing resync or repair. Otherwise, leave | |
2904 | * this device alone for this sync request. | |
2905 | */ | |
c34b7ac6 | 2906 | bio->bi_opf = REQ_OP_WRITE; |
d57368af AL |
2907 | bio->bi_end_io = end_sync_write; |
2908 | write_targets++; | |
3e198f78 | 2909 | } |
3e198f78 | 2910 | } |
028288df | 2911 | if (rdev && bio->bi_end_io) { |
06f60385 | 2912 | atomic_inc(&rdev->nr_pending); |
4f024f37 | 2913 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
74d46992 | 2914 | bio_set_dev(bio, rdev->bdev); |
2e52d449 N |
2915 | if (test_bit(FailFast, &rdev->flags)) |
2916 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2917 | } |
1da177e4 | 2918 | } |
3e198f78 N |
2919 | if (disk < 0) |
2920 | disk = wonly; | |
2921 | r1_bio->read_disk = disk; | |
191ea9b2 | 2922 | |
06f60385 N |
2923 | if (read_targets == 0 && min_bad > 0) { |
2924 | /* These sectors are bad on all InSync devices, so we | |
2925 | * need to mark them bad on all write targets | |
2926 | */ | |
2927 | int ok = 1; | |
8f19ccb2 | 2928 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2929 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2930 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2931 | ok = rdev_set_badblocks(rdev, sector_nr, |
2932 | min_bad, 0 | |
2933 | ) && ok; | |
2934 | } | |
2953079c | 2935 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2936 | *skipped = 1; |
2937 | put_buf(r1_bio); | |
2938 | ||
2939 | if (!ok) { | |
2940 | /* Cannot record the badblocks, so need to | |
2941 | * abort the resync. | |
2942 | * If there are multiple read targets, could just | |
2943 | * fail the really bad ones ??? | |
2944 | */ | |
2945 | conf->recovery_disabled = mddev->recovery_disabled; | |
2946 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2947 | return 0; | |
2948 | } else | |
2949 | return min_bad; | |
2950 | ||
2951 | } | |
2952 | if (min_bad > 0 && min_bad < good_sectors) { | |
2953 | /* only resync enough to reach the next bad->good | |
2954 | * transition */ | |
2955 | good_sectors = min_bad; | |
2956 | } | |
2957 | ||
3e198f78 N |
2958 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2959 | /* extra read targets are also write targets */ | |
2960 | write_targets += read_targets-1; | |
2961 | ||
2962 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2963 | /* There is nowhere to write, so all non-sync |
2964 | * drives must be failed - so we are finished | |
2965 | */ | |
b7219ccb N |
2966 | sector_t rv; |
2967 | if (min_bad > 0) | |
2968 | max_sector = sector_nr + min_bad; | |
2969 | rv = max_sector - sector_nr; | |
57afd89f | 2970 | *skipped = 1; |
1da177e4 | 2971 | put_buf(r1_bio); |
1da177e4 LT |
2972 | return rv; |
2973 | } | |
2974 | ||
c6207277 N |
2975 | if (max_sector > mddev->resync_max) |
2976 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2977 | if (max_sector > sector_nr + good_sectors) |
2978 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2979 | nr_sectors = 0; |
289e99e8 | 2980 | sync_blocks = 0; |
1da177e4 LT |
2981 | do { |
2982 | struct page *page; | |
2983 | int len = PAGE_SIZE; | |
2984 | if (sector_nr + (len>>9) > max_sector) | |
2985 | len = (max_sector - sector_nr) << 9; | |
2986 | if (len == 0) | |
2987 | break; | |
6a806c51 | 2988 | if (sync_blocks == 0) { |
e64e4018 AS |
2989 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
2990 | &sync_blocks, still_degraded) && | |
e5de485f N |
2991 | !conf->fullsync && |
2992 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2993 | break; |
7571ae88 | 2994 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2995 | len = sync_blocks<<9; |
ab7a30c7 | 2996 | } |
191ea9b2 | 2997 | |
8f19ccb2 | 2998 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
98d30c58 ML |
2999 | struct resync_pages *rp; |
3000 | ||
1da177e4 | 3001 | bio = r1_bio->bios[i]; |
98d30c58 | 3002 | rp = get_resync_pages(bio); |
1da177e4 | 3003 | if (bio->bi_end_io) { |
022e510f | 3004 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
3005 | |
3006 | /* | |
3007 | * won't fail because the vec table is big | |
3008 | * enough to hold all these pages | |
3009 | */ | |
f8312322 | 3010 | __bio_add_page(bio, page, len, 0); |
1da177e4 LT |
3011 | } |
3012 | } | |
3013 | nr_sectors += len>>9; | |
3014 | sector_nr += len>>9; | |
191ea9b2 | 3015 | sync_blocks -= (len>>9); |
022e510f | 3016 | } while (++page_idx < RESYNC_PAGES); |
98d30c58 | 3017 | |
1da177e4 LT |
3018 | r1_bio->sectors = nr_sectors; |
3019 | ||
c40f341f GR |
3020 | if (mddev_is_clustered(mddev) && |
3021 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
3022 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
3023 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
3024 | /* Send resync message */ | |
3025 | md_cluster_ops->resync_info_update(mddev, | |
3026 | conf->cluster_sync_low, | |
3027 | conf->cluster_sync_high); | |
3028 | } | |
3029 | ||
d11c171e N |
3030 | /* For a user-requested sync, we read all readable devices and do a |
3031 | * compare | |
3032 | */ | |
3033 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
3034 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 3035 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
3036 | bio = r1_bio->bios[i]; |
3037 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 3038 | read_targets--; |
74d46992 | 3039 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
3040 | if (read_targets == 1) |
3041 | bio->bi_opf &= ~MD_FAILFAST; | |
ed00aabd | 3042 | submit_bio_noacct(bio); |
d11c171e N |
3043 | } |
3044 | } | |
3045 | } else { | |
3046 | atomic_set(&r1_bio->remaining, 1); | |
3047 | bio = r1_bio->bios[r1_bio->read_disk]; | |
74d46992 | 3048 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
3049 | if (read_targets == 1) |
3050 | bio->bi_opf &= ~MD_FAILFAST; | |
ed00aabd | 3051 | submit_bio_noacct(bio); |
d11c171e | 3052 | } |
1da177e4 LT |
3053 | return nr_sectors; |
3054 | } | |
3055 | ||
fd01b88c | 3056 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3057 | { |
3058 | if (sectors) | |
3059 | return sectors; | |
3060 | ||
3061 | return mddev->dev_sectors; | |
3062 | } | |
3063 | ||
e8096360 | 3064 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3065 | { |
e8096360 | 3066 | struct r1conf *conf; |
709ae487 | 3067 | int i; |
0eaf822c | 3068 | struct raid1_info *disk; |
3cb03002 | 3069 | struct md_rdev *rdev; |
709ae487 | 3070 | int err = -ENOMEM; |
1da177e4 | 3071 | |
e8096360 | 3072 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 3073 | if (!conf) |
709ae487 | 3074 | goto abort; |
1da177e4 | 3075 | |
fd76863e | 3076 | conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, |
824e47da | 3077 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3078 | if (!conf->nr_pending) |
3079 | goto abort; | |
3080 | ||
3081 | conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 3082 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3083 | if (!conf->nr_waiting) |
3084 | goto abort; | |
3085 | ||
3086 | conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 3087 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3088 | if (!conf->nr_queued) |
3089 | goto abort; | |
3090 | ||
3091 | conf->barrier = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 3092 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 3093 | if (!conf->barrier) |
3094 | goto abort; | |
3095 | ||
6396bb22 KC |
3096 | conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
3097 | mddev->raid_disks, 2), | |
3098 | GFP_KERNEL); | |
1da177e4 | 3099 | if (!conf->mirrors) |
709ae487 | 3100 | goto abort; |
1da177e4 | 3101 | |
ddaf22ab N |
3102 | conf->tmppage = alloc_page(GFP_KERNEL); |
3103 | if (!conf->tmppage) | |
709ae487 | 3104 | goto abort; |
ddaf22ab | 3105 | |
709ae487 | 3106 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 3107 | if (!conf->poolinfo) |
709ae487 | 3108 | goto abort; |
8f19ccb2 | 3109 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
3f677f9c | 3110 | err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 3111 | rbio_pool_free, conf->poolinfo); |
afeee514 | 3112 | if (err) |
709ae487 N |
3113 | goto abort; |
3114 | ||
afeee514 KO |
3115 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
3116 | if (err) | |
c230e7e5 N |
3117 | goto abort; |
3118 | ||
ed9bfdf1 | 3119 | conf->poolinfo->mddev = mddev; |
1da177e4 | 3120 | |
c19d5798 | 3121 | err = -EINVAL; |
e7e72bf6 | 3122 | spin_lock_init(&conf->device_lock); |
969d6589 | 3123 | conf->raid_disks = mddev->raid_disks; |
dafb20fa | 3124 | rdev_for_each(rdev, mddev) { |
709ae487 | 3125 | int disk_idx = rdev->raid_disk; |
969d6589 YK |
3126 | |
3127 | if (disk_idx >= conf->raid_disks || disk_idx < 0) | |
1da177e4 | 3128 | continue; |
1da177e4 | 3129 | |
969d6589 YK |
3130 | if (!raid1_add_conf(conf, rdev, disk_idx, |
3131 | test_bit(Replacement, &rdev->flags))) | |
c19d5798 | 3132 | goto abort; |
1da177e4 | 3133 | } |
1da177e4 | 3134 | conf->mddev = mddev; |
1da177e4 | 3135 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 3136 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
3137 | |
3138 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 3139 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 3140 | |
191ea9b2 | 3141 | bio_list_init(&conf->pending_bio_list); |
d890fa2b | 3142 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 3143 | |
c19d5798 | 3144 | err = -EIO; |
8f19ccb2 | 3145 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
3146 | |
3147 | disk = conf->mirrors + i; | |
3148 | ||
c19d5798 N |
3149 | if (i < conf->raid_disks && |
3150 | disk[conf->raid_disks].rdev) { | |
3151 | /* This slot has a replacement. */ | |
3152 | if (!disk->rdev) { | |
3153 | /* No original, just make the replacement | |
3154 | * a recovering spare | |
3155 | */ | |
3156 | disk->rdev = | |
3157 | disk[conf->raid_disks].rdev; | |
3158 | disk[conf->raid_disks].rdev = NULL; | |
3159 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
3160 | /* Original is not in_sync - bad */ | |
3161 | goto abort; | |
3162 | } | |
3163 | ||
5fd6c1dc N |
3164 | if (!disk->rdev || |
3165 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 3166 | disk->head_position = 0; |
4f0a5e01 JB |
3167 | if (disk->rdev && |
3168 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 3169 | conf->fullsync = 1; |
be4d3280 | 3170 | } |
1da177e4 | 3171 | } |
709ae487 | 3172 | |
709ae487 | 3173 | err = -ENOMEM; |
44693154 YK |
3174 | rcu_assign_pointer(conf->thread, |
3175 | md_register_thread(raid1d, mddev, "raid1")); | |
1d41c216 | 3176 | if (!conf->thread) |
709ae487 | 3177 | goto abort; |
1da177e4 | 3178 | |
709ae487 N |
3179 | return conf; |
3180 | ||
3181 | abort: | |
3182 | if (conf) { | |
afeee514 | 3183 | mempool_exit(&conf->r1bio_pool); |
709ae487 N |
3184 | kfree(conf->mirrors); |
3185 | safe_put_page(conf->tmppage); | |
3186 | kfree(conf->poolinfo); | |
fd76863e | 3187 | kfree(conf->nr_pending); |
3188 | kfree(conf->nr_waiting); | |
3189 | kfree(conf->nr_queued); | |
3190 | kfree(conf->barrier); | |
afeee514 | 3191 | bioset_exit(&conf->bio_split); |
709ae487 N |
3192 | kfree(conf); |
3193 | } | |
3194 | return ERR_PTR(err); | |
3195 | } | |
3196 | ||
97894f7d CH |
3197 | static int raid1_set_limits(struct mddev *mddev) |
3198 | { | |
3199 | struct queue_limits lim; | |
3200 | ||
3201 | blk_set_stacking_limits(&lim); | |
3202 | lim.max_write_zeroes_sectors = 0; | |
3203 | mddev_stack_rdev_limits(mddev, &lim); | |
396799eb | 3204 | return queue_limits_set(mddev->gendisk->queue, &lim); |
97894f7d CH |
3205 | } |
3206 | ||
afa0f557 | 3207 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 3208 | static int raid1_run(struct mddev *mddev) |
709ae487 | 3209 | { |
e8096360 | 3210 | struct r1conf *conf; |
709ae487 | 3211 | int i; |
5220ea1e | 3212 | int ret; |
709ae487 N |
3213 | |
3214 | if (mddev->level != 1) { | |
1d41c216 N |
3215 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3216 | mdname(mddev), mddev->level); | |
709ae487 N |
3217 | return -EIO; |
3218 | } | |
3219 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3220 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3221 | mdname(mddev)); | |
709ae487 N |
3222 | return -EIO; |
3223 | } | |
b8494823 | 3224 | |
1da177e4 | 3225 | /* |
709ae487 N |
3226 | * copy the already verified devices into our private RAID1 |
3227 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3228 | * should be freed in raid1_free()] |
1da177e4 | 3229 | */ |
709ae487 N |
3230 | if (mddev->private == NULL) |
3231 | conf = setup_conf(mddev); | |
3232 | else | |
3233 | conf = mddev->private; | |
1da177e4 | 3234 | |
709ae487 N |
3235 | if (IS_ERR(conf)) |
3236 | return PTR_ERR(conf); | |
1da177e4 | 3237 | |
176df894 | 3238 | if (!mddev_is_dm(mddev)) { |
97894f7d CH |
3239 | ret = raid1_set_limits(mddev); |
3240 | if (ret) | |
3241 | goto abort; | |
1da177e4 | 3242 | } |
191ea9b2 | 3243 | |
709ae487 | 3244 | mddev->degraded = 0; |
ebfeb444 | 3245 | for (i = 0; i < conf->raid_disks; i++) |
709ae487 N |
3246 | if (conf->mirrors[i].rdev == NULL || |
3247 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3248 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3249 | mddev->degraded++; | |
07f1a685 YY |
3250 | /* |
3251 | * RAID1 needs at least one disk in active | |
3252 | */ | |
3253 | if (conf->raid_disks - mddev->degraded < 1) { | |
7eb8ff02 | 3254 | md_unregister_thread(mddev, &conf->thread); |
07f1a685 YY |
3255 | ret = -EINVAL; |
3256 | goto abort; | |
3257 | } | |
709ae487 N |
3258 | |
3259 | if (conf->raid_disks - mddev->degraded == 1) | |
3260 | mddev->recovery_cp = MaxSector; | |
3261 | ||
8c6ac868 | 3262 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3263 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3264 | mdname(mddev)); | |
3265 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3266 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3267 | mddev->raid_disks); |
709ae487 | 3268 | |
1da177e4 LT |
3269 | /* |
3270 | * Ok, everything is just fine now | |
3271 | */ | |
44693154 YK |
3272 | rcu_assign_pointer(mddev->thread, conf->thread); |
3273 | rcu_assign_pointer(conf->thread, NULL); | |
709ae487 | 3274 | mddev->private = conf; |
46533ff7 | 3275 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3276 | |
1f403624 | 3277 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3278 | |
ebfeb444 | 3279 | ret = md_integrity_register(mddev); |
5aa61f42 | 3280 | if (ret) { |
7eb8ff02 | 3281 | md_unregister_thread(mddev, &mddev->thread); |
07f1a685 | 3282 | goto abort; |
5aa61f42 | 3283 | } |
07f1a685 YY |
3284 | return 0; |
3285 | ||
3286 | abort: | |
3287 | raid1_free(mddev, conf); | |
5220ea1e | 3288 | return ret; |
1da177e4 LT |
3289 | } |
3290 | ||
afa0f557 | 3291 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3292 | { |
afa0f557 | 3293 | struct r1conf *conf = priv; |
409c57f3 | 3294 | |
afeee514 | 3295 | mempool_exit(&conf->r1bio_pool); |
990a8baf | 3296 | kfree(conf->mirrors); |
0fea7ed8 | 3297 | safe_put_page(conf->tmppage); |
990a8baf | 3298 | kfree(conf->poolinfo); |
fd76863e | 3299 | kfree(conf->nr_pending); |
3300 | kfree(conf->nr_waiting); | |
3301 | kfree(conf->nr_queued); | |
3302 | kfree(conf->barrier); | |
afeee514 | 3303 | bioset_exit(&conf->bio_split); |
1da177e4 | 3304 | kfree(conf); |
1da177e4 LT |
3305 | } |
3306 | ||
fd01b88c | 3307 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3308 | { |
3309 | /* no resync is happening, and there is enough space | |
3310 | * on all devices, so we can resize. | |
3311 | * We need to make sure resync covers any new space. | |
3312 | * If the array is shrinking we should possibly wait until | |
3313 | * any io in the removed space completes, but it hardly seems | |
3314 | * worth it. | |
3315 | */ | |
a4a6125a N |
3316 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3317 | if (mddev->external_size && | |
3318 | mddev->array_sectors > newsize) | |
b522adcd | 3319 | return -EINVAL; |
a4a6125a | 3320 | if (mddev->bitmap) { |
e64e4018 | 3321 | int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); |
a4a6125a N |
3322 | if (ret) |
3323 | return ret; | |
3324 | } | |
3325 | md_set_array_sectors(mddev, newsize); | |
b522adcd | 3326 | if (sectors > mddev->dev_sectors && |
b098636c | 3327 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3328 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3329 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3330 | } | |
b522adcd | 3331 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3332 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3333 | return 0; |
3334 | } | |
3335 | ||
fd01b88c | 3336 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3337 | { |
3338 | /* We need to: | |
3339 | * 1/ resize the r1bio_pool | |
3340 | * 2/ resize conf->mirrors | |
3341 | * | |
3342 | * We allocate a new r1bio_pool if we can. | |
3343 | * Then raise a device barrier and wait until all IO stops. | |
3344 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3345 | * |
3346 | * At the same time, we "pack" the devices so that all the missing | |
3347 | * devices have the higher raid_disk numbers. | |
1da177e4 | 3348 | */ |
afeee514 | 3349 | mempool_t newpool, oldpool; |
1da177e4 | 3350 | struct pool_info *newpoolinfo; |
0eaf822c | 3351 | struct raid1_info *newmirrors; |
e8096360 | 3352 | struct r1conf *conf = mddev->private; |
63c70c4f | 3353 | int cnt, raid_disks; |
c04be0aa | 3354 | unsigned long flags; |
2214c260 | 3355 | int d, d2; |
afeee514 KO |
3356 | int ret; |
3357 | ||
3358 | memset(&newpool, 0, sizeof(newpool)); | |
3359 | memset(&oldpool, 0, sizeof(oldpool)); | |
1da177e4 | 3360 | |
63c70c4f | 3361 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3362 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3363 | mddev->layout != mddev->new_layout || |
3364 | mddev->level != mddev->new_level) { | |
664e7c41 | 3365 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3366 | mddev->new_layout = mddev->layout; |
3367 | mddev->new_level = mddev->level; | |
3368 | return -EINVAL; | |
3369 | } | |
3370 | ||
2214c260 AP |
3371 | if (!mddev_is_clustered(mddev)) |
3372 | md_allow_write(mddev); | |
2a2275d6 | 3373 | |
63c70c4f N |
3374 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3375 | ||
6ea9c07c N |
3376 | if (raid_disks < conf->raid_disks) { |
3377 | cnt=0; | |
3378 | for (d= 0; d < conf->raid_disks; d++) | |
3379 | if (conf->mirrors[d].rdev) | |
3380 | cnt++; | |
3381 | if (cnt > raid_disks) | |
1da177e4 | 3382 | return -EBUSY; |
6ea9c07c | 3383 | } |
1da177e4 LT |
3384 | |
3385 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3386 | if (!newpoolinfo) | |
3387 | return -ENOMEM; | |
3388 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3389 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 | 3390 | |
3f677f9c | 3391 | ret = mempool_init(&newpool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 3392 | rbio_pool_free, newpoolinfo); |
afeee514 | 3393 | if (ret) { |
1da177e4 | 3394 | kfree(newpoolinfo); |
afeee514 | 3395 | return ret; |
1da177e4 | 3396 | } |
6396bb22 KC |
3397 | newmirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
3398 | raid_disks, 2), | |
8f19ccb2 | 3399 | GFP_KERNEL); |
1da177e4 LT |
3400 | if (!newmirrors) { |
3401 | kfree(newpoolinfo); | |
afeee514 | 3402 | mempool_exit(&newpool); |
1da177e4 LT |
3403 | return -ENOMEM; |
3404 | } | |
1da177e4 | 3405 | |
e2d59925 | 3406 | freeze_array(conf, 0); |
1da177e4 LT |
3407 | |
3408 | /* ok, everything is stopped */ | |
3409 | oldpool = conf->r1bio_pool; | |
3410 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3411 | |
a88aa786 | 3412 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3413 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3414 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3415 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3416 | rdev->raid_disk = d2; |
36fad858 NK |
3417 | sysfs_unlink_rdev(mddev, rdev); |
3418 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3419 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3420 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3421 | } |
a88aa786 N |
3422 | if (rdev) |
3423 | newmirrors[d2++].rdev = rdev; | |
3424 | } | |
1da177e4 LT |
3425 | kfree(conf->mirrors); |
3426 | conf->mirrors = newmirrors; | |
3427 | kfree(conf->poolinfo); | |
3428 | conf->poolinfo = newpoolinfo; | |
3429 | ||
c04be0aa | 3430 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3431 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3432 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3433 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3434 | mddev->delta_disks = 0; |
1da177e4 | 3435 | |
e2d59925 | 3436 | unfreeze_array(conf); |
1da177e4 | 3437 | |
985ca973 | 3438 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3439 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3440 | md_wakeup_thread(mddev->thread); | |
3441 | ||
afeee514 | 3442 | mempool_exit(&oldpool); |
1da177e4 LT |
3443 | return 0; |
3444 | } | |
3445 | ||
b03e0ccb | 3446 | static void raid1_quiesce(struct mddev *mddev, int quiesce) |
36fa3063 | 3447 | { |
e8096360 | 3448 | struct r1conf *conf = mddev->private; |
36fa3063 | 3449 | |
b03e0ccb | 3450 | if (quiesce) |
07169fd4 | 3451 | freeze_array(conf, 0); |
b03e0ccb | 3452 | else |
07169fd4 | 3453 | unfreeze_array(conf); |
36fa3063 N |
3454 | } |
3455 | ||
fd01b88c | 3456 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3457 | { |
3458 | /* raid1 can take over: | |
3459 | * raid5 with 2 devices, any layout or chunk size | |
3460 | */ | |
3461 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3462 | struct r1conf *conf; |
709ae487 N |
3463 | mddev->new_level = 1; |
3464 | mddev->new_layout = 0; | |
3465 | mddev->new_chunk_sectors = 0; | |
3466 | conf = setup_conf(mddev); | |
6995f0b2 | 3467 | if (!IS_ERR(conf)) { |
07169fd4 | 3468 | /* Array must appear to be quiesced */ |
3469 | conf->array_frozen = 1; | |
394ed8e4 SL |
3470 | mddev_clear_unsupported_flags(mddev, |
3471 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3472 | } |
709ae487 N |
3473 | return conf; |
3474 | } | |
3475 | return ERR_PTR(-EINVAL); | |
3476 | } | |
1da177e4 | 3477 | |
84fc4b56 | 3478 | static struct md_personality raid1_personality = |
1da177e4 LT |
3479 | { |
3480 | .name = "raid1", | |
2604b703 | 3481 | .level = 1, |
1da177e4 | 3482 | .owner = THIS_MODULE, |
849674e4 SL |
3483 | .make_request = raid1_make_request, |
3484 | .run = raid1_run, | |
afa0f557 | 3485 | .free = raid1_free, |
849674e4 SL |
3486 | .status = raid1_status, |
3487 | .error_handler = raid1_error, | |
1da177e4 LT |
3488 | .hot_add_disk = raid1_add_disk, |
3489 | .hot_remove_disk= raid1_remove_disk, | |
3490 | .spare_active = raid1_spare_active, | |
849674e4 | 3491 | .sync_request = raid1_sync_request, |
1da177e4 | 3492 | .resize = raid1_resize, |
80c3a6ce | 3493 | .size = raid1_size, |
63c70c4f | 3494 | .check_reshape = raid1_reshape, |
36fa3063 | 3495 | .quiesce = raid1_quiesce, |
709ae487 | 3496 | .takeover = raid1_takeover, |
1da177e4 LT |
3497 | }; |
3498 | ||
3499 | static int __init raid_init(void) | |
3500 | { | |
2604b703 | 3501 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3502 | } |
3503 | ||
3504 | static void raid_exit(void) | |
3505 | { | |
2604b703 | 3506 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3507 | } |
3508 | ||
3509 | module_init(raid_init); | |
3510 | module_exit(raid_exit); | |
3511 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3512 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3513 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3514 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3515 | MODULE_ALIAS("md-level-1"); |