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 | |
578b54ad N |
49 | #define raid1_log(md, fmt, args...) \ |
50 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) | |
51 | ||
fb0eb5df ML |
52 | #include "raid1-10.c" |
53 | ||
69b00b5b GJ |
54 | #define START(node) ((node)->start) |
55 | #define LAST(node) ((node)->last) | |
56 | INTERVAL_TREE_DEFINE(struct serial_info, node, sector_t, _subtree_last, | |
57 | START, LAST, static inline, raid1_rb); | |
58 | ||
d0d2d8ba GJ |
59 | static int check_and_add_serial(struct md_rdev *rdev, struct r1bio *r1_bio, |
60 | struct serial_info *si, int idx) | |
3e148a32 | 61 | { |
3e148a32 GJ |
62 | unsigned long flags; |
63 | int ret = 0; | |
d0d2d8ba GJ |
64 | sector_t lo = r1_bio->sector; |
65 | sector_t hi = lo + r1_bio->sectors; | |
025471f9 | 66 | struct serial_in_rdev *serial = &rdev->serial[idx]; |
3e148a32 | 67 | |
69b00b5b GJ |
68 | spin_lock_irqsave(&serial->serial_lock, flags); |
69 | /* collision happened */ | |
70 | if (raid1_rb_iter_first(&serial->serial_rb, lo, hi)) | |
71 | ret = -EBUSY; | |
d0d2d8ba | 72 | else { |
69b00b5b GJ |
73 | si->start = lo; |
74 | si->last = hi; | |
75 | raid1_rb_insert(si, &serial->serial_rb); | |
d0d2d8ba | 76 | } |
69b00b5b | 77 | spin_unlock_irqrestore(&serial->serial_lock, flags); |
3e148a32 GJ |
78 | |
79 | return ret; | |
80 | } | |
81 | ||
d0d2d8ba GJ |
82 | static void wait_for_serialization(struct md_rdev *rdev, struct r1bio *r1_bio) |
83 | { | |
84 | struct mddev *mddev = rdev->mddev; | |
85 | struct serial_info *si; | |
86 | int idx = sector_to_idx(r1_bio->sector); | |
87 | struct serial_in_rdev *serial = &rdev->serial[idx]; | |
88 | ||
89 | if (WARN_ON(!mddev->serial_info_pool)) | |
90 | return; | |
91 | si = mempool_alloc(mddev->serial_info_pool, GFP_NOIO); | |
92 | wait_event(serial->serial_io_wait, | |
93 | check_and_add_serial(rdev, r1_bio, si, idx) == 0); | |
94 | } | |
95 | ||
404659cf | 96 | static void remove_serial(struct md_rdev *rdev, sector_t lo, sector_t hi) |
3e148a32 | 97 | { |
69b00b5b | 98 | struct serial_info *si; |
3e148a32 GJ |
99 | unsigned long flags; |
100 | int found = 0; | |
101 | struct mddev *mddev = rdev->mddev; | |
025471f9 GJ |
102 | int idx = sector_to_idx(lo); |
103 | struct serial_in_rdev *serial = &rdev->serial[idx]; | |
69b00b5b GJ |
104 | |
105 | spin_lock_irqsave(&serial->serial_lock, flags); | |
106 | for (si = raid1_rb_iter_first(&serial->serial_rb, lo, hi); | |
107 | si; si = raid1_rb_iter_next(si, lo, hi)) { | |
108 | if (si->start == lo && si->last == hi) { | |
109 | raid1_rb_remove(si, &serial->serial_rb); | |
110 | mempool_free(si, mddev->serial_info_pool); | |
3e148a32 GJ |
111 | found = 1; |
112 | break; | |
113 | } | |
69b00b5b | 114 | } |
3e148a32 | 115 | if (!found) |
404659cf | 116 | WARN(1, "The write IO is not recorded for serialization\n"); |
69b00b5b GJ |
117 | spin_unlock_irqrestore(&serial->serial_lock, flags); |
118 | wake_up(&serial->serial_io_wait); | |
3e148a32 GJ |
119 | } |
120 | ||
98d30c58 ML |
121 | /* |
122 | * for resync bio, r1bio pointer can be retrieved from the per-bio | |
123 | * 'struct resync_pages'. | |
124 | */ | |
125 | static inline struct r1bio *get_resync_r1bio(struct bio *bio) | |
126 | { | |
127 | return get_resync_pages(bio)->raid_bio; | |
128 | } | |
129 | ||
dd0fc66f | 130 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
131 | { |
132 | struct pool_info *pi = data; | |
9f2c9d12 | 133 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
134 | |
135 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 136 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
137 | } |
138 | ||
8e005f7c | 139 | #define RESYNC_DEPTH 32 |
1da177e4 | 140 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
8e005f7c | 141 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
142 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
143 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
144 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
1da177e4 | 145 | |
dd0fc66f | 146 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
147 | { |
148 | struct pool_info *pi = data; | |
9f2c9d12 | 149 | struct r1bio *r1_bio; |
1da177e4 | 150 | struct bio *bio; |
da1aab3d | 151 | int need_pages; |
98d30c58 ML |
152 | int j; |
153 | struct resync_pages *rps; | |
1da177e4 LT |
154 | |
155 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 156 | if (!r1_bio) |
1da177e4 | 157 | return NULL; |
1da177e4 | 158 | |
6da2ec56 KC |
159 | rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages), |
160 | gfp_flags); | |
98d30c58 ML |
161 | if (!rps) |
162 | goto out_free_r1bio; | |
163 | ||
1da177e4 LT |
164 | /* |
165 | * Allocate bios : 1 for reading, n-1 for writing | |
166 | */ | |
167 | for (j = pi->raid_disks ; j-- ; ) { | |
066ff571 | 168 | bio = bio_kmalloc(RESYNC_PAGES, gfp_flags); |
1da177e4 LT |
169 | if (!bio) |
170 | goto out_free_bio; | |
066ff571 | 171 | bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); |
1da177e4 LT |
172 | r1_bio->bios[j] = bio; |
173 | } | |
174 | /* | |
175 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
176 | * the first bio. |
177 | * If this is a user-requested check/repair, allocate | |
178 | * RESYNC_PAGES for each bio. | |
1da177e4 | 179 | */ |
d11c171e | 180 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 181 | need_pages = pi->raid_disks; |
d11c171e | 182 | else |
da1aab3d | 183 | need_pages = 1; |
98d30c58 ML |
184 | for (j = 0; j < pi->raid_disks; j++) { |
185 | struct resync_pages *rp = &rps[j]; | |
186 | ||
d11c171e | 187 | bio = r1_bio->bios[j]; |
d11c171e | 188 | |
98d30c58 ML |
189 | if (j < need_pages) { |
190 | if (resync_alloc_pages(rp, gfp_flags)) | |
191 | goto out_free_pages; | |
192 | } else { | |
193 | memcpy(rp, &rps[0], sizeof(*rp)); | |
194 | resync_get_all_pages(rp); | |
195 | } | |
196 | ||
98d30c58 ML |
197 | rp->raid_bio = r1_bio; |
198 | bio->bi_private = rp; | |
1da177e4 LT |
199 | } |
200 | ||
201 | r1_bio->master_bio = NULL; | |
202 | ||
203 | return r1_bio; | |
204 | ||
da1aab3d | 205 | out_free_pages: |
491221f8 | 206 | while (--j >= 0) |
98d30c58 | 207 | resync_free_pages(&rps[j]); |
da1aab3d | 208 | |
1da177e4 | 209 | out_free_bio: |
066ff571 CH |
210 | while (++j < pi->raid_disks) { |
211 | bio_uninit(r1_bio->bios[j]); | |
212 | kfree(r1_bio->bios[j]); | |
213 | } | |
98d30c58 ML |
214 | kfree(rps); |
215 | ||
216 | out_free_r1bio: | |
c7afa803 | 217 | rbio_pool_free(r1_bio, data); |
1da177e4 LT |
218 | return NULL; |
219 | } | |
220 | ||
221 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
222 | { | |
223 | struct pool_info *pi = data; | |
98d30c58 | 224 | int i; |
9f2c9d12 | 225 | struct r1bio *r1bio = __r1_bio; |
98d30c58 | 226 | struct resync_pages *rp = NULL; |
1da177e4 | 227 | |
98d30c58 ML |
228 | for (i = pi->raid_disks; i--; ) { |
229 | rp = get_resync_pages(r1bio->bios[i]); | |
230 | resync_free_pages(rp); | |
066ff571 CH |
231 | bio_uninit(r1bio->bios[i]); |
232 | kfree(r1bio->bios[i]); | |
98d30c58 ML |
233 | } |
234 | ||
235 | /* resync pages array stored in the 1st bio's .bi_private */ | |
236 | kfree(rp); | |
1da177e4 | 237 | |
c7afa803 | 238 | rbio_pool_free(r1bio, data); |
1da177e4 LT |
239 | } |
240 | ||
e8096360 | 241 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
242 | { |
243 | int i; | |
244 | ||
8f19ccb2 | 245 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 246 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 247 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
248 | bio_put(*bio); |
249 | *bio = NULL; | |
250 | } | |
251 | } | |
252 | ||
9f2c9d12 | 253 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 254 | { |
e8096360 | 255 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 256 | |
1da177e4 | 257 | put_all_bios(conf, r1_bio); |
afeee514 | 258 | mempool_free(r1_bio, &conf->r1bio_pool); |
1da177e4 LT |
259 | } |
260 | ||
9f2c9d12 | 261 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 262 | { |
e8096360 | 263 | struct r1conf *conf = r1_bio->mddev->private; |
af5f42a7 | 264 | sector_t sect = r1_bio->sector; |
3e198f78 N |
265 | int i; |
266 | ||
8f19ccb2 | 267 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
268 | struct bio *bio = r1_bio->bios[i]; |
269 | if (bio->bi_end_io) | |
270 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
271 | } | |
1da177e4 | 272 | |
afeee514 | 273 | mempool_free(r1_bio, &conf->r1buf_pool); |
1da177e4 | 274 | |
af5f42a7 | 275 | lower_barrier(conf, sect); |
1da177e4 LT |
276 | } |
277 | ||
9f2c9d12 | 278 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
279 | { |
280 | unsigned long flags; | |
fd01b88c | 281 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 282 | struct r1conf *conf = mddev->private; |
fd76863e | 283 | int idx; |
1da177e4 | 284 | |
fd76863e | 285 | idx = sector_to_idx(r1_bio->sector); |
1da177e4 LT |
286 | spin_lock_irqsave(&conf->device_lock, flags); |
287 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
824e47da | 288 | atomic_inc(&conf->nr_queued[idx]); |
1da177e4 LT |
289 | spin_unlock_irqrestore(&conf->device_lock, flags); |
290 | ||
17999be4 | 291 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
292 | md_wakeup_thread(mddev->thread); |
293 | } | |
294 | ||
295 | /* | |
296 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
297 | * operation and are ready to return a success/failure code to the buffer | |
298 | * cache layer. | |
299 | */ | |
9f2c9d12 | 300 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
301 | { |
302 | struct bio *bio = r1_bio->master_bio; | |
d2eb35ac N |
303 | |
304 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4e4cbee9 | 305 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 306 | |
37011e3a | 307 | bio_endio(bio); |
d2eb35ac N |
308 | } |
309 | ||
9f2c9d12 | 310 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
311 | { |
312 | struct bio *bio = r1_bio->master_bio; | |
c91114c2 | 313 | struct r1conf *conf = r1_bio->mddev->private; |
c5d736f5 | 314 | sector_t sector = r1_bio->sector; |
1da177e4 | 315 | |
4b6d287f N |
316 | /* if nobody has done the final endio yet, do it now */ |
317 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
318 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
319 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
320 | (unsigned long long) bio->bi_iter.bi_sector, |
321 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 322 | |
d2eb35ac | 323 | call_bio_endio(r1_bio); |
4b6d287f | 324 | } |
c5d736f5 XH |
325 | |
326 | free_r1bio(r1_bio); | |
c91114c2 DJ |
327 | /* |
328 | * Wake up any possible resync thread that waits for the device | |
329 | * to go idle. All I/Os, even write-behind writes, are done. | |
330 | */ | |
c5d736f5 | 331 | allow_barrier(conf, sector); |
1da177e4 LT |
332 | } |
333 | ||
334 | /* | |
335 | * Update disk head position estimator based on IRQ completion info. | |
336 | */ | |
9f2c9d12 | 337 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 338 | { |
e8096360 | 339 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
340 | |
341 | conf->mirrors[disk].head_position = | |
342 | r1_bio->sector + (r1_bio->sectors); | |
343 | } | |
344 | ||
ba3ae3be NK |
345 | /* |
346 | * Find the disk number which triggered given bio | |
347 | */ | |
9f2c9d12 | 348 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
349 | { |
350 | int mirror; | |
30194636 N |
351 | struct r1conf *conf = r1_bio->mddev->private; |
352 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 353 | |
8f19ccb2 | 354 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
355 | if (r1_bio->bios[mirror] == bio) |
356 | break; | |
357 | ||
8f19ccb2 | 358 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
359 | update_head_pos(mirror, r1_bio); |
360 | ||
361 | return mirror; | |
362 | } | |
363 | ||
4246a0b6 | 364 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 365 | { |
4e4cbee9 | 366 | int uptodate = !bio->bi_status; |
9f2c9d12 | 367 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 368 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 369 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 370 | |
1da177e4 LT |
371 | /* |
372 | * this branch is our 'one mirror IO has finished' event handler: | |
373 | */ | |
e5872d58 | 374 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 375 | |
dd00a99e N |
376 | if (uptodate) |
377 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
378 | else if (test_bit(FailFast, &rdev->flags) && |
379 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
380 | /* This was a fail-fast read so we definitely | |
381 | * want to retry */ | |
382 | ; | |
dd00a99e N |
383 | else { |
384 | /* If all other devices have failed, we want to return | |
385 | * the error upwards rather than fail the last device. | |
386 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 387 | */ |
dd00a99e N |
388 | unsigned long flags; |
389 | spin_lock_irqsave(&conf->device_lock, flags); | |
390 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
391 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 392 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
393 | uptodate = 1; |
394 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
395 | } | |
1da177e4 | 396 | |
7ad4d4a6 | 397 | if (uptodate) { |
1da177e4 | 398 | raid_end_bio_io(r1_bio); |
e5872d58 | 399 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 400 | } else { |
1da177e4 LT |
401 | /* |
402 | * oops, read error: | |
403 | */ | |
913cce5a | 404 | pr_err_ratelimited("md/raid1:%s: %pg: rescheduling sector %llu\n", |
1d41c216 | 405 | mdname(conf->mddev), |
913cce5a | 406 | rdev->bdev, |
1d41c216 | 407 | (unsigned long long)r1_bio->sector); |
d2eb35ac | 408 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 409 | reschedule_retry(r1_bio); |
7ad4d4a6 | 410 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 411 | } |
1da177e4 LT |
412 | } |
413 | ||
9f2c9d12 | 414 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
415 | { |
416 | /* it really is the end of this request */ | |
417 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
841c1316 ML |
418 | bio_free_pages(r1_bio->behind_master_bio); |
419 | bio_put(r1_bio->behind_master_bio); | |
420 | r1_bio->behind_master_bio = NULL; | |
cd5ff9a1 N |
421 | } |
422 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
423 | md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
424 | r1_bio->sectors, | |
425 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
426 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
cd5ff9a1 N |
427 | md_write_end(r1_bio->mddev); |
428 | } | |
429 | ||
9f2c9d12 | 430 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 431 | { |
cd5ff9a1 N |
432 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
433 | return; | |
434 | ||
435 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
436 | reschedule_retry(r1_bio); | |
437 | else { | |
438 | close_write(r1_bio); | |
4367af55 N |
439 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
440 | reschedule_retry(r1_bio); | |
441 | else | |
442 | raid_end_bio_io(r1_bio); | |
4e78064f N |
443 | } |
444 | } | |
445 | ||
4246a0b6 | 446 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 447 | { |
9f2c9d12 | 448 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 449 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 450 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 451 | struct bio *to_put = NULL; |
e5872d58 N |
452 | int mirror = find_bio_disk(r1_bio, bio); |
453 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd | 454 | bool discard_error; |
69df9cfc GJ |
455 | sector_t lo = r1_bio->sector; |
456 | sector_t hi = r1_bio->sector + r1_bio->sectors; | |
e3f948cd | 457 | |
4e4cbee9 | 458 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 459 | |
e9c7469b TH |
460 | /* |
461 | * 'one mirror IO has finished' event handler: | |
462 | */ | |
4e4cbee9 | 463 | if (bio->bi_status && !discard_error) { |
e5872d58 N |
464 | set_bit(WriteErrorSeen, &rdev->flags); |
465 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
466 | set_bit(MD_RECOVERY_NEEDED, & |
467 | conf->mddev->recovery); | |
468 | ||
212e7eb7 N |
469 | if (test_bit(FailFast, &rdev->flags) && |
470 | (bio->bi_opf & MD_FAILFAST) && | |
471 | /* We never try FailFast to WriteMostly devices */ | |
472 | !test_bit(WriteMostly, &rdev->flags)) { | |
473 | md_error(r1_bio->mddev, rdev); | |
eeba6809 YY |
474 | } |
475 | ||
476 | /* | |
477 | * When the device is faulty, it is not necessary to | |
478 | * handle write error. | |
eeba6809 YY |
479 | */ |
480 | if (!test_bit(Faulty, &rdev->flags)) | |
212e7eb7 | 481 | set_bit(R1BIO_WriteError, &r1_bio->state); |
eeba6809 | 482 | else { |
2417b986 PC |
483 | /* Fail the request */ |
484 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
eeba6809 YY |
485 | /* Finished with this branch */ |
486 | r1_bio->bios[mirror] = NULL; | |
487 | to_put = bio; | |
488 | } | |
4367af55 | 489 | } else { |
1da177e4 | 490 | /* |
e9c7469b TH |
491 | * Set R1BIO_Uptodate in our master bio, so that we |
492 | * will return a good error code for to the higher | |
493 | * levels even if IO on some other mirrored buffer | |
494 | * fails. | |
495 | * | |
496 | * The 'master' represents the composite IO operation | |
497 | * to user-side. So if something waits for IO, then it | |
498 | * will wait for the 'master' bio. | |
1da177e4 | 499 | */ |
4367af55 N |
500 | sector_t first_bad; |
501 | int bad_sectors; | |
502 | ||
cd5ff9a1 N |
503 | r1_bio->bios[mirror] = NULL; |
504 | to_put = bio; | |
3056e3ae AL |
505 | /* |
506 | * Do not set R1BIO_Uptodate if the current device is | |
507 | * rebuilding or Faulty. This is because we cannot use | |
508 | * such device for properly reading the data back (we could | |
509 | * potentially use it, if the current write would have felt | |
510 | * before rdev->recovery_offset, but for simplicity we don't | |
511 | * check this here. | |
512 | */ | |
e5872d58 N |
513 | if (test_bit(In_sync, &rdev->flags) && |
514 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 515 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 516 | |
4367af55 | 517 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 518 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 519 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
520 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
521 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
522 | } | |
523 | } | |
524 | ||
e9c7469b | 525 | if (behind) { |
69df9cfc | 526 | if (test_bit(CollisionCheck, &rdev->flags)) |
404659cf | 527 | remove_serial(rdev, lo, hi); |
e5872d58 | 528 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
529 | atomic_dec(&r1_bio->behind_remaining); |
530 | ||
531 | /* | |
532 | * In behind mode, we ACK the master bio once the I/O | |
533 | * has safely reached all non-writemostly | |
534 | * disks. Setting the Returned bit ensures that this | |
535 | * gets done only once -- we don't ever want to return | |
536 | * -EIO here, instead we'll wait | |
537 | */ | |
538 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
539 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
540 | /* Maybe we can return now */ | |
541 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
542 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
543 | pr_debug("raid1: behind end write sectors" |
544 | " %llu-%llu\n", | |
4f024f37 KO |
545 | (unsigned long long) mbio->bi_iter.bi_sector, |
546 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 547 | call_bio_endio(r1_bio); |
4b6d287f N |
548 | } |
549 | } | |
69df9cfc GJ |
550 | } else if (rdev->mddev->serialize_policy) |
551 | remove_serial(rdev, lo, hi); | |
4367af55 | 552 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 553 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 554 | |
1da177e4 | 555 | /* |
1da177e4 LT |
556 | * Let's see if all mirrored write operations have finished |
557 | * already. | |
558 | */ | |
af6d7b76 | 559 | r1_bio_write_done(r1_bio); |
c70810b3 | 560 | |
04b857f7 N |
561 | if (to_put) |
562 | bio_put(to_put); | |
1da177e4 LT |
563 | } |
564 | ||
fd76863e | 565 | static sector_t align_to_barrier_unit_end(sector_t start_sector, |
566 | sector_t sectors) | |
567 | { | |
568 | sector_t len; | |
569 | ||
570 | WARN_ON(sectors == 0); | |
571 | /* | |
572 | * len is the number of sectors from start_sector to end of the | |
573 | * barrier unit which start_sector belongs to. | |
574 | */ | |
575 | len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - | |
576 | start_sector; | |
577 | ||
578 | if (len > sectors) | |
579 | len = sectors; | |
580 | ||
581 | return len; | |
582 | } | |
583 | ||
1da177e4 LT |
584 | /* |
585 | * This routine returns the disk from which the requested read should | |
586 | * be done. There is a per-array 'next expected sequential IO' sector | |
587 | * number - if this matches on the next IO then we use the last disk. | |
588 | * There is also a per-disk 'last know head position' sector that is | |
589 | * maintained from IRQ contexts, both the normal and the resync IO | |
590 | * completion handlers update this position correctly. If there is no | |
591 | * perfect sequential match then we pick the disk whose head is closest. | |
592 | * | |
593 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
594 | * because position is mirror, not device based. | |
595 | * | |
596 | * The rdev for the device selected will have nr_pending incremented. | |
597 | */ | |
e8096360 | 598 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 599 | { |
af3a2cd6 | 600 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
601 | int sectors; |
602 | int best_good_sectors; | |
9dedf603 SL |
603 | int best_disk, best_dist_disk, best_pending_disk; |
604 | int has_nonrot_disk; | |
be4d3280 | 605 | int disk; |
76073054 | 606 | sector_t best_dist; |
9dedf603 | 607 | unsigned int min_pending; |
3cb03002 | 608 | struct md_rdev *rdev; |
f3ac8bf7 | 609 | int choose_first; |
12cee5a8 | 610 | int choose_next_idle; |
1da177e4 LT |
611 | |
612 | rcu_read_lock(); | |
613 | /* | |
8ddf9efe | 614 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
615 | * device if no resync is going on, or below the resync window. |
616 | * We take the first readable disk when above the resync window. | |
617 | */ | |
618 | retry: | |
d2eb35ac | 619 | sectors = r1_bio->sectors; |
76073054 | 620 | best_disk = -1; |
9dedf603 | 621 | best_dist_disk = -1; |
76073054 | 622 | best_dist = MaxSector; |
9dedf603 SL |
623 | best_pending_disk = -1; |
624 | min_pending = UINT_MAX; | |
d2eb35ac | 625 | best_good_sectors = 0; |
9dedf603 | 626 | has_nonrot_disk = 0; |
12cee5a8 | 627 | choose_next_idle = 0; |
2e52d449 | 628 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 629 | |
7d49ffcf GR |
630 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
631 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 632 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
633 | this_sector + sectors))) |
634 | choose_first = 1; | |
635 | else | |
636 | choose_first = 0; | |
1da177e4 | 637 | |
be4d3280 | 638 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 639 | sector_t dist; |
d2eb35ac N |
640 | sector_t first_bad; |
641 | int bad_sectors; | |
9dedf603 | 642 | unsigned int pending; |
12cee5a8 | 643 | bool nonrot; |
d2eb35ac | 644 | |
f3ac8bf7 N |
645 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
646 | if (r1_bio->bios[disk] == IO_BLOCKED | |
647 | || rdev == NULL | |
76073054 | 648 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 649 | continue; |
76073054 N |
650 | if (!test_bit(In_sync, &rdev->flags) && |
651 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 652 | continue; |
76073054 N |
653 | if (test_bit(WriteMostly, &rdev->flags)) { |
654 | /* Don't balance among write-mostly, just | |
655 | * use the first as a last resort */ | |
d1901ef0 | 656 | if (best_dist_disk < 0) { |
307729c8 N |
657 | if (is_badblock(rdev, this_sector, sectors, |
658 | &first_bad, &bad_sectors)) { | |
816b0acf | 659 | if (first_bad <= this_sector) |
307729c8 N |
660 | /* Cannot use this */ |
661 | continue; | |
662 | best_good_sectors = first_bad - this_sector; | |
663 | } else | |
664 | best_good_sectors = sectors; | |
d1901ef0 TH |
665 | best_dist_disk = disk; |
666 | best_pending_disk = disk; | |
307729c8 | 667 | } |
76073054 N |
668 | continue; |
669 | } | |
670 | /* This is a reasonable device to use. It might | |
671 | * even be best. | |
672 | */ | |
d2eb35ac N |
673 | if (is_badblock(rdev, this_sector, sectors, |
674 | &first_bad, &bad_sectors)) { | |
675 | if (best_dist < MaxSector) | |
676 | /* already have a better device */ | |
677 | continue; | |
678 | if (first_bad <= this_sector) { | |
679 | /* cannot read here. If this is the 'primary' | |
680 | * device, then we must not read beyond | |
681 | * bad_sectors from another device.. | |
682 | */ | |
683 | bad_sectors -= (this_sector - first_bad); | |
684 | if (choose_first && sectors > bad_sectors) | |
685 | sectors = bad_sectors; | |
686 | if (best_good_sectors > sectors) | |
687 | best_good_sectors = sectors; | |
688 | ||
689 | } else { | |
690 | sector_t good_sectors = first_bad - this_sector; | |
691 | if (good_sectors > best_good_sectors) { | |
692 | best_good_sectors = good_sectors; | |
693 | best_disk = disk; | |
694 | } | |
695 | if (choose_first) | |
696 | break; | |
697 | } | |
698 | continue; | |
d82dd0e3 TM |
699 | } else { |
700 | if ((sectors > best_good_sectors) && (best_disk >= 0)) | |
701 | best_disk = -1; | |
d2eb35ac | 702 | best_good_sectors = sectors; |
d82dd0e3 | 703 | } |
d2eb35ac | 704 | |
2e52d449 N |
705 | if (best_disk >= 0) |
706 | /* At least two disks to choose from so failfast is OK */ | |
707 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
708 | ||
10f0d2a5 | 709 | nonrot = bdev_nonrot(rdev->bdev); |
12cee5a8 | 710 | has_nonrot_disk |= nonrot; |
9dedf603 | 711 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 712 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 713 | if (choose_first) { |
76073054 | 714 | best_disk = disk; |
1da177e4 LT |
715 | break; |
716 | } | |
12cee5a8 SL |
717 | /* Don't change to another disk for sequential reads */ |
718 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
719 | || dist == 0) { | |
720 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
721 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
722 | ||
723 | best_disk = disk; | |
724 | /* | |
725 | * If buffered sequential IO size exceeds optimal | |
726 | * iosize, check if there is idle disk. If yes, choose | |
727 | * the idle disk. read_balance could already choose an | |
728 | * idle disk before noticing it's a sequential IO in | |
729 | * this disk. This doesn't matter because this disk | |
730 | * will idle, next time it will be utilized after the | |
731 | * first disk has IO size exceeds optimal iosize. In | |
732 | * this way, iosize of the first disk will be optimal | |
733 | * iosize at least. iosize of the second disk might be | |
734 | * small, but not a big deal since when the second disk | |
735 | * starts IO, the first disk is likely still busy. | |
736 | */ | |
737 | if (nonrot && opt_iosize > 0 && | |
738 | mirror->seq_start != MaxSector && | |
739 | mirror->next_seq_sect > opt_iosize && | |
740 | mirror->next_seq_sect - opt_iosize >= | |
741 | mirror->seq_start) { | |
742 | choose_next_idle = 1; | |
743 | continue; | |
744 | } | |
745 | break; | |
746 | } | |
12cee5a8 SL |
747 | |
748 | if (choose_next_idle) | |
749 | continue; | |
9dedf603 SL |
750 | |
751 | if (min_pending > pending) { | |
752 | min_pending = pending; | |
753 | best_pending_disk = disk; | |
754 | } | |
755 | ||
76073054 N |
756 | if (dist < best_dist) { |
757 | best_dist = dist; | |
9dedf603 | 758 | best_dist_disk = disk; |
1da177e4 | 759 | } |
f3ac8bf7 | 760 | } |
1da177e4 | 761 | |
9dedf603 SL |
762 | /* |
763 | * If all disks are rotational, choose the closest disk. If any disk is | |
764 | * non-rotational, choose the disk with less pending request even the | |
765 | * disk is rotational, which might/might not be optimal for raids with | |
766 | * mixed ratation/non-rotational disks depending on workload. | |
767 | */ | |
768 | if (best_disk == -1) { | |
2e52d449 | 769 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
770 | best_disk = best_pending_disk; |
771 | else | |
772 | best_disk = best_dist_disk; | |
773 | } | |
774 | ||
76073054 N |
775 | if (best_disk >= 0) { |
776 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
777 | if (!rdev) |
778 | goto retry; | |
779 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 780 | sectors = best_good_sectors; |
12cee5a8 SL |
781 | |
782 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
783 | conf->mirrors[best_disk].seq_start = this_sector; | |
784 | ||
be4d3280 | 785 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
786 | } |
787 | rcu_read_unlock(); | |
d2eb35ac | 788 | *max_sectors = sectors; |
1da177e4 | 789 | |
76073054 | 790 | return best_disk; |
1da177e4 LT |
791 | } |
792 | ||
21bd9a68 JW |
793 | static void wake_up_barrier(struct r1conf *conf) |
794 | { | |
795 | if (wq_has_sleeper(&conf->wait_barrier)) | |
796 | wake_up(&conf->wait_barrier); | |
797 | } | |
798 | ||
673ca68d N |
799 | static void flush_bio_list(struct r1conf *conf, struct bio *bio) |
800 | { | |
801 | /* flush any pending bitmap writes to disk before proceeding w/ I/O */ | |
9efcc2c3 | 802 | raid1_prepare_flush_writes(conf->mddev->bitmap); |
21bd9a68 | 803 | wake_up_barrier(conf); |
673ca68d N |
804 | |
805 | while (bio) { /* submit pending writes */ | |
806 | struct bio *next = bio->bi_next; | |
8295efbe YK |
807 | |
808 | raid1_submit_write(bio); | |
673ca68d | 809 | bio = next; |
5fa4f8ba | 810 | cond_resched(); |
673ca68d N |
811 | } |
812 | } | |
813 | ||
e8096360 | 814 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
815 | { |
816 | /* Any writes that have been queued but are awaiting | |
817 | * bitmap updates get flushed here. | |
a35e63ef | 818 | */ |
a35e63ef N |
819 | spin_lock_irq(&conf->device_lock); |
820 | ||
821 | if (conf->pending_bio_list.head) { | |
18022a1b | 822 | struct blk_plug plug; |
a35e63ef | 823 | struct bio *bio; |
18022a1b | 824 | |
a35e63ef | 825 | bio = bio_list_get(&conf->pending_bio_list); |
a35e63ef | 826 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
827 | |
828 | /* | |
829 | * As this is called in a wait_event() loop (see freeze_array), | |
830 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
831 | * cause a warning when we prepare to wait again. As it is | |
832 | * rare that this path is taken, it is perfectly safe to force | |
833 | * us to go around the wait_event() loop again, so the warning | |
834 | * is a false-positive. Silence the warning by resetting | |
835 | * thread state | |
836 | */ | |
837 | __set_current_state(TASK_RUNNING); | |
18022a1b | 838 | blk_start_plug(&plug); |
673ca68d | 839 | flush_bio_list(conf, bio); |
18022a1b | 840 | blk_finish_plug(&plug); |
a35e63ef N |
841 | } else |
842 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
843 | } |
844 | ||
17999be4 N |
845 | /* Barriers.... |
846 | * Sometimes we need to suspend IO while we do something else, | |
847 | * either some resync/recovery, or reconfigure the array. | |
848 | * To do this we raise a 'barrier'. | |
849 | * The 'barrier' is a counter that can be raised multiple times | |
850 | * to count how many activities are happening which preclude | |
851 | * normal IO. | |
852 | * We can only raise the barrier if there is no pending IO. | |
853 | * i.e. if nr_pending == 0. | |
854 | * We choose only to raise the barrier if no-one is waiting for the | |
855 | * barrier to go down. This means that as soon as an IO request | |
856 | * is ready, no other operations which require a barrier will start | |
857 | * until the IO request has had a chance. | |
858 | * | |
859 | * So: regular IO calls 'wait_barrier'. When that returns there | |
860 | * is no backgroup IO happening, It must arrange to call | |
861 | * allow_barrier when it has finished its IO. | |
862 | * backgroup IO calls must call raise_barrier. Once that returns | |
863 | * there is no normal IO happeing. It must arrange to call | |
864 | * lower_barrier when the particular background IO completes. | |
4675719d HT |
865 | * |
866 | * If resync/recovery is interrupted, returns -EINTR; | |
867 | * Otherwise, returns 0. | |
1da177e4 | 868 | */ |
4675719d | 869 | static int raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 | 870 | { |
fd76863e | 871 | int idx = sector_to_idx(sector_nr); |
872 | ||
1da177e4 | 873 | spin_lock_irq(&conf->resync_lock); |
17999be4 N |
874 | |
875 | /* Wait until no block IO is waiting */ | |
824e47da | 876 | wait_event_lock_irq(conf->wait_barrier, |
877 | !atomic_read(&conf->nr_waiting[idx]), | |
eed8c02e | 878 | conf->resync_lock); |
17999be4 N |
879 | |
880 | /* block any new IO from starting */ | |
824e47da | 881 | atomic_inc(&conf->barrier[idx]); |
882 | /* | |
883 | * In raise_barrier() we firstly increase conf->barrier[idx] then | |
884 | * check conf->nr_pending[idx]. In _wait_barrier() we firstly | |
885 | * increase conf->nr_pending[idx] then check conf->barrier[idx]. | |
886 | * A memory barrier here to make sure conf->nr_pending[idx] won't | |
887 | * be fetched before conf->barrier[idx] is increased. Otherwise | |
888 | * there will be a race between raise_barrier() and _wait_barrier(). | |
889 | */ | |
890 | smp_mb__after_atomic(); | |
17999be4 | 891 | |
79ef3a8a | 892 | /* For these conditions we must wait: |
893 | * A: while the array is in frozen state | |
fd76863e | 894 | * B: while conf->nr_pending[idx] is not 0, meaning regular I/O |
895 | * existing in corresponding I/O barrier bucket. | |
896 | * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches | |
897 | * max resync count which allowed on current I/O barrier bucket. | |
79ef3a8a | 898 | */ |
17999be4 | 899 | wait_event_lock_irq(conf->wait_barrier, |
8c242593 | 900 | (!conf->array_frozen && |
824e47da | 901 | !atomic_read(&conf->nr_pending[idx]) && |
8c242593 YY |
902 | atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || |
903 | test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), | |
eed8c02e | 904 | conf->resync_lock); |
17999be4 | 905 | |
8c242593 YY |
906 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
907 | atomic_dec(&conf->barrier[idx]); | |
908 | spin_unlock_irq(&conf->resync_lock); | |
909 | wake_up(&conf->wait_barrier); | |
910 | return -EINTR; | |
911 | } | |
912 | ||
43ac9b84 | 913 | atomic_inc(&conf->nr_sync_pending); |
17999be4 | 914 | spin_unlock_irq(&conf->resync_lock); |
8c242593 YY |
915 | |
916 | return 0; | |
17999be4 N |
917 | } |
918 | ||
fd76863e | 919 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 920 | { |
fd76863e | 921 | int idx = sector_to_idx(sector_nr); |
922 | ||
824e47da | 923 | BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); |
fd76863e | 924 | |
824e47da | 925 | atomic_dec(&conf->barrier[idx]); |
43ac9b84 | 926 | atomic_dec(&conf->nr_sync_pending); |
17999be4 N |
927 | wake_up(&conf->wait_barrier); |
928 | } | |
929 | ||
5aa70503 | 930 | static bool _wait_barrier(struct r1conf *conf, int idx, bool nowait) |
17999be4 | 931 | { |
5aa70503 VV |
932 | bool ret = true; |
933 | ||
824e47da | 934 | /* |
935 | * We need to increase conf->nr_pending[idx] very early here, | |
936 | * then raise_barrier() can be blocked when it waits for | |
937 | * conf->nr_pending[idx] to be 0. Then we can avoid holding | |
938 | * conf->resync_lock when there is no barrier raised in same | |
939 | * barrier unit bucket. Also if the array is frozen, I/O | |
940 | * should be blocked until array is unfrozen. | |
941 | */ | |
942 | atomic_inc(&conf->nr_pending[idx]); | |
943 | /* | |
944 | * In _wait_barrier() we firstly increase conf->nr_pending[idx], then | |
945 | * check conf->barrier[idx]. In raise_barrier() we firstly increase | |
946 | * conf->barrier[idx], then check conf->nr_pending[idx]. A memory | |
947 | * barrier is necessary here to make sure conf->barrier[idx] won't be | |
948 | * fetched before conf->nr_pending[idx] is increased. Otherwise there | |
949 | * will be a race between _wait_barrier() and raise_barrier(). | |
950 | */ | |
951 | smp_mb__after_atomic(); | |
79ef3a8a | 952 | |
824e47da | 953 | /* |
954 | * Don't worry about checking two atomic_t variables at same time | |
955 | * here. If during we check conf->barrier[idx], the array is | |
956 | * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is | |
957 | * 0, it is safe to return and make the I/O continue. Because the | |
958 | * array is frozen, all I/O returned here will eventually complete | |
959 | * or be queued, no race will happen. See code comment in | |
960 | * frozen_array(). | |
961 | */ | |
962 | if (!READ_ONCE(conf->array_frozen) && | |
963 | !atomic_read(&conf->barrier[idx])) | |
5aa70503 | 964 | return ret; |
79ef3a8a | 965 | |
824e47da | 966 | /* |
967 | * After holding conf->resync_lock, conf->nr_pending[idx] | |
968 | * should be decreased before waiting for barrier to drop. | |
969 | * Otherwise, we may encounter a race condition because | |
970 | * raise_barrer() might be waiting for conf->nr_pending[idx] | |
971 | * to be 0 at same time. | |
972 | */ | |
973 | spin_lock_irq(&conf->resync_lock); | |
974 | atomic_inc(&conf->nr_waiting[idx]); | |
975 | atomic_dec(&conf->nr_pending[idx]); | |
976 | /* | |
977 | * In case freeze_array() is waiting for | |
978 | * get_unqueued_pending() == extra | |
979 | */ | |
21bd9a68 | 980 | wake_up_barrier(conf); |
824e47da | 981 | /* Wait for the barrier in same barrier unit bucket to drop. */ |
5aa70503 VV |
982 | |
983 | /* Return false when nowait flag is set */ | |
984 | if (nowait) { | |
985 | ret = false; | |
986 | } else { | |
987 | wait_event_lock_irq(conf->wait_barrier, | |
988 | !conf->array_frozen && | |
989 | !atomic_read(&conf->barrier[idx]), | |
990 | conf->resync_lock); | |
991 | atomic_inc(&conf->nr_pending[idx]); | |
992 | } | |
993 | ||
824e47da | 994 | atomic_dec(&conf->nr_waiting[idx]); |
fd76863e | 995 | spin_unlock_irq(&conf->resync_lock); |
5aa70503 | 996 | return ret; |
79ef3a8a | 997 | } |
998 | ||
5aa70503 | 999 | static bool wait_read_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait) |
79ef3a8a | 1000 | { |
fd76863e | 1001 | int idx = sector_to_idx(sector_nr); |
5aa70503 | 1002 | bool ret = true; |
79ef3a8a | 1003 | |
824e47da | 1004 | /* |
1005 | * Very similar to _wait_barrier(). The difference is, for read | |
1006 | * I/O we don't need wait for sync I/O, but if the whole array | |
1007 | * is frozen, the read I/O still has to wait until the array is | |
1008 | * unfrozen. Since there is no ordering requirement with | |
1009 | * conf->barrier[idx] here, memory barrier is unnecessary as well. | |
1010 | */ | |
1011 | atomic_inc(&conf->nr_pending[idx]); | |
79ef3a8a | 1012 | |
824e47da | 1013 | if (!READ_ONCE(conf->array_frozen)) |
5aa70503 | 1014 | return ret; |
824e47da | 1015 | |
1016 | spin_lock_irq(&conf->resync_lock); | |
1017 | atomic_inc(&conf->nr_waiting[idx]); | |
1018 | atomic_dec(&conf->nr_pending[idx]); | |
1019 | /* | |
1020 | * In case freeze_array() is waiting for | |
1021 | * get_unqueued_pending() == extra | |
1022 | */ | |
21bd9a68 | 1023 | wake_up_barrier(conf); |
824e47da | 1024 | /* Wait for array to be unfrozen */ |
5aa70503 VV |
1025 | |
1026 | /* Return false when nowait flag is set */ | |
1027 | if (nowait) { | |
1028 | /* Return false when nowait flag is set */ | |
1029 | ret = false; | |
1030 | } else { | |
1031 | wait_event_lock_irq(conf->wait_barrier, | |
1032 | !conf->array_frozen, | |
1033 | conf->resync_lock); | |
1034 | atomic_inc(&conf->nr_pending[idx]); | |
1035 | } | |
1036 | ||
824e47da | 1037 | atomic_dec(&conf->nr_waiting[idx]); |
1da177e4 | 1038 | spin_unlock_irq(&conf->resync_lock); |
5aa70503 | 1039 | return ret; |
1da177e4 LT |
1040 | } |
1041 | ||
5aa70503 | 1042 | static bool wait_barrier(struct r1conf *conf, sector_t sector_nr, bool nowait) |
17999be4 | 1043 | { |
fd76863e | 1044 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 1045 | |
5aa70503 | 1046 | return _wait_barrier(conf, idx, nowait); |
fd76863e | 1047 | } |
1048 | ||
fd76863e | 1049 | static void _allow_barrier(struct r1conf *conf, int idx) |
17999be4 | 1050 | { |
824e47da | 1051 | atomic_dec(&conf->nr_pending[idx]); |
21bd9a68 | 1052 | wake_up_barrier(conf); |
17999be4 N |
1053 | } |
1054 | ||
fd76863e | 1055 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr) |
1056 | { | |
1057 | int idx = sector_to_idx(sector_nr); | |
1058 | ||
1059 | _allow_barrier(conf, idx); | |
1060 | } | |
1061 | ||
fd76863e | 1062 | /* conf->resync_lock should be held */ |
1063 | static int get_unqueued_pending(struct r1conf *conf) | |
1064 | { | |
1065 | int idx, ret; | |
1066 | ||
43ac9b84 XN |
1067 | ret = atomic_read(&conf->nr_sync_pending); |
1068 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
824e47da | 1069 | ret += atomic_read(&conf->nr_pending[idx]) - |
1070 | atomic_read(&conf->nr_queued[idx]); | |
fd76863e | 1071 | |
1072 | return ret; | |
1073 | } | |
1074 | ||
e2d59925 | 1075 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab | 1076 | { |
fd76863e | 1077 | /* Stop sync I/O and normal I/O and wait for everything to |
11353b9d | 1078 | * go quiet. |
fd76863e | 1079 | * This is called in two situations: |
1080 | * 1) management command handlers (reshape, remove disk, quiesce). | |
1081 | * 2) one normal I/O request failed. | |
1082 | ||
1083 | * After array_frozen is set to 1, new sync IO will be blocked at | |
1084 | * raise_barrier(), and new normal I/O will blocked at _wait_barrier() | |
1085 | * or wait_read_barrier(). The flying I/Os will either complete or be | |
1086 | * queued. When everything goes quite, there are only queued I/Os left. | |
1087 | ||
1088 | * Every flying I/O contributes to a conf->nr_pending[idx], idx is the | |
1089 | * barrier bucket index which this I/O request hits. When all sync and | |
1090 | * normal I/O are queued, sum of all conf->nr_pending[] will match sum | |
1091 | * of all conf->nr_queued[]. But normal I/O failure is an exception, | |
1092 | * in handle_read_error(), we may call freeze_array() before trying to | |
1093 | * fix the read error. In this case, the error read I/O is not queued, | |
1094 | * so get_unqueued_pending() == 1. | |
1095 | * | |
1096 | * Therefore before this function returns, we need to wait until | |
1097 | * get_unqueued_pendings(conf) gets equal to extra. For | |
1098 | * normal I/O context, extra is 1, in rested situations extra is 0. | |
ddaf22ab N |
1099 | */ |
1100 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1101 | conf->array_frozen = 1; |
578b54ad | 1102 | raid1_log(conf->mddev, "wait freeze"); |
fd76863e | 1103 | wait_event_lock_irq_cmd( |
1104 | conf->wait_barrier, | |
1105 | get_unqueued_pending(conf) == extra, | |
1106 | conf->resync_lock, | |
1107 | flush_pending_writes(conf)); | |
ddaf22ab N |
1108 | spin_unlock_irq(&conf->resync_lock); |
1109 | } | |
e8096360 | 1110 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
1111 | { |
1112 | /* reverse the effect of the freeze */ | |
1113 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1114 | conf->array_frozen = 0; |
ddaf22ab | 1115 | spin_unlock_irq(&conf->resync_lock); |
824e47da | 1116 | wake_up(&conf->wait_barrier); |
ddaf22ab N |
1117 | } |
1118 | ||
16d56e2f | 1119 | static void alloc_behind_master_bio(struct r1bio *r1_bio, |
cb83efcf | 1120 | struct bio *bio) |
4b6d287f | 1121 | { |
cb83efcf | 1122 | int size = bio->bi_iter.bi_size; |
841c1316 ML |
1123 | unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1124 | int i = 0; | |
1125 | struct bio *behind_bio = NULL; | |
1126 | ||
609be106 CH |
1127 | behind_bio = bio_alloc_bioset(NULL, vcnt, 0, GFP_NOIO, |
1128 | &r1_bio->mddev->bio_set); | |
841c1316 | 1129 | if (!behind_bio) |
16d56e2f | 1130 | return; |
4b6d287f | 1131 | |
41743c1f | 1132 | /* discard op, we don't support writezero/writesame yet */ |
16d56e2f SL |
1133 | if (!bio_has_data(bio)) { |
1134 | behind_bio->bi_iter.bi_size = size; | |
41743c1f | 1135 | goto skip_copy; |
16d56e2f | 1136 | } |
41743c1f | 1137 | |
841c1316 ML |
1138 | while (i < vcnt && size) { |
1139 | struct page *page; | |
1140 | int len = min_t(int, PAGE_SIZE, size); | |
1141 | ||
1142 | page = alloc_page(GFP_NOIO); | |
1143 | if (unlikely(!page)) | |
1144 | goto free_pages; | |
1145 | ||
b42473cd JT |
1146 | if (!bio_add_page(behind_bio, page, len, 0)) { |
1147 | put_page(page); | |
1148 | goto free_pages; | |
1149 | } | |
841c1316 ML |
1150 | |
1151 | size -= len; | |
1152 | i++; | |
4b6d287f | 1153 | } |
841c1316 | 1154 | |
cb83efcf | 1155 | bio_copy_data(behind_bio, bio); |
41743c1f | 1156 | skip_copy: |
56a64c17 | 1157 | r1_bio->behind_master_bio = behind_bio; |
af6d7b76 | 1158 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
4b6d287f | 1159 | |
16d56e2f | 1160 | return; |
841c1316 ML |
1161 | |
1162 | free_pages: | |
4f024f37 KO |
1163 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1164 | bio->bi_iter.bi_size); | |
841c1316 | 1165 | bio_free_pages(behind_bio); |
16d56e2f | 1166 | bio_put(behind_bio); |
4b6d287f N |
1167 | } |
1168 | ||
f54a9d0e N |
1169 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) |
1170 | { | |
1171 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1172 | cb); | |
1173 | struct mddev *mddev = plug->cb.data; | |
1174 | struct r1conf *conf = mddev->private; | |
1175 | struct bio *bio; | |
1176 | ||
9efcc2c3 | 1177 | if (from_schedule) { |
f54a9d0e N |
1178 | spin_lock_irq(&conf->device_lock); |
1179 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
f54a9d0e | 1180 | spin_unlock_irq(&conf->device_lock); |
21bd9a68 | 1181 | wake_up_barrier(conf); |
f54a9d0e N |
1182 | md_wakeup_thread(mddev->thread); |
1183 | kfree(plug); | |
1184 | return; | |
1185 | } | |
1186 | ||
1187 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1188 | bio = bio_list_get(&plug->pending); | |
673ca68d | 1189 | flush_bio_list(conf, bio); |
f54a9d0e N |
1190 | kfree(plug); |
1191 | } | |
1192 | ||
689389a0 N |
1193 | static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio) |
1194 | { | |
1195 | r1_bio->master_bio = bio; | |
1196 | r1_bio->sectors = bio_sectors(bio); | |
1197 | r1_bio->state = 0; | |
1198 | r1_bio->mddev = mddev; | |
1199 | r1_bio->sector = bio->bi_iter.bi_sector; | |
1200 | } | |
1201 | ||
fd76863e | 1202 | static inline struct r1bio * |
689389a0 | 1203 | alloc_r1bio(struct mddev *mddev, struct bio *bio) |
fd76863e | 1204 | { |
1205 | struct r1conf *conf = mddev->private; | |
1206 | struct r1bio *r1_bio; | |
1207 | ||
afeee514 | 1208 | r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); |
689389a0 N |
1209 | /* Ensure no bio records IO_BLOCKED */ |
1210 | memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); | |
1211 | init_r1bio(r1_bio, mddev, bio); | |
fd76863e | 1212 | return r1_bio; |
1213 | } | |
1214 | ||
c230e7e5 | 1215 | static void raid1_read_request(struct mddev *mddev, struct bio *bio, |
689389a0 | 1216 | int max_read_sectors, struct r1bio *r1_bio) |
1da177e4 | 1217 | { |
e8096360 | 1218 | struct r1conf *conf = mddev->private; |
0eaf822c | 1219 | struct raid1_info *mirror; |
1da177e4 | 1220 | struct bio *read_bio; |
3b046a97 | 1221 | struct bitmap *bitmap = mddev->bitmap; |
3c5e514d BVA |
1222 | const enum req_op op = bio_op(bio); |
1223 | const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; | |
3b046a97 RL |
1224 | int max_sectors; |
1225 | int rdisk; | |
9b8ae7b9 | 1226 | bool r1bio_existed = !!r1_bio; |
689389a0 | 1227 | char b[BDEVNAME_SIZE]; |
3b046a97 | 1228 | |
fd76863e | 1229 | /* |
689389a0 N |
1230 | * If r1_bio is set, we are blocking the raid1d thread |
1231 | * so there is a tiny risk of deadlock. So ask for | |
1232 | * emergency memory if needed. | |
fd76863e | 1233 | */ |
689389a0 | 1234 | gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO; |
fd76863e | 1235 | |
9b8ae7b9 | 1236 | if (r1bio_existed) { |
689389a0 N |
1237 | /* Need to get the block device name carefully */ |
1238 | struct md_rdev *rdev; | |
1239 | rcu_read_lock(); | |
1240 | rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); | |
1241 | if (rdev) | |
900d156b | 1242 | snprintf(b, sizeof(b), "%pg", rdev->bdev); |
689389a0 N |
1243 | else |
1244 | strcpy(b, "???"); | |
1245 | rcu_read_unlock(); | |
1246 | } | |
3b046a97 | 1247 | |
fd76863e | 1248 | /* |
fd76863e | 1249 | * Still need barrier for READ in case that whole |
1250 | * array is frozen. | |
fd76863e | 1251 | */ |
5aa70503 VV |
1252 | if (!wait_read_barrier(conf, bio->bi_iter.bi_sector, |
1253 | bio->bi_opf & REQ_NOWAIT)) { | |
1254 | bio_wouldblock_error(bio); | |
1255 | return; | |
1256 | } | |
fd76863e | 1257 | |
689389a0 N |
1258 | if (!r1_bio) |
1259 | r1_bio = alloc_r1bio(mddev, bio); | |
1260 | else | |
1261 | init_r1bio(r1_bio, mddev, bio); | |
c230e7e5 | 1262 | r1_bio->sectors = max_read_sectors; |
fd76863e | 1263 | |
1264 | /* | |
1265 | * make_request() can abort the operation when read-ahead is being | |
1266 | * used and no empty request is available. | |
1267 | */ | |
3b046a97 RL |
1268 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
1269 | ||
1270 | if (rdisk < 0) { | |
1271 | /* couldn't find anywhere to read from */ | |
9b8ae7b9 | 1272 | if (r1bio_existed) { |
689389a0 N |
1273 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
1274 | mdname(mddev), | |
1275 | b, | |
1276 | (unsigned long long)r1_bio->sector); | |
1277 | } | |
3b046a97 RL |
1278 | raid_end_bio_io(r1_bio); |
1279 | return; | |
1280 | } | |
1281 | mirror = conf->mirrors + rdisk; | |
1282 | ||
9b8ae7b9 | 1283 | if (r1bio_existed) |
913cce5a | 1284 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n", |
689389a0 N |
1285 | mdname(mddev), |
1286 | (unsigned long long)r1_bio->sector, | |
913cce5a | 1287 | mirror->rdev->bdev); |
689389a0 | 1288 | |
3b046a97 RL |
1289 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1290 | bitmap) { | |
1291 | /* | |
1292 | * Reading from a write-mostly device must take care not to | |
1293 | * over-take any writes that are 'behind' | |
1294 | */ | |
1295 | raid1_log(mddev, "wait behind writes"); | |
1296 | wait_event(bitmap->behind_wait, | |
1297 | atomic_read(&bitmap->behind_writes) == 0); | |
1298 | } | |
c230e7e5 N |
1299 | |
1300 | if (max_sectors < bio_sectors(bio)) { | |
1301 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1302 | gfp, &conf->bio_split); |
c230e7e5 | 1303 | bio_chain(split, bio); |
ed00aabd | 1304 | submit_bio_noacct(bio); |
c230e7e5 N |
1305 | bio = split; |
1306 | r1_bio->master_bio = bio; | |
1307 | r1_bio->sectors = max_sectors; | |
1308 | } | |
1309 | ||
3b046a97 | 1310 | r1_bio->read_disk = rdisk; |
bb2a9ace YK |
1311 | if (!r1bio_existed) { |
1312 | md_account_bio(mddev, &bio); | |
1313 | r1_bio->master_bio = bio; | |
1314 | } | |
abfc426d CH |
1315 | read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp, |
1316 | &mddev->bio_set); | |
3b046a97 RL |
1317 | |
1318 | r1_bio->bios[rdisk] = read_bio; | |
1319 | ||
1320 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1321 | mirror->rdev->data_offset; | |
3b046a97 | 1322 | read_bio->bi_end_io = raid1_end_read_request; |
c34b7ac6 | 1323 | read_bio->bi_opf = op | do_sync; |
3b046a97 RL |
1324 | if (test_bit(FailFast, &mirror->rdev->flags) && |
1325 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1326 | read_bio->bi_opf |= MD_FAILFAST; | |
1327 | read_bio->bi_private = r1_bio; | |
1328 | ||
1329 | if (mddev->gendisk) | |
1c02fca6 CH |
1330 | trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), |
1331 | r1_bio->sector); | |
3b046a97 | 1332 | |
ed00aabd | 1333 | submit_bio_noacct(read_bio); |
3b046a97 RL |
1334 | } |
1335 | ||
c230e7e5 N |
1336 | static void raid1_write_request(struct mddev *mddev, struct bio *bio, |
1337 | int max_write_sectors) | |
3b046a97 RL |
1338 | { |
1339 | struct r1conf *conf = mddev->private; | |
fd76863e | 1340 | struct r1bio *r1_bio; |
1f68f0c4 | 1341 | int i, disks; |
3b046a97 | 1342 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1343 | unsigned long flags; |
3cb03002 | 1344 | struct md_rdev *blocked_rdev; |
1f68f0c4 | 1345 | int first_clone; |
1f68f0c4 | 1346 | int max_sectors; |
6607cd31 | 1347 | bool write_behind = false; |
191ea9b2 | 1348 | |
b3143b9a | 1349 | if (mddev_is_clustered(mddev) && |
90382ed9 | 1350 | md_cluster_ops->area_resyncing(mddev, WRITE, |
b3143b9a | 1351 | bio->bi_iter.bi_sector, bio_end_sector(bio))) { |
3b046a97 | 1352 | |
6eef4b21 | 1353 | DEFINE_WAIT(w); |
5aa70503 VV |
1354 | if (bio->bi_opf & REQ_NOWAIT) { |
1355 | bio_wouldblock_error(bio); | |
1356 | return; | |
1357 | } | |
6eef4b21 | 1358 | for (;;) { |
6eef4b21 | 1359 | prepare_to_wait(&conf->wait_barrier, |
ae89fd3d | 1360 | &w, TASK_IDLE); |
f81f7302 | 1361 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, |
385f4d7f | 1362 | bio->bi_iter.bi_sector, |
b3143b9a | 1363 | bio_end_sector(bio))) |
6eef4b21 N |
1364 | break; |
1365 | schedule(); | |
1366 | } | |
1367 | finish_wait(&conf->wait_barrier, &w); | |
1368 | } | |
f81f7302 GJ |
1369 | |
1370 | /* | |
1371 | * Register the new request and wait if the reconstruction | |
1372 | * thread has put up a bar for new requests. | |
1373 | * Continue immediately if no resync is active currently. | |
1374 | */ | |
5aa70503 VV |
1375 | if (!wait_barrier(conf, bio->bi_iter.bi_sector, |
1376 | bio->bi_opf & REQ_NOWAIT)) { | |
1377 | bio_wouldblock_error(bio); | |
1378 | return; | |
1379 | } | |
fd76863e | 1380 | |
992db13a | 1381 | retry_write: |
689389a0 | 1382 | r1_bio = alloc_r1bio(mddev, bio); |
c230e7e5 | 1383 | r1_bio->sectors = max_write_sectors; |
1da177e4 | 1384 | |
1f68f0c4 | 1385 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1386 | * inc refcount on their rdev. Record them by setting |
1387 | * bios[x] to bio | |
1f68f0c4 N |
1388 | * If there are known/acknowledged bad blocks on any device on |
1389 | * which we have seen a write error, we want to avoid writing those | |
1390 | * blocks. | |
1391 | * This potentially requires several writes to write around | |
1392 | * the bad blocks. Each set of writes gets it's own r1bio | |
1393 | * with a set of bios attached. | |
1da177e4 | 1394 | */ |
c3b328ac | 1395 | |
8f19ccb2 | 1396 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1397 | blocked_rdev = NULL; |
1da177e4 | 1398 | rcu_read_lock(); |
1f68f0c4 | 1399 | max_sectors = r1_bio->sectors; |
1da177e4 | 1400 | for (i = 0; i < disks; i++) { |
3cb03002 | 1401 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6607cd31 GJ |
1402 | |
1403 | /* | |
1404 | * The write-behind io is only attempted on drives marked as | |
1405 | * write-mostly, which means we could allocate write behind | |
1406 | * bio later. | |
1407 | */ | |
1408 | if (rdev && test_bit(WriteMostly, &rdev->flags)) | |
1409 | write_behind = true; | |
1410 | ||
6bfe0b49 DW |
1411 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1412 | atomic_inc(&rdev->nr_pending); | |
1413 | blocked_rdev = rdev; | |
1414 | break; | |
1415 | } | |
1f68f0c4 | 1416 | r1_bio->bios[i] = NULL; |
8ae12666 | 1417 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1418 | if (i < conf->raid_disks) |
1419 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1420 | continue; |
1421 | } | |
1422 | ||
1423 | atomic_inc(&rdev->nr_pending); | |
1424 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1425 | sector_t first_bad; | |
1426 | int bad_sectors; | |
1427 | int is_bad; | |
1428 | ||
3b046a97 | 1429 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1430 | &first_bad, &bad_sectors); |
1431 | if (is_bad < 0) { | |
1432 | /* mustn't write here until the bad block is | |
1433 | * acknowledged*/ | |
1434 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1435 | blocked_rdev = rdev; | |
1436 | break; | |
1437 | } | |
1438 | if (is_bad && first_bad <= r1_bio->sector) { | |
1439 | /* Cannot write here at all */ | |
1440 | bad_sectors -= (r1_bio->sector - first_bad); | |
1441 | if (bad_sectors < max_sectors) | |
1442 | /* mustn't write more than bad_sectors | |
1443 | * to other devices yet | |
1444 | */ | |
1445 | max_sectors = bad_sectors; | |
03c902e1 | 1446 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1447 | /* We don't set R1BIO_Degraded as that |
1448 | * only applies if the disk is | |
1449 | * missing, so it might be re-added, | |
1450 | * and we want to know to recover this | |
1451 | * chunk. | |
1452 | * In this case the device is here, | |
1453 | * and the fact that this chunk is not | |
1454 | * in-sync is recorded in the bad | |
1455 | * block log | |
1456 | */ | |
1457 | continue; | |
964147d5 | 1458 | } |
1f68f0c4 N |
1459 | if (is_bad) { |
1460 | int good_sectors = first_bad - r1_bio->sector; | |
1461 | if (good_sectors < max_sectors) | |
1462 | max_sectors = good_sectors; | |
1463 | } | |
1464 | } | |
1465 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1466 | } |
1467 | rcu_read_unlock(); | |
1468 | ||
6bfe0b49 DW |
1469 | if (unlikely(blocked_rdev)) { |
1470 | /* Wait for this device to become unblocked */ | |
1471 | int j; | |
1472 | ||
1473 | for (j = 0; j < i; j++) | |
1474 | if (r1_bio->bios[j]) | |
1475 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
992db13a | 1476 | free_r1bio(r1_bio); |
fd76863e | 1477 | allow_barrier(conf, bio->bi_iter.bi_sector); |
5aa70503 VV |
1478 | |
1479 | if (bio->bi_opf & REQ_NOWAIT) { | |
1480 | bio_wouldblock_error(bio); | |
1481 | return; | |
1482 | } | |
578b54ad | 1483 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1484 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
5aa70503 | 1485 | wait_barrier(conf, bio->bi_iter.bi_sector, false); |
6bfe0b49 DW |
1486 | goto retry_write; |
1487 | } | |
1488 | ||
6607cd31 GJ |
1489 | /* |
1490 | * When using a bitmap, we may call alloc_behind_master_bio below. | |
1491 | * alloc_behind_master_bio allocates a copy of the data payload a page | |
1492 | * at a time and thus needs a new bio that can fit the whole payload | |
1493 | * this bio in page sized chunks. | |
1494 | */ | |
1495 | if (write_behind && bitmap) | |
1496 | max_sectors = min_t(int, max_sectors, | |
1497 | BIO_MAX_VECS * (PAGE_SIZE >> 9)); | |
c230e7e5 N |
1498 | if (max_sectors < bio_sectors(bio)) { |
1499 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1500 | GFP_NOIO, &conf->bio_split); |
c230e7e5 | 1501 | bio_chain(split, bio); |
ed00aabd | 1502 | submit_bio_noacct(bio); |
c230e7e5 N |
1503 | bio = split; |
1504 | r1_bio->master_bio = bio; | |
1f68f0c4 | 1505 | r1_bio->sectors = max_sectors; |
191ea9b2 | 1506 | } |
4b6d287f | 1507 | |
bb2a9ace YK |
1508 | md_account_bio(mddev, &bio); |
1509 | r1_bio->master_bio = bio; | |
4e78064f | 1510 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1511 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1512 | |
1f68f0c4 | 1513 | first_clone = 1; |
d8c84c4f | 1514 | |
1da177e4 | 1515 | for (i = 0; i < disks; i++) { |
8e58e327 | 1516 | struct bio *mbio = NULL; |
69df9cfc | 1517 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
1da177e4 LT |
1518 | if (!r1_bio->bios[i]) |
1519 | continue; | |
1520 | ||
46669e86 | 1521 | if (first_clone) { |
1f68f0c4 N |
1522 | /* do behind I/O ? |
1523 | * Not if there are too many, or cannot | |
1524 | * allocate memory, or a reader on WriteMostly | |
1525 | * is waiting for behind writes to flush */ | |
6b2460e6 | 1526 | if (bitmap && write_behind && |
1f68f0c4 N |
1527 | (atomic_read(&bitmap->behind_writes) |
1528 | < mddev->bitmap_info.max_write_behind) && | |
8e58e327 | 1529 | !waitqueue_active(&bitmap->behind_wait)) { |
16d56e2f | 1530 | alloc_behind_master_bio(r1_bio, bio); |
8e58e327 | 1531 | } |
1f68f0c4 | 1532 | |
e64e4018 AS |
1533 | md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, |
1534 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
1f68f0c4 N |
1535 | first_clone = 0; |
1536 | } | |
8e58e327 | 1537 | |
841c1316 | 1538 | if (r1_bio->behind_master_bio) { |
abfc426d CH |
1539 | mbio = bio_alloc_clone(rdev->bdev, |
1540 | r1_bio->behind_master_bio, | |
1541 | GFP_NOIO, &mddev->bio_set); | |
69df9cfc | 1542 | if (test_bit(CollisionCheck, &rdev->flags)) |
d0d2d8ba | 1543 | wait_for_serialization(rdev, r1_bio); |
3e148a32 | 1544 | if (test_bit(WriteMostly, &rdev->flags)) |
4b6d287f | 1545 | atomic_inc(&r1_bio->behind_remaining); |
abfc426d CH |
1546 | } else { |
1547 | mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, | |
1548 | &mddev->bio_set); | |
1549 | ||
1550 | if (mddev->serialize_policy) | |
1551 | wait_for_serialization(rdev, r1_bio); | |
1552 | } | |
4b6d287f | 1553 | |
1f68f0c4 N |
1554 | r1_bio->bios[i] = mbio; |
1555 | ||
2e94275e | 1556 | mbio->bi_iter.bi_sector = (r1_bio->sector + rdev->data_offset); |
1f68f0c4 | 1557 | mbio->bi_end_io = raid1_end_write_request; |
a682e003 | 1558 | mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); |
2e94275e GJ |
1559 | if (test_bit(FailFast, &rdev->flags) && |
1560 | !test_bit(WriteMostly, &rdev->flags) && | |
212e7eb7 N |
1561 | conf->raid_disks - mddev->degraded > 1) |
1562 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1563 | mbio->bi_private = r1_bio; |
1564 | ||
1da177e4 | 1565 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1566 | |
109e3765 | 1567 | if (mddev->gendisk) |
1c02fca6 | 1568 | trace_block_bio_remap(mbio, disk_devt(mddev->gendisk), |
109e3765 N |
1569 | r1_bio->sector); |
1570 | /* flush_pending_writes() needs access to the rdev so...*/ | |
2e94275e | 1571 | mbio->bi_bdev = (void *)rdev; |
460af1f9 | 1572 | if (!raid1_add_bio_to_plug(mddev, mbio, raid1_unplug, disks)) { |
23b245c0 | 1573 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e | 1574 | bio_list_add(&conf->pending_bio_list, mbio); |
23b245c0 | 1575 | spin_unlock_irqrestore(&conf->device_lock, flags); |
b357f04a | 1576 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1577 | } |
1da177e4 | 1578 | } |
1f68f0c4 | 1579 | |
079fa166 N |
1580 | r1_bio_write_done(r1_bio); |
1581 | ||
1582 | /* In case raid1d snuck in to freeze_array */ | |
21bd9a68 | 1583 | wake_up_barrier(conf); |
1da177e4 LT |
1584 | } |
1585 | ||
cc27b0c7 | 1586 | static bool raid1_make_request(struct mddev *mddev, struct bio *bio) |
3b046a97 | 1587 | { |
fd76863e | 1588 | sector_t sectors; |
3b046a97 | 1589 | |
775d7831 DJ |
1590 | if (unlikely(bio->bi_opf & REQ_PREFLUSH) |
1591 | && md_flush_request(mddev, bio)) | |
cc27b0c7 | 1592 | return true; |
3b046a97 | 1593 | |
c230e7e5 N |
1594 | /* |
1595 | * There is a limit to the maximum size, but | |
1596 | * the read/write handler might find a lower limit | |
1597 | * due to bad blocks. To avoid multiple splits, | |
1598 | * we pass the maximum number of sectors down | |
1599 | * and let the lower level perform the split. | |
1600 | */ | |
1601 | sectors = align_to_barrier_unit_end( | |
1602 | bio->bi_iter.bi_sector, bio_sectors(bio)); | |
61eb2b43 | 1603 | |
c230e7e5 | 1604 | if (bio_data_dir(bio) == READ) |
689389a0 | 1605 | raid1_read_request(mddev, bio, sectors, NULL); |
cc27b0c7 N |
1606 | else { |
1607 | if (!md_write_start(mddev,bio)) | |
1608 | return false; | |
c230e7e5 | 1609 | raid1_write_request(mddev, bio, sectors); |
cc27b0c7 N |
1610 | } |
1611 | return true; | |
3b046a97 RL |
1612 | } |
1613 | ||
849674e4 | 1614 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1615 | { |
e8096360 | 1616 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1617 | int i; |
1618 | ||
1619 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1620 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1621 | rcu_read_lock(); |
1622 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1623 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1624 | seq_printf(seq, "%s", |
ddac7c7e N |
1625 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1626 | } | |
1627 | rcu_read_unlock(); | |
1da177e4 LT |
1628 | seq_printf(seq, "]"); |
1629 | } | |
1630 | ||
9631abdb MT |
1631 | /** |
1632 | * raid1_error() - RAID1 error handler. | |
1633 | * @mddev: affected md device. | |
1634 | * @rdev: member device to fail. | |
1635 | * | |
1636 | * The routine acknowledges &rdev failure and determines new @mddev state. | |
1637 | * If it failed, then: | |
1638 | * - &MD_BROKEN flag is set in &mddev->flags. | |
1639 | * - recovery is disabled. | |
1640 | * Otherwise, it must be degraded: | |
1641 | * - recovery is interrupted. | |
1642 | * - &mddev->degraded is bumped. | |
1643 | * | |
1644 | * @rdev is marked as &Faulty excluding case when array is failed and | |
1645 | * &mddev->fail_last_dev is off. | |
1646 | */ | |
849674e4 | 1647 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1648 | { |
e8096360 | 1649 | struct r1conf *conf = mddev->private; |
423f04d6 | 1650 | unsigned long flags; |
1da177e4 | 1651 | |
2e52d449 | 1652 | spin_lock_irqsave(&conf->device_lock, flags); |
9631abdb MT |
1653 | |
1654 | if (test_bit(In_sync, &rdev->flags) && | |
1655 | (conf->raid_disks - mddev->degraded) == 1) { | |
1656 | set_bit(MD_BROKEN, &mddev->flags); | |
1657 | ||
1658 | if (!mddev->fail_last_dev) { | |
1659 | conf->recovery_disabled = mddev->recovery_disabled; | |
1660 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1661 | return; | |
1662 | } | |
4044ba58 | 1663 | } |
de393cde | 1664 | set_bit(Blocked, &rdev->flags); |
ebda52fa | 1665 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1666 | mddev->degraded++; |
ebda52fa | 1667 | set_bit(Faulty, &rdev->flags); |
423f04d6 | 1668 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1669 | /* |
1670 | * if recovery is running, make sure it aborts. | |
1671 | */ | |
1672 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1673 | set_mask_bits(&mddev->sb_flags, 0, |
1674 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
913cce5a | 1675 | pr_crit("md/raid1:%s: Disk failure on %pg, disabling device.\n" |
1d41c216 | 1676 | "md/raid1:%s: Operation continuing on %d devices.\n", |
913cce5a | 1677 | mdname(mddev), rdev->bdev, |
1d41c216 | 1678 | mdname(mddev), conf->raid_disks - mddev->degraded); |
1da177e4 LT |
1679 | } |
1680 | ||
e8096360 | 1681 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1682 | { |
1683 | int i; | |
1da177e4 | 1684 | |
1d41c216 | 1685 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1686 | if (!conf) { |
1d41c216 | 1687 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1688 | return; |
1689 | } | |
1d41c216 N |
1690 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1691 | conf->raid_disks); | |
1da177e4 | 1692 | |
ddac7c7e | 1693 | rcu_read_lock(); |
1da177e4 | 1694 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1695 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1696 | if (rdev) |
913cce5a | 1697 | pr_debug(" disk %d, wo:%d, o:%d, dev:%pg\n", |
1d41c216 N |
1698 | i, !test_bit(In_sync, &rdev->flags), |
1699 | !test_bit(Faulty, &rdev->flags), | |
913cce5a | 1700 | rdev->bdev); |
1da177e4 | 1701 | } |
ddac7c7e | 1702 | rcu_read_unlock(); |
1da177e4 LT |
1703 | } |
1704 | ||
e8096360 | 1705 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1706 | { |
f6eca2d4 ND |
1707 | int idx; |
1708 | ||
1709 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) { | |
5aa70503 | 1710 | _wait_barrier(conf, idx, false); |
f6eca2d4 ND |
1711 | _allow_barrier(conf, idx); |
1712 | } | |
1da177e4 | 1713 | |
afeee514 | 1714 | mempool_exit(&conf->r1buf_pool); |
1da177e4 LT |
1715 | } |
1716 | ||
fd01b88c | 1717 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1718 | { |
1719 | int i; | |
e8096360 | 1720 | struct r1conf *conf = mddev->private; |
6b965620 N |
1721 | int count = 0; |
1722 | unsigned long flags; | |
1da177e4 LT |
1723 | |
1724 | /* | |
f72ffdd6 | 1725 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1726 | * and mark them readable. |
1727 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1728 | * device_lock used to avoid races with raid1_end_read_request |
1729 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1730 | */ |
423f04d6 | 1731 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1732 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1733 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1734 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1735 | if (repl | |
1aee41f6 | 1736 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1737 | && repl->recovery_offset == MaxSector |
1738 | && !test_bit(Faulty, &repl->flags) | |
1739 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1740 | /* replacement has just become active */ | |
1741 | if (!rdev || | |
1742 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1743 | count++; | |
1744 | if (rdev) { | |
1745 | /* Replaced device not technically | |
1746 | * faulty, but we need to be sure | |
1747 | * it gets removed and never re-added | |
1748 | */ | |
1749 | set_bit(Faulty, &rdev->flags); | |
1750 | sysfs_notify_dirent_safe( | |
1751 | rdev->sysfs_state); | |
1752 | } | |
1753 | } | |
ddac7c7e | 1754 | if (rdev |
61e4947c | 1755 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1756 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1757 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1758 | count++; |
654e8b5a | 1759 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1760 | } |
1761 | } | |
6b965620 N |
1762 | mddev->degraded -= count; |
1763 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1764 | |
1765 | print_conf(conf); | |
6b965620 | 1766 | return count; |
1da177e4 LT |
1767 | } |
1768 | ||
fd01b88c | 1769 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1770 | { |
e8096360 | 1771 | struct r1conf *conf = mddev->private; |
199050ea | 1772 | int err = -EEXIST; |
ffb1e7a0 | 1773 | int mirror = 0, repl_slot = -1; |
0eaf822c | 1774 | struct raid1_info *p; |
6c2fce2e | 1775 | int first = 0; |
30194636 | 1776 | int last = conf->raid_disks - 1; |
1da177e4 | 1777 | |
5389042f N |
1778 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1779 | return -EBUSY; | |
1780 | ||
1501efad DW |
1781 | if (md_integrity_add_rdev(rdev, mddev)) |
1782 | return -ENXIO; | |
1783 | ||
6c2fce2e NB |
1784 | if (rdev->raid_disk >= 0) |
1785 | first = last = rdev->raid_disk; | |
1786 | ||
70bcecdb GR |
1787 | /* |
1788 | * find the disk ... but prefer rdev->saved_raid_disk | |
1789 | * if possible. | |
1790 | */ | |
1791 | if (rdev->saved_raid_disk >= 0 && | |
1792 | rdev->saved_raid_disk >= first && | |
9e753ba9 | 1793 | rdev->saved_raid_disk < conf->raid_disks && |
70bcecdb GR |
1794 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1795 | first = last = rdev->saved_raid_disk; | |
1796 | ||
7ef449d1 | 1797 | for (mirror = first; mirror <= last; mirror++) { |
ebfeb444 | 1798 | p = conf->mirrors + mirror; |
7ef449d1 | 1799 | if (!p->rdev) { |
9092c02d JB |
1800 | if (mddev->gendisk) |
1801 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1802 | rdev->data_offset << 9); | |
1da177e4 LT |
1803 | |
1804 | p->head_position = 0; | |
1805 | rdev->raid_disk = mirror; | |
199050ea | 1806 | err = 0; |
6aea114a N |
1807 | /* As all devices are equivalent, we don't need a full recovery |
1808 | * if this was recently any drive of the array | |
1809 | */ | |
1810 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1811 | conf->fullsync = 1; |
d6065f7b | 1812 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1813 | break; |
1814 | } | |
7ef449d1 | 1815 | if (test_bit(WantReplacement, &p->rdev->flags) && |
ffb1e7a0 LN |
1816 | p[conf->raid_disks].rdev == NULL && repl_slot < 0) |
1817 | repl_slot = mirror; | |
7ef449d1 | 1818 | } |
ffb1e7a0 LN |
1819 | |
1820 | if (err && repl_slot >= 0) { | |
1821 | /* Add this device as a replacement */ | |
1822 | p = conf->mirrors + repl_slot; | |
1823 | clear_bit(In_sync, &rdev->flags); | |
1824 | set_bit(Replacement, &rdev->flags); | |
1825 | rdev->raid_disk = repl_slot; | |
1826 | err = 0; | |
1827 | conf->fullsync = 1; | |
1828 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1829 | } | |
1830 | ||
1da177e4 | 1831 | print_conf(conf); |
199050ea | 1832 | return err; |
1da177e4 LT |
1833 | } |
1834 | ||
b8321b68 | 1835 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1836 | { |
e8096360 | 1837 | struct r1conf *conf = mddev->private; |
1da177e4 | 1838 | int err = 0; |
b8321b68 | 1839 | int number = rdev->raid_disk; |
8b0472b5 ZS |
1840 | |
1841 | if (unlikely(number >= conf->raid_disks)) | |
1842 | goto abort; | |
1843 | ||
0eaf822c | 1844 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1845 | |
b014f14c N |
1846 | if (rdev != p->rdev) |
1847 | p = conf->mirrors + conf->raid_disks + number; | |
1848 | ||
1da177e4 | 1849 | print_conf(conf); |
b8321b68 | 1850 | if (rdev == p->rdev) { |
b2d444d7 | 1851 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1852 | atomic_read(&rdev->nr_pending)) { |
1853 | err = -EBUSY; | |
1854 | goto abort; | |
1855 | } | |
046abeed | 1856 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1857 | * is not possible. |
1858 | */ | |
1859 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1860 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1861 | mddev->degraded < conf->raid_disks) { |
1862 | err = -EBUSY; | |
1863 | goto abort; | |
1864 | } | |
1da177e4 | 1865 | p->rdev = NULL; |
d787be40 N |
1866 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1867 | synchronize_rcu(); | |
1868 | if (atomic_read(&rdev->nr_pending)) { | |
1869 | /* lost the race, try later */ | |
1870 | err = -EBUSY; | |
1871 | p->rdev = rdev; | |
1872 | goto abort; | |
1873 | } | |
1874 | } | |
1875 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1876 | /* We just removed a device that is being replaced. |
1877 | * Move down the replacement. We drain all IO before | |
1878 | * doing this to avoid confusion. | |
1879 | */ | |
1880 | struct md_rdev *repl = | |
1881 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1882 | freeze_array(conf, 0); |
3de59bb9 YY |
1883 | if (atomic_read(&repl->nr_pending)) { |
1884 | /* It means that some queued IO of retry_list | |
1885 | * hold repl. Thus, we cannot set replacement | |
1886 | * as NULL, avoiding rdev NULL pointer | |
1887 | * dereference in sync_request_write and | |
1888 | * handle_write_finished. | |
1889 | */ | |
1890 | err = -EBUSY; | |
1891 | unfreeze_array(conf); | |
1892 | goto abort; | |
1893 | } | |
8c7a2c2b N |
1894 | clear_bit(Replacement, &repl->flags); |
1895 | p->rdev = repl; | |
1896 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1897 | unfreeze_array(conf); |
e5bc9c3c GJ |
1898 | } |
1899 | ||
1900 | clear_bit(WantReplacement, &rdev->flags); | |
a91a2785 | 1901 | err = md_integrity_register(mddev); |
1da177e4 LT |
1902 | } |
1903 | abort: | |
1904 | ||
1905 | print_conf(conf); | |
1906 | return err; | |
1907 | } | |
1908 | ||
4246a0b6 | 1909 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1910 | { |
98d30c58 | 1911 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
1da177e4 | 1912 | |
0fc280f6 | 1913 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1914 | |
1da177e4 LT |
1915 | /* |
1916 | * we have read a block, now it needs to be re-written, | |
1917 | * or re-read if the read failed. | |
1918 | * We don't do much here, just schedule handling by raid1d | |
1919 | */ | |
4e4cbee9 | 1920 | if (!bio->bi_status) |
1da177e4 | 1921 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1922 | |
1923 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1924 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1925 | } |
1926 | ||
dfcc34c9 ND |
1927 | static void abort_sync_write(struct mddev *mddev, struct r1bio *r1_bio) |
1928 | { | |
1929 | sector_t sync_blocks = 0; | |
1930 | sector_t s = r1_bio->sector; | |
1931 | long sectors_to_go = r1_bio->sectors; | |
1932 | ||
1933 | /* make sure these bits don't get cleared. */ | |
1934 | do { | |
1935 | md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); | |
1936 | s += sync_blocks; | |
1937 | sectors_to_go -= sync_blocks; | |
1938 | } while (sectors_to_go > 0); | |
1939 | } | |
1940 | ||
449808a2 HT |
1941 | static void put_sync_write_buf(struct r1bio *r1_bio, int uptodate) |
1942 | { | |
1943 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
1944 | struct mddev *mddev = r1_bio->mddev; | |
1945 | int s = r1_bio->sectors; | |
1946 | ||
1947 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
1948 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
1949 | reschedule_retry(r1_bio); | |
1950 | else { | |
1951 | put_buf(r1_bio); | |
1952 | md_done_sync(mddev, s, uptodate); | |
1953 | } | |
1954 | } | |
1955 | } | |
1956 | ||
4246a0b6 | 1957 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1958 | { |
4e4cbee9 | 1959 | int uptodate = !bio->bi_status; |
98d30c58 | 1960 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
fd01b88c | 1961 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1962 | struct r1conf *conf = mddev->private; |
4367af55 N |
1963 | sector_t first_bad; |
1964 | int bad_sectors; | |
854abd75 | 1965 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1966 | |
6b1117d5 | 1967 | if (!uptodate) { |
dfcc34c9 | 1968 | abort_sync_write(mddev, r1_bio); |
854abd75 N |
1969 | set_bit(WriteErrorSeen, &rdev->flags); |
1970 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1971 | set_bit(MD_RECOVERY_NEEDED, & |
1972 | mddev->recovery); | |
d8f05d29 | 1973 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1974 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1975 | &first_bad, &bad_sectors) && |
1976 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1977 | r1_bio->sector, | |
1978 | r1_bio->sectors, | |
1979 | &first_bad, &bad_sectors) | |
1980 | ) | |
4367af55 | 1981 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1982 | |
449808a2 | 1983 | put_sync_write_buf(r1_bio, uptodate); |
1da177e4 LT |
1984 | } |
1985 | ||
3cb03002 | 1986 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
4ce4c73f | 1987 | int sectors, struct page *page, int rw) |
d8f05d29 | 1988 | { |
4ce4c73f | 1989 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, false)) |
d8f05d29 N |
1990 | /* success */ |
1991 | return 1; | |
19d67169 | 1992 | if (rw == WRITE) { |
d8f05d29 | 1993 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1994 | if (!test_and_set_bit(WantReplacement, |
1995 | &rdev->flags)) | |
1996 | set_bit(MD_RECOVERY_NEEDED, & | |
1997 | rdev->mddev->recovery); | |
1998 | } | |
d8f05d29 N |
1999 | /* need to record an error - either for the block or the device */ |
2000 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2001 | md_error(rdev->mddev, rdev); | |
2002 | return 0; | |
2003 | } | |
2004 | ||
9f2c9d12 | 2005 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 2006 | { |
a68e5870 N |
2007 | /* Try some synchronous reads of other devices to get |
2008 | * good data, much like with normal read errors. Only | |
2009 | * read into the pages we already have so we don't | |
2010 | * need to re-issue the read request. | |
2011 | * We don't need to freeze the array, because being in an | |
2012 | * active sync request, there is no normal IO, and | |
2013 | * no overlapping syncs. | |
06f60385 N |
2014 | * We don't need to check is_badblock() again as we |
2015 | * made sure that anything with a bad block in range | |
2016 | * will have bi_end_io clear. | |
a68e5870 | 2017 | */ |
fd01b88c | 2018 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2019 | struct r1conf *conf = mddev->private; |
a68e5870 | 2020 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
44cf0f4d | 2021 | struct page **pages = get_resync_pages(bio)->pages; |
a68e5870 N |
2022 | sector_t sect = r1_bio->sector; |
2023 | int sectors = r1_bio->sectors; | |
2024 | int idx = 0; | |
2e52d449 N |
2025 | struct md_rdev *rdev; |
2026 | ||
2027 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
2028 | if (test_bit(FailFast, &rdev->flags)) { | |
2029 | /* Don't try recovering from here - just fail it | |
2030 | * ... unless it is the last working device of course */ | |
2031 | md_error(mddev, rdev); | |
2032 | if (test_bit(Faulty, &rdev->flags)) | |
2033 | /* Don't try to read from here, but make sure | |
2034 | * put_buf does it's thing | |
2035 | */ | |
2036 | bio->bi_end_io = end_sync_write; | |
2037 | } | |
a68e5870 N |
2038 | |
2039 | while(sectors) { | |
2040 | int s = sectors; | |
2041 | int d = r1_bio->read_disk; | |
2042 | int success = 0; | |
78d7f5f7 | 2043 | int start; |
a68e5870 N |
2044 | |
2045 | if (s > (PAGE_SIZE>>9)) | |
2046 | s = PAGE_SIZE >> 9; | |
2047 | do { | |
2048 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
2049 | /* No rcu protection needed here devices | |
2050 | * can only be removed when no resync is | |
2051 | * active, and resync is currently active | |
2052 | */ | |
2053 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 2054 | if (sync_page_io(rdev, sect, s<<9, |
44cf0f4d | 2055 | pages[idx], |
4ce4c73f | 2056 | REQ_OP_READ, false)) { |
a68e5870 N |
2057 | success = 1; |
2058 | break; | |
2059 | } | |
2060 | } | |
2061 | d++; | |
8f19ccb2 | 2062 | if (d == conf->raid_disks * 2) |
a68e5870 N |
2063 | d = 0; |
2064 | } while (!success && d != r1_bio->read_disk); | |
2065 | ||
78d7f5f7 | 2066 | if (!success) { |
3a9f28a5 N |
2067 | int abort = 0; |
2068 | /* Cannot read from anywhere, this block is lost. | |
2069 | * Record a bad block on each device. If that doesn't | |
2070 | * work just disable and interrupt the recovery. | |
2071 | * Don't fail devices as that won't really help. | |
2072 | */ | |
ac483eb3 CH |
2073 | pr_crit_ratelimited("md/raid1:%s: %pg: unrecoverable I/O read error for block %llu\n", |
2074 | mdname(mddev), bio->bi_bdev, | |
1d41c216 | 2075 | (unsigned long long)r1_bio->sector); |
8f19ccb2 | 2076 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
2077 | rdev = conf->mirrors[d].rdev; |
2078 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
2079 | continue; | |
2080 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
2081 | abort = 1; | |
2082 | } | |
2083 | if (abort) { | |
d890fa2b N |
2084 | conf->recovery_disabled = |
2085 | mddev->recovery_disabled; | |
3a9f28a5 N |
2086 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
2087 | md_done_sync(mddev, r1_bio->sectors, 0); | |
2088 | put_buf(r1_bio); | |
2089 | return 0; | |
2090 | } | |
2091 | /* Try next page */ | |
2092 | sectors -= s; | |
2093 | sect += s; | |
2094 | idx++; | |
2095 | continue; | |
d11c171e | 2096 | } |
78d7f5f7 N |
2097 | |
2098 | start = d; | |
2099 | /* write it back and re-read */ | |
2100 | while (d != r1_bio->read_disk) { | |
2101 | if (d == 0) | |
8f19ccb2 | 2102 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2103 | d--; |
2104 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2105 | continue; | |
2106 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2107 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2108 | pages[idx], |
d8f05d29 | 2109 | WRITE) == 0) { |
78d7f5f7 N |
2110 | r1_bio->bios[d]->bi_end_io = NULL; |
2111 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 2112 | } |
78d7f5f7 N |
2113 | } |
2114 | d = start; | |
2115 | while (d != r1_bio->read_disk) { | |
2116 | if (d == 0) | |
8f19ccb2 | 2117 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2118 | d--; |
2119 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2120 | continue; | |
2121 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2122 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2123 | pages[idx], |
d8f05d29 | 2124 | READ) != 0) |
9d3d8011 | 2125 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 2126 | } |
a68e5870 N |
2127 | sectors -= s; |
2128 | sect += s; | |
2129 | idx ++; | |
2130 | } | |
78d7f5f7 | 2131 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4e4cbee9 | 2132 | bio->bi_status = 0; |
a68e5870 N |
2133 | return 1; |
2134 | } | |
2135 | ||
c95e6385 | 2136 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
2137 | { |
2138 | /* We have read all readable devices. If we haven't | |
2139 | * got the block, then there is no hope left. | |
2140 | * If we have, then we want to do a comparison | |
2141 | * and skip the write if everything is the same. | |
2142 | * If any blocks failed to read, then we need to | |
2143 | * attempt an over-write | |
2144 | */ | |
fd01b88c | 2145 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2146 | struct r1conf *conf = mddev->private; |
a68e5870 N |
2147 | int primary; |
2148 | int i; | |
f4380a91 | 2149 | int vcnt; |
a68e5870 | 2150 | |
30bc9b53 N |
2151 | /* Fix variable parts of all bios */ |
2152 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
2153 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
4e4cbee9 | 2154 | blk_status_t status; |
30bc9b53 | 2155 | struct bio *b = r1_bio->bios[i]; |
98d30c58 | 2156 | struct resync_pages *rp = get_resync_pages(b); |
30bc9b53 N |
2157 | if (b->bi_end_io != end_sync_read) |
2158 | continue; | |
4246a0b6 | 2159 | /* fixup the bio for reuse, but preserve errno */ |
4e4cbee9 | 2160 | status = b->bi_status; |
a7c50c94 | 2161 | bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ); |
4e4cbee9 | 2162 | b->bi_status = status; |
4f024f37 | 2163 | b->bi_iter.bi_sector = r1_bio->sector + |
30bc9b53 | 2164 | conf->mirrors[i].rdev->data_offset; |
30bc9b53 | 2165 | b->bi_end_io = end_sync_read; |
98d30c58 ML |
2166 | rp->raid_bio = r1_bio; |
2167 | b->bi_private = rp; | |
30bc9b53 | 2168 | |
fb0eb5df ML |
2169 | /* initialize bvec table again */ |
2170 | md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); | |
30bc9b53 | 2171 | } |
8f19ccb2 | 2172 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2173 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4e4cbee9 | 2174 | !r1_bio->bios[primary]->bi_status) { |
a68e5870 N |
2175 | r1_bio->bios[primary]->bi_end_io = NULL; |
2176 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2177 | break; | |
2178 | } | |
2179 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2180 | for (i = 0; i < conf->raid_disks * 2; i++) { |
2b070cfe | 2181 | int j = 0; |
78d7f5f7 N |
2182 | struct bio *pbio = r1_bio->bios[primary]; |
2183 | struct bio *sbio = r1_bio->bios[i]; | |
4e4cbee9 | 2184 | blk_status_t status = sbio->bi_status; |
44cf0f4d ML |
2185 | struct page **ppages = get_resync_pages(pbio)->pages; |
2186 | struct page **spages = get_resync_pages(sbio)->pages; | |
60928a91 | 2187 | struct bio_vec *bi; |
8fc04e6e | 2188 | int page_len[RESYNC_PAGES] = { 0 }; |
6dc4f100 | 2189 | struct bvec_iter_all iter_all; |
a68e5870 | 2190 | |
2aabaa65 | 2191 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2192 | continue; |
4246a0b6 | 2193 | /* Now we can 'fixup' the error value */ |
4e4cbee9 | 2194 | sbio->bi_status = 0; |
78d7f5f7 | 2195 | |
2b070cfe CH |
2196 | bio_for_each_segment_all(bi, sbio, iter_all) |
2197 | page_len[j++] = bi->bv_len; | |
60928a91 | 2198 | |
4e4cbee9 | 2199 | if (!status) { |
78d7f5f7 | 2200 | for (j = vcnt; j-- ; ) { |
44cf0f4d ML |
2201 | if (memcmp(page_address(ppages[j]), |
2202 | page_address(spages[j]), | |
60928a91 | 2203 | page_len[j])) |
78d7f5f7 | 2204 | break; |
69382e85 | 2205 | } |
78d7f5f7 N |
2206 | } else |
2207 | j = 0; | |
2208 | if (j >= 0) | |
7f7583d4 | 2209 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2210 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4e4cbee9 | 2211 | && !status)) { |
78d7f5f7 N |
2212 | /* No need to write to this device. */ |
2213 | sbio->bi_end_io = NULL; | |
2214 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2215 | continue; | |
2216 | } | |
d3b45c2a KO |
2217 | |
2218 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2219 | } |
a68e5870 N |
2220 | } |
2221 | ||
9f2c9d12 | 2222 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2223 | { |
e8096360 | 2224 | struct r1conf *conf = mddev->private; |
a68e5870 | 2225 | int i; |
8f19ccb2 | 2226 | int disks = conf->raid_disks * 2; |
037d2ff6 | 2227 | struct bio *wbio; |
a68e5870 | 2228 | |
a68e5870 N |
2229 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2230 | /* ouch - failed to read all of that. */ | |
2231 | if (!fix_sync_read_error(r1_bio)) | |
2232 | return; | |
7ca78d57 N |
2233 | |
2234 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2235 | process_checks(r1_bio); |
2236 | ||
d11c171e N |
2237 | /* |
2238 | * schedule writes | |
2239 | */ | |
1da177e4 LT |
2240 | atomic_set(&r1_bio->remaining, 1); |
2241 | for (i = 0; i < disks ; i++) { | |
2242 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2243 | if (wbio->bi_end_io == NULL || |
2244 | (wbio->bi_end_io == end_sync_read && | |
2245 | (i == r1_bio->read_disk || | |
2246 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 | 2247 | continue; |
dfcc34c9 ND |
2248 | if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { |
2249 | abort_sync_write(mddev, r1_bio); | |
0c9d5b12 | 2250 | continue; |
dfcc34c9 | 2251 | } |
1da177e4 | 2252 | |
c34b7ac6 | 2253 | wbio->bi_opf = REQ_OP_WRITE; |
212e7eb7 N |
2254 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2255 | wbio->bi_opf |= MD_FAILFAST; | |
2256 | ||
3e198f78 | 2257 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2258 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2259 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2260 | |
ed00aabd | 2261 | submit_bio_noacct(wbio); |
1da177e4 LT |
2262 | } |
2263 | ||
449808a2 | 2264 | put_sync_write_buf(r1_bio, 1); |
1da177e4 LT |
2265 | } |
2266 | ||
2267 | /* | |
2268 | * This is a kernel thread which: | |
2269 | * | |
2270 | * 1. Retries failed read operations on working mirrors. | |
2271 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2272 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2273 | */ |
2274 | ||
e8096360 | 2275 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2276 | sector_t sect, int sectors) |
2277 | { | |
fd01b88c | 2278 | struct mddev *mddev = conf->mddev; |
867868fb N |
2279 | while(sectors) { |
2280 | int s = sectors; | |
2281 | int d = read_disk; | |
2282 | int success = 0; | |
2283 | int start; | |
3cb03002 | 2284 | struct md_rdev *rdev; |
867868fb N |
2285 | |
2286 | if (s > (PAGE_SIZE>>9)) | |
2287 | s = PAGE_SIZE >> 9; | |
2288 | ||
2289 | do { | |
d2eb35ac N |
2290 | sector_t first_bad; |
2291 | int bad_sectors; | |
2292 | ||
707a6a42 N |
2293 | rcu_read_lock(); |
2294 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2295 | if (rdev && |
da8840a7 | 2296 | (test_bit(In_sync, &rdev->flags) || |
2297 | (!test_bit(Faulty, &rdev->flags) && | |
2298 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2299 | is_badblock(rdev, sect, s, |
707a6a42 N |
2300 | &first_bad, &bad_sectors) == 0) { |
2301 | atomic_inc(&rdev->nr_pending); | |
2302 | rcu_read_unlock(); | |
2303 | if (sync_page_io(rdev, sect, s<<9, | |
4ce4c73f | 2304 | conf->tmppage, REQ_OP_READ, false)) |
707a6a42 N |
2305 | success = 1; |
2306 | rdev_dec_pending(rdev, mddev); | |
2307 | if (success) | |
2308 | break; | |
2309 | } else | |
2310 | rcu_read_unlock(); | |
2311 | d++; | |
2312 | if (d == conf->raid_disks * 2) | |
2313 | d = 0; | |
02c67a3b | 2314 | } while (d != read_disk); |
867868fb N |
2315 | |
2316 | if (!success) { | |
d8f05d29 | 2317 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2318 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2319 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2320 | md_error(mddev, rdev); | |
867868fb N |
2321 | break; |
2322 | } | |
2323 | /* write it back and re-read */ | |
2324 | start = d; | |
2325 | while (d != read_disk) { | |
2326 | if (d==0) | |
8f19ccb2 | 2327 | d = conf->raid_disks * 2; |
867868fb | 2328 | d--; |
707a6a42 N |
2329 | rcu_read_lock(); |
2330 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2331 | if (rdev && |
707a6a42 N |
2332 | !test_bit(Faulty, &rdev->flags)) { |
2333 | atomic_inc(&rdev->nr_pending); | |
2334 | rcu_read_unlock(); | |
d8f05d29 N |
2335 | r1_sync_page_io(rdev, sect, s, |
2336 | conf->tmppage, WRITE); | |
707a6a42 N |
2337 | rdev_dec_pending(rdev, mddev); |
2338 | } else | |
2339 | rcu_read_unlock(); | |
867868fb N |
2340 | } |
2341 | d = start; | |
2342 | while (d != read_disk) { | |
867868fb | 2343 | if (d==0) |
8f19ccb2 | 2344 | d = conf->raid_disks * 2; |
867868fb | 2345 | d--; |
707a6a42 N |
2346 | rcu_read_lock(); |
2347 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2348 | if (rdev && |
b8cb6b4c | 2349 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2350 | atomic_inc(&rdev->nr_pending); |
2351 | rcu_read_unlock(); | |
d8f05d29 N |
2352 | if (r1_sync_page_io(rdev, sect, s, |
2353 | conf->tmppage, READ)) { | |
867868fb | 2354 | atomic_add(s, &rdev->corrected_errors); |
913cce5a | 2355 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %pg)\n", |
1d41c216 N |
2356 | mdname(mddev), s, |
2357 | (unsigned long long)(sect + | |
2358 | rdev->data_offset), | |
913cce5a | 2359 | rdev->bdev); |
867868fb | 2360 | } |
707a6a42 N |
2361 | rdev_dec_pending(rdev, mddev); |
2362 | } else | |
2363 | rcu_read_unlock(); | |
867868fb N |
2364 | } |
2365 | sectors -= s; | |
2366 | sect += s; | |
2367 | } | |
2368 | } | |
2369 | ||
9f2c9d12 | 2370 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2371 | { |
fd01b88c | 2372 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2373 | struct r1conf *conf = mddev->private; |
3cb03002 | 2374 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2375 | |
2376 | /* bio has the data to be written to device 'i' where | |
2377 | * we just recently had a write error. | |
2378 | * We repeatedly clone the bio and trim down to one block, | |
2379 | * then try the write. Where the write fails we record | |
2380 | * a bad block. | |
2381 | * It is conceivable that the bio doesn't exactly align with | |
2382 | * blocks. We must handle this somehow. | |
2383 | * | |
2384 | * We currently own a reference on the rdev. | |
2385 | */ | |
2386 | ||
2387 | int block_sectors; | |
2388 | sector_t sector; | |
2389 | int sectors; | |
2390 | int sect_to_write = r1_bio->sectors; | |
2391 | int ok = 1; | |
2392 | ||
2393 | if (rdev->badblocks.shift < 0) | |
2394 | return 0; | |
2395 | ||
ab713cdc ND |
2396 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2397 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2398 | sector = r1_bio->sector; |
2399 | sectors = ((sector + block_sectors) | |
2400 | & ~(sector_t)(block_sectors - 1)) | |
2401 | - sector; | |
2402 | ||
cd5ff9a1 N |
2403 | while (sect_to_write) { |
2404 | struct bio *wbio; | |
2405 | if (sectors > sect_to_write) | |
2406 | sectors = sect_to_write; | |
2407 | /* Write at 'sector' for 'sectors'*/ | |
2408 | ||
b783863f | 2409 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
abfc426d CH |
2410 | wbio = bio_alloc_clone(rdev->bdev, |
2411 | r1_bio->behind_master_bio, | |
2412 | GFP_NOIO, &mddev->bio_set); | |
b783863f | 2413 | } else { |
abfc426d CH |
2414 | wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio, |
2415 | GFP_NOIO, &mddev->bio_set); | |
b783863f KO |
2416 | } |
2417 | ||
c34b7ac6 | 2418 | wbio->bi_opf = REQ_OP_WRITE; |
4f024f37 KO |
2419 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2420 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2421 | |
6678d83f | 2422 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2423 | wbio->bi_iter.bi_sector += rdev->data_offset; |
4e49ea4a MC |
2424 | |
2425 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2426 | /* failure! */ |
2427 | ok = rdev_set_badblocks(rdev, sector, | |
2428 | sectors, 0) | |
2429 | && ok; | |
2430 | ||
2431 | bio_put(wbio); | |
2432 | sect_to_write -= sectors; | |
2433 | sector += sectors; | |
2434 | sectors = block_sectors; | |
2435 | } | |
2436 | return ok; | |
2437 | } | |
2438 | ||
e8096360 | 2439 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2440 | { |
2441 | int m; | |
2442 | int s = r1_bio->sectors; | |
8f19ccb2 | 2443 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2444 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2445 | struct bio *bio = r1_bio->bios[m]; |
2446 | if (bio->bi_end_io == NULL) | |
2447 | continue; | |
4e4cbee9 | 2448 | if (!bio->bi_status && |
62096bce | 2449 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2450 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2451 | } |
4e4cbee9 | 2452 | if (bio->bi_status && |
62096bce N |
2453 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2454 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2455 | md_error(conf->mddev, rdev); | |
2456 | } | |
2457 | } | |
2458 | put_buf(r1_bio); | |
2459 | md_done_sync(conf->mddev, s, 1); | |
2460 | } | |
2461 | ||
e8096360 | 2462 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2463 | { |
fd76863e | 2464 | int m, idx; |
55ce74d4 | 2465 | bool fail = false; |
fd76863e | 2466 | |
8f19ccb2 | 2467 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2468 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2469 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2470 | rdev_clear_badblocks(rdev, |
2471 | r1_bio->sector, | |
c6563a8c | 2472 | r1_bio->sectors, 0); |
62096bce N |
2473 | rdev_dec_pending(rdev, conf->mddev); |
2474 | } else if (r1_bio->bios[m] != NULL) { | |
2475 | /* This drive got a write error. We need to | |
2476 | * narrow down and record precise write | |
2477 | * errors. | |
2478 | */ | |
55ce74d4 | 2479 | fail = true; |
62096bce N |
2480 | if (!narrow_write_error(r1_bio, m)) { |
2481 | md_error(conf->mddev, | |
2482 | conf->mirrors[m].rdev); | |
2483 | /* an I/O failed, we can't clear the bitmap */ | |
2484 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2485 | } | |
2486 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2487 | conf->mddev); | |
2488 | } | |
55ce74d4 N |
2489 | if (fail) { |
2490 | spin_lock_irq(&conf->device_lock); | |
2491 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
fd76863e | 2492 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2493 | atomic_inc(&conf->nr_queued[idx]); |
55ce74d4 | 2494 | spin_unlock_irq(&conf->device_lock); |
824e47da | 2495 | /* |
2496 | * In case freeze_array() is waiting for condition | |
2497 | * get_unqueued_pending() == extra to be true. | |
2498 | */ | |
2499 | wake_up(&conf->wait_barrier); | |
55ce74d4 | 2500 | md_wakeup_thread(conf->mddev->thread); |
bd8688a1 N |
2501 | } else { |
2502 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2503 | close_write(r1_bio); | |
55ce74d4 | 2504 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2505 | } |
62096bce N |
2506 | } |
2507 | ||
e8096360 | 2508 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2509 | { |
fd01b88c | 2510 | struct mddev *mddev = conf->mddev; |
62096bce | 2511 | struct bio *bio; |
3cb03002 | 2512 | struct md_rdev *rdev; |
c069da44 | 2513 | sector_t sector; |
62096bce N |
2514 | |
2515 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2516 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2517 | * the block and we can fix it. | |
2518 | * We freeze all other IO, and try reading the block from | |
2519 | * other devices. When we find one, we re-write | |
2520 | * and check it that fixes the read error. | |
2521 | * This is all done synchronously while the array is | |
2522 | * frozen | |
2523 | */ | |
7449f699 TM |
2524 | |
2525 | bio = r1_bio->bios[r1_bio->read_disk]; | |
7449f699 TM |
2526 | bio_put(bio); |
2527 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2528 | ||
2e52d449 N |
2529 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2530 | if (mddev->ro == 0 | |
2531 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2532 | freeze_array(conf, 1); |
62096bce N |
2533 | fix_read_error(conf, r1_bio->read_disk, |
2534 | r1_bio->sector, r1_bio->sectors); | |
2535 | unfreeze_array(conf); | |
b33d1062 GK |
2536 | } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { |
2537 | md_error(mddev, rdev); | |
7449f699 TM |
2538 | } else { |
2539 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2540 | } | |
2541 | ||
2e52d449 | 2542 | rdev_dec_pending(rdev, conf->mddev); |
c069da44 | 2543 | sector = r1_bio->sector; |
689389a0 | 2544 | bio = r1_bio->master_bio; |
62096bce | 2545 | |
689389a0 N |
2546 | /* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */ |
2547 | r1_bio->state = 0; | |
2548 | raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); | |
c069da44 | 2549 | allow_barrier(conf, sector); |
62096bce N |
2550 | } |
2551 | ||
4ed8731d | 2552 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2553 | { |
4ed8731d | 2554 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2555 | struct r1bio *r1_bio; |
1da177e4 | 2556 | unsigned long flags; |
e8096360 | 2557 | struct r1conf *conf = mddev->private; |
1da177e4 | 2558 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2559 | struct blk_plug plug; |
fd76863e | 2560 | int idx; |
1da177e4 LT |
2561 | |
2562 | md_check_recovery(mddev); | |
e1dfa0a2 | 2563 | |
55ce74d4 | 2564 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2565 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2566 | LIST_HEAD(tmp); |
2567 | spin_lock_irqsave(&conf->device_lock, flags); | |
fd76863e | 2568 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
2569 | list_splice_init(&conf->bio_end_io_list, &tmp); | |
55ce74d4 N |
2570 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2571 | while (!list_empty(&tmp)) { | |
a452744b MP |
2572 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2573 | retry_list); | |
55ce74d4 | 2574 | list_del(&r1_bio->retry_list); |
fd76863e | 2575 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2576 | atomic_dec(&conf->nr_queued[idx]); |
bd8688a1 N |
2577 | if (mddev->degraded) |
2578 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2579 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2580 | close_write(r1_bio); | |
55ce74d4 N |
2581 | raid_end_bio_io(r1_bio); |
2582 | } | |
2583 | } | |
2584 | ||
e1dfa0a2 | 2585 | blk_start_plug(&plug); |
1da177e4 | 2586 | for (;;) { |
191ea9b2 | 2587 | |
0021b7bc | 2588 | flush_pending_writes(conf); |
191ea9b2 | 2589 | |
a35e63ef N |
2590 | spin_lock_irqsave(&conf->device_lock, flags); |
2591 | if (list_empty(head)) { | |
2592 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2593 | break; |
a35e63ef | 2594 | } |
9f2c9d12 | 2595 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2596 | list_del(head->prev); |
fd76863e | 2597 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2598 | atomic_dec(&conf->nr_queued[idx]); |
1da177e4 LT |
2599 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2600 | ||
2601 | mddev = r1_bio->mddev; | |
070ec55d | 2602 | conf = mddev->private; |
4367af55 | 2603 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2604 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2605 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2606 | handle_sync_write_finished(conf, r1_bio); | |
2607 | else | |
4367af55 | 2608 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2609 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2610 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2611 | handle_write_finished(conf, r1_bio); | |
2612 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2613 | handle_read_error(conf, r1_bio); | |
2614 | else | |
c230e7e5 | 2615 | WARN_ON_ONCE(1); |
62096bce | 2616 | |
1d9d5241 | 2617 | cond_resched(); |
2953079c | 2618 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2619 | md_check_recovery(mddev); |
1da177e4 | 2620 | } |
e1dfa0a2 | 2621 | blk_finish_plug(&plug); |
1da177e4 LT |
2622 | } |
2623 | ||
e8096360 | 2624 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2625 | { |
2626 | int buffs; | |
2627 | ||
2628 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 KO |
2629 | BUG_ON(mempool_initialized(&conf->r1buf_pool)); |
2630 | ||
2631 | return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, | |
2632 | r1buf_pool_free, conf->poolinfo); | |
1da177e4 LT |
2633 | } |
2634 | ||
208410b5 SL |
2635 | static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf) |
2636 | { | |
afeee514 | 2637 | struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); |
208410b5 SL |
2638 | struct resync_pages *rps; |
2639 | struct bio *bio; | |
2640 | int i; | |
2641 | ||
2642 | for (i = conf->poolinfo->raid_disks; i--; ) { | |
2643 | bio = r1bio->bios[i]; | |
2644 | rps = bio->bi_private; | |
a7c50c94 | 2645 | bio_reset(bio, NULL, 0); |
208410b5 SL |
2646 | bio->bi_private = rps; |
2647 | } | |
2648 | r1bio->master_bio = NULL; | |
2649 | return r1bio; | |
2650 | } | |
2651 | ||
1da177e4 LT |
2652 | /* |
2653 | * perform a "sync" on one "block" | |
2654 | * | |
2655 | * We need to make sure that no normal I/O request - particularly write | |
2656 | * requests - conflict with active sync requests. | |
2657 | * | |
2658 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2659 | * that can be installed to exclude normal IO requests. | |
2660 | */ | |
2661 | ||
849674e4 SL |
2662 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2663 | int *skipped) | |
1da177e4 | 2664 | { |
e8096360 | 2665 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2666 | struct r1bio *r1_bio; |
1da177e4 LT |
2667 | struct bio *bio; |
2668 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2669 | int disk = -1; |
1da177e4 | 2670 | int i; |
3e198f78 N |
2671 | int wonly = -1; |
2672 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2673 | sector_t sync_blocks; |
e3b9703e | 2674 | int still_degraded = 0; |
06f60385 N |
2675 | int good_sectors = RESYNC_SECTORS; |
2676 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
fd76863e | 2677 | int idx = sector_to_idx(sector_nr); |
022e510f | 2678 | int page_idx = 0; |
1da177e4 | 2679 | |
afeee514 | 2680 | if (!mempool_initialized(&conf->r1buf_pool)) |
1da177e4 | 2681 | if (init_resync(conf)) |
57afd89f | 2682 | return 0; |
1da177e4 | 2683 | |
58c0fed4 | 2684 | max_sector = mddev->dev_sectors; |
1da177e4 | 2685 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2686 | /* If we aborted, we need to abort the |
2687 | * sync on the 'current' bitmap chunk (there will | |
2688 | * only be one in raid1 resync. | |
2689 | * We can find the current addess in mddev->curr_resync | |
2690 | */ | |
6a806c51 | 2691 | if (mddev->curr_resync < max_sector) /* aborted */ |
e64e4018 AS |
2692 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
2693 | &sync_blocks, 1); | |
6a806c51 | 2694 | else /* completed sync */ |
191ea9b2 | 2695 | conf->fullsync = 0; |
6a806c51 | 2696 | |
e64e4018 | 2697 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2698 | close_sync(conf); |
c40f341f GR |
2699 | |
2700 | if (mddev_is_clustered(mddev)) { | |
2701 | conf->cluster_sync_low = 0; | |
2702 | conf->cluster_sync_high = 0; | |
c40f341f | 2703 | } |
1da177e4 LT |
2704 | return 0; |
2705 | } | |
2706 | ||
07d84d10 N |
2707 | if (mddev->bitmap == NULL && |
2708 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2709 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2710 | conf->fullsync == 0) { |
2711 | *skipped = 1; | |
2712 | return max_sector - sector_nr; | |
2713 | } | |
6394cca5 N |
2714 | /* before building a request, check if we can skip these blocks.. |
2715 | * This call the bitmap_start_sync doesn't actually record anything | |
2716 | */ | |
e64e4018 | 2717 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2718 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2719 | /* We can skip this block, and probably several more */ |
2720 | *skipped = 1; | |
2721 | return sync_blocks; | |
2722 | } | |
17999be4 | 2723 | |
7ac50447 TM |
2724 | /* |
2725 | * If there is non-resync activity waiting for a turn, then let it | |
2726 | * though before starting on this new sync request. | |
2727 | */ | |
824e47da | 2728 | if (atomic_read(&conf->nr_waiting[idx])) |
7ac50447 TM |
2729 | schedule_timeout_uninterruptible(1); |
2730 | ||
c40f341f GR |
2731 | /* we are incrementing sector_nr below. To be safe, we check against |
2732 | * sector_nr + two times RESYNC_SECTORS | |
2733 | */ | |
2734 | ||
e64e4018 | 2735 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
c40f341f | 2736 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); |
17999be4 | 2737 | |
8c242593 YY |
2738 | |
2739 | if (raise_barrier(conf, sector_nr)) | |
2740 | return 0; | |
2741 | ||
2742 | r1_bio = raid1_alloc_init_r1buf(conf); | |
1da177e4 | 2743 | |
3e198f78 | 2744 | rcu_read_lock(); |
1da177e4 | 2745 | /* |
3e198f78 N |
2746 | * If we get a correctably read error during resync or recovery, |
2747 | * we might want to read from a different device. So we | |
2748 | * flag all drives that could conceivably be read from for READ, | |
2749 | * and any others (which will be non-In_sync devices) for WRITE. | |
2750 | * If a read fails, we try reading from something else for which READ | |
2751 | * is OK. | |
1da177e4 | 2752 | */ |
1da177e4 | 2753 | |
1da177e4 LT |
2754 | r1_bio->mddev = mddev; |
2755 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2756 | r1_bio->state = 0; |
1da177e4 | 2757 | set_bit(R1BIO_IsSync, &r1_bio->state); |
fd76863e | 2758 | /* make sure good_sectors won't go across barrier unit boundary */ |
2759 | good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); | |
1da177e4 | 2760 | |
8f19ccb2 | 2761 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2762 | struct md_rdev *rdev; |
1da177e4 | 2763 | bio = r1_bio->bios[i]; |
1da177e4 | 2764 | |
3e198f78 N |
2765 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2766 | if (rdev == NULL || | |
06f60385 | 2767 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2768 | if (i < conf->raid_disks) |
2769 | still_degraded = 1; | |
3e198f78 | 2770 | } else if (!test_bit(In_sync, &rdev->flags)) { |
c34b7ac6 | 2771 | bio->bi_opf = REQ_OP_WRITE; |
1da177e4 LT |
2772 | bio->bi_end_io = end_sync_write; |
2773 | write_targets ++; | |
3e198f78 N |
2774 | } else { |
2775 | /* may need to read from here */ | |
06f60385 N |
2776 | sector_t first_bad = MaxSector; |
2777 | int bad_sectors; | |
2778 | ||
2779 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2780 | &first_bad, &bad_sectors)) { | |
2781 | if (first_bad > sector_nr) | |
2782 | good_sectors = first_bad - sector_nr; | |
2783 | else { | |
2784 | bad_sectors -= (sector_nr - first_bad); | |
2785 | if (min_bad == 0 || | |
2786 | min_bad > bad_sectors) | |
2787 | min_bad = bad_sectors; | |
2788 | } | |
2789 | } | |
2790 | if (sector_nr < first_bad) { | |
2791 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2792 | if (wonly < 0) | |
2793 | wonly = i; | |
2794 | } else { | |
2795 | if (disk < 0) | |
2796 | disk = i; | |
2797 | } | |
c34b7ac6 | 2798 | bio->bi_opf = REQ_OP_READ; |
06f60385 N |
2799 | bio->bi_end_io = end_sync_read; |
2800 | read_targets++; | |
d57368af AL |
2801 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2802 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2803 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2804 | /* | |
2805 | * The device is suitable for reading (InSync), | |
2806 | * but has bad block(s) here. Let's try to correct them, | |
2807 | * if we are doing resync or repair. Otherwise, leave | |
2808 | * this device alone for this sync request. | |
2809 | */ | |
c34b7ac6 | 2810 | bio->bi_opf = REQ_OP_WRITE; |
d57368af AL |
2811 | bio->bi_end_io = end_sync_write; |
2812 | write_targets++; | |
3e198f78 | 2813 | } |
3e198f78 | 2814 | } |
028288df | 2815 | if (rdev && bio->bi_end_io) { |
06f60385 | 2816 | atomic_inc(&rdev->nr_pending); |
4f024f37 | 2817 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
74d46992 | 2818 | bio_set_dev(bio, rdev->bdev); |
2e52d449 N |
2819 | if (test_bit(FailFast, &rdev->flags)) |
2820 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2821 | } |
1da177e4 | 2822 | } |
3e198f78 N |
2823 | rcu_read_unlock(); |
2824 | if (disk < 0) | |
2825 | disk = wonly; | |
2826 | r1_bio->read_disk = disk; | |
191ea9b2 | 2827 | |
06f60385 N |
2828 | if (read_targets == 0 && min_bad > 0) { |
2829 | /* These sectors are bad on all InSync devices, so we | |
2830 | * need to mark them bad on all write targets | |
2831 | */ | |
2832 | int ok = 1; | |
8f19ccb2 | 2833 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2834 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2835 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2836 | ok = rdev_set_badblocks(rdev, sector_nr, |
2837 | min_bad, 0 | |
2838 | ) && ok; | |
2839 | } | |
2953079c | 2840 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2841 | *skipped = 1; |
2842 | put_buf(r1_bio); | |
2843 | ||
2844 | if (!ok) { | |
2845 | /* Cannot record the badblocks, so need to | |
2846 | * abort the resync. | |
2847 | * If there are multiple read targets, could just | |
2848 | * fail the really bad ones ??? | |
2849 | */ | |
2850 | conf->recovery_disabled = mddev->recovery_disabled; | |
2851 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2852 | return 0; | |
2853 | } else | |
2854 | return min_bad; | |
2855 | ||
2856 | } | |
2857 | if (min_bad > 0 && min_bad < good_sectors) { | |
2858 | /* only resync enough to reach the next bad->good | |
2859 | * transition */ | |
2860 | good_sectors = min_bad; | |
2861 | } | |
2862 | ||
3e198f78 N |
2863 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2864 | /* extra read targets are also write targets */ | |
2865 | write_targets += read_targets-1; | |
2866 | ||
2867 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2868 | /* There is nowhere to write, so all non-sync |
2869 | * drives must be failed - so we are finished | |
2870 | */ | |
b7219ccb N |
2871 | sector_t rv; |
2872 | if (min_bad > 0) | |
2873 | max_sector = sector_nr + min_bad; | |
2874 | rv = max_sector - sector_nr; | |
57afd89f | 2875 | *skipped = 1; |
1da177e4 | 2876 | put_buf(r1_bio); |
1da177e4 LT |
2877 | return rv; |
2878 | } | |
2879 | ||
c6207277 N |
2880 | if (max_sector > mddev->resync_max) |
2881 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2882 | if (max_sector > sector_nr + good_sectors) |
2883 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2884 | nr_sectors = 0; |
289e99e8 | 2885 | sync_blocks = 0; |
1da177e4 LT |
2886 | do { |
2887 | struct page *page; | |
2888 | int len = PAGE_SIZE; | |
2889 | if (sector_nr + (len>>9) > max_sector) | |
2890 | len = (max_sector - sector_nr) << 9; | |
2891 | if (len == 0) | |
2892 | break; | |
6a806c51 | 2893 | if (sync_blocks == 0) { |
e64e4018 AS |
2894 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
2895 | &sync_blocks, still_degraded) && | |
e5de485f N |
2896 | !conf->fullsync && |
2897 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2898 | break; |
7571ae88 | 2899 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2900 | len = sync_blocks<<9; |
ab7a30c7 | 2901 | } |
191ea9b2 | 2902 | |
8f19ccb2 | 2903 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
98d30c58 ML |
2904 | struct resync_pages *rp; |
2905 | ||
1da177e4 | 2906 | bio = r1_bio->bios[i]; |
98d30c58 | 2907 | rp = get_resync_pages(bio); |
1da177e4 | 2908 | if (bio->bi_end_io) { |
022e510f | 2909 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
2910 | |
2911 | /* | |
2912 | * won't fail because the vec table is big | |
2913 | * enough to hold all these pages | |
2914 | */ | |
f8312322 | 2915 | __bio_add_page(bio, page, len, 0); |
1da177e4 LT |
2916 | } |
2917 | } | |
2918 | nr_sectors += len>>9; | |
2919 | sector_nr += len>>9; | |
191ea9b2 | 2920 | sync_blocks -= (len>>9); |
022e510f | 2921 | } while (++page_idx < RESYNC_PAGES); |
98d30c58 | 2922 | |
1da177e4 LT |
2923 | r1_bio->sectors = nr_sectors; |
2924 | ||
c40f341f GR |
2925 | if (mddev_is_clustered(mddev) && |
2926 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2927 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2928 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2929 | /* Send resync message */ | |
2930 | md_cluster_ops->resync_info_update(mddev, | |
2931 | conf->cluster_sync_low, | |
2932 | conf->cluster_sync_high); | |
2933 | } | |
2934 | ||
d11c171e N |
2935 | /* For a user-requested sync, we read all readable devices and do a |
2936 | * compare | |
2937 | */ | |
2938 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2939 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2940 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2941 | bio = r1_bio->bios[i]; |
2942 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2943 | read_targets--; |
74d46992 | 2944 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2945 | if (read_targets == 1) |
2946 | bio->bi_opf &= ~MD_FAILFAST; | |
ed00aabd | 2947 | submit_bio_noacct(bio); |
d11c171e N |
2948 | } |
2949 | } | |
2950 | } else { | |
2951 | atomic_set(&r1_bio->remaining, 1); | |
2952 | bio = r1_bio->bios[r1_bio->read_disk]; | |
74d46992 | 2953 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2954 | if (read_targets == 1) |
2955 | bio->bi_opf &= ~MD_FAILFAST; | |
ed00aabd | 2956 | submit_bio_noacct(bio); |
d11c171e | 2957 | } |
1da177e4 LT |
2958 | return nr_sectors; |
2959 | } | |
2960 | ||
fd01b88c | 2961 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2962 | { |
2963 | if (sectors) | |
2964 | return sectors; | |
2965 | ||
2966 | return mddev->dev_sectors; | |
2967 | } | |
2968 | ||
e8096360 | 2969 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2970 | { |
e8096360 | 2971 | struct r1conf *conf; |
709ae487 | 2972 | int i; |
0eaf822c | 2973 | struct raid1_info *disk; |
3cb03002 | 2974 | struct md_rdev *rdev; |
709ae487 | 2975 | int err = -ENOMEM; |
1da177e4 | 2976 | |
e8096360 | 2977 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2978 | if (!conf) |
709ae487 | 2979 | goto abort; |
1da177e4 | 2980 | |
fd76863e | 2981 | conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, |
824e47da | 2982 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2983 | if (!conf->nr_pending) |
2984 | goto abort; | |
2985 | ||
2986 | conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2987 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2988 | if (!conf->nr_waiting) |
2989 | goto abort; | |
2990 | ||
2991 | conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2992 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2993 | if (!conf->nr_queued) |
2994 | goto abort; | |
2995 | ||
2996 | conf->barrier = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2997 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2998 | if (!conf->barrier) |
2999 | goto abort; | |
3000 | ||
6396bb22 KC |
3001 | conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
3002 | mddev->raid_disks, 2), | |
3003 | GFP_KERNEL); | |
1da177e4 | 3004 | if (!conf->mirrors) |
709ae487 | 3005 | goto abort; |
1da177e4 | 3006 | |
ddaf22ab N |
3007 | conf->tmppage = alloc_page(GFP_KERNEL); |
3008 | if (!conf->tmppage) | |
709ae487 | 3009 | goto abort; |
ddaf22ab | 3010 | |
709ae487 | 3011 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 3012 | if (!conf->poolinfo) |
709ae487 | 3013 | goto abort; |
8f19ccb2 | 3014 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
3f677f9c | 3015 | err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 3016 | rbio_pool_free, conf->poolinfo); |
afeee514 | 3017 | if (err) |
709ae487 N |
3018 | goto abort; |
3019 | ||
afeee514 KO |
3020 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
3021 | if (err) | |
c230e7e5 N |
3022 | goto abort; |
3023 | ||
ed9bfdf1 | 3024 | conf->poolinfo->mddev = mddev; |
1da177e4 | 3025 | |
c19d5798 | 3026 | err = -EINVAL; |
e7e72bf6 | 3027 | spin_lock_init(&conf->device_lock); |
dafb20fa | 3028 | rdev_for_each(rdev, mddev) { |
709ae487 | 3029 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
3030 | if (disk_idx >= mddev->raid_disks |
3031 | || disk_idx < 0) | |
3032 | continue; | |
c19d5798 | 3033 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 3034 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
3035 | else |
3036 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 3037 | |
c19d5798 N |
3038 | if (disk->rdev) |
3039 | goto abort; | |
1da177e4 | 3040 | disk->rdev = rdev; |
1da177e4 | 3041 | disk->head_position = 0; |
12cee5a8 | 3042 | disk->seq_start = MaxSector; |
1da177e4 LT |
3043 | } |
3044 | conf->raid_disks = mddev->raid_disks; | |
3045 | conf->mddev = mddev; | |
1da177e4 | 3046 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 3047 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
3048 | |
3049 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 3050 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 3051 | |
191ea9b2 | 3052 | bio_list_init(&conf->pending_bio_list); |
d890fa2b | 3053 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 3054 | |
c19d5798 | 3055 | err = -EIO; |
8f19ccb2 | 3056 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
3057 | |
3058 | disk = conf->mirrors + i; | |
3059 | ||
c19d5798 N |
3060 | if (i < conf->raid_disks && |
3061 | disk[conf->raid_disks].rdev) { | |
3062 | /* This slot has a replacement. */ | |
3063 | if (!disk->rdev) { | |
3064 | /* No original, just make the replacement | |
3065 | * a recovering spare | |
3066 | */ | |
3067 | disk->rdev = | |
3068 | disk[conf->raid_disks].rdev; | |
3069 | disk[conf->raid_disks].rdev = NULL; | |
3070 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
3071 | /* Original is not in_sync - bad */ | |
3072 | goto abort; | |
3073 | } | |
3074 | ||
5fd6c1dc N |
3075 | if (!disk->rdev || |
3076 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 3077 | disk->head_position = 0; |
4f0a5e01 JB |
3078 | if (disk->rdev && |
3079 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 3080 | conf->fullsync = 1; |
be4d3280 | 3081 | } |
1da177e4 | 3082 | } |
709ae487 | 3083 | |
709ae487 | 3084 | err = -ENOMEM; |
44693154 YK |
3085 | rcu_assign_pointer(conf->thread, |
3086 | md_register_thread(raid1d, mddev, "raid1")); | |
1d41c216 | 3087 | if (!conf->thread) |
709ae487 | 3088 | goto abort; |
1da177e4 | 3089 | |
709ae487 N |
3090 | return conf; |
3091 | ||
3092 | abort: | |
3093 | if (conf) { | |
afeee514 | 3094 | mempool_exit(&conf->r1bio_pool); |
709ae487 N |
3095 | kfree(conf->mirrors); |
3096 | safe_put_page(conf->tmppage); | |
3097 | kfree(conf->poolinfo); | |
fd76863e | 3098 | kfree(conf->nr_pending); |
3099 | kfree(conf->nr_waiting); | |
3100 | kfree(conf->nr_queued); | |
3101 | kfree(conf->barrier); | |
afeee514 | 3102 | bioset_exit(&conf->bio_split); |
709ae487 N |
3103 | kfree(conf); |
3104 | } | |
3105 | return ERR_PTR(err); | |
3106 | } | |
3107 | ||
afa0f557 | 3108 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 3109 | static int raid1_run(struct mddev *mddev) |
709ae487 | 3110 | { |
e8096360 | 3111 | struct r1conf *conf; |
709ae487 | 3112 | int i; |
3cb03002 | 3113 | struct md_rdev *rdev; |
5220ea1e | 3114 | int ret; |
709ae487 N |
3115 | |
3116 | if (mddev->level != 1) { | |
1d41c216 N |
3117 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3118 | mdname(mddev), mddev->level); | |
709ae487 N |
3119 | return -EIO; |
3120 | } | |
3121 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3122 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3123 | mdname(mddev)); | |
709ae487 N |
3124 | return -EIO; |
3125 | } | |
a415c0f1 N |
3126 | if (mddev_init_writes_pending(mddev) < 0) |
3127 | return -ENOMEM; | |
1da177e4 | 3128 | /* |
709ae487 N |
3129 | * copy the already verified devices into our private RAID1 |
3130 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3131 | * should be freed in raid1_free()] |
1da177e4 | 3132 | */ |
709ae487 N |
3133 | if (mddev->private == NULL) |
3134 | conf = setup_conf(mddev); | |
3135 | else | |
3136 | conf = mddev->private; | |
1da177e4 | 3137 | |
709ae487 N |
3138 | if (IS_ERR(conf)) |
3139 | return PTR_ERR(conf); | |
1da177e4 | 3140 | |
10fa225c | 3141 | if (mddev->queue) |
3deff1a7 | 3142 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
5026d7a9 | 3143 | |
dafb20fa | 3144 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3145 | if (!mddev->gendisk) |
3146 | continue; | |
709ae487 N |
3147 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3148 | rdev->data_offset << 9); | |
1da177e4 | 3149 | } |
191ea9b2 | 3150 | |
709ae487 | 3151 | mddev->degraded = 0; |
ebfeb444 | 3152 | for (i = 0; i < conf->raid_disks; i++) |
709ae487 N |
3153 | if (conf->mirrors[i].rdev == NULL || |
3154 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3155 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3156 | mddev->degraded++; | |
07f1a685 YY |
3157 | /* |
3158 | * RAID1 needs at least one disk in active | |
3159 | */ | |
3160 | if (conf->raid_disks - mddev->degraded < 1) { | |
7eb8ff02 | 3161 | md_unregister_thread(mddev, &conf->thread); |
07f1a685 YY |
3162 | ret = -EINVAL; |
3163 | goto abort; | |
3164 | } | |
709ae487 N |
3165 | |
3166 | if (conf->raid_disks - mddev->degraded == 1) | |
3167 | mddev->recovery_cp = MaxSector; | |
3168 | ||
8c6ac868 | 3169 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3170 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3171 | mdname(mddev)); | |
3172 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3173 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3174 | mddev->raid_disks); |
709ae487 | 3175 | |
1da177e4 LT |
3176 | /* |
3177 | * Ok, everything is just fine now | |
3178 | */ | |
44693154 YK |
3179 | rcu_assign_pointer(mddev->thread, conf->thread); |
3180 | rcu_assign_pointer(conf->thread, NULL); | |
709ae487 | 3181 | mddev->private = conf; |
46533ff7 | 3182 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3183 | |
1f403624 | 3184 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3185 | |
ebfeb444 | 3186 | ret = md_integrity_register(mddev); |
5aa61f42 | 3187 | if (ret) { |
7eb8ff02 | 3188 | md_unregister_thread(mddev, &mddev->thread); |
07f1a685 | 3189 | goto abort; |
5aa61f42 | 3190 | } |
07f1a685 YY |
3191 | return 0; |
3192 | ||
3193 | abort: | |
3194 | raid1_free(mddev, conf); | |
5220ea1e | 3195 | return ret; |
1da177e4 LT |
3196 | } |
3197 | ||
afa0f557 | 3198 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3199 | { |
afa0f557 | 3200 | struct r1conf *conf = priv; |
409c57f3 | 3201 | |
afeee514 | 3202 | mempool_exit(&conf->r1bio_pool); |
990a8baf | 3203 | kfree(conf->mirrors); |
0fea7ed8 | 3204 | safe_put_page(conf->tmppage); |
990a8baf | 3205 | kfree(conf->poolinfo); |
fd76863e | 3206 | kfree(conf->nr_pending); |
3207 | kfree(conf->nr_waiting); | |
3208 | kfree(conf->nr_queued); | |
3209 | kfree(conf->barrier); | |
afeee514 | 3210 | bioset_exit(&conf->bio_split); |
1da177e4 | 3211 | kfree(conf); |
1da177e4 LT |
3212 | } |
3213 | ||
fd01b88c | 3214 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3215 | { |
3216 | /* no resync is happening, and there is enough space | |
3217 | * on all devices, so we can resize. | |
3218 | * We need to make sure resync covers any new space. | |
3219 | * If the array is shrinking we should possibly wait until | |
3220 | * any io in the removed space completes, but it hardly seems | |
3221 | * worth it. | |
3222 | */ | |
a4a6125a N |
3223 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3224 | if (mddev->external_size && | |
3225 | mddev->array_sectors > newsize) | |
b522adcd | 3226 | return -EINVAL; |
a4a6125a | 3227 | if (mddev->bitmap) { |
e64e4018 | 3228 | int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); |
a4a6125a N |
3229 | if (ret) |
3230 | return ret; | |
3231 | } | |
3232 | md_set_array_sectors(mddev, newsize); | |
b522adcd | 3233 | if (sectors > mddev->dev_sectors && |
b098636c | 3234 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3235 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3236 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3237 | } | |
b522adcd | 3238 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3239 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3240 | return 0; |
3241 | } | |
3242 | ||
fd01b88c | 3243 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3244 | { |
3245 | /* We need to: | |
3246 | * 1/ resize the r1bio_pool | |
3247 | * 2/ resize conf->mirrors | |
3248 | * | |
3249 | * We allocate a new r1bio_pool if we can. | |
3250 | * Then raise a device barrier and wait until all IO stops. | |
3251 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3252 | * |
3253 | * At the same time, we "pack" the devices so that all the missing | |
3254 | * devices have the higher raid_disk numbers. | |
1da177e4 | 3255 | */ |
afeee514 | 3256 | mempool_t newpool, oldpool; |
1da177e4 | 3257 | struct pool_info *newpoolinfo; |
0eaf822c | 3258 | struct raid1_info *newmirrors; |
e8096360 | 3259 | struct r1conf *conf = mddev->private; |
63c70c4f | 3260 | int cnt, raid_disks; |
c04be0aa | 3261 | unsigned long flags; |
2214c260 | 3262 | int d, d2; |
afeee514 KO |
3263 | int ret; |
3264 | ||
3265 | memset(&newpool, 0, sizeof(newpool)); | |
3266 | memset(&oldpool, 0, sizeof(oldpool)); | |
1da177e4 | 3267 | |
63c70c4f | 3268 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3269 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3270 | mddev->layout != mddev->new_layout || |
3271 | mddev->level != mddev->new_level) { | |
664e7c41 | 3272 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3273 | mddev->new_layout = mddev->layout; |
3274 | mddev->new_level = mddev->level; | |
3275 | return -EINVAL; | |
3276 | } | |
3277 | ||
2214c260 AP |
3278 | if (!mddev_is_clustered(mddev)) |
3279 | md_allow_write(mddev); | |
2a2275d6 | 3280 | |
63c70c4f N |
3281 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3282 | ||
6ea9c07c N |
3283 | if (raid_disks < conf->raid_disks) { |
3284 | cnt=0; | |
3285 | for (d= 0; d < conf->raid_disks; d++) | |
3286 | if (conf->mirrors[d].rdev) | |
3287 | cnt++; | |
3288 | if (cnt > raid_disks) | |
1da177e4 | 3289 | return -EBUSY; |
6ea9c07c | 3290 | } |
1da177e4 LT |
3291 | |
3292 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3293 | if (!newpoolinfo) | |
3294 | return -ENOMEM; | |
3295 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3296 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 | 3297 | |
3f677f9c | 3298 | ret = mempool_init(&newpool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 3299 | rbio_pool_free, newpoolinfo); |
afeee514 | 3300 | if (ret) { |
1da177e4 | 3301 | kfree(newpoolinfo); |
afeee514 | 3302 | return ret; |
1da177e4 | 3303 | } |
6396bb22 KC |
3304 | newmirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
3305 | raid_disks, 2), | |
8f19ccb2 | 3306 | GFP_KERNEL); |
1da177e4 LT |
3307 | if (!newmirrors) { |
3308 | kfree(newpoolinfo); | |
afeee514 | 3309 | mempool_exit(&newpool); |
1da177e4 LT |
3310 | return -ENOMEM; |
3311 | } | |
1da177e4 | 3312 | |
e2d59925 | 3313 | freeze_array(conf, 0); |
1da177e4 LT |
3314 | |
3315 | /* ok, everything is stopped */ | |
3316 | oldpool = conf->r1bio_pool; | |
3317 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3318 | |
a88aa786 | 3319 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3320 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3321 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3322 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3323 | rdev->raid_disk = d2; |
36fad858 NK |
3324 | sysfs_unlink_rdev(mddev, rdev); |
3325 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3326 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3327 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3328 | } |
a88aa786 N |
3329 | if (rdev) |
3330 | newmirrors[d2++].rdev = rdev; | |
3331 | } | |
1da177e4 LT |
3332 | kfree(conf->mirrors); |
3333 | conf->mirrors = newmirrors; | |
3334 | kfree(conf->poolinfo); | |
3335 | conf->poolinfo = newpoolinfo; | |
3336 | ||
c04be0aa | 3337 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3338 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3339 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3340 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3341 | mddev->delta_disks = 0; |
1da177e4 | 3342 | |
e2d59925 | 3343 | unfreeze_array(conf); |
1da177e4 | 3344 | |
985ca973 | 3345 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3346 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3347 | md_wakeup_thread(mddev->thread); | |
3348 | ||
afeee514 | 3349 | mempool_exit(&oldpool); |
1da177e4 LT |
3350 | return 0; |
3351 | } | |
3352 | ||
b03e0ccb | 3353 | static void raid1_quiesce(struct mddev *mddev, int quiesce) |
36fa3063 | 3354 | { |
e8096360 | 3355 | struct r1conf *conf = mddev->private; |
36fa3063 | 3356 | |
b03e0ccb | 3357 | if (quiesce) |
07169fd4 | 3358 | freeze_array(conf, 0); |
b03e0ccb | 3359 | else |
07169fd4 | 3360 | unfreeze_array(conf); |
36fa3063 N |
3361 | } |
3362 | ||
fd01b88c | 3363 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3364 | { |
3365 | /* raid1 can take over: | |
3366 | * raid5 with 2 devices, any layout or chunk size | |
3367 | */ | |
3368 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3369 | struct r1conf *conf; |
709ae487 N |
3370 | mddev->new_level = 1; |
3371 | mddev->new_layout = 0; | |
3372 | mddev->new_chunk_sectors = 0; | |
3373 | conf = setup_conf(mddev); | |
6995f0b2 | 3374 | if (!IS_ERR(conf)) { |
07169fd4 | 3375 | /* Array must appear to be quiesced */ |
3376 | conf->array_frozen = 1; | |
394ed8e4 SL |
3377 | mddev_clear_unsupported_flags(mddev, |
3378 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3379 | } |
709ae487 N |
3380 | return conf; |
3381 | } | |
3382 | return ERR_PTR(-EINVAL); | |
3383 | } | |
1da177e4 | 3384 | |
84fc4b56 | 3385 | static struct md_personality raid1_personality = |
1da177e4 LT |
3386 | { |
3387 | .name = "raid1", | |
2604b703 | 3388 | .level = 1, |
1da177e4 | 3389 | .owner = THIS_MODULE, |
849674e4 SL |
3390 | .make_request = raid1_make_request, |
3391 | .run = raid1_run, | |
afa0f557 | 3392 | .free = raid1_free, |
849674e4 SL |
3393 | .status = raid1_status, |
3394 | .error_handler = raid1_error, | |
1da177e4 LT |
3395 | .hot_add_disk = raid1_add_disk, |
3396 | .hot_remove_disk= raid1_remove_disk, | |
3397 | .spare_active = raid1_spare_active, | |
849674e4 | 3398 | .sync_request = raid1_sync_request, |
1da177e4 | 3399 | .resize = raid1_resize, |
80c3a6ce | 3400 | .size = raid1_size, |
63c70c4f | 3401 | .check_reshape = raid1_reshape, |
36fa3063 | 3402 | .quiesce = raid1_quiesce, |
709ae487 | 3403 | .takeover = raid1_takeover, |
1da177e4 LT |
3404 | }; |
3405 | ||
3406 | static int __init raid_init(void) | |
3407 | { | |
2604b703 | 3408 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3409 | } |
3410 | ||
3411 | static void raid_exit(void) | |
3412 | { | |
2604b703 | 3413 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3414 | } |
3415 | ||
3416 | module_init(raid_init); | |
3417 | module_exit(raid_exit); | |
3418 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3419 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3420 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3421 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3422 | MODULE_ALIAS("md-level-1"); |