]>
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-- ; ) { | |
6746557f | 168 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
169 | if (!bio) |
170 | goto out_free_bio; | |
171 | r1_bio->bios[j] = bio; | |
172 | } | |
173 | /* | |
174 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
175 | * the first bio. |
176 | * If this is a user-requested check/repair, allocate | |
177 | * RESYNC_PAGES for each bio. | |
1da177e4 | 178 | */ |
d11c171e | 179 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 180 | need_pages = pi->raid_disks; |
d11c171e | 181 | else |
da1aab3d | 182 | need_pages = 1; |
98d30c58 ML |
183 | for (j = 0; j < pi->raid_disks; j++) { |
184 | struct resync_pages *rp = &rps[j]; | |
185 | ||
d11c171e | 186 | bio = r1_bio->bios[j]; |
d11c171e | 187 | |
98d30c58 ML |
188 | if (j < need_pages) { |
189 | if (resync_alloc_pages(rp, gfp_flags)) | |
190 | goto out_free_pages; | |
191 | } else { | |
192 | memcpy(rp, &rps[0], sizeof(*rp)); | |
193 | resync_get_all_pages(rp); | |
194 | } | |
195 | ||
98d30c58 ML |
196 | rp->raid_bio = r1_bio; |
197 | bio->bi_private = rp; | |
1da177e4 LT |
198 | } |
199 | ||
200 | r1_bio->master_bio = NULL; | |
201 | ||
202 | return r1_bio; | |
203 | ||
da1aab3d | 204 | out_free_pages: |
491221f8 | 205 | while (--j >= 0) |
98d30c58 | 206 | resync_free_pages(&rps[j]); |
da1aab3d | 207 | |
1da177e4 | 208 | out_free_bio: |
8f19ccb2 | 209 | while (++j < pi->raid_disks) |
1da177e4 | 210 | bio_put(r1_bio->bios[j]); |
98d30c58 ML |
211 | kfree(rps); |
212 | ||
213 | out_free_r1bio: | |
c7afa803 | 214 | rbio_pool_free(r1_bio, data); |
1da177e4 LT |
215 | return NULL; |
216 | } | |
217 | ||
218 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
219 | { | |
220 | struct pool_info *pi = data; | |
98d30c58 | 221 | int i; |
9f2c9d12 | 222 | struct r1bio *r1bio = __r1_bio; |
98d30c58 | 223 | struct resync_pages *rp = NULL; |
1da177e4 | 224 | |
98d30c58 ML |
225 | for (i = pi->raid_disks; i--; ) { |
226 | rp = get_resync_pages(r1bio->bios[i]); | |
227 | resync_free_pages(rp); | |
1da177e4 | 228 | bio_put(r1bio->bios[i]); |
98d30c58 ML |
229 | } |
230 | ||
231 | /* resync pages array stored in the 1st bio's .bi_private */ | |
232 | kfree(rp); | |
1da177e4 | 233 | |
c7afa803 | 234 | rbio_pool_free(r1bio, data); |
1da177e4 LT |
235 | } |
236 | ||
e8096360 | 237 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
238 | { |
239 | int i; | |
240 | ||
8f19ccb2 | 241 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 242 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 243 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
244 | bio_put(*bio); |
245 | *bio = NULL; | |
246 | } | |
247 | } | |
248 | ||
9f2c9d12 | 249 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 250 | { |
e8096360 | 251 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 252 | |
1da177e4 | 253 | put_all_bios(conf, r1_bio); |
afeee514 | 254 | mempool_free(r1_bio, &conf->r1bio_pool); |
1da177e4 LT |
255 | } |
256 | ||
9f2c9d12 | 257 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 258 | { |
e8096360 | 259 | struct r1conf *conf = r1_bio->mddev->private; |
af5f42a7 | 260 | sector_t sect = r1_bio->sector; |
3e198f78 N |
261 | int i; |
262 | ||
8f19ccb2 | 263 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
264 | struct bio *bio = r1_bio->bios[i]; |
265 | if (bio->bi_end_io) | |
266 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
267 | } | |
1da177e4 | 268 | |
afeee514 | 269 | mempool_free(r1_bio, &conf->r1buf_pool); |
1da177e4 | 270 | |
af5f42a7 | 271 | lower_barrier(conf, sect); |
1da177e4 LT |
272 | } |
273 | ||
9f2c9d12 | 274 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
275 | { |
276 | unsigned long flags; | |
fd01b88c | 277 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 278 | struct r1conf *conf = mddev->private; |
fd76863e | 279 | int idx; |
1da177e4 | 280 | |
fd76863e | 281 | idx = sector_to_idx(r1_bio->sector); |
1da177e4 LT |
282 | spin_lock_irqsave(&conf->device_lock, flags); |
283 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
824e47da | 284 | atomic_inc(&conf->nr_queued[idx]); |
1da177e4 LT |
285 | spin_unlock_irqrestore(&conf->device_lock, flags); |
286 | ||
17999be4 | 287 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
288 | md_wakeup_thread(mddev->thread); |
289 | } | |
290 | ||
291 | /* | |
292 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
293 | * operation and are ready to return a success/failure code to the buffer | |
294 | * cache layer. | |
295 | */ | |
9f2c9d12 | 296 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
297 | { |
298 | struct bio *bio = r1_bio->master_bio; | |
d2eb35ac N |
299 | |
300 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4e4cbee9 | 301 | bio->bi_status = BLK_STS_IOERR; |
4246a0b6 | 302 | |
37011e3a | 303 | bio_endio(bio); |
d2eb35ac N |
304 | } |
305 | ||
9f2c9d12 | 306 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
307 | { |
308 | struct bio *bio = r1_bio->master_bio; | |
c91114c2 | 309 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 310 | |
4b6d287f N |
311 | /* if nobody has done the final endio yet, do it now */ |
312 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
313 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
314 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
315 | (unsigned long long) bio->bi_iter.bi_sector, |
316 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 317 | |
d2eb35ac | 318 | call_bio_endio(r1_bio); |
4b6d287f | 319 | } |
c91114c2 DJ |
320 | /* |
321 | * Wake up any possible resync thread that waits for the device | |
322 | * to go idle. All I/Os, even write-behind writes, are done. | |
323 | */ | |
324 | allow_barrier(conf, r1_bio->sector); | |
325 | ||
1da177e4 LT |
326 | free_r1bio(r1_bio); |
327 | } | |
328 | ||
329 | /* | |
330 | * Update disk head position estimator based on IRQ completion info. | |
331 | */ | |
9f2c9d12 | 332 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 333 | { |
e8096360 | 334 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
335 | |
336 | conf->mirrors[disk].head_position = | |
337 | r1_bio->sector + (r1_bio->sectors); | |
338 | } | |
339 | ||
ba3ae3be NK |
340 | /* |
341 | * Find the disk number which triggered given bio | |
342 | */ | |
9f2c9d12 | 343 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
344 | { |
345 | int mirror; | |
30194636 N |
346 | struct r1conf *conf = r1_bio->mddev->private; |
347 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 348 | |
8f19ccb2 | 349 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
350 | if (r1_bio->bios[mirror] == bio) |
351 | break; | |
352 | ||
8f19ccb2 | 353 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
354 | update_head_pos(mirror, r1_bio); |
355 | ||
356 | return mirror; | |
357 | } | |
358 | ||
4246a0b6 | 359 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 360 | { |
4e4cbee9 | 361 | int uptodate = !bio->bi_status; |
9f2c9d12 | 362 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 363 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 364 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 365 | |
1da177e4 LT |
366 | /* |
367 | * this branch is our 'one mirror IO has finished' event handler: | |
368 | */ | |
e5872d58 | 369 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 370 | |
dd00a99e N |
371 | if (uptodate) |
372 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
373 | else if (test_bit(FailFast, &rdev->flags) && |
374 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
375 | /* This was a fail-fast read so we definitely | |
376 | * want to retry */ | |
377 | ; | |
dd00a99e N |
378 | else { |
379 | /* If all other devices have failed, we want to return | |
380 | * the error upwards rather than fail the last device. | |
381 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 382 | */ |
dd00a99e N |
383 | unsigned long flags; |
384 | spin_lock_irqsave(&conf->device_lock, flags); | |
385 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
386 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 387 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
388 | uptodate = 1; |
389 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
390 | } | |
1da177e4 | 391 | |
7ad4d4a6 | 392 | if (uptodate) { |
1da177e4 | 393 | raid_end_bio_io(r1_bio); |
e5872d58 | 394 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 395 | } else { |
1da177e4 LT |
396 | /* |
397 | * oops, read error: | |
398 | */ | |
399 | char b[BDEVNAME_SIZE]; | |
1d41c216 N |
400 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
401 | mdname(conf->mddev), | |
402 | bdevname(rdev->bdev, b), | |
403 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 404 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 405 | reschedule_retry(r1_bio); |
7ad4d4a6 | 406 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 407 | } |
1da177e4 LT |
408 | } |
409 | ||
9f2c9d12 | 410 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
411 | { |
412 | /* it really is the end of this request */ | |
413 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
841c1316 ML |
414 | bio_free_pages(r1_bio->behind_master_bio); |
415 | bio_put(r1_bio->behind_master_bio); | |
416 | r1_bio->behind_master_bio = NULL; | |
cd5ff9a1 N |
417 | } |
418 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
419 | md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, |
420 | r1_bio->sectors, | |
421 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
422 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
cd5ff9a1 N |
423 | md_write_end(r1_bio->mddev); |
424 | } | |
425 | ||
9f2c9d12 | 426 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 427 | { |
cd5ff9a1 N |
428 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
429 | return; | |
430 | ||
431 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
432 | reschedule_retry(r1_bio); | |
433 | else { | |
434 | close_write(r1_bio); | |
4367af55 N |
435 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
436 | reschedule_retry(r1_bio); | |
437 | else | |
438 | raid_end_bio_io(r1_bio); | |
4e78064f N |
439 | } |
440 | } | |
441 | ||
4246a0b6 | 442 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 443 | { |
9f2c9d12 | 444 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 445 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 446 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 447 | struct bio *to_put = NULL; |
e5872d58 N |
448 | int mirror = find_bio_disk(r1_bio, bio); |
449 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd | 450 | bool discard_error; |
69df9cfc GJ |
451 | sector_t lo = r1_bio->sector; |
452 | sector_t hi = r1_bio->sector + r1_bio->sectors; | |
e3f948cd | 453 | |
4e4cbee9 | 454 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 455 | |
e9c7469b TH |
456 | /* |
457 | * 'one mirror IO has finished' event handler: | |
458 | */ | |
4e4cbee9 | 459 | if (bio->bi_status && !discard_error) { |
e5872d58 N |
460 | set_bit(WriteErrorSeen, &rdev->flags); |
461 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
462 | set_bit(MD_RECOVERY_NEEDED, & |
463 | conf->mddev->recovery); | |
464 | ||
212e7eb7 N |
465 | if (test_bit(FailFast, &rdev->flags) && |
466 | (bio->bi_opf & MD_FAILFAST) && | |
467 | /* We never try FailFast to WriteMostly devices */ | |
468 | !test_bit(WriteMostly, &rdev->flags)) { | |
469 | md_error(r1_bio->mddev, rdev); | |
eeba6809 YY |
470 | } |
471 | ||
472 | /* | |
473 | * When the device is faulty, it is not necessary to | |
474 | * handle write error. | |
475 | * For failfast, this is the only remaining device, | |
476 | * We need to retry the write without FailFast. | |
477 | */ | |
478 | if (!test_bit(Faulty, &rdev->flags)) | |
212e7eb7 | 479 | set_bit(R1BIO_WriteError, &r1_bio->state); |
eeba6809 | 480 | else { |
2417b986 PC |
481 | /* Fail the request */ |
482 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
eeba6809 YY |
483 | /* Finished with this branch */ |
484 | r1_bio->bios[mirror] = NULL; | |
485 | to_put = bio; | |
486 | } | |
4367af55 | 487 | } else { |
1da177e4 | 488 | /* |
e9c7469b TH |
489 | * Set R1BIO_Uptodate in our master bio, so that we |
490 | * will return a good error code for to the higher | |
491 | * levels even if IO on some other mirrored buffer | |
492 | * fails. | |
493 | * | |
494 | * The 'master' represents the composite IO operation | |
495 | * to user-side. So if something waits for IO, then it | |
496 | * will wait for the 'master' bio. | |
1da177e4 | 497 | */ |
4367af55 N |
498 | sector_t first_bad; |
499 | int bad_sectors; | |
500 | ||
cd5ff9a1 N |
501 | r1_bio->bios[mirror] = NULL; |
502 | to_put = bio; | |
3056e3ae AL |
503 | /* |
504 | * Do not set R1BIO_Uptodate if the current device is | |
505 | * rebuilding or Faulty. This is because we cannot use | |
506 | * such device for properly reading the data back (we could | |
507 | * potentially use it, if the current write would have felt | |
508 | * before rdev->recovery_offset, but for simplicity we don't | |
509 | * check this here. | |
510 | */ | |
e5872d58 N |
511 | if (test_bit(In_sync, &rdev->flags) && |
512 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 513 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 514 | |
4367af55 | 515 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 516 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 517 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
518 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
519 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
520 | } | |
521 | } | |
522 | ||
e9c7469b | 523 | if (behind) { |
69df9cfc | 524 | if (test_bit(CollisionCheck, &rdev->flags)) |
404659cf | 525 | remove_serial(rdev, lo, hi); |
e5872d58 | 526 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
527 | atomic_dec(&r1_bio->behind_remaining); |
528 | ||
529 | /* | |
530 | * In behind mode, we ACK the master bio once the I/O | |
531 | * has safely reached all non-writemostly | |
532 | * disks. Setting the Returned bit ensures that this | |
533 | * gets done only once -- we don't ever want to return | |
534 | * -EIO here, instead we'll wait | |
535 | */ | |
536 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
537 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
538 | /* Maybe we can return now */ | |
539 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
540 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
541 | pr_debug("raid1: behind end write sectors" |
542 | " %llu-%llu\n", | |
4f024f37 KO |
543 | (unsigned long long) mbio->bi_iter.bi_sector, |
544 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 545 | call_bio_endio(r1_bio); |
4b6d287f N |
546 | } |
547 | } | |
69df9cfc GJ |
548 | } else if (rdev->mddev->serialize_policy) |
549 | remove_serial(rdev, lo, hi); | |
4367af55 | 550 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 551 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 552 | |
1da177e4 | 553 | /* |
1da177e4 LT |
554 | * Let's see if all mirrored write operations have finished |
555 | * already. | |
556 | */ | |
af6d7b76 | 557 | r1_bio_write_done(r1_bio); |
c70810b3 | 558 | |
04b857f7 N |
559 | if (to_put) |
560 | bio_put(to_put); | |
1da177e4 LT |
561 | } |
562 | ||
fd76863e | 563 | static sector_t align_to_barrier_unit_end(sector_t start_sector, |
564 | sector_t sectors) | |
565 | { | |
566 | sector_t len; | |
567 | ||
568 | WARN_ON(sectors == 0); | |
569 | /* | |
570 | * len is the number of sectors from start_sector to end of the | |
571 | * barrier unit which start_sector belongs to. | |
572 | */ | |
573 | len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - | |
574 | start_sector; | |
575 | ||
576 | if (len > sectors) | |
577 | len = sectors; | |
578 | ||
579 | return len; | |
580 | } | |
581 | ||
1da177e4 LT |
582 | /* |
583 | * This routine returns the disk from which the requested read should | |
584 | * be done. There is a per-array 'next expected sequential IO' sector | |
585 | * number - if this matches on the next IO then we use the last disk. | |
586 | * There is also a per-disk 'last know head position' sector that is | |
587 | * maintained from IRQ contexts, both the normal and the resync IO | |
588 | * completion handlers update this position correctly. If there is no | |
589 | * perfect sequential match then we pick the disk whose head is closest. | |
590 | * | |
591 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
592 | * because position is mirror, not device based. | |
593 | * | |
594 | * The rdev for the device selected will have nr_pending incremented. | |
595 | */ | |
e8096360 | 596 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 597 | { |
af3a2cd6 | 598 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
599 | int sectors; |
600 | int best_good_sectors; | |
9dedf603 SL |
601 | int best_disk, best_dist_disk, best_pending_disk; |
602 | int has_nonrot_disk; | |
be4d3280 | 603 | int disk; |
76073054 | 604 | sector_t best_dist; |
9dedf603 | 605 | unsigned int min_pending; |
3cb03002 | 606 | struct md_rdev *rdev; |
f3ac8bf7 | 607 | int choose_first; |
12cee5a8 | 608 | int choose_next_idle; |
1da177e4 LT |
609 | |
610 | rcu_read_lock(); | |
611 | /* | |
8ddf9efe | 612 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
613 | * device if no resync is going on, or below the resync window. |
614 | * We take the first readable disk when above the resync window. | |
615 | */ | |
616 | retry: | |
d2eb35ac | 617 | sectors = r1_bio->sectors; |
76073054 | 618 | best_disk = -1; |
9dedf603 | 619 | best_dist_disk = -1; |
76073054 | 620 | best_dist = MaxSector; |
9dedf603 SL |
621 | best_pending_disk = -1; |
622 | min_pending = UINT_MAX; | |
d2eb35ac | 623 | best_good_sectors = 0; |
9dedf603 | 624 | has_nonrot_disk = 0; |
12cee5a8 | 625 | choose_next_idle = 0; |
2e52d449 | 626 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 627 | |
7d49ffcf GR |
628 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
629 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 630 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
631 | this_sector + sectors))) |
632 | choose_first = 1; | |
633 | else | |
634 | choose_first = 0; | |
1da177e4 | 635 | |
be4d3280 | 636 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 637 | sector_t dist; |
d2eb35ac N |
638 | sector_t first_bad; |
639 | int bad_sectors; | |
9dedf603 | 640 | unsigned int pending; |
12cee5a8 | 641 | bool nonrot; |
d2eb35ac | 642 | |
f3ac8bf7 N |
643 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
644 | if (r1_bio->bios[disk] == IO_BLOCKED | |
645 | || rdev == NULL | |
76073054 | 646 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 647 | continue; |
76073054 N |
648 | if (!test_bit(In_sync, &rdev->flags) && |
649 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 650 | continue; |
76073054 N |
651 | if (test_bit(WriteMostly, &rdev->flags)) { |
652 | /* Don't balance among write-mostly, just | |
653 | * use the first as a last resort */ | |
d1901ef0 | 654 | if (best_dist_disk < 0) { |
307729c8 N |
655 | if (is_badblock(rdev, this_sector, sectors, |
656 | &first_bad, &bad_sectors)) { | |
816b0acf | 657 | if (first_bad <= this_sector) |
307729c8 N |
658 | /* Cannot use this */ |
659 | continue; | |
660 | best_good_sectors = first_bad - this_sector; | |
661 | } else | |
662 | best_good_sectors = sectors; | |
d1901ef0 TH |
663 | best_dist_disk = disk; |
664 | best_pending_disk = disk; | |
307729c8 | 665 | } |
76073054 N |
666 | continue; |
667 | } | |
668 | /* This is a reasonable device to use. It might | |
669 | * even be best. | |
670 | */ | |
d2eb35ac N |
671 | if (is_badblock(rdev, this_sector, sectors, |
672 | &first_bad, &bad_sectors)) { | |
673 | if (best_dist < MaxSector) | |
674 | /* already have a better device */ | |
675 | continue; | |
676 | if (first_bad <= this_sector) { | |
677 | /* cannot read here. If this is the 'primary' | |
678 | * device, then we must not read beyond | |
679 | * bad_sectors from another device.. | |
680 | */ | |
681 | bad_sectors -= (this_sector - first_bad); | |
682 | if (choose_first && sectors > bad_sectors) | |
683 | sectors = bad_sectors; | |
684 | if (best_good_sectors > sectors) | |
685 | best_good_sectors = sectors; | |
686 | ||
687 | } else { | |
688 | sector_t good_sectors = first_bad - this_sector; | |
689 | if (good_sectors > best_good_sectors) { | |
690 | best_good_sectors = good_sectors; | |
691 | best_disk = disk; | |
692 | } | |
693 | if (choose_first) | |
694 | break; | |
695 | } | |
696 | continue; | |
d82dd0e3 TM |
697 | } else { |
698 | if ((sectors > best_good_sectors) && (best_disk >= 0)) | |
699 | best_disk = -1; | |
d2eb35ac | 700 | best_good_sectors = sectors; |
d82dd0e3 | 701 | } |
d2eb35ac | 702 | |
2e52d449 N |
703 | if (best_disk >= 0) |
704 | /* At least two disks to choose from so failfast is OK */ | |
705 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
706 | ||
12cee5a8 SL |
707 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
708 | has_nonrot_disk |= nonrot; | |
9dedf603 | 709 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 710 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 711 | if (choose_first) { |
76073054 | 712 | best_disk = disk; |
1da177e4 LT |
713 | break; |
714 | } | |
12cee5a8 SL |
715 | /* Don't change to another disk for sequential reads */ |
716 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
717 | || dist == 0) { | |
718 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
719 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
720 | ||
721 | best_disk = disk; | |
722 | /* | |
723 | * If buffered sequential IO size exceeds optimal | |
724 | * iosize, check if there is idle disk. If yes, choose | |
725 | * the idle disk. read_balance could already choose an | |
726 | * idle disk before noticing it's a sequential IO in | |
727 | * this disk. This doesn't matter because this disk | |
728 | * will idle, next time it will be utilized after the | |
729 | * first disk has IO size exceeds optimal iosize. In | |
730 | * this way, iosize of the first disk will be optimal | |
731 | * iosize at least. iosize of the second disk might be | |
732 | * small, but not a big deal since when the second disk | |
733 | * starts IO, the first disk is likely still busy. | |
734 | */ | |
735 | if (nonrot && opt_iosize > 0 && | |
736 | mirror->seq_start != MaxSector && | |
737 | mirror->next_seq_sect > opt_iosize && | |
738 | mirror->next_seq_sect - opt_iosize >= | |
739 | mirror->seq_start) { | |
740 | choose_next_idle = 1; | |
741 | continue; | |
742 | } | |
743 | break; | |
744 | } | |
12cee5a8 SL |
745 | |
746 | if (choose_next_idle) | |
747 | continue; | |
9dedf603 SL |
748 | |
749 | if (min_pending > pending) { | |
750 | min_pending = pending; | |
751 | best_pending_disk = disk; | |
752 | } | |
753 | ||
76073054 N |
754 | if (dist < best_dist) { |
755 | best_dist = dist; | |
9dedf603 | 756 | best_dist_disk = disk; |
1da177e4 | 757 | } |
f3ac8bf7 | 758 | } |
1da177e4 | 759 | |
9dedf603 SL |
760 | /* |
761 | * If all disks are rotational, choose the closest disk. If any disk is | |
762 | * non-rotational, choose the disk with less pending request even the | |
763 | * disk is rotational, which might/might not be optimal for raids with | |
764 | * mixed ratation/non-rotational disks depending on workload. | |
765 | */ | |
766 | if (best_disk == -1) { | |
2e52d449 | 767 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
768 | best_disk = best_pending_disk; |
769 | else | |
770 | best_disk = best_dist_disk; | |
771 | } | |
772 | ||
76073054 N |
773 | if (best_disk >= 0) { |
774 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
775 | if (!rdev) |
776 | goto retry; | |
777 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 778 | sectors = best_good_sectors; |
12cee5a8 SL |
779 | |
780 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
781 | conf->mirrors[best_disk].seq_start = this_sector; | |
782 | ||
be4d3280 | 783 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
784 | } |
785 | rcu_read_unlock(); | |
d2eb35ac | 786 | *max_sectors = sectors; |
1da177e4 | 787 | |
76073054 | 788 | return best_disk; |
1da177e4 LT |
789 | } |
790 | ||
673ca68d N |
791 | static void flush_bio_list(struct r1conf *conf, struct bio *bio) |
792 | { | |
793 | /* flush any pending bitmap writes to disk before proceeding w/ I/O */ | |
e64e4018 | 794 | md_bitmap_unplug(conf->mddev->bitmap); |
673ca68d N |
795 | wake_up(&conf->wait_barrier); |
796 | ||
797 | while (bio) { /* submit pending writes */ | |
798 | struct bio *next = bio->bi_next; | |
309dca30 | 799 | struct md_rdev *rdev = (void *)bio->bi_bdev; |
673ca68d | 800 | bio->bi_next = NULL; |
74d46992 | 801 | bio_set_dev(bio, rdev->bdev); |
673ca68d | 802 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 803 | bio_io_error(bio); |
673ca68d | 804 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
309dca30 | 805 | !blk_queue_discard(bio->bi_bdev->bd_disk->queue))) |
673ca68d N |
806 | /* Just ignore it */ |
807 | bio_endio(bio); | |
808 | else | |
ed00aabd | 809 | submit_bio_noacct(bio); |
673ca68d | 810 | bio = next; |
5fa4f8ba | 811 | cond_resched(); |
673ca68d N |
812 | } |
813 | } | |
814 | ||
e8096360 | 815 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
816 | { |
817 | /* Any writes that have been queued but are awaiting | |
818 | * bitmap updates get flushed here. | |
a35e63ef | 819 | */ |
a35e63ef N |
820 | spin_lock_irq(&conf->device_lock); |
821 | ||
822 | if (conf->pending_bio_list.head) { | |
18022a1b | 823 | struct blk_plug plug; |
a35e63ef | 824 | struct bio *bio; |
18022a1b | 825 | |
a35e63ef | 826 | bio = bio_list_get(&conf->pending_bio_list); |
34db0cd6 | 827 | conf->pending_count = 0; |
a35e63ef | 828 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
829 | |
830 | /* | |
831 | * As this is called in a wait_event() loop (see freeze_array), | |
832 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
833 | * cause a warning when we prepare to wait again. As it is | |
834 | * rare that this path is taken, it is perfectly safe to force | |
835 | * us to go around the wait_event() loop again, so the warning | |
836 | * is a false-positive. Silence the warning by resetting | |
837 | * thread state | |
838 | */ | |
839 | __set_current_state(TASK_RUNNING); | |
18022a1b | 840 | blk_start_plug(&plug); |
673ca68d | 841 | flush_bio_list(conf, bio); |
18022a1b | 842 | blk_finish_plug(&plug); |
a35e63ef N |
843 | } else |
844 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
845 | } |
846 | ||
17999be4 N |
847 | /* Barriers.... |
848 | * Sometimes we need to suspend IO while we do something else, | |
849 | * either some resync/recovery, or reconfigure the array. | |
850 | * To do this we raise a 'barrier'. | |
851 | * The 'barrier' is a counter that can be raised multiple times | |
852 | * to count how many activities are happening which preclude | |
853 | * normal IO. | |
854 | * We can only raise the barrier if there is no pending IO. | |
855 | * i.e. if nr_pending == 0. | |
856 | * We choose only to raise the barrier if no-one is waiting for the | |
857 | * barrier to go down. This means that as soon as an IO request | |
858 | * is ready, no other operations which require a barrier will start | |
859 | * until the IO request has had a chance. | |
860 | * | |
861 | * So: regular IO calls 'wait_barrier'. When that returns there | |
862 | * is no backgroup IO happening, It must arrange to call | |
863 | * allow_barrier when it has finished its IO. | |
864 | * backgroup IO calls must call raise_barrier. Once that returns | |
865 | * there is no normal IO happeing. It must arrange to call | |
866 | * lower_barrier when the particular background IO completes. | |
4675719d HT |
867 | * |
868 | * If resync/recovery is interrupted, returns -EINTR; | |
869 | * Otherwise, returns 0. | |
1da177e4 | 870 | */ |
4675719d | 871 | static int raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 | 872 | { |
fd76863e | 873 | int idx = sector_to_idx(sector_nr); |
874 | ||
1da177e4 | 875 | spin_lock_irq(&conf->resync_lock); |
17999be4 N |
876 | |
877 | /* Wait until no block IO is waiting */ | |
824e47da | 878 | wait_event_lock_irq(conf->wait_barrier, |
879 | !atomic_read(&conf->nr_waiting[idx]), | |
eed8c02e | 880 | conf->resync_lock); |
17999be4 N |
881 | |
882 | /* block any new IO from starting */ | |
824e47da | 883 | atomic_inc(&conf->barrier[idx]); |
884 | /* | |
885 | * In raise_barrier() we firstly increase conf->barrier[idx] then | |
886 | * check conf->nr_pending[idx]. In _wait_barrier() we firstly | |
887 | * increase conf->nr_pending[idx] then check conf->barrier[idx]. | |
888 | * A memory barrier here to make sure conf->nr_pending[idx] won't | |
889 | * be fetched before conf->barrier[idx] is increased. Otherwise | |
890 | * there will be a race between raise_barrier() and _wait_barrier(). | |
891 | */ | |
892 | smp_mb__after_atomic(); | |
17999be4 | 893 | |
79ef3a8a | 894 | /* For these conditions we must wait: |
895 | * A: while the array is in frozen state | |
fd76863e | 896 | * B: while conf->nr_pending[idx] is not 0, meaning regular I/O |
897 | * existing in corresponding I/O barrier bucket. | |
898 | * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches | |
899 | * max resync count which allowed on current I/O barrier bucket. | |
79ef3a8a | 900 | */ |
17999be4 | 901 | wait_event_lock_irq(conf->wait_barrier, |
8c242593 | 902 | (!conf->array_frozen && |
824e47da | 903 | !atomic_read(&conf->nr_pending[idx]) && |
8c242593 YY |
904 | atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || |
905 | test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), | |
eed8c02e | 906 | conf->resync_lock); |
17999be4 | 907 | |
8c242593 YY |
908 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
909 | atomic_dec(&conf->barrier[idx]); | |
910 | spin_unlock_irq(&conf->resync_lock); | |
911 | wake_up(&conf->wait_barrier); | |
912 | return -EINTR; | |
913 | } | |
914 | ||
43ac9b84 | 915 | atomic_inc(&conf->nr_sync_pending); |
17999be4 | 916 | spin_unlock_irq(&conf->resync_lock); |
8c242593 YY |
917 | |
918 | return 0; | |
17999be4 N |
919 | } |
920 | ||
fd76863e | 921 | static void lower_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 922 | { |
fd76863e | 923 | int idx = sector_to_idx(sector_nr); |
924 | ||
824e47da | 925 | BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); |
fd76863e | 926 | |
824e47da | 927 | atomic_dec(&conf->barrier[idx]); |
43ac9b84 | 928 | atomic_dec(&conf->nr_sync_pending); |
17999be4 N |
929 | wake_up(&conf->wait_barrier); |
930 | } | |
931 | ||
fd76863e | 932 | static void _wait_barrier(struct r1conf *conf, int idx) |
17999be4 | 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])) | |
964 | return; | |
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 | */ | |
980 | wake_up(&conf->wait_barrier); | |
981 | /* Wait for the barrier in same barrier unit bucket to drop. */ | |
982 | wait_event_lock_irq(conf->wait_barrier, | |
983 | !conf->array_frozen && | |
984 | !atomic_read(&conf->barrier[idx]), | |
985 | conf->resync_lock); | |
986 | atomic_inc(&conf->nr_pending[idx]); | |
987 | atomic_dec(&conf->nr_waiting[idx]); | |
fd76863e | 988 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 989 | } |
990 | ||
fd76863e | 991 | static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr) |
79ef3a8a | 992 | { |
fd76863e | 993 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 994 | |
824e47da | 995 | /* |
996 | * Very similar to _wait_barrier(). The difference is, for read | |
997 | * I/O we don't need wait for sync I/O, but if the whole array | |
998 | * is frozen, the read I/O still has to wait until the array is | |
999 | * unfrozen. Since there is no ordering requirement with | |
1000 | * conf->barrier[idx] here, memory barrier is unnecessary as well. | |
1001 | */ | |
1002 | atomic_inc(&conf->nr_pending[idx]); | |
79ef3a8a | 1003 | |
824e47da | 1004 | if (!READ_ONCE(conf->array_frozen)) |
1005 | return; | |
1006 | ||
1007 | spin_lock_irq(&conf->resync_lock); | |
1008 | atomic_inc(&conf->nr_waiting[idx]); | |
1009 | atomic_dec(&conf->nr_pending[idx]); | |
1010 | /* | |
1011 | * In case freeze_array() is waiting for | |
1012 | * get_unqueued_pending() == extra | |
1013 | */ | |
1014 | wake_up(&conf->wait_barrier); | |
1015 | /* Wait for array to be unfrozen */ | |
1016 | wait_event_lock_irq(conf->wait_barrier, | |
1017 | !conf->array_frozen, | |
1018 | conf->resync_lock); | |
1019 | atomic_inc(&conf->nr_pending[idx]); | |
1020 | atomic_dec(&conf->nr_waiting[idx]); | |
1da177e4 LT |
1021 | spin_unlock_irq(&conf->resync_lock); |
1022 | } | |
1023 | ||
fd76863e | 1024 | static void wait_barrier(struct r1conf *conf, sector_t sector_nr) |
17999be4 | 1025 | { |
fd76863e | 1026 | int idx = sector_to_idx(sector_nr); |
79ef3a8a | 1027 | |
fd76863e | 1028 | _wait_barrier(conf, idx); |
1029 | } | |
1030 | ||
fd76863e | 1031 | static void _allow_barrier(struct r1conf *conf, int idx) |
17999be4 | 1032 | { |
824e47da | 1033 | atomic_dec(&conf->nr_pending[idx]); |
17999be4 N |
1034 | wake_up(&conf->wait_barrier); |
1035 | } | |
1036 | ||
fd76863e | 1037 | static void allow_barrier(struct r1conf *conf, sector_t sector_nr) |
1038 | { | |
1039 | int idx = sector_to_idx(sector_nr); | |
1040 | ||
1041 | _allow_barrier(conf, idx); | |
1042 | } | |
1043 | ||
fd76863e | 1044 | /* conf->resync_lock should be held */ |
1045 | static int get_unqueued_pending(struct r1conf *conf) | |
1046 | { | |
1047 | int idx, ret; | |
1048 | ||
43ac9b84 XN |
1049 | ret = atomic_read(&conf->nr_sync_pending); |
1050 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) | |
824e47da | 1051 | ret += atomic_read(&conf->nr_pending[idx]) - |
1052 | atomic_read(&conf->nr_queued[idx]); | |
fd76863e | 1053 | |
1054 | return ret; | |
1055 | } | |
1056 | ||
e2d59925 | 1057 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab | 1058 | { |
fd76863e | 1059 | /* Stop sync I/O and normal I/O and wait for everything to |
11353b9d | 1060 | * go quiet. |
fd76863e | 1061 | * This is called in two situations: |
1062 | * 1) management command handlers (reshape, remove disk, quiesce). | |
1063 | * 2) one normal I/O request failed. | |
1064 | ||
1065 | * After array_frozen is set to 1, new sync IO will be blocked at | |
1066 | * raise_barrier(), and new normal I/O will blocked at _wait_barrier() | |
1067 | * or wait_read_barrier(). The flying I/Os will either complete or be | |
1068 | * queued. When everything goes quite, there are only queued I/Os left. | |
1069 | ||
1070 | * Every flying I/O contributes to a conf->nr_pending[idx], idx is the | |
1071 | * barrier bucket index which this I/O request hits. When all sync and | |
1072 | * normal I/O are queued, sum of all conf->nr_pending[] will match sum | |
1073 | * of all conf->nr_queued[]. But normal I/O failure is an exception, | |
1074 | * in handle_read_error(), we may call freeze_array() before trying to | |
1075 | * fix the read error. In this case, the error read I/O is not queued, | |
1076 | * so get_unqueued_pending() == 1. | |
1077 | * | |
1078 | * Therefore before this function returns, we need to wait until | |
1079 | * get_unqueued_pendings(conf) gets equal to extra. For | |
1080 | * normal I/O context, extra is 1, in rested situations extra is 0. | |
ddaf22ab N |
1081 | */ |
1082 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1083 | conf->array_frozen = 1; |
578b54ad | 1084 | raid1_log(conf->mddev, "wait freeze"); |
fd76863e | 1085 | wait_event_lock_irq_cmd( |
1086 | conf->wait_barrier, | |
1087 | get_unqueued_pending(conf) == extra, | |
1088 | conf->resync_lock, | |
1089 | flush_pending_writes(conf)); | |
ddaf22ab N |
1090 | spin_unlock_irq(&conf->resync_lock); |
1091 | } | |
e8096360 | 1092 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
1093 | { |
1094 | /* reverse the effect of the freeze */ | |
1095 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 1096 | conf->array_frozen = 0; |
ddaf22ab | 1097 | spin_unlock_irq(&conf->resync_lock); |
824e47da | 1098 | wake_up(&conf->wait_barrier); |
ddaf22ab N |
1099 | } |
1100 | ||
16d56e2f | 1101 | static void alloc_behind_master_bio(struct r1bio *r1_bio, |
cb83efcf | 1102 | struct bio *bio) |
4b6d287f | 1103 | { |
cb83efcf | 1104 | int size = bio->bi_iter.bi_size; |
841c1316 ML |
1105 | unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1106 | int i = 0; | |
1107 | struct bio *behind_bio = NULL; | |
1108 | ||
a78f18da | 1109 | behind_bio = bio_alloc_bioset(GFP_NOIO, vcnt, &r1_bio->mddev->bio_set); |
841c1316 | 1110 | if (!behind_bio) |
16d56e2f | 1111 | return; |
4b6d287f | 1112 | |
41743c1f | 1113 | /* discard op, we don't support writezero/writesame yet */ |
16d56e2f SL |
1114 | if (!bio_has_data(bio)) { |
1115 | behind_bio->bi_iter.bi_size = size; | |
41743c1f | 1116 | goto skip_copy; |
16d56e2f | 1117 | } |
41743c1f | 1118 | |
dba40d46 MD |
1119 | behind_bio->bi_write_hint = bio->bi_write_hint; |
1120 | ||
841c1316 ML |
1121 | while (i < vcnt && size) { |
1122 | struct page *page; | |
1123 | int len = min_t(int, PAGE_SIZE, size); | |
1124 | ||
1125 | page = alloc_page(GFP_NOIO); | |
1126 | if (unlikely(!page)) | |
1127 | goto free_pages; | |
1128 | ||
1129 | bio_add_page(behind_bio, page, len, 0); | |
1130 | ||
1131 | size -= len; | |
1132 | i++; | |
4b6d287f | 1133 | } |
841c1316 | 1134 | |
cb83efcf | 1135 | bio_copy_data(behind_bio, bio); |
41743c1f | 1136 | skip_copy: |
56a64c17 | 1137 | r1_bio->behind_master_bio = behind_bio; |
af6d7b76 | 1138 | set_bit(R1BIO_BehindIO, &r1_bio->state); |
4b6d287f | 1139 | |
16d56e2f | 1140 | return; |
841c1316 ML |
1141 | |
1142 | free_pages: | |
4f024f37 KO |
1143 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1144 | bio->bi_iter.bi_size); | |
841c1316 | 1145 | bio_free_pages(behind_bio); |
16d56e2f | 1146 | bio_put(behind_bio); |
4b6d287f N |
1147 | } |
1148 | ||
f54a9d0e N |
1149 | struct raid1_plug_cb { |
1150 | struct blk_plug_cb cb; | |
1151 | struct bio_list pending; | |
1152 | int pending_cnt; | |
1153 | }; | |
1154 | ||
1155 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1156 | { | |
1157 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1158 | cb); | |
1159 | struct mddev *mddev = plug->cb.data; | |
1160 | struct r1conf *conf = mddev->private; | |
1161 | struct bio *bio; | |
1162 | ||
874807a8 | 1163 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1164 | spin_lock_irq(&conf->device_lock); |
1165 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1166 | conf->pending_count += plug->pending_cnt; | |
1167 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1168 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1169 | md_wakeup_thread(mddev->thread); |
1170 | kfree(plug); | |
1171 | return; | |
1172 | } | |
1173 | ||
1174 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1175 | bio = bio_list_get(&plug->pending); | |
673ca68d | 1176 | flush_bio_list(conf, bio); |
f54a9d0e N |
1177 | kfree(plug); |
1178 | } | |
1179 | ||
689389a0 N |
1180 | static void init_r1bio(struct r1bio *r1_bio, struct mddev *mddev, struct bio *bio) |
1181 | { | |
1182 | r1_bio->master_bio = bio; | |
1183 | r1_bio->sectors = bio_sectors(bio); | |
1184 | r1_bio->state = 0; | |
1185 | r1_bio->mddev = mddev; | |
1186 | r1_bio->sector = bio->bi_iter.bi_sector; | |
1187 | } | |
1188 | ||
fd76863e | 1189 | static inline struct r1bio * |
689389a0 | 1190 | alloc_r1bio(struct mddev *mddev, struct bio *bio) |
fd76863e | 1191 | { |
1192 | struct r1conf *conf = mddev->private; | |
1193 | struct r1bio *r1_bio; | |
1194 | ||
afeee514 | 1195 | r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); |
689389a0 N |
1196 | /* Ensure no bio records IO_BLOCKED */ |
1197 | memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); | |
1198 | init_r1bio(r1_bio, mddev, bio); | |
fd76863e | 1199 | return r1_bio; |
1200 | } | |
1201 | ||
c230e7e5 | 1202 | static void raid1_read_request(struct mddev *mddev, struct bio *bio, |
689389a0 | 1203 | int max_read_sectors, struct r1bio *r1_bio) |
1da177e4 | 1204 | { |
e8096360 | 1205 | struct r1conf *conf = mddev->private; |
0eaf822c | 1206 | struct raid1_info *mirror; |
1da177e4 | 1207 | struct bio *read_bio; |
3b046a97 RL |
1208 | struct bitmap *bitmap = mddev->bitmap; |
1209 | const int op = bio_op(bio); | |
1210 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
3b046a97 RL |
1211 | int max_sectors; |
1212 | int rdisk; | |
689389a0 N |
1213 | bool print_msg = !!r1_bio; |
1214 | char b[BDEVNAME_SIZE]; | |
3b046a97 | 1215 | |
fd76863e | 1216 | /* |
689389a0 N |
1217 | * If r1_bio is set, we are blocking the raid1d thread |
1218 | * so there is a tiny risk of deadlock. So ask for | |
1219 | * emergency memory if needed. | |
fd76863e | 1220 | */ |
689389a0 | 1221 | gfp_t gfp = r1_bio ? (GFP_NOIO | __GFP_HIGH) : GFP_NOIO; |
fd76863e | 1222 | |
689389a0 N |
1223 | if (print_msg) { |
1224 | /* Need to get the block device name carefully */ | |
1225 | struct md_rdev *rdev; | |
1226 | rcu_read_lock(); | |
1227 | rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); | |
1228 | if (rdev) | |
1229 | bdevname(rdev->bdev, b); | |
1230 | else | |
1231 | strcpy(b, "???"); | |
1232 | rcu_read_unlock(); | |
1233 | } | |
3b046a97 | 1234 | |
fd76863e | 1235 | /* |
fd76863e | 1236 | * Still need barrier for READ in case that whole |
1237 | * array is frozen. | |
fd76863e | 1238 | */ |
fd76863e | 1239 | wait_read_barrier(conf, bio->bi_iter.bi_sector); |
1240 | ||
689389a0 N |
1241 | if (!r1_bio) |
1242 | r1_bio = alloc_r1bio(mddev, bio); | |
1243 | else | |
1244 | init_r1bio(r1_bio, mddev, bio); | |
c230e7e5 | 1245 | r1_bio->sectors = max_read_sectors; |
fd76863e | 1246 | |
1247 | /* | |
1248 | * make_request() can abort the operation when read-ahead is being | |
1249 | * used and no empty request is available. | |
1250 | */ | |
3b046a97 RL |
1251 | rdisk = read_balance(conf, r1_bio, &max_sectors); |
1252 | ||
1253 | if (rdisk < 0) { | |
1254 | /* couldn't find anywhere to read from */ | |
689389a0 N |
1255 | if (print_msg) { |
1256 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1257 | mdname(mddev), | |
1258 | b, | |
1259 | (unsigned long long)r1_bio->sector); | |
1260 | } | |
3b046a97 RL |
1261 | raid_end_bio_io(r1_bio); |
1262 | return; | |
1263 | } | |
1264 | mirror = conf->mirrors + rdisk; | |
1265 | ||
689389a0 N |
1266 | if (print_msg) |
1267 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", | |
1268 | mdname(mddev), | |
1269 | (unsigned long long)r1_bio->sector, | |
1270 | bdevname(mirror->rdev->bdev, b)); | |
1271 | ||
3b046a97 RL |
1272 | if (test_bit(WriteMostly, &mirror->rdev->flags) && |
1273 | bitmap) { | |
1274 | /* | |
1275 | * Reading from a write-mostly device must take care not to | |
1276 | * over-take any writes that are 'behind' | |
1277 | */ | |
1278 | raid1_log(mddev, "wait behind writes"); | |
1279 | wait_event(bitmap->behind_wait, | |
1280 | atomic_read(&bitmap->behind_writes) == 0); | |
1281 | } | |
c230e7e5 N |
1282 | |
1283 | if (max_sectors < bio_sectors(bio)) { | |
1284 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1285 | gfp, &conf->bio_split); |
c230e7e5 | 1286 | bio_chain(split, bio); |
ed00aabd | 1287 | submit_bio_noacct(bio); |
c230e7e5 N |
1288 | bio = split; |
1289 | r1_bio->master_bio = bio; | |
1290 | r1_bio->sectors = max_sectors; | |
1291 | } | |
1292 | ||
3b046a97 | 1293 | r1_bio->read_disk = rdisk; |
3b046a97 | 1294 | |
afeee514 | 1295 | read_bio = bio_clone_fast(bio, gfp, &mddev->bio_set); |
3b046a97 RL |
1296 | |
1297 | r1_bio->bios[rdisk] = read_bio; | |
1298 | ||
1299 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1300 | mirror->rdev->data_offset; | |
74d46992 | 1301 | bio_set_dev(read_bio, mirror->rdev->bdev); |
3b046a97 RL |
1302 | read_bio->bi_end_io = raid1_end_read_request; |
1303 | bio_set_op_attrs(read_bio, op, do_sync); | |
1304 | if (test_bit(FailFast, &mirror->rdev->flags) && | |
1305 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1306 | read_bio->bi_opf |= MD_FAILFAST; | |
1307 | read_bio->bi_private = r1_bio; | |
1308 | ||
1309 | if (mddev->gendisk) | |
1c02fca6 CH |
1310 | trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), |
1311 | r1_bio->sector); | |
3b046a97 | 1312 | |
ed00aabd | 1313 | submit_bio_noacct(read_bio); |
3b046a97 RL |
1314 | } |
1315 | ||
c230e7e5 N |
1316 | static void raid1_write_request(struct mddev *mddev, struct bio *bio, |
1317 | int max_write_sectors) | |
3b046a97 RL |
1318 | { |
1319 | struct r1conf *conf = mddev->private; | |
fd76863e | 1320 | struct r1bio *r1_bio; |
1f68f0c4 | 1321 | int i, disks; |
3b046a97 | 1322 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1323 | unsigned long flags; |
3cb03002 | 1324 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1325 | struct blk_plug_cb *cb; |
1326 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 | 1327 | int first_clone; |
1f68f0c4 | 1328 | int max_sectors; |
191ea9b2 | 1329 | |
b3143b9a | 1330 | if (mddev_is_clustered(mddev) && |
90382ed9 | 1331 | md_cluster_ops->area_resyncing(mddev, WRITE, |
b3143b9a | 1332 | bio->bi_iter.bi_sector, bio_end_sector(bio))) { |
3b046a97 | 1333 | |
6eef4b21 N |
1334 | DEFINE_WAIT(w); |
1335 | for (;;) { | |
6eef4b21 | 1336 | prepare_to_wait(&conf->wait_barrier, |
ae89fd3d | 1337 | &w, TASK_IDLE); |
f81f7302 | 1338 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, |
385f4d7f | 1339 | bio->bi_iter.bi_sector, |
b3143b9a | 1340 | bio_end_sector(bio))) |
6eef4b21 N |
1341 | break; |
1342 | schedule(); | |
1343 | } | |
1344 | finish_wait(&conf->wait_barrier, &w); | |
1345 | } | |
f81f7302 GJ |
1346 | |
1347 | /* | |
1348 | * Register the new request and wait if the reconstruction | |
1349 | * thread has put up a bar for new requests. | |
1350 | * Continue immediately if no resync is active currently. | |
1351 | */ | |
fd76863e | 1352 | wait_barrier(conf, bio->bi_iter.bi_sector); |
1353 | ||
689389a0 | 1354 | r1_bio = alloc_r1bio(mddev, bio); |
c230e7e5 | 1355 | r1_bio->sectors = max_write_sectors; |
1da177e4 | 1356 | |
34db0cd6 N |
1357 | if (conf->pending_count >= max_queued_requests) { |
1358 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1359 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1360 | wait_event(conf->wait_barrier, |
1361 | conf->pending_count < max_queued_requests); | |
1362 | } | |
1f68f0c4 | 1363 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1364 | * inc refcount on their rdev. Record them by setting |
1365 | * bios[x] to bio | |
1f68f0c4 N |
1366 | * If there are known/acknowledged bad blocks on any device on |
1367 | * which we have seen a write error, we want to avoid writing those | |
1368 | * blocks. | |
1369 | * This potentially requires several writes to write around | |
1370 | * the bad blocks. Each set of writes gets it's own r1bio | |
1371 | * with a set of bios attached. | |
1da177e4 | 1372 | */ |
c3b328ac | 1373 | |
8f19ccb2 | 1374 | disks = conf->raid_disks * 2; |
6bfe0b49 DW |
1375 | retry_write: |
1376 | blocked_rdev = NULL; | |
1da177e4 | 1377 | rcu_read_lock(); |
1f68f0c4 | 1378 | max_sectors = r1_bio->sectors; |
1da177e4 | 1379 | for (i = 0; i < disks; i++) { |
3cb03002 | 1380 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1381 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1382 | atomic_inc(&rdev->nr_pending); | |
1383 | blocked_rdev = rdev; | |
1384 | break; | |
1385 | } | |
1f68f0c4 | 1386 | r1_bio->bios[i] = NULL; |
8ae12666 | 1387 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1388 | if (i < conf->raid_disks) |
1389 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1390 | continue; |
1391 | } | |
1392 | ||
1393 | atomic_inc(&rdev->nr_pending); | |
1394 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1395 | sector_t first_bad; | |
1396 | int bad_sectors; | |
1397 | int is_bad; | |
1398 | ||
3b046a97 | 1399 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1400 | &first_bad, &bad_sectors); |
1401 | if (is_bad < 0) { | |
1402 | /* mustn't write here until the bad block is | |
1403 | * acknowledged*/ | |
1404 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1405 | blocked_rdev = rdev; | |
1406 | break; | |
1407 | } | |
1408 | if (is_bad && first_bad <= r1_bio->sector) { | |
1409 | /* Cannot write here at all */ | |
1410 | bad_sectors -= (r1_bio->sector - first_bad); | |
1411 | if (bad_sectors < max_sectors) | |
1412 | /* mustn't write more than bad_sectors | |
1413 | * to other devices yet | |
1414 | */ | |
1415 | max_sectors = bad_sectors; | |
03c902e1 | 1416 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1417 | /* We don't set R1BIO_Degraded as that |
1418 | * only applies if the disk is | |
1419 | * missing, so it might be re-added, | |
1420 | * and we want to know to recover this | |
1421 | * chunk. | |
1422 | * In this case the device is here, | |
1423 | * and the fact that this chunk is not | |
1424 | * in-sync is recorded in the bad | |
1425 | * block log | |
1426 | */ | |
1427 | continue; | |
964147d5 | 1428 | } |
1f68f0c4 N |
1429 | if (is_bad) { |
1430 | int good_sectors = first_bad - r1_bio->sector; | |
1431 | if (good_sectors < max_sectors) | |
1432 | max_sectors = good_sectors; | |
1433 | } | |
1434 | } | |
1435 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1436 | } |
1437 | rcu_read_unlock(); | |
1438 | ||
6bfe0b49 DW |
1439 | if (unlikely(blocked_rdev)) { |
1440 | /* Wait for this device to become unblocked */ | |
1441 | int j; | |
1442 | ||
1443 | for (j = 0; j < i; j++) | |
1444 | if (r1_bio->bios[j]) | |
1445 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1446 | r1_bio->state = 0; |
fd76863e | 1447 | allow_barrier(conf, bio->bi_iter.bi_sector); |
578b54ad | 1448 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1449 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
fd76863e | 1450 | wait_barrier(conf, bio->bi_iter.bi_sector); |
6bfe0b49 DW |
1451 | goto retry_write; |
1452 | } | |
1453 | ||
c230e7e5 N |
1454 | if (max_sectors < bio_sectors(bio)) { |
1455 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1456 | GFP_NOIO, &conf->bio_split); |
c230e7e5 | 1457 | bio_chain(split, bio); |
ed00aabd | 1458 | submit_bio_noacct(bio); |
c230e7e5 N |
1459 | bio = split; |
1460 | r1_bio->master_bio = bio; | |
1f68f0c4 | 1461 | r1_bio->sectors = max_sectors; |
191ea9b2 | 1462 | } |
4b6d287f | 1463 | |
4e78064f | 1464 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1465 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1466 | |
1f68f0c4 | 1467 | first_clone = 1; |
d8c84c4f | 1468 | |
1da177e4 | 1469 | for (i = 0; i < disks; i++) { |
8e58e327 | 1470 | struct bio *mbio = NULL; |
69df9cfc | 1471 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
1da177e4 LT |
1472 | if (!r1_bio->bios[i]) |
1473 | continue; | |
1474 | ||
1f68f0c4 N |
1475 | if (first_clone) { |
1476 | /* do behind I/O ? | |
1477 | * Not if there are too many, or cannot | |
1478 | * allocate memory, or a reader on WriteMostly | |
1479 | * is waiting for behind writes to flush */ | |
1480 | if (bitmap && | |
1481 | (atomic_read(&bitmap->behind_writes) | |
1482 | < mddev->bitmap_info.max_write_behind) && | |
8e58e327 | 1483 | !waitqueue_active(&bitmap->behind_wait)) { |
16d56e2f | 1484 | alloc_behind_master_bio(r1_bio, bio); |
8e58e327 | 1485 | } |
1f68f0c4 | 1486 | |
e64e4018 AS |
1487 | md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, |
1488 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
1f68f0c4 N |
1489 | first_clone = 0; |
1490 | } | |
8e58e327 | 1491 | |
16d56e2f SL |
1492 | if (r1_bio->behind_master_bio) |
1493 | mbio = bio_clone_fast(r1_bio->behind_master_bio, | |
afeee514 | 1494 | GFP_NOIO, &mddev->bio_set); |
16d56e2f | 1495 | else |
afeee514 | 1496 | mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); |
8e58e327 | 1497 | |
841c1316 | 1498 | if (r1_bio->behind_master_bio) { |
69df9cfc | 1499 | if (test_bit(CollisionCheck, &rdev->flags)) |
d0d2d8ba | 1500 | wait_for_serialization(rdev, r1_bio); |
3e148a32 | 1501 | if (test_bit(WriteMostly, &rdev->flags)) |
4b6d287f | 1502 | atomic_inc(&r1_bio->behind_remaining); |
69df9cfc | 1503 | } else if (mddev->serialize_policy) |
d0d2d8ba | 1504 | wait_for_serialization(rdev, r1_bio); |
4b6d287f | 1505 | |
1f68f0c4 N |
1506 | r1_bio->bios[i] = mbio; |
1507 | ||
4f024f37 | 1508 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1509 | conf->mirrors[i].rdev->data_offset); |
74d46992 | 1510 | bio_set_dev(mbio, conf->mirrors[i].rdev->bdev); |
1f68f0c4 | 1511 | mbio->bi_end_io = raid1_end_write_request; |
a682e003 | 1512 | mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); |
212e7eb7 N |
1513 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && |
1514 | !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && | |
1515 | conf->raid_disks - mddev->degraded > 1) | |
1516 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1517 | mbio->bi_private = r1_bio; |
1518 | ||
1da177e4 | 1519 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1520 | |
109e3765 | 1521 | if (mddev->gendisk) |
1c02fca6 | 1522 | trace_block_bio_remap(mbio, disk_devt(mddev->gendisk), |
109e3765 N |
1523 | r1_bio->sector); |
1524 | /* flush_pending_writes() needs access to the rdev so...*/ | |
309dca30 | 1525 | mbio->bi_bdev = (void *)conf->mirrors[i].rdev; |
109e3765 | 1526 | |
f54a9d0e N |
1527 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1528 | if (cb) | |
1529 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1530 | else | |
1531 | plug = NULL; | |
f54a9d0e N |
1532 | if (plug) { |
1533 | bio_list_add(&plug->pending, mbio); | |
1534 | plug->pending_cnt++; | |
1535 | } else { | |
23b245c0 | 1536 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1537 | bio_list_add(&conf->pending_bio_list, mbio); |
1538 | conf->pending_count++; | |
23b245c0 | 1539 | spin_unlock_irqrestore(&conf->device_lock, flags); |
b357f04a | 1540 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1541 | } |
1da177e4 | 1542 | } |
1f68f0c4 | 1543 | |
079fa166 N |
1544 | r1_bio_write_done(r1_bio); |
1545 | ||
1546 | /* In case raid1d snuck in to freeze_array */ | |
1547 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1548 | } |
1549 | ||
cc27b0c7 | 1550 | static bool raid1_make_request(struct mddev *mddev, struct bio *bio) |
3b046a97 | 1551 | { |
fd76863e | 1552 | sector_t sectors; |
3b046a97 | 1553 | |
775d7831 DJ |
1554 | if (unlikely(bio->bi_opf & REQ_PREFLUSH) |
1555 | && md_flush_request(mddev, bio)) | |
cc27b0c7 | 1556 | return true; |
3b046a97 | 1557 | |
c230e7e5 N |
1558 | /* |
1559 | * There is a limit to the maximum size, but | |
1560 | * the read/write handler might find a lower limit | |
1561 | * due to bad blocks. To avoid multiple splits, | |
1562 | * we pass the maximum number of sectors down | |
1563 | * and let the lower level perform the split. | |
1564 | */ | |
1565 | sectors = align_to_barrier_unit_end( | |
1566 | bio->bi_iter.bi_sector, bio_sectors(bio)); | |
61eb2b43 | 1567 | |
c230e7e5 | 1568 | if (bio_data_dir(bio) == READ) |
689389a0 | 1569 | raid1_read_request(mddev, bio, sectors, NULL); |
cc27b0c7 N |
1570 | else { |
1571 | if (!md_write_start(mddev,bio)) | |
1572 | return false; | |
c230e7e5 | 1573 | raid1_write_request(mddev, bio, sectors); |
cc27b0c7 N |
1574 | } |
1575 | return true; | |
3b046a97 RL |
1576 | } |
1577 | ||
849674e4 | 1578 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1579 | { |
e8096360 | 1580 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1581 | int i; |
1582 | ||
1583 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1584 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1585 | rcu_read_lock(); |
1586 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1587 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1588 | seq_printf(seq, "%s", |
ddac7c7e N |
1589 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1590 | } | |
1591 | rcu_read_unlock(); | |
1da177e4 LT |
1592 | seq_printf(seq, "]"); |
1593 | } | |
1594 | ||
849674e4 | 1595 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1596 | { |
1597 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1598 | struct r1conf *conf = mddev->private; |
423f04d6 | 1599 | unsigned long flags; |
1da177e4 LT |
1600 | |
1601 | /* | |
1602 | * If it is not operational, then we have already marked it as dead | |
9a567843 GJ |
1603 | * else if it is the last working disks with "fail_last_dev == false", |
1604 | * ignore the error, let the next level up know. | |
1da177e4 LT |
1605 | * else mark the drive as failed |
1606 | */ | |
2e52d449 | 1607 | spin_lock_irqsave(&conf->device_lock, flags); |
9a567843 | 1608 | if (test_bit(In_sync, &rdev->flags) && !mddev->fail_last_dev |
4044ba58 | 1609 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1610 | /* |
1611 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1612 | * normal single drive. |
1613 | * However don't try a recovery from this drive as | |
1614 | * it is very likely to fail. | |
1da177e4 | 1615 | */ |
5389042f | 1616 | conf->recovery_disabled = mddev->recovery_disabled; |
2e52d449 | 1617 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1618 | return; |
4044ba58 | 1619 | } |
de393cde | 1620 | set_bit(Blocked, &rdev->flags); |
ebda52fa | 1621 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1622 | mddev->degraded++; |
ebda52fa | 1623 | set_bit(Faulty, &rdev->flags); |
423f04d6 | 1624 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1625 | /* |
1626 | * if recovery is running, make sure it aborts. | |
1627 | */ | |
1628 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1629 | set_mask_bits(&mddev->sb_flags, 0, |
1630 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
1d41c216 N |
1631 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1632 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1633 | mdname(mddev), bdevname(rdev->bdev, b), | |
1634 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1635 | } |
1636 | ||
e8096360 | 1637 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1638 | { |
1639 | int i; | |
1da177e4 | 1640 | |
1d41c216 | 1641 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1642 | if (!conf) { |
1d41c216 | 1643 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1644 | return; |
1645 | } | |
1d41c216 N |
1646 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1647 | conf->raid_disks); | |
1da177e4 | 1648 | |
ddac7c7e | 1649 | rcu_read_lock(); |
1da177e4 LT |
1650 | for (i = 0; i < conf->raid_disks; i++) { |
1651 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1652 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1653 | if (rdev) |
1d41c216 N |
1654 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1655 | i, !test_bit(In_sync, &rdev->flags), | |
1656 | !test_bit(Faulty, &rdev->flags), | |
1657 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1658 | } |
ddac7c7e | 1659 | rcu_read_unlock(); |
1da177e4 LT |
1660 | } |
1661 | ||
e8096360 | 1662 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1663 | { |
f6eca2d4 ND |
1664 | int idx; |
1665 | ||
1666 | for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++) { | |
1667 | _wait_barrier(conf, idx); | |
1668 | _allow_barrier(conf, idx); | |
1669 | } | |
1da177e4 | 1670 | |
afeee514 | 1671 | mempool_exit(&conf->r1buf_pool); |
1da177e4 LT |
1672 | } |
1673 | ||
fd01b88c | 1674 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1675 | { |
1676 | int i; | |
e8096360 | 1677 | struct r1conf *conf = mddev->private; |
6b965620 N |
1678 | int count = 0; |
1679 | unsigned long flags; | |
1da177e4 LT |
1680 | |
1681 | /* | |
f72ffdd6 | 1682 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1683 | * and mark them readable. |
1684 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1685 | * device_lock used to avoid races with raid1_end_read_request |
1686 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1687 | */ |
423f04d6 | 1688 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1689 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1690 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1691 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1692 | if (repl | |
1aee41f6 | 1693 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1694 | && repl->recovery_offset == MaxSector |
1695 | && !test_bit(Faulty, &repl->flags) | |
1696 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1697 | /* replacement has just become active */ | |
1698 | if (!rdev || | |
1699 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1700 | count++; | |
1701 | if (rdev) { | |
1702 | /* Replaced device not technically | |
1703 | * faulty, but we need to be sure | |
1704 | * it gets removed and never re-added | |
1705 | */ | |
1706 | set_bit(Faulty, &rdev->flags); | |
1707 | sysfs_notify_dirent_safe( | |
1708 | rdev->sysfs_state); | |
1709 | } | |
1710 | } | |
ddac7c7e | 1711 | if (rdev |
61e4947c | 1712 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1713 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1714 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1715 | count++; |
654e8b5a | 1716 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1717 | } |
1718 | } | |
6b965620 N |
1719 | mddev->degraded -= count; |
1720 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1721 | |
1722 | print_conf(conf); | |
6b965620 | 1723 | return count; |
1da177e4 LT |
1724 | } |
1725 | ||
fd01b88c | 1726 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1727 | { |
e8096360 | 1728 | struct r1conf *conf = mddev->private; |
199050ea | 1729 | int err = -EEXIST; |
41158c7e | 1730 | int mirror = 0; |
0eaf822c | 1731 | struct raid1_info *p; |
6c2fce2e | 1732 | int first = 0; |
30194636 | 1733 | int last = conf->raid_disks - 1; |
1da177e4 | 1734 | |
5389042f N |
1735 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1736 | return -EBUSY; | |
1737 | ||
1501efad DW |
1738 | if (md_integrity_add_rdev(rdev, mddev)) |
1739 | return -ENXIO; | |
1740 | ||
6c2fce2e NB |
1741 | if (rdev->raid_disk >= 0) |
1742 | first = last = rdev->raid_disk; | |
1743 | ||
70bcecdb GR |
1744 | /* |
1745 | * find the disk ... but prefer rdev->saved_raid_disk | |
1746 | * if possible. | |
1747 | */ | |
1748 | if (rdev->saved_raid_disk >= 0 && | |
1749 | rdev->saved_raid_disk >= first && | |
9e753ba9 | 1750 | rdev->saved_raid_disk < conf->raid_disks && |
70bcecdb GR |
1751 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1752 | first = last = rdev->saved_raid_disk; | |
1753 | ||
7ef449d1 | 1754 | for (mirror = first; mirror <= last; mirror++) { |
ebfeb444 | 1755 | p = conf->mirrors + mirror; |
7ef449d1 | 1756 | if (!p->rdev) { |
9092c02d JB |
1757 | if (mddev->gendisk) |
1758 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1759 | rdev->data_offset << 9); | |
1da177e4 LT |
1760 | |
1761 | p->head_position = 0; | |
1762 | rdev->raid_disk = mirror; | |
199050ea | 1763 | err = 0; |
6aea114a N |
1764 | /* As all devices are equivalent, we don't need a full recovery |
1765 | * if this was recently any drive of the array | |
1766 | */ | |
1767 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1768 | conf->fullsync = 1; |
d6065f7b | 1769 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1770 | break; |
1771 | } | |
7ef449d1 N |
1772 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1773 | p[conf->raid_disks].rdev == NULL) { | |
1774 | /* Add this device as a replacement */ | |
1775 | clear_bit(In_sync, &rdev->flags); | |
1776 | set_bit(Replacement, &rdev->flags); | |
1777 | rdev->raid_disk = mirror; | |
1778 | err = 0; | |
1779 | conf->fullsync = 1; | |
1780 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1781 | break; | |
1782 | } | |
1783 | } | |
9092c02d | 1784 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
8b904b5b | 1785 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1786 | print_conf(conf); |
199050ea | 1787 | return err; |
1da177e4 LT |
1788 | } |
1789 | ||
b8321b68 | 1790 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1791 | { |
e8096360 | 1792 | struct r1conf *conf = mddev->private; |
1da177e4 | 1793 | int err = 0; |
b8321b68 | 1794 | int number = rdev->raid_disk; |
0eaf822c | 1795 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1796 | |
b014f14c N |
1797 | if (rdev != p->rdev) |
1798 | p = conf->mirrors + conf->raid_disks + number; | |
1799 | ||
1da177e4 | 1800 | print_conf(conf); |
b8321b68 | 1801 | if (rdev == p->rdev) { |
b2d444d7 | 1802 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1803 | atomic_read(&rdev->nr_pending)) { |
1804 | err = -EBUSY; | |
1805 | goto abort; | |
1806 | } | |
046abeed | 1807 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1808 | * is not possible. |
1809 | */ | |
1810 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1811 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1812 | mddev->degraded < conf->raid_disks) { |
1813 | err = -EBUSY; | |
1814 | goto abort; | |
1815 | } | |
1da177e4 | 1816 | p->rdev = NULL; |
d787be40 N |
1817 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1818 | synchronize_rcu(); | |
1819 | if (atomic_read(&rdev->nr_pending)) { | |
1820 | /* lost the race, try later */ | |
1821 | err = -EBUSY; | |
1822 | p->rdev = rdev; | |
1823 | goto abort; | |
1824 | } | |
1825 | } | |
1826 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1827 | /* We just removed a device that is being replaced. |
1828 | * Move down the replacement. We drain all IO before | |
1829 | * doing this to avoid confusion. | |
1830 | */ | |
1831 | struct md_rdev *repl = | |
1832 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1833 | freeze_array(conf, 0); |
3de59bb9 YY |
1834 | if (atomic_read(&repl->nr_pending)) { |
1835 | /* It means that some queued IO of retry_list | |
1836 | * hold repl. Thus, we cannot set replacement | |
1837 | * as NULL, avoiding rdev NULL pointer | |
1838 | * dereference in sync_request_write and | |
1839 | * handle_write_finished. | |
1840 | */ | |
1841 | err = -EBUSY; | |
1842 | unfreeze_array(conf); | |
1843 | goto abort; | |
1844 | } | |
8c7a2c2b N |
1845 | clear_bit(Replacement, &repl->flags); |
1846 | p->rdev = repl; | |
1847 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1848 | unfreeze_array(conf); |
e5bc9c3c GJ |
1849 | } |
1850 | ||
1851 | clear_bit(WantReplacement, &rdev->flags); | |
a91a2785 | 1852 | err = md_integrity_register(mddev); |
1da177e4 LT |
1853 | } |
1854 | abort: | |
1855 | ||
1856 | print_conf(conf); | |
1857 | return err; | |
1858 | } | |
1859 | ||
4246a0b6 | 1860 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1861 | { |
98d30c58 | 1862 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
1da177e4 | 1863 | |
0fc280f6 | 1864 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1865 | |
1da177e4 LT |
1866 | /* |
1867 | * we have read a block, now it needs to be re-written, | |
1868 | * or re-read if the read failed. | |
1869 | * We don't do much here, just schedule handling by raid1d | |
1870 | */ | |
4e4cbee9 | 1871 | if (!bio->bi_status) |
1da177e4 | 1872 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1873 | |
1874 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1875 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1876 | } |
1877 | ||
dfcc34c9 ND |
1878 | static void abort_sync_write(struct mddev *mddev, struct r1bio *r1_bio) |
1879 | { | |
1880 | sector_t sync_blocks = 0; | |
1881 | sector_t s = r1_bio->sector; | |
1882 | long sectors_to_go = r1_bio->sectors; | |
1883 | ||
1884 | /* make sure these bits don't get cleared. */ | |
1885 | do { | |
1886 | md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); | |
1887 | s += sync_blocks; | |
1888 | sectors_to_go -= sync_blocks; | |
1889 | } while (sectors_to_go > 0); | |
1890 | } | |
1891 | ||
449808a2 HT |
1892 | static void put_sync_write_buf(struct r1bio *r1_bio, int uptodate) |
1893 | { | |
1894 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
1895 | struct mddev *mddev = r1_bio->mddev; | |
1896 | int s = r1_bio->sectors; | |
1897 | ||
1898 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
1899 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
1900 | reschedule_retry(r1_bio); | |
1901 | else { | |
1902 | put_buf(r1_bio); | |
1903 | md_done_sync(mddev, s, uptodate); | |
1904 | } | |
1905 | } | |
1906 | } | |
1907 | ||
4246a0b6 | 1908 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1909 | { |
4e4cbee9 | 1910 | int uptodate = !bio->bi_status; |
98d30c58 | 1911 | struct r1bio *r1_bio = get_resync_r1bio(bio); |
fd01b88c | 1912 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1913 | struct r1conf *conf = mddev->private; |
4367af55 N |
1914 | sector_t first_bad; |
1915 | int bad_sectors; | |
854abd75 | 1916 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1917 | |
6b1117d5 | 1918 | if (!uptodate) { |
dfcc34c9 | 1919 | abort_sync_write(mddev, r1_bio); |
854abd75 N |
1920 | set_bit(WriteErrorSeen, &rdev->flags); |
1921 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1922 | set_bit(MD_RECOVERY_NEEDED, & |
1923 | mddev->recovery); | |
d8f05d29 | 1924 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1925 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1926 | &first_bad, &bad_sectors) && |
1927 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1928 | r1_bio->sector, | |
1929 | r1_bio->sectors, | |
1930 | &first_bad, &bad_sectors) | |
1931 | ) | |
4367af55 | 1932 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1933 | |
449808a2 | 1934 | put_sync_write_buf(r1_bio, uptodate); |
1da177e4 LT |
1935 | } |
1936 | ||
3cb03002 | 1937 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1938 | int sectors, struct page *page, int rw) |
1939 | { | |
796a5cf0 | 1940 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1941 | /* success */ |
1942 | return 1; | |
19d67169 | 1943 | if (rw == WRITE) { |
d8f05d29 | 1944 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1945 | if (!test_and_set_bit(WantReplacement, |
1946 | &rdev->flags)) | |
1947 | set_bit(MD_RECOVERY_NEEDED, & | |
1948 | rdev->mddev->recovery); | |
1949 | } | |
d8f05d29 N |
1950 | /* need to record an error - either for the block or the device */ |
1951 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1952 | md_error(rdev->mddev, rdev); | |
1953 | return 0; | |
1954 | } | |
1955 | ||
9f2c9d12 | 1956 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1957 | { |
a68e5870 N |
1958 | /* Try some synchronous reads of other devices to get |
1959 | * good data, much like with normal read errors. Only | |
1960 | * read into the pages we already have so we don't | |
1961 | * need to re-issue the read request. | |
1962 | * We don't need to freeze the array, because being in an | |
1963 | * active sync request, there is no normal IO, and | |
1964 | * no overlapping syncs. | |
06f60385 N |
1965 | * We don't need to check is_badblock() again as we |
1966 | * made sure that anything with a bad block in range | |
1967 | * will have bi_end_io clear. | |
a68e5870 | 1968 | */ |
fd01b88c | 1969 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1970 | struct r1conf *conf = mddev->private; |
a68e5870 | 1971 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
44cf0f4d | 1972 | struct page **pages = get_resync_pages(bio)->pages; |
a68e5870 N |
1973 | sector_t sect = r1_bio->sector; |
1974 | int sectors = r1_bio->sectors; | |
1975 | int idx = 0; | |
2e52d449 N |
1976 | struct md_rdev *rdev; |
1977 | ||
1978 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
1979 | if (test_bit(FailFast, &rdev->flags)) { | |
1980 | /* Don't try recovering from here - just fail it | |
1981 | * ... unless it is the last working device of course */ | |
1982 | md_error(mddev, rdev); | |
1983 | if (test_bit(Faulty, &rdev->flags)) | |
1984 | /* Don't try to read from here, but make sure | |
1985 | * put_buf does it's thing | |
1986 | */ | |
1987 | bio->bi_end_io = end_sync_write; | |
1988 | } | |
a68e5870 N |
1989 | |
1990 | while(sectors) { | |
1991 | int s = sectors; | |
1992 | int d = r1_bio->read_disk; | |
1993 | int success = 0; | |
78d7f5f7 | 1994 | int start; |
a68e5870 N |
1995 | |
1996 | if (s > (PAGE_SIZE>>9)) | |
1997 | s = PAGE_SIZE >> 9; | |
1998 | do { | |
1999 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
2000 | /* No rcu protection needed here devices | |
2001 | * can only be removed when no resync is | |
2002 | * active, and resync is currently active | |
2003 | */ | |
2004 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 2005 | if (sync_page_io(rdev, sect, s<<9, |
44cf0f4d | 2006 | pages[idx], |
796a5cf0 | 2007 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
2008 | success = 1; |
2009 | break; | |
2010 | } | |
2011 | } | |
2012 | d++; | |
8f19ccb2 | 2013 | if (d == conf->raid_disks * 2) |
a68e5870 N |
2014 | d = 0; |
2015 | } while (!success && d != r1_bio->read_disk); | |
2016 | ||
78d7f5f7 | 2017 | if (!success) { |
a68e5870 | 2018 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
2019 | int abort = 0; |
2020 | /* Cannot read from anywhere, this block is lost. | |
2021 | * Record a bad block on each device. If that doesn't | |
2022 | * work just disable and interrupt the recovery. | |
2023 | * Don't fail devices as that won't really help. | |
2024 | */ | |
1d41c216 | 2025 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
74d46992 | 2026 | mdname(mddev), bio_devname(bio, b), |
1d41c216 | 2027 | (unsigned long long)r1_bio->sector); |
8f19ccb2 | 2028 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
2029 | rdev = conf->mirrors[d].rdev; |
2030 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
2031 | continue; | |
2032 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
2033 | abort = 1; | |
2034 | } | |
2035 | if (abort) { | |
d890fa2b N |
2036 | conf->recovery_disabled = |
2037 | mddev->recovery_disabled; | |
3a9f28a5 N |
2038 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
2039 | md_done_sync(mddev, r1_bio->sectors, 0); | |
2040 | put_buf(r1_bio); | |
2041 | return 0; | |
2042 | } | |
2043 | /* Try next page */ | |
2044 | sectors -= s; | |
2045 | sect += s; | |
2046 | idx++; | |
2047 | continue; | |
d11c171e | 2048 | } |
78d7f5f7 N |
2049 | |
2050 | start = d; | |
2051 | /* write it back and re-read */ | |
2052 | while (d != r1_bio->read_disk) { | |
2053 | if (d == 0) | |
8f19ccb2 | 2054 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2055 | d--; |
2056 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2057 | continue; | |
2058 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2059 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2060 | pages[idx], |
d8f05d29 | 2061 | WRITE) == 0) { |
78d7f5f7 N |
2062 | r1_bio->bios[d]->bi_end_io = NULL; |
2063 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 2064 | } |
78d7f5f7 N |
2065 | } |
2066 | d = start; | |
2067 | while (d != r1_bio->read_disk) { | |
2068 | if (d == 0) | |
8f19ccb2 | 2069 | d = conf->raid_disks * 2; |
78d7f5f7 N |
2070 | d--; |
2071 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
2072 | continue; | |
2073 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 | 2074 | if (r1_sync_page_io(rdev, sect, s, |
44cf0f4d | 2075 | pages[idx], |
d8f05d29 | 2076 | READ) != 0) |
9d3d8011 | 2077 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 2078 | } |
a68e5870 N |
2079 | sectors -= s; |
2080 | sect += s; | |
2081 | idx ++; | |
2082 | } | |
78d7f5f7 | 2083 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4e4cbee9 | 2084 | bio->bi_status = 0; |
a68e5870 N |
2085 | return 1; |
2086 | } | |
2087 | ||
c95e6385 | 2088 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
2089 | { |
2090 | /* We have read all readable devices. If we haven't | |
2091 | * got the block, then there is no hope left. | |
2092 | * If we have, then we want to do a comparison | |
2093 | * and skip the write if everything is the same. | |
2094 | * If any blocks failed to read, then we need to | |
2095 | * attempt an over-write | |
2096 | */ | |
fd01b88c | 2097 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2098 | struct r1conf *conf = mddev->private; |
a68e5870 N |
2099 | int primary; |
2100 | int i; | |
f4380a91 | 2101 | int vcnt; |
a68e5870 | 2102 | |
30bc9b53 N |
2103 | /* Fix variable parts of all bios */ |
2104 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
2105 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
4e4cbee9 | 2106 | blk_status_t status; |
30bc9b53 | 2107 | struct bio *b = r1_bio->bios[i]; |
98d30c58 | 2108 | struct resync_pages *rp = get_resync_pages(b); |
30bc9b53 N |
2109 | if (b->bi_end_io != end_sync_read) |
2110 | continue; | |
4246a0b6 | 2111 | /* fixup the bio for reuse, but preserve errno */ |
4e4cbee9 | 2112 | status = b->bi_status; |
30bc9b53 | 2113 | bio_reset(b); |
4e4cbee9 | 2114 | b->bi_status = status; |
4f024f37 | 2115 | b->bi_iter.bi_sector = r1_bio->sector + |
30bc9b53 | 2116 | conf->mirrors[i].rdev->data_offset; |
74d46992 | 2117 | bio_set_dev(b, conf->mirrors[i].rdev->bdev); |
30bc9b53 | 2118 | b->bi_end_io = end_sync_read; |
98d30c58 ML |
2119 | rp->raid_bio = r1_bio; |
2120 | b->bi_private = rp; | |
30bc9b53 | 2121 | |
fb0eb5df ML |
2122 | /* initialize bvec table again */ |
2123 | md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); | |
30bc9b53 | 2124 | } |
8f19ccb2 | 2125 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2126 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4e4cbee9 | 2127 | !r1_bio->bios[primary]->bi_status) { |
a68e5870 N |
2128 | r1_bio->bios[primary]->bi_end_io = NULL; |
2129 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2130 | break; | |
2131 | } | |
2132 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2133 | for (i = 0; i < conf->raid_disks * 2; i++) { |
2b070cfe | 2134 | int j = 0; |
78d7f5f7 N |
2135 | struct bio *pbio = r1_bio->bios[primary]; |
2136 | struct bio *sbio = r1_bio->bios[i]; | |
4e4cbee9 | 2137 | blk_status_t status = sbio->bi_status; |
44cf0f4d ML |
2138 | struct page **ppages = get_resync_pages(pbio)->pages; |
2139 | struct page **spages = get_resync_pages(sbio)->pages; | |
60928a91 | 2140 | struct bio_vec *bi; |
8fc04e6e | 2141 | int page_len[RESYNC_PAGES] = { 0 }; |
6dc4f100 | 2142 | struct bvec_iter_all iter_all; |
a68e5870 | 2143 | |
2aabaa65 | 2144 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2145 | continue; |
4246a0b6 | 2146 | /* Now we can 'fixup' the error value */ |
4e4cbee9 | 2147 | sbio->bi_status = 0; |
78d7f5f7 | 2148 | |
2b070cfe CH |
2149 | bio_for_each_segment_all(bi, sbio, iter_all) |
2150 | page_len[j++] = bi->bv_len; | |
60928a91 | 2151 | |
4e4cbee9 | 2152 | if (!status) { |
78d7f5f7 | 2153 | for (j = vcnt; j-- ; ) { |
44cf0f4d ML |
2154 | if (memcmp(page_address(ppages[j]), |
2155 | page_address(spages[j]), | |
60928a91 | 2156 | page_len[j])) |
78d7f5f7 | 2157 | break; |
69382e85 | 2158 | } |
78d7f5f7 N |
2159 | } else |
2160 | j = 0; | |
2161 | if (j >= 0) | |
7f7583d4 | 2162 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2163 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4e4cbee9 | 2164 | && !status)) { |
78d7f5f7 N |
2165 | /* No need to write to this device. */ |
2166 | sbio->bi_end_io = NULL; | |
2167 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2168 | continue; | |
2169 | } | |
d3b45c2a KO |
2170 | |
2171 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2172 | } |
a68e5870 N |
2173 | } |
2174 | ||
9f2c9d12 | 2175 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2176 | { |
e8096360 | 2177 | struct r1conf *conf = mddev->private; |
a68e5870 | 2178 | int i; |
8f19ccb2 | 2179 | int disks = conf->raid_disks * 2; |
037d2ff6 | 2180 | struct bio *wbio; |
a68e5870 | 2181 | |
a68e5870 N |
2182 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2183 | /* ouch - failed to read all of that. */ | |
2184 | if (!fix_sync_read_error(r1_bio)) | |
2185 | return; | |
7ca78d57 N |
2186 | |
2187 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2188 | process_checks(r1_bio); |
2189 | ||
d11c171e N |
2190 | /* |
2191 | * schedule writes | |
2192 | */ | |
1da177e4 LT |
2193 | atomic_set(&r1_bio->remaining, 1); |
2194 | for (i = 0; i < disks ; i++) { | |
2195 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2196 | if (wbio->bi_end_io == NULL || |
2197 | (wbio->bi_end_io == end_sync_read && | |
2198 | (i == r1_bio->read_disk || | |
2199 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 | 2200 | continue; |
dfcc34c9 ND |
2201 | if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { |
2202 | abort_sync_write(mddev, r1_bio); | |
0c9d5b12 | 2203 | continue; |
dfcc34c9 | 2204 | } |
1da177e4 | 2205 | |
796a5cf0 | 2206 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
212e7eb7 N |
2207 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2208 | wbio->bi_opf |= MD_FAILFAST; | |
2209 | ||
3e198f78 | 2210 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2211 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2212 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2213 | |
ed00aabd | 2214 | submit_bio_noacct(wbio); |
1da177e4 LT |
2215 | } |
2216 | ||
449808a2 | 2217 | put_sync_write_buf(r1_bio, 1); |
1da177e4 LT |
2218 | } |
2219 | ||
2220 | /* | |
2221 | * This is a kernel thread which: | |
2222 | * | |
2223 | * 1. Retries failed read operations on working mirrors. | |
2224 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2225 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2226 | */ |
2227 | ||
e8096360 | 2228 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2229 | sector_t sect, int sectors) |
2230 | { | |
fd01b88c | 2231 | struct mddev *mddev = conf->mddev; |
867868fb N |
2232 | while(sectors) { |
2233 | int s = sectors; | |
2234 | int d = read_disk; | |
2235 | int success = 0; | |
2236 | int start; | |
3cb03002 | 2237 | struct md_rdev *rdev; |
867868fb N |
2238 | |
2239 | if (s > (PAGE_SIZE>>9)) | |
2240 | s = PAGE_SIZE >> 9; | |
2241 | ||
2242 | do { | |
d2eb35ac N |
2243 | sector_t first_bad; |
2244 | int bad_sectors; | |
2245 | ||
707a6a42 N |
2246 | rcu_read_lock(); |
2247 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2248 | if (rdev && |
da8840a7 | 2249 | (test_bit(In_sync, &rdev->flags) || |
2250 | (!test_bit(Faulty, &rdev->flags) && | |
2251 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2252 | is_badblock(rdev, sect, s, |
707a6a42 N |
2253 | &first_bad, &bad_sectors) == 0) { |
2254 | atomic_inc(&rdev->nr_pending); | |
2255 | rcu_read_unlock(); | |
2256 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2257 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2258 | success = 1; |
2259 | rdev_dec_pending(rdev, mddev); | |
2260 | if (success) | |
2261 | break; | |
2262 | } else | |
2263 | rcu_read_unlock(); | |
2264 | d++; | |
2265 | if (d == conf->raid_disks * 2) | |
2266 | d = 0; | |
867868fb N |
2267 | } while (!success && d != read_disk); |
2268 | ||
2269 | if (!success) { | |
d8f05d29 | 2270 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2271 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2272 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2273 | md_error(mddev, rdev); | |
867868fb N |
2274 | break; |
2275 | } | |
2276 | /* write it back and re-read */ | |
2277 | start = d; | |
2278 | while (d != read_disk) { | |
2279 | if (d==0) | |
8f19ccb2 | 2280 | d = conf->raid_disks * 2; |
867868fb | 2281 | d--; |
707a6a42 N |
2282 | rcu_read_lock(); |
2283 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2284 | if (rdev && |
707a6a42 N |
2285 | !test_bit(Faulty, &rdev->flags)) { |
2286 | atomic_inc(&rdev->nr_pending); | |
2287 | rcu_read_unlock(); | |
d8f05d29 N |
2288 | r1_sync_page_io(rdev, sect, s, |
2289 | conf->tmppage, WRITE); | |
707a6a42 N |
2290 | rdev_dec_pending(rdev, mddev); |
2291 | } else | |
2292 | rcu_read_unlock(); | |
867868fb N |
2293 | } |
2294 | d = start; | |
2295 | while (d != read_disk) { | |
2296 | char b[BDEVNAME_SIZE]; | |
2297 | if (d==0) | |
8f19ccb2 | 2298 | d = conf->raid_disks * 2; |
867868fb | 2299 | d--; |
707a6a42 N |
2300 | rcu_read_lock(); |
2301 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2302 | if (rdev && |
b8cb6b4c | 2303 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2304 | atomic_inc(&rdev->nr_pending); |
2305 | rcu_read_unlock(); | |
d8f05d29 N |
2306 | if (r1_sync_page_io(rdev, sect, s, |
2307 | conf->tmppage, READ)) { | |
867868fb | 2308 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2309 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2310 | mdname(mddev), s, | |
2311 | (unsigned long long)(sect + | |
2312 | rdev->data_offset), | |
2313 | bdevname(rdev->bdev, b)); | |
867868fb | 2314 | } |
707a6a42 N |
2315 | rdev_dec_pending(rdev, mddev); |
2316 | } else | |
2317 | rcu_read_unlock(); | |
867868fb N |
2318 | } |
2319 | sectors -= s; | |
2320 | sect += s; | |
2321 | } | |
2322 | } | |
2323 | ||
9f2c9d12 | 2324 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2325 | { |
fd01b88c | 2326 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2327 | struct r1conf *conf = mddev->private; |
3cb03002 | 2328 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2329 | |
2330 | /* bio has the data to be written to device 'i' where | |
2331 | * we just recently had a write error. | |
2332 | * We repeatedly clone the bio and trim down to one block, | |
2333 | * then try the write. Where the write fails we record | |
2334 | * a bad block. | |
2335 | * It is conceivable that the bio doesn't exactly align with | |
2336 | * blocks. We must handle this somehow. | |
2337 | * | |
2338 | * We currently own a reference on the rdev. | |
2339 | */ | |
2340 | ||
2341 | int block_sectors; | |
2342 | sector_t sector; | |
2343 | int sectors; | |
2344 | int sect_to_write = r1_bio->sectors; | |
2345 | int ok = 1; | |
2346 | ||
2347 | if (rdev->badblocks.shift < 0) | |
2348 | return 0; | |
2349 | ||
ab713cdc ND |
2350 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2351 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2352 | sector = r1_bio->sector; |
2353 | sectors = ((sector + block_sectors) | |
2354 | & ~(sector_t)(block_sectors - 1)) | |
2355 | - sector; | |
2356 | ||
cd5ff9a1 N |
2357 | while (sect_to_write) { |
2358 | struct bio *wbio; | |
2359 | if (sectors > sect_to_write) | |
2360 | sectors = sect_to_write; | |
2361 | /* Write at 'sector' for 'sectors'*/ | |
2362 | ||
b783863f | 2363 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
841c1316 ML |
2364 | wbio = bio_clone_fast(r1_bio->behind_master_bio, |
2365 | GFP_NOIO, | |
afeee514 | 2366 | &mddev->bio_set); |
b783863f | 2367 | } else { |
d7a10308 | 2368 | wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, |
afeee514 | 2369 | &mddev->bio_set); |
b783863f KO |
2370 | } |
2371 | ||
796a5cf0 | 2372 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2373 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2374 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2375 | |
6678d83f | 2376 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2377 | wbio->bi_iter.bi_sector += rdev->data_offset; |
74d46992 | 2378 | bio_set_dev(wbio, rdev->bdev); |
4e49ea4a MC |
2379 | |
2380 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2381 | /* failure! */ |
2382 | ok = rdev_set_badblocks(rdev, sector, | |
2383 | sectors, 0) | |
2384 | && ok; | |
2385 | ||
2386 | bio_put(wbio); | |
2387 | sect_to_write -= sectors; | |
2388 | sector += sectors; | |
2389 | sectors = block_sectors; | |
2390 | } | |
2391 | return ok; | |
2392 | } | |
2393 | ||
e8096360 | 2394 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2395 | { |
2396 | int m; | |
2397 | int s = r1_bio->sectors; | |
8f19ccb2 | 2398 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2399 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2400 | struct bio *bio = r1_bio->bios[m]; |
2401 | if (bio->bi_end_io == NULL) | |
2402 | continue; | |
4e4cbee9 | 2403 | if (!bio->bi_status && |
62096bce | 2404 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2405 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2406 | } |
4e4cbee9 | 2407 | if (bio->bi_status && |
62096bce N |
2408 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2409 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2410 | md_error(conf->mddev, rdev); | |
2411 | } | |
2412 | } | |
2413 | put_buf(r1_bio); | |
2414 | md_done_sync(conf->mddev, s, 1); | |
2415 | } | |
2416 | ||
e8096360 | 2417 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2418 | { |
fd76863e | 2419 | int m, idx; |
55ce74d4 | 2420 | bool fail = false; |
fd76863e | 2421 | |
8f19ccb2 | 2422 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2423 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2424 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2425 | rdev_clear_badblocks(rdev, |
2426 | r1_bio->sector, | |
c6563a8c | 2427 | r1_bio->sectors, 0); |
62096bce N |
2428 | rdev_dec_pending(rdev, conf->mddev); |
2429 | } else if (r1_bio->bios[m] != NULL) { | |
2430 | /* This drive got a write error. We need to | |
2431 | * narrow down and record precise write | |
2432 | * errors. | |
2433 | */ | |
55ce74d4 | 2434 | fail = true; |
62096bce N |
2435 | if (!narrow_write_error(r1_bio, m)) { |
2436 | md_error(conf->mddev, | |
2437 | conf->mirrors[m].rdev); | |
2438 | /* an I/O failed, we can't clear the bitmap */ | |
2439 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2440 | } | |
2441 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2442 | conf->mddev); | |
2443 | } | |
55ce74d4 N |
2444 | if (fail) { |
2445 | spin_lock_irq(&conf->device_lock); | |
2446 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
fd76863e | 2447 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2448 | atomic_inc(&conf->nr_queued[idx]); |
55ce74d4 | 2449 | spin_unlock_irq(&conf->device_lock); |
824e47da | 2450 | /* |
2451 | * In case freeze_array() is waiting for condition | |
2452 | * get_unqueued_pending() == extra to be true. | |
2453 | */ | |
2454 | wake_up(&conf->wait_barrier); | |
55ce74d4 | 2455 | md_wakeup_thread(conf->mddev->thread); |
bd8688a1 N |
2456 | } else { |
2457 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2458 | close_write(r1_bio); | |
55ce74d4 | 2459 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2460 | } |
62096bce N |
2461 | } |
2462 | ||
e8096360 | 2463 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce | 2464 | { |
fd01b88c | 2465 | struct mddev *mddev = conf->mddev; |
62096bce | 2466 | struct bio *bio; |
3cb03002 | 2467 | struct md_rdev *rdev; |
62096bce N |
2468 | |
2469 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2470 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2471 | * the block and we can fix it. | |
2472 | * We freeze all other IO, and try reading the block from | |
2473 | * other devices. When we find one, we re-write | |
2474 | * and check it that fixes the read error. | |
2475 | * This is all done synchronously while the array is | |
2476 | * frozen | |
2477 | */ | |
7449f699 TM |
2478 | |
2479 | bio = r1_bio->bios[r1_bio->read_disk]; | |
7449f699 TM |
2480 | bio_put(bio); |
2481 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2482 | ||
2e52d449 N |
2483 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2484 | if (mddev->ro == 0 | |
2485 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2486 | freeze_array(conf, 1); |
62096bce N |
2487 | fix_read_error(conf, r1_bio->read_disk, |
2488 | r1_bio->sector, r1_bio->sectors); | |
2489 | unfreeze_array(conf); | |
b33d1062 GK |
2490 | } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { |
2491 | md_error(mddev, rdev); | |
7449f699 TM |
2492 | } else { |
2493 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2494 | } | |
2495 | ||
2e52d449 | 2496 | rdev_dec_pending(rdev, conf->mddev); |
689389a0 N |
2497 | allow_barrier(conf, r1_bio->sector); |
2498 | bio = r1_bio->master_bio; | |
62096bce | 2499 | |
689389a0 N |
2500 | /* Reuse the old r1_bio so that the IO_BLOCKED settings are preserved */ |
2501 | r1_bio->state = 0; | |
2502 | raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); | |
62096bce N |
2503 | } |
2504 | ||
4ed8731d | 2505 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2506 | { |
4ed8731d | 2507 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2508 | struct r1bio *r1_bio; |
1da177e4 | 2509 | unsigned long flags; |
e8096360 | 2510 | struct r1conf *conf = mddev->private; |
1da177e4 | 2511 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2512 | struct blk_plug plug; |
fd76863e | 2513 | int idx; |
1da177e4 LT |
2514 | |
2515 | md_check_recovery(mddev); | |
e1dfa0a2 | 2516 | |
55ce74d4 | 2517 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2518 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2519 | LIST_HEAD(tmp); |
2520 | spin_lock_irqsave(&conf->device_lock, flags); | |
fd76863e | 2521 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) |
2522 | list_splice_init(&conf->bio_end_io_list, &tmp); | |
55ce74d4 N |
2523 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2524 | while (!list_empty(&tmp)) { | |
a452744b MP |
2525 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2526 | retry_list); | |
55ce74d4 | 2527 | list_del(&r1_bio->retry_list); |
fd76863e | 2528 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2529 | atomic_dec(&conf->nr_queued[idx]); |
bd8688a1 N |
2530 | if (mddev->degraded) |
2531 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2532 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2533 | close_write(r1_bio); | |
55ce74d4 N |
2534 | raid_end_bio_io(r1_bio); |
2535 | } | |
2536 | } | |
2537 | ||
e1dfa0a2 | 2538 | blk_start_plug(&plug); |
1da177e4 | 2539 | for (;;) { |
191ea9b2 | 2540 | |
0021b7bc | 2541 | flush_pending_writes(conf); |
191ea9b2 | 2542 | |
a35e63ef N |
2543 | spin_lock_irqsave(&conf->device_lock, flags); |
2544 | if (list_empty(head)) { | |
2545 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2546 | break; |
a35e63ef | 2547 | } |
9f2c9d12 | 2548 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2549 | list_del(head->prev); |
fd76863e | 2550 | idx = sector_to_idx(r1_bio->sector); |
824e47da | 2551 | atomic_dec(&conf->nr_queued[idx]); |
1da177e4 LT |
2552 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2553 | ||
2554 | mddev = r1_bio->mddev; | |
070ec55d | 2555 | conf = mddev->private; |
4367af55 | 2556 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2557 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2558 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2559 | handle_sync_write_finished(conf, r1_bio); | |
2560 | else | |
4367af55 | 2561 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2562 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2563 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2564 | handle_write_finished(conf, r1_bio); | |
2565 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2566 | handle_read_error(conf, r1_bio); | |
2567 | else | |
c230e7e5 | 2568 | WARN_ON_ONCE(1); |
62096bce | 2569 | |
1d9d5241 | 2570 | cond_resched(); |
2953079c | 2571 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2572 | md_check_recovery(mddev); |
1da177e4 | 2573 | } |
e1dfa0a2 | 2574 | blk_finish_plug(&plug); |
1da177e4 LT |
2575 | } |
2576 | ||
e8096360 | 2577 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2578 | { |
2579 | int buffs; | |
2580 | ||
2581 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 KO |
2582 | BUG_ON(mempool_initialized(&conf->r1buf_pool)); |
2583 | ||
2584 | return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, | |
2585 | r1buf_pool_free, conf->poolinfo); | |
1da177e4 LT |
2586 | } |
2587 | ||
208410b5 SL |
2588 | static struct r1bio *raid1_alloc_init_r1buf(struct r1conf *conf) |
2589 | { | |
afeee514 | 2590 | struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); |
208410b5 SL |
2591 | struct resync_pages *rps; |
2592 | struct bio *bio; | |
2593 | int i; | |
2594 | ||
2595 | for (i = conf->poolinfo->raid_disks; i--; ) { | |
2596 | bio = r1bio->bios[i]; | |
2597 | rps = bio->bi_private; | |
2598 | bio_reset(bio); | |
2599 | bio->bi_private = rps; | |
2600 | } | |
2601 | r1bio->master_bio = NULL; | |
2602 | return r1bio; | |
2603 | } | |
2604 | ||
1da177e4 LT |
2605 | /* |
2606 | * perform a "sync" on one "block" | |
2607 | * | |
2608 | * We need to make sure that no normal I/O request - particularly write | |
2609 | * requests - conflict with active sync requests. | |
2610 | * | |
2611 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2612 | * that can be installed to exclude normal IO requests. | |
2613 | */ | |
2614 | ||
849674e4 SL |
2615 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2616 | int *skipped) | |
1da177e4 | 2617 | { |
e8096360 | 2618 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2619 | struct r1bio *r1_bio; |
1da177e4 LT |
2620 | struct bio *bio; |
2621 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2622 | int disk = -1; |
1da177e4 | 2623 | int i; |
3e198f78 N |
2624 | int wonly = -1; |
2625 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2626 | sector_t sync_blocks; |
e3b9703e | 2627 | int still_degraded = 0; |
06f60385 N |
2628 | int good_sectors = RESYNC_SECTORS; |
2629 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
fd76863e | 2630 | int idx = sector_to_idx(sector_nr); |
022e510f | 2631 | int page_idx = 0; |
1da177e4 | 2632 | |
afeee514 | 2633 | if (!mempool_initialized(&conf->r1buf_pool)) |
1da177e4 | 2634 | if (init_resync(conf)) |
57afd89f | 2635 | return 0; |
1da177e4 | 2636 | |
58c0fed4 | 2637 | max_sector = mddev->dev_sectors; |
1da177e4 | 2638 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2639 | /* If we aborted, we need to abort the |
2640 | * sync on the 'current' bitmap chunk (there will | |
2641 | * only be one in raid1 resync. | |
2642 | * We can find the current addess in mddev->curr_resync | |
2643 | */ | |
6a806c51 | 2644 | if (mddev->curr_resync < max_sector) /* aborted */ |
e64e4018 AS |
2645 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
2646 | &sync_blocks, 1); | |
6a806c51 | 2647 | else /* completed sync */ |
191ea9b2 | 2648 | conf->fullsync = 0; |
6a806c51 | 2649 | |
e64e4018 | 2650 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 2651 | close_sync(conf); |
c40f341f GR |
2652 | |
2653 | if (mddev_is_clustered(mddev)) { | |
2654 | conf->cluster_sync_low = 0; | |
2655 | conf->cluster_sync_high = 0; | |
c40f341f | 2656 | } |
1da177e4 LT |
2657 | return 0; |
2658 | } | |
2659 | ||
07d84d10 N |
2660 | if (mddev->bitmap == NULL && |
2661 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2662 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2663 | conf->fullsync == 0) { |
2664 | *skipped = 1; | |
2665 | return max_sector - sector_nr; | |
2666 | } | |
6394cca5 N |
2667 | /* before building a request, check if we can skip these blocks.. |
2668 | * This call the bitmap_start_sync doesn't actually record anything | |
2669 | */ | |
e64e4018 | 2670 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2671 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2672 | /* We can skip this block, and probably several more */ |
2673 | *skipped = 1; | |
2674 | return sync_blocks; | |
2675 | } | |
17999be4 | 2676 | |
7ac50447 TM |
2677 | /* |
2678 | * If there is non-resync activity waiting for a turn, then let it | |
2679 | * though before starting on this new sync request. | |
2680 | */ | |
824e47da | 2681 | if (atomic_read(&conf->nr_waiting[idx])) |
7ac50447 TM |
2682 | schedule_timeout_uninterruptible(1); |
2683 | ||
c40f341f GR |
2684 | /* we are incrementing sector_nr below. To be safe, we check against |
2685 | * sector_nr + two times RESYNC_SECTORS | |
2686 | */ | |
2687 | ||
e64e4018 | 2688 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
c40f341f | 2689 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); |
17999be4 | 2690 | |
8c242593 YY |
2691 | |
2692 | if (raise_barrier(conf, sector_nr)) | |
2693 | return 0; | |
2694 | ||
2695 | r1_bio = raid1_alloc_init_r1buf(conf); | |
1da177e4 | 2696 | |
3e198f78 | 2697 | rcu_read_lock(); |
1da177e4 | 2698 | /* |
3e198f78 N |
2699 | * If we get a correctably read error during resync or recovery, |
2700 | * we might want to read from a different device. So we | |
2701 | * flag all drives that could conceivably be read from for READ, | |
2702 | * and any others (which will be non-In_sync devices) for WRITE. | |
2703 | * If a read fails, we try reading from something else for which READ | |
2704 | * is OK. | |
1da177e4 | 2705 | */ |
1da177e4 | 2706 | |
1da177e4 LT |
2707 | r1_bio->mddev = mddev; |
2708 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2709 | r1_bio->state = 0; |
1da177e4 | 2710 | set_bit(R1BIO_IsSync, &r1_bio->state); |
fd76863e | 2711 | /* make sure good_sectors won't go across barrier unit boundary */ |
2712 | good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors); | |
1da177e4 | 2713 | |
8f19ccb2 | 2714 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2715 | struct md_rdev *rdev; |
1da177e4 | 2716 | bio = r1_bio->bios[i]; |
1da177e4 | 2717 | |
3e198f78 N |
2718 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2719 | if (rdev == NULL || | |
06f60385 | 2720 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2721 | if (i < conf->raid_disks) |
2722 | still_degraded = 1; | |
3e198f78 | 2723 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2724 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2725 | bio->bi_end_io = end_sync_write; |
2726 | write_targets ++; | |
3e198f78 N |
2727 | } else { |
2728 | /* may need to read from here */ | |
06f60385 N |
2729 | sector_t first_bad = MaxSector; |
2730 | int bad_sectors; | |
2731 | ||
2732 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2733 | &first_bad, &bad_sectors)) { | |
2734 | if (first_bad > sector_nr) | |
2735 | good_sectors = first_bad - sector_nr; | |
2736 | else { | |
2737 | bad_sectors -= (sector_nr - first_bad); | |
2738 | if (min_bad == 0 || | |
2739 | min_bad > bad_sectors) | |
2740 | min_bad = bad_sectors; | |
2741 | } | |
2742 | } | |
2743 | if (sector_nr < first_bad) { | |
2744 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2745 | if (wonly < 0) | |
2746 | wonly = i; | |
2747 | } else { | |
2748 | if (disk < 0) | |
2749 | disk = i; | |
2750 | } | |
796a5cf0 | 2751 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2752 | bio->bi_end_io = end_sync_read; |
2753 | read_targets++; | |
d57368af AL |
2754 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2755 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2756 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2757 | /* | |
2758 | * The device is suitable for reading (InSync), | |
2759 | * but has bad block(s) here. Let's try to correct them, | |
2760 | * if we are doing resync or repair. Otherwise, leave | |
2761 | * this device alone for this sync request. | |
2762 | */ | |
796a5cf0 | 2763 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2764 | bio->bi_end_io = end_sync_write; |
2765 | write_targets++; | |
3e198f78 | 2766 | } |
3e198f78 | 2767 | } |
028288df | 2768 | if (rdev && bio->bi_end_io) { |
06f60385 | 2769 | atomic_inc(&rdev->nr_pending); |
4f024f37 | 2770 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
74d46992 | 2771 | bio_set_dev(bio, rdev->bdev); |
2e52d449 N |
2772 | if (test_bit(FailFast, &rdev->flags)) |
2773 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2774 | } |
1da177e4 | 2775 | } |
3e198f78 N |
2776 | rcu_read_unlock(); |
2777 | if (disk < 0) | |
2778 | disk = wonly; | |
2779 | r1_bio->read_disk = disk; | |
191ea9b2 | 2780 | |
06f60385 N |
2781 | if (read_targets == 0 && min_bad > 0) { |
2782 | /* These sectors are bad on all InSync devices, so we | |
2783 | * need to mark them bad on all write targets | |
2784 | */ | |
2785 | int ok = 1; | |
8f19ccb2 | 2786 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2787 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2788 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2789 | ok = rdev_set_badblocks(rdev, sector_nr, |
2790 | min_bad, 0 | |
2791 | ) && ok; | |
2792 | } | |
2953079c | 2793 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2794 | *skipped = 1; |
2795 | put_buf(r1_bio); | |
2796 | ||
2797 | if (!ok) { | |
2798 | /* Cannot record the badblocks, so need to | |
2799 | * abort the resync. | |
2800 | * If there are multiple read targets, could just | |
2801 | * fail the really bad ones ??? | |
2802 | */ | |
2803 | conf->recovery_disabled = mddev->recovery_disabled; | |
2804 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2805 | return 0; | |
2806 | } else | |
2807 | return min_bad; | |
2808 | ||
2809 | } | |
2810 | if (min_bad > 0 && min_bad < good_sectors) { | |
2811 | /* only resync enough to reach the next bad->good | |
2812 | * transition */ | |
2813 | good_sectors = min_bad; | |
2814 | } | |
2815 | ||
3e198f78 N |
2816 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2817 | /* extra read targets are also write targets */ | |
2818 | write_targets += read_targets-1; | |
2819 | ||
2820 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2821 | /* There is nowhere to write, so all non-sync |
2822 | * drives must be failed - so we are finished | |
2823 | */ | |
b7219ccb N |
2824 | sector_t rv; |
2825 | if (min_bad > 0) | |
2826 | max_sector = sector_nr + min_bad; | |
2827 | rv = max_sector - sector_nr; | |
57afd89f | 2828 | *skipped = 1; |
1da177e4 | 2829 | put_buf(r1_bio); |
1da177e4 LT |
2830 | return rv; |
2831 | } | |
2832 | ||
c6207277 N |
2833 | if (max_sector > mddev->resync_max) |
2834 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2835 | if (max_sector > sector_nr + good_sectors) |
2836 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2837 | nr_sectors = 0; |
289e99e8 | 2838 | sync_blocks = 0; |
1da177e4 LT |
2839 | do { |
2840 | struct page *page; | |
2841 | int len = PAGE_SIZE; | |
2842 | if (sector_nr + (len>>9) > max_sector) | |
2843 | len = (max_sector - sector_nr) << 9; | |
2844 | if (len == 0) | |
2845 | break; | |
6a806c51 | 2846 | if (sync_blocks == 0) { |
e64e4018 AS |
2847 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
2848 | &sync_blocks, still_degraded) && | |
e5de485f N |
2849 | !conf->fullsync && |
2850 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2851 | break; |
7571ae88 | 2852 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2853 | len = sync_blocks<<9; |
ab7a30c7 | 2854 | } |
191ea9b2 | 2855 | |
8f19ccb2 | 2856 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
98d30c58 ML |
2857 | struct resync_pages *rp; |
2858 | ||
1da177e4 | 2859 | bio = r1_bio->bios[i]; |
98d30c58 | 2860 | rp = get_resync_pages(bio); |
1da177e4 | 2861 | if (bio->bi_end_io) { |
022e510f | 2862 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
2863 | |
2864 | /* | |
2865 | * won't fail because the vec table is big | |
2866 | * enough to hold all these pages | |
2867 | */ | |
2868 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
2869 | } |
2870 | } | |
2871 | nr_sectors += len>>9; | |
2872 | sector_nr += len>>9; | |
191ea9b2 | 2873 | sync_blocks -= (len>>9); |
022e510f | 2874 | } while (++page_idx < RESYNC_PAGES); |
98d30c58 | 2875 | |
1da177e4 LT |
2876 | r1_bio->sectors = nr_sectors; |
2877 | ||
c40f341f GR |
2878 | if (mddev_is_clustered(mddev) && |
2879 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2880 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2881 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2882 | /* Send resync message */ | |
2883 | md_cluster_ops->resync_info_update(mddev, | |
2884 | conf->cluster_sync_low, | |
2885 | conf->cluster_sync_high); | |
2886 | } | |
2887 | ||
d11c171e N |
2888 | /* For a user-requested sync, we read all readable devices and do a |
2889 | * compare | |
2890 | */ | |
2891 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2892 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2893 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2894 | bio = r1_bio->bios[i]; |
2895 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2896 | read_targets--; |
74d46992 | 2897 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2898 | if (read_targets == 1) |
2899 | bio->bi_opf &= ~MD_FAILFAST; | |
ed00aabd | 2900 | submit_bio_noacct(bio); |
d11c171e N |
2901 | } |
2902 | } | |
2903 | } else { | |
2904 | atomic_set(&r1_bio->remaining, 1); | |
2905 | bio = r1_bio->bios[r1_bio->read_disk]; | |
74d46992 | 2906 | md_sync_acct_bio(bio, nr_sectors); |
2e52d449 N |
2907 | if (read_targets == 1) |
2908 | bio->bi_opf &= ~MD_FAILFAST; | |
ed00aabd | 2909 | submit_bio_noacct(bio); |
d11c171e | 2910 | } |
1da177e4 LT |
2911 | return nr_sectors; |
2912 | } | |
2913 | ||
fd01b88c | 2914 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2915 | { |
2916 | if (sectors) | |
2917 | return sectors; | |
2918 | ||
2919 | return mddev->dev_sectors; | |
2920 | } | |
2921 | ||
e8096360 | 2922 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2923 | { |
e8096360 | 2924 | struct r1conf *conf; |
709ae487 | 2925 | int i; |
0eaf822c | 2926 | struct raid1_info *disk; |
3cb03002 | 2927 | struct md_rdev *rdev; |
709ae487 | 2928 | int err = -ENOMEM; |
1da177e4 | 2929 | |
e8096360 | 2930 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2931 | if (!conf) |
709ae487 | 2932 | goto abort; |
1da177e4 | 2933 | |
fd76863e | 2934 | conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, |
824e47da | 2935 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2936 | if (!conf->nr_pending) |
2937 | goto abort; | |
2938 | ||
2939 | conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2940 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2941 | if (!conf->nr_waiting) |
2942 | goto abort; | |
2943 | ||
2944 | conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2945 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2946 | if (!conf->nr_queued) |
2947 | goto abort; | |
2948 | ||
2949 | conf->barrier = kcalloc(BARRIER_BUCKETS_NR, | |
824e47da | 2950 | sizeof(atomic_t), GFP_KERNEL); |
fd76863e | 2951 | if (!conf->barrier) |
2952 | goto abort; | |
2953 | ||
6396bb22 KC |
2954 | conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
2955 | mddev->raid_disks, 2), | |
2956 | GFP_KERNEL); | |
1da177e4 | 2957 | if (!conf->mirrors) |
709ae487 | 2958 | goto abort; |
1da177e4 | 2959 | |
ddaf22ab N |
2960 | conf->tmppage = alloc_page(GFP_KERNEL); |
2961 | if (!conf->tmppage) | |
709ae487 | 2962 | goto abort; |
ddaf22ab | 2963 | |
709ae487 | 2964 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2965 | if (!conf->poolinfo) |
709ae487 | 2966 | goto abort; |
8f19ccb2 | 2967 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
3f677f9c | 2968 | err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 2969 | rbio_pool_free, conf->poolinfo); |
afeee514 | 2970 | if (err) |
709ae487 N |
2971 | goto abort; |
2972 | ||
afeee514 KO |
2973 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
2974 | if (err) | |
c230e7e5 N |
2975 | goto abort; |
2976 | ||
ed9bfdf1 | 2977 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2978 | |
c19d5798 | 2979 | err = -EINVAL; |
e7e72bf6 | 2980 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2981 | rdev_for_each(rdev, mddev) { |
709ae487 | 2982 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2983 | if (disk_idx >= mddev->raid_disks |
2984 | || disk_idx < 0) | |
2985 | continue; | |
c19d5798 | 2986 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2987 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2988 | else |
2989 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2990 | |
c19d5798 N |
2991 | if (disk->rdev) |
2992 | goto abort; | |
1da177e4 | 2993 | disk->rdev = rdev; |
1da177e4 | 2994 | disk->head_position = 0; |
12cee5a8 | 2995 | disk->seq_start = MaxSector; |
1da177e4 LT |
2996 | } |
2997 | conf->raid_disks = mddev->raid_disks; | |
2998 | conf->mddev = mddev; | |
1da177e4 | 2999 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 3000 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
3001 | |
3002 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 3003 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 3004 | |
191ea9b2 | 3005 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 3006 | conf->pending_count = 0; |
d890fa2b | 3007 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 3008 | |
c19d5798 | 3009 | err = -EIO; |
8f19ccb2 | 3010 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
3011 | |
3012 | disk = conf->mirrors + i; | |
3013 | ||
c19d5798 N |
3014 | if (i < conf->raid_disks && |
3015 | disk[conf->raid_disks].rdev) { | |
3016 | /* This slot has a replacement. */ | |
3017 | if (!disk->rdev) { | |
3018 | /* No original, just make the replacement | |
3019 | * a recovering spare | |
3020 | */ | |
3021 | disk->rdev = | |
3022 | disk[conf->raid_disks].rdev; | |
3023 | disk[conf->raid_disks].rdev = NULL; | |
3024 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
3025 | /* Original is not in_sync - bad */ | |
3026 | goto abort; | |
3027 | } | |
3028 | ||
5fd6c1dc N |
3029 | if (!disk->rdev || |
3030 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 3031 | disk->head_position = 0; |
4f0a5e01 JB |
3032 | if (disk->rdev && |
3033 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 3034 | conf->fullsync = 1; |
be4d3280 | 3035 | } |
1da177e4 | 3036 | } |
709ae487 | 3037 | |
709ae487 | 3038 | err = -ENOMEM; |
0232605d | 3039 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 3040 | if (!conf->thread) |
709ae487 | 3041 | goto abort; |
1da177e4 | 3042 | |
709ae487 N |
3043 | return conf; |
3044 | ||
3045 | abort: | |
3046 | if (conf) { | |
afeee514 | 3047 | mempool_exit(&conf->r1bio_pool); |
709ae487 N |
3048 | kfree(conf->mirrors); |
3049 | safe_put_page(conf->tmppage); | |
3050 | kfree(conf->poolinfo); | |
fd76863e | 3051 | kfree(conf->nr_pending); |
3052 | kfree(conf->nr_waiting); | |
3053 | kfree(conf->nr_queued); | |
3054 | kfree(conf->barrier); | |
afeee514 | 3055 | bioset_exit(&conf->bio_split); |
709ae487 N |
3056 | kfree(conf); |
3057 | } | |
3058 | return ERR_PTR(err); | |
3059 | } | |
3060 | ||
afa0f557 | 3061 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 3062 | static int raid1_run(struct mddev *mddev) |
709ae487 | 3063 | { |
e8096360 | 3064 | struct r1conf *conf; |
709ae487 | 3065 | int i; |
3cb03002 | 3066 | struct md_rdev *rdev; |
5220ea1e | 3067 | int ret; |
2ff8cc2c | 3068 | bool discard_supported = false; |
709ae487 N |
3069 | |
3070 | if (mddev->level != 1) { | |
1d41c216 N |
3071 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3072 | mdname(mddev), mddev->level); | |
709ae487 N |
3073 | return -EIO; |
3074 | } | |
3075 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3076 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3077 | mdname(mddev)); | |
709ae487 N |
3078 | return -EIO; |
3079 | } | |
a415c0f1 N |
3080 | if (mddev_init_writes_pending(mddev) < 0) |
3081 | return -ENOMEM; | |
1da177e4 | 3082 | /* |
709ae487 N |
3083 | * copy the already verified devices into our private RAID1 |
3084 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3085 | * should be freed in raid1_free()] |
1da177e4 | 3086 | */ |
709ae487 N |
3087 | if (mddev->private == NULL) |
3088 | conf = setup_conf(mddev); | |
3089 | else | |
3090 | conf = mddev->private; | |
1da177e4 | 3091 | |
709ae487 N |
3092 | if (IS_ERR(conf)) |
3093 | return PTR_ERR(conf); | |
1da177e4 | 3094 | |
3deff1a7 | 3095 | if (mddev->queue) { |
5026d7a9 | 3096 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 CH |
3097 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
3098 | } | |
5026d7a9 | 3099 | |
dafb20fa | 3100 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3101 | if (!mddev->gendisk) |
3102 | continue; | |
709ae487 N |
3103 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3104 | rdev->data_offset << 9); | |
2ff8cc2c SL |
3105 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
3106 | discard_supported = true; | |
1da177e4 | 3107 | } |
191ea9b2 | 3108 | |
709ae487 | 3109 | mddev->degraded = 0; |
ebfeb444 | 3110 | for (i = 0; i < conf->raid_disks; i++) |
709ae487 N |
3111 | if (conf->mirrors[i].rdev == NULL || |
3112 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3113 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3114 | mddev->degraded++; | |
07f1a685 YY |
3115 | /* |
3116 | * RAID1 needs at least one disk in active | |
3117 | */ | |
3118 | if (conf->raid_disks - mddev->degraded < 1) { | |
3119 | ret = -EINVAL; | |
3120 | goto abort; | |
3121 | } | |
709ae487 N |
3122 | |
3123 | if (conf->raid_disks - mddev->degraded == 1) | |
3124 | mddev->recovery_cp = MaxSector; | |
3125 | ||
8c6ac868 | 3126 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3127 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3128 | mdname(mddev)); | |
3129 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3130 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3131 | mddev->raid_disks); |
709ae487 | 3132 | |
1da177e4 LT |
3133 | /* |
3134 | * Ok, everything is just fine now | |
3135 | */ | |
709ae487 N |
3136 | mddev->thread = conf->thread; |
3137 | conf->thread = NULL; | |
3138 | mddev->private = conf; | |
46533ff7 | 3139 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3140 | |
1f403624 | 3141 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3142 | |
1ed7242e | 3143 | if (mddev->queue) { |
2ff8cc2c | 3144 | if (discard_supported) |
8b904b5b | 3145 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
2ff8cc2c SL |
3146 | mddev->queue); |
3147 | else | |
8b904b5b | 3148 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
2ff8cc2c | 3149 | mddev->queue); |
1ed7242e | 3150 | } |
5220ea1e | 3151 | |
ebfeb444 | 3152 | ret = md_integrity_register(mddev); |
5aa61f42 N |
3153 | if (ret) { |
3154 | md_unregister_thread(&mddev->thread); | |
07f1a685 | 3155 | goto abort; |
5aa61f42 | 3156 | } |
07f1a685 YY |
3157 | return 0; |
3158 | ||
3159 | abort: | |
3160 | raid1_free(mddev, conf); | |
5220ea1e | 3161 | return ret; |
1da177e4 LT |
3162 | } |
3163 | ||
afa0f557 | 3164 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3165 | { |
afa0f557 | 3166 | struct r1conf *conf = priv; |
409c57f3 | 3167 | |
afeee514 | 3168 | mempool_exit(&conf->r1bio_pool); |
990a8baf | 3169 | kfree(conf->mirrors); |
0fea7ed8 | 3170 | safe_put_page(conf->tmppage); |
990a8baf | 3171 | kfree(conf->poolinfo); |
fd76863e | 3172 | kfree(conf->nr_pending); |
3173 | kfree(conf->nr_waiting); | |
3174 | kfree(conf->nr_queued); | |
3175 | kfree(conf->barrier); | |
afeee514 | 3176 | bioset_exit(&conf->bio_split); |
1da177e4 | 3177 | kfree(conf); |
1da177e4 LT |
3178 | } |
3179 | ||
fd01b88c | 3180 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3181 | { |
3182 | /* no resync is happening, and there is enough space | |
3183 | * on all devices, so we can resize. | |
3184 | * We need to make sure resync covers any new space. | |
3185 | * If the array is shrinking we should possibly wait until | |
3186 | * any io in the removed space completes, but it hardly seems | |
3187 | * worth it. | |
3188 | */ | |
a4a6125a N |
3189 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3190 | if (mddev->external_size && | |
3191 | mddev->array_sectors > newsize) | |
b522adcd | 3192 | return -EINVAL; |
a4a6125a | 3193 | if (mddev->bitmap) { |
e64e4018 | 3194 | int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); |
a4a6125a N |
3195 | if (ret) |
3196 | return ret; | |
3197 | } | |
3198 | md_set_array_sectors(mddev, newsize); | |
b522adcd | 3199 | if (sectors > mddev->dev_sectors && |
b098636c | 3200 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3201 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3202 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3203 | } | |
b522adcd | 3204 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3205 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3206 | return 0; |
3207 | } | |
3208 | ||
fd01b88c | 3209 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3210 | { |
3211 | /* We need to: | |
3212 | * 1/ resize the r1bio_pool | |
3213 | * 2/ resize conf->mirrors | |
3214 | * | |
3215 | * We allocate a new r1bio_pool if we can. | |
3216 | * Then raise a device barrier and wait until all IO stops. | |
3217 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3218 | * |
3219 | * At the same time, we "pack" the devices so that all the missing | |
3220 | * devices have the higher raid_disk numbers. | |
1da177e4 | 3221 | */ |
afeee514 | 3222 | mempool_t newpool, oldpool; |
1da177e4 | 3223 | struct pool_info *newpoolinfo; |
0eaf822c | 3224 | struct raid1_info *newmirrors; |
e8096360 | 3225 | struct r1conf *conf = mddev->private; |
63c70c4f | 3226 | int cnt, raid_disks; |
c04be0aa | 3227 | unsigned long flags; |
2214c260 | 3228 | int d, d2; |
afeee514 KO |
3229 | int ret; |
3230 | ||
3231 | memset(&newpool, 0, sizeof(newpool)); | |
3232 | memset(&oldpool, 0, sizeof(oldpool)); | |
1da177e4 | 3233 | |
63c70c4f | 3234 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3235 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3236 | mddev->layout != mddev->new_layout || |
3237 | mddev->level != mddev->new_level) { | |
664e7c41 | 3238 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3239 | mddev->new_layout = mddev->layout; |
3240 | mddev->new_level = mddev->level; | |
3241 | return -EINVAL; | |
3242 | } | |
3243 | ||
2214c260 AP |
3244 | if (!mddev_is_clustered(mddev)) |
3245 | md_allow_write(mddev); | |
2a2275d6 | 3246 | |
63c70c4f N |
3247 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3248 | ||
6ea9c07c N |
3249 | if (raid_disks < conf->raid_disks) { |
3250 | cnt=0; | |
3251 | for (d= 0; d < conf->raid_disks; d++) | |
3252 | if (conf->mirrors[d].rdev) | |
3253 | cnt++; | |
3254 | if (cnt > raid_disks) | |
1da177e4 | 3255 | return -EBUSY; |
6ea9c07c | 3256 | } |
1da177e4 LT |
3257 | |
3258 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3259 | if (!newpoolinfo) | |
3260 | return -ENOMEM; | |
3261 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3262 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 | 3263 | |
3f677f9c | 3264 | ret = mempool_init(&newpool, NR_RAID_BIOS, r1bio_pool_alloc, |
c7afa803 | 3265 | rbio_pool_free, newpoolinfo); |
afeee514 | 3266 | if (ret) { |
1da177e4 | 3267 | kfree(newpoolinfo); |
afeee514 | 3268 | return ret; |
1da177e4 | 3269 | } |
6396bb22 KC |
3270 | newmirrors = kzalloc(array3_size(sizeof(struct raid1_info), |
3271 | raid_disks, 2), | |
8f19ccb2 | 3272 | GFP_KERNEL); |
1da177e4 LT |
3273 | if (!newmirrors) { |
3274 | kfree(newpoolinfo); | |
afeee514 | 3275 | mempool_exit(&newpool); |
1da177e4 LT |
3276 | return -ENOMEM; |
3277 | } | |
1da177e4 | 3278 | |
e2d59925 | 3279 | freeze_array(conf, 0); |
1da177e4 LT |
3280 | |
3281 | /* ok, everything is stopped */ | |
3282 | oldpool = conf->r1bio_pool; | |
3283 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3284 | |
a88aa786 | 3285 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3286 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3287 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3288 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3289 | rdev->raid_disk = d2; |
36fad858 NK |
3290 | sysfs_unlink_rdev(mddev, rdev); |
3291 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3292 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3293 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3294 | } |
a88aa786 N |
3295 | if (rdev) |
3296 | newmirrors[d2++].rdev = rdev; | |
3297 | } | |
1da177e4 LT |
3298 | kfree(conf->mirrors); |
3299 | conf->mirrors = newmirrors; | |
3300 | kfree(conf->poolinfo); | |
3301 | conf->poolinfo = newpoolinfo; | |
3302 | ||
c04be0aa | 3303 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3304 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3305 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3306 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3307 | mddev->delta_disks = 0; |
1da177e4 | 3308 | |
e2d59925 | 3309 | unfreeze_array(conf); |
1da177e4 | 3310 | |
985ca973 | 3311 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3312 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3313 | md_wakeup_thread(mddev->thread); | |
3314 | ||
afeee514 | 3315 | mempool_exit(&oldpool); |
1da177e4 LT |
3316 | return 0; |
3317 | } | |
3318 | ||
b03e0ccb | 3319 | static void raid1_quiesce(struct mddev *mddev, int quiesce) |
36fa3063 | 3320 | { |
e8096360 | 3321 | struct r1conf *conf = mddev->private; |
36fa3063 | 3322 | |
b03e0ccb | 3323 | if (quiesce) |
07169fd4 | 3324 | freeze_array(conf, 0); |
b03e0ccb | 3325 | else |
07169fd4 | 3326 | unfreeze_array(conf); |
36fa3063 N |
3327 | } |
3328 | ||
fd01b88c | 3329 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3330 | { |
3331 | /* raid1 can take over: | |
3332 | * raid5 with 2 devices, any layout or chunk size | |
3333 | */ | |
3334 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3335 | struct r1conf *conf; |
709ae487 N |
3336 | mddev->new_level = 1; |
3337 | mddev->new_layout = 0; | |
3338 | mddev->new_chunk_sectors = 0; | |
3339 | conf = setup_conf(mddev); | |
6995f0b2 | 3340 | if (!IS_ERR(conf)) { |
07169fd4 | 3341 | /* Array must appear to be quiesced */ |
3342 | conf->array_frozen = 1; | |
394ed8e4 SL |
3343 | mddev_clear_unsupported_flags(mddev, |
3344 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3345 | } |
709ae487 N |
3346 | return conf; |
3347 | } | |
3348 | return ERR_PTR(-EINVAL); | |
3349 | } | |
1da177e4 | 3350 | |
84fc4b56 | 3351 | static struct md_personality raid1_personality = |
1da177e4 LT |
3352 | { |
3353 | .name = "raid1", | |
2604b703 | 3354 | .level = 1, |
1da177e4 | 3355 | .owner = THIS_MODULE, |
849674e4 SL |
3356 | .make_request = raid1_make_request, |
3357 | .run = raid1_run, | |
afa0f557 | 3358 | .free = raid1_free, |
849674e4 SL |
3359 | .status = raid1_status, |
3360 | .error_handler = raid1_error, | |
1da177e4 LT |
3361 | .hot_add_disk = raid1_add_disk, |
3362 | .hot_remove_disk= raid1_remove_disk, | |
3363 | .spare_active = raid1_spare_active, | |
849674e4 | 3364 | .sync_request = raid1_sync_request, |
1da177e4 | 3365 | .resize = raid1_resize, |
80c3a6ce | 3366 | .size = raid1_size, |
63c70c4f | 3367 | .check_reshape = raid1_reshape, |
36fa3063 | 3368 | .quiesce = raid1_quiesce, |
709ae487 | 3369 | .takeover = raid1_takeover, |
1da177e4 LT |
3370 | }; |
3371 | ||
3372 | static int __init raid_init(void) | |
3373 | { | |
2604b703 | 3374 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3375 | } |
3376 | ||
3377 | static void raid_exit(void) | |
3378 | { | |
2604b703 | 3379 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3380 | } |
3381 | ||
3382 | module_init(raid_init); | |
3383 | module_exit(raid_exit); | |
3384 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3385 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3386 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3387 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3388 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3389 | |
3390 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |