]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid1.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat | |
5 | * | |
6 | * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
7 | * | |
8 | * RAID-1 management functions. | |
9 | * | |
10 | * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 | |
11 | * | |
96de0e25 | 12 | * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> |
1da177e4 LT |
13 | * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> |
14 | * | |
191ea9b2 N |
15 | * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support |
16 | * bitmapped intelligence in resync: | |
17 | * | |
18 | * - bitmap marked during normal i/o | |
19 | * - bitmap used to skip nondirty blocks during sync | |
20 | * | |
21 | * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology: | |
22 | * - persistent bitmap code | |
23 | * | |
1da177e4 LT |
24 | * This program is free software; you can redistribute it and/or modify |
25 | * it under the terms of the GNU General Public License as published by | |
26 | * the Free Software Foundation; either version 2, or (at your option) | |
27 | * any later version. | |
28 | * | |
29 | * You should have received a copy of the GNU General Public License | |
30 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
31 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
32 | */ | |
33 | ||
5a0e3ad6 | 34 | #include <linux/slab.h> |
25570727 | 35 | #include <linux/delay.h> |
bff61975 | 36 | #include <linux/blkdev.h> |
056075c7 | 37 | #include <linux/module.h> |
bff61975 | 38 | #include <linux/seq_file.h> |
8bda470e | 39 | #include <linux/ratelimit.h> |
109e3765 | 40 | #include <trace/events/block.h> |
43b2e5d8 | 41 | #include "md.h" |
ef740c37 CH |
42 | #include "raid1.h" |
43 | #include "bitmap.h" | |
191ea9b2 | 44 | |
394ed8e4 SL |
45 | #define UNSUPPORTED_MDDEV_FLAGS \ |
46 | ((1L << MD_HAS_JOURNAL) | \ | |
47 | (1L << MD_JOURNAL_CLEAN)) | |
48 | ||
1da177e4 LT |
49 | /* |
50 | * Number of guaranteed r1bios in case of extreme VM load: | |
51 | */ | |
52 | #define NR_RAID1_BIOS 256 | |
53 | ||
473e87ce JB |
54 | /* when we get a read error on a read-only array, we redirect to another |
55 | * device without failing the first device, or trying to over-write to | |
56 | * correct the read error. To keep track of bad blocks on a per-bio | |
57 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
58 | */ | |
59 | #define IO_BLOCKED ((struct bio *)1) | |
60 | /* When we successfully write to a known bad-block, we need to remove the | |
61 | * bad-block marking which must be done from process context. So we record | |
62 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
63 | */ | |
64 | #define IO_MADE_GOOD ((struct bio *)2) | |
65 | ||
66 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
67 | ||
34db0cd6 N |
68 | /* When there are this many requests queue to be written by |
69 | * the raid1 thread, we become 'congested' to provide back-pressure | |
70 | * for writeback. | |
71 | */ | |
72 | static int max_queued_requests = 1024; | |
1da177e4 | 73 | |
79ef3a8a | 74 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
75 | sector_t bi_sector); | |
e8096360 | 76 | static void lower_barrier(struct r1conf *conf); |
1da177e4 | 77 | |
578b54ad N |
78 | #define raid1_log(md, fmt, args...) \ |
79 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0) | |
80 | ||
dd0fc66f | 81 | static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
82 | { |
83 | struct pool_info *pi = data; | |
9f2c9d12 | 84 | int size = offsetof(struct r1bio, bios[pi->raid_disks]); |
1da177e4 LT |
85 | |
86 | /* allocate a r1bio with room for raid_disks entries in the bios array */ | |
7eaceacc | 87 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
88 | } |
89 | ||
90 | static void r1bio_pool_free(void *r1_bio, void *data) | |
91 | { | |
92 | kfree(r1_bio); | |
93 | } | |
94 | ||
95 | #define RESYNC_BLOCK_SIZE (64*1024) | |
8e005f7c | 96 | #define RESYNC_DEPTH 32 |
1da177e4 LT |
97 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
98 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) | |
8e005f7c | 99 | #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH) |
100 | #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9) | |
c40f341f GR |
101 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
102 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
8e005f7c | 103 | #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS) |
1da177e4 | 104 | |
dd0fc66f | 105 | static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 LT |
106 | { |
107 | struct pool_info *pi = data; | |
9f2c9d12 | 108 | struct r1bio *r1_bio; |
1da177e4 | 109 | struct bio *bio; |
da1aab3d | 110 | int need_pages; |
1da177e4 LT |
111 | int i, j; |
112 | ||
113 | r1_bio = r1bio_pool_alloc(gfp_flags, pi); | |
7eaceacc | 114 | if (!r1_bio) |
1da177e4 | 115 | return NULL; |
1da177e4 LT |
116 | |
117 | /* | |
118 | * Allocate bios : 1 for reading, n-1 for writing | |
119 | */ | |
120 | for (j = pi->raid_disks ; j-- ; ) { | |
6746557f | 121 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
122 | if (!bio) |
123 | goto out_free_bio; | |
124 | r1_bio->bios[j] = bio; | |
125 | } | |
126 | /* | |
127 | * Allocate RESYNC_PAGES data pages and attach them to | |
d11c171e N |
128 | * the first bio. |
129 | * If this is a user-requested check/repair, allocate | |
130 | * RESYNC_PAGES for each bio. | |
1da177e4 | 131 | */ |
d11c171e | 132 | if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) |
da1aab3d | 133 | need_pages = pi->raid_disks; |
d11c171e | 134 | else |
da1aab3d N |
135 | need_pages = 1; |
136 | for (j = 0; j < need_pages; j++) { | |
d11c171e | 137 | bio = r1_bio->bios[j]; |
a0787606 | 138 | bio->bi_vcnt = RESYNC_PAGES; |
d11c171e | 139 | |
a0787606 | 140 | if (bio_alloc_pages(bio, gfp_flags)) |
da1aab3d | 141 | goto out_free_pages; |
d11c171e N |
142 | } |
143 | /* If not user-requests, copy the page pointers to all bios */ | |
144 | if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) { | |
145 | for (i=0; i<RESYNC_PAGES ; i++) | |
146 | for (j=1; j<pi->raid_disks; j++) | |
147 | r1_bio->bios[j]->bi_io_vec[i].bv_page = | |
148 | r1_bio->bios[0]->bi_io_vec[i].bv_page; | |
1da177e4 LT |
149 | } |
150 | ||
151 | r1_bio->master_bio = NULL; | |
152 | ||
153 | return r1_bio; | |
154 | ||
da1aab3d | 155 | out_free_pages: |
491221f8 GJ |
156 | while (--j >= 0) |
157 | bio_free_pages(r1_bio->bios[j]); | |
da1aab3d | 158 | |
1da177e4 | 159 | out_free_bio: |
8f19ccb2 | 160 | while (++j < pi->raid_disks) |
1da177e4 LT |
161 | bio_put(r1_bio->bios[j]); |
162 | r1bio_pool_free(r1_bio, data); | |
163 | return NULL; | |
164 | } | |
165 | ||
166 | static void r1buf_pool_free(void *__r1_bio, void *data) | |
167 | { | |
168 | struct pool_info *pi = data; | |
d11c171e | 169 | int i,j; |
9f2c9d12 | 170 | struct r1bio *r1bio = __r1_bio; |
1da177e4 | 171 | |
d11c171e N |
172 | for (i = 0; i < RESYNC_PAGES; i++) |
173 | for (j = pi->raid_disks; j-- ;) { | |
174 | if (j == 0 || | |
175 | r1bio->bios[j]->bi_io_vec[i].bv_page != | |
176 | r1bio->bios[0]->bi_io_vec[i].bv_page) | |
1345b1d8 | 177 | safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page); |
d11c171e | 178 | } |
1da177e4 LT |
179 | for (i=0 ; i < pi->raid_disks; i++) |
180 | bio_put(r1bio->bios[i]); | |
181 | ||
182 | r1bio_pool_free(r1bio, data); | |
183 | } | |
184 | ||
e8096360 | 185 | static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio) |
1da177e4 LT |
186 | { |
187 | int i; | |
188 | ||
8f19ccb2 | 189 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 | 190 | struct bio **bio = r1_bio->bios + i; |
4367af55 | 191 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
192 | bio_put(*bio); |
193 | *bio = NULL; | |
194 | } | |
195 | } | |
196 | ||
9f2c9d12 | 197 | static void free_r1bio(struct r1bio *r1_bio) |
1da177e4 | 198 | { |
e8096360 | 199 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 | 200 | |
1da177e4 LT |
201 | put_all_bios(conf, r1_bio); |
202 | mempool_free(r1_bio, conf->r1bio_pool); | |
203 | } | |
204 | ||
9f2c9d12 | 205 | static void put_buf(struct r1bio *r1_bio) |
1da177e4 | 206 | { |
e8096360 | 207 | struct r1conf *conf = r1_bio->mddev->private; |
3e198f78 N |
208 | int i; |
209 | ||
8f19ccb2 | 210 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3e198f78 N |
211 | struct bio *bio = r1_bio->bios[i]; |
212 | if (bio->bi_end_io) | |
213 | rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); | |
214 | } | |
1da177e4 LT |
215 | |
216 | mempool_free(r1_bio, conf->r1buf_pool); | |
217 | ||
17999be4 | 218 | lower_barrier(conf); |
1da177e4 LT |
219 | } |
220 | ||
9f2c9d12 | 221 | static void reschedule_retry(struct r1bio *r1_bio) |
1da177e4 LT |
222 | { |
223 | unsigned long flags; | |
fd01b88c | 224 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 225 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
226 | |
227 | spin_lock_irqsave(&conf->device_lock, flags); | |
228 | list_add(&r1_bio->retry_list, &conf->retry_list); | |
ddaf22ab | 229 | conf->nr_queued ++; |
1da177e4 LT |
230 | spin_unlock_irqrestore(&conf->device_lock, flags); |
231 | ||
17999be4 | 232 | wake_up(&conf->wait_barrier); |
1da177e4 LT |
233 | md_wakeup_thread(mddev->thread); |
234 | } | |
235 | ||
236 | /* | |
237 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
238 | * operation and are ready to return a success/failure code to the buffer | |
239 | * cache layer. | |
240 | */ | |
9f2c9d12 | 241 | static void call_bio_endio(struct r1bio *r1_bio) |
d2eb35ac N |
242 | { |
243 | struct bio *bio = r1_bio->master_bio; | |
244 | int done; | |
e8096360 | 245 | struct r1conf *conf = r1_bio->mddev->private; |
79ef3a8a | 246 | sector_t start_next_window = r1_bio->start_next_window; |
4f024f37 | 247 | sector_t bi_sector = bio->bi_iter.bi_sector; |
d2eb35ac N |
248 | |
249 | if (bio->bi_phys_segments) { | |
250 | unsigned long flags; | |
251 | spin_lock_irqsave(&conf->device_lock, flags); | |
252 | bio->bi_phys_segments--; | |
253 | done = (bio->bi_phys_segments == 0); | |
254 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
79ef3a8a | 255 | /* |
256 | * make_request() might be waiting for | |
257 | * bi_phys_segments to decrease | |
258 | */ | |
259 | wake_up(&conf->wait_barrier); | |
d2eb35ac N |
260 | } else |
261 | done = 1; | |
262 | ||
263 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) | |
4246a0b6 CH |
264 | bio->bi_error = -EIO; |
265 | ||
d2eb35ac | 266 | if (done) { |
4246a0b6 | 267 | bio_endio(bio); |
d2eb35ac N |
268 | /* |
269 | * Wake up any possible resync thread that waits for the device | |
270 | * to go idle. | |
271 | */ | |
79ef3a8a | 272 | allow_barrier(conf, start_next_window, bi_sector); |
d2eb35ac N |
273 | } |
274 | } | |
275 | ||
9f2c9d12 | 276 | static void raid_end_bio_io(struct r1bio *r1_bio) |
1da177e4 LT |
277 | { |
278 | struct bio *bio = r1_bio->master_bio; | |
279 | ||
4b6d287f N |
280 | /* if nobody has done the final endio yet, do it now */ |
281 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
36a4e1fe N |
282 | pr_debug("raid1: sync end %s on sectors %llu-%llu\n", |
283 | (bio_data_dir(bio) == WRITE) ? "write" : "read", | |
4f024f37 KO |
284 | (unsigned long long) bio->bi_iter.bi_sector, |
285 | (unsigned long long) bio_end_sector(bio) - 1); | |
4b6d287f | 286 | |
d2eb35ac | 287 | call_bio_endio(r1_bio); |
4b6d287f | 288 | } |
1da177e4 LT |
289 | free_r1bio(r1_bio); |
290 | } | |
291 | ||
292 | /* | |
293 | * Update disk head position estimator based on IRQ completion info. | |
294 | */ | |
9f2c9d12 | 295 | static inline void update_head_pos(int disk, struct r1bio *r1_bio) |
1da177e4 | 296 | { |
e8096360 | 297 | struct r1conf *conf = r1_bio->mddev->private; |
1da177e4 LT |
298 | |
299 | conf->mirrors[disk].head_position = | |
300 | r1_bio->sector + (r1_bio->sectors); | |
301 | } | |
302 | ||
ba3ae3be NK |
303 | /* |
304 | * Find the disk number which triggered given bio | |
305 | */ | |
9f2c9d12 | 306 | static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio) |
ba3ae3be NK |
307 | { |
308 | int mirror; | |
30194636 N |
309 | struct r1conf *conf = r1_bio->mddev->private; |
310 | int raid_disks = conf->raid_disks; | |
ba3ae3be | 311 | |
8f19ccb2 | 312 | for (mirror = 0; mirror < raid_disks * 2; mirror++) |
ba3ae3be NK |
313 | if (r1_bio->bios[mirror] == bio) |
314 | break; | |
315 | ||
8f19ccb2 | 316 | BUG_ON(mirror == raid_disks * 2); |
ba3ae3be NK |
317 | update_head_pos(mirror, r1_bio); |
318 | ||
319 | return mirror; | |
320 | } | |
321 | ||
4246a0b6 | 322 | static void raid1_end_read_request(struct bio *bio) |
1da177e4 | 323 | { |
4246a0b6 | 324 | int uptodate = !bio->bi_error; |
9f2c9d12 | 325 | struct r1bio *r1_bio = bio->bi_private; |
e8096360 | 326 | struct r1conf *conf = r1_bio->mddev->private; |
e5872d58 | 327 | struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; |
1da177e4 | 328 | |
1da177e4 LT |
329 | /* |
330 | * this branch is our 'one mirror IO has finished' event handler: | |
331 | */ | |
e5872d58 | 332 | update_head_pos(r1_bio->read_disk, r1_bio); |
ddaf22ab | 333 | |
dd00a99e N |
334 | if (uptodate) |
335 | set_bit(R1BIO_Uptodate, &r1_bio->state); | |
2e52d449 N |
336 | else if (test_bit(FailFast, &rdev->flags) && |
337 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
338 | /* This was a fail-fast read so we definitely | |
339 | * want to retry */ | |
340 | ; | |
dd00a99e N |
341 | else { |
342 | /* If all other devices have failed, we want to return | |
343 | * the error upwards rather than fail the last device. | |
344 | * Here we redefine "uptodate" to mean "Don't want to retry" | |
1da177e4 | 345 | */ |
dd00a99e N |
346 | unsigned long flags; |
347 | spin_lock_irqsave(&conf->device_lock, flags); | |
348 | if (r1_bio->mddev->degraded == conf->raid_disks || | |
349 | (r1_bio->mddev->degraded == conf->raid_disks-1 && | |
e5872d58 | 350 | test_bit(In_sync, &rdev->flags))) |
dd00a99e N |
351 | uptodate = 1; |
352 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
353 | } | |
1da177e4 | 354 | |
7ad4d4a6 | 355 | if (uptodate) { |
1da177e4 | 356 | raid_end_bio_io(r1_bio); |
e5872d58 | 357 | rdev_dec_pending(rdev, conf->mddev); |
7ad4d4a6 | 358 | } else { |
1da177e4 LT |
359 | /* |
360 | * oops, read error: | |
361 | */ | |
362 | char b[BDEVNAME_SIZE]; | |
1d41c216 N |
363 | pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n", |
364 | mdname(conf->mddev), | |
365 | bdevname(rdev->bdev, b), | |
366 | (unsigned long long)r1_bio->sector); | |
d2eb35ac | 367 | set_bit(R1BIO_ReadError, &r1_bio->state); |
1da177e4 | 368 | reschedule_retry(r1_bio); |
7ad4d4a6 | 369 | /* don't drop the reference on read_disk yet */ |
1da177e4 | 370 | } |
1da177e4 LT |
371 | } |
372 | ||
9f2c9d12 | 373 | static void close_write(struct r1bio *r1_bio) |
cd5ff9a1 N |
374 | { |
375 | /* it really is the end of this request */ | |
376 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { | |
377 | /* free extra copy of the data pages */ | |
378 | int i = r1_bio->behind_page_count; | |
379 | while (i--) | |
380 | safe_put_page(r1_bio->behind_bvecs[i].bv_page); | |
381 | kfree(r1_bio->behind_bvecs); | |
382 | r1_bio->behind_bvecs = NULL; | |
383 | } | |
384 | /* clear the bitmap if all writes complete successfully */ | |
385 | bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, | |
386 | r1_bio->sectors, | |
387 | !test_bit(R1BIO_Degraded, &r1_bio->state), | |
388 | test_bit(R1BIO_BehindIO, &r1_bio->state)); | |
389 | md_write_end(r1_bio->mddev); | |
390 | } | |
391 | ||
9f2c9d12 | 392 | static void r1_bio_write_done(struct r1bio *r1_bio) |
4e78064f | 393 | { |
cd5ff9a1 N |
394 | if (!atomic_dec_and_test(&r1_bio->remaining)) |
395 | return; | |
396 | ||
397 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
398 | reschedule_retry(r1_bio); | |
399 | else { | |
400 | close_write(r1_bio); | |
4367af55 N |
401 | if (test_bit(R1BIO_MadeGood, &r1_bio->state)) |
402 | reschedule_retry(r1_bio); | |
403 | else | |
404 | raid_end_bio_io(r1_bio); | |
4e78064f N |
405 | } |
406 | } | |
407 | ||
4246a0b6 | 408 | static void raid1_end_write_request(struct bio *bio) |
1da177e4 | 409 | { |
9f2c9d12 | 410 | struct r1bio *r1_bio = bio->bi_private; |
e5872d58 | 411 | int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); |
e8096360 | 412 | struct r1conf *conf = r1_bio->mddev->private; |
04b857f7 | 413 | struct bio *to_put = NULL; |
e5872d58 N |
414 | int mirror = find_bio_disk(r1_bio, bio); |
415 | struct md_rdev *rdev = conf->mirrors[mirror].rdev; | |
e3f948cd SL |
416 | bool discard_error; |
417 | ||
418 | discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; | |
1da177e4 | 419 | |
e9c7469b TH |
420 | /* |
421 | * 'one mirror IO has finished' event handler: | |
422 | */ | |
e3f948cd | 423 | if (bio->bi_error && !discard_error) { |
e5872d58 N |
424 | set_bit(WriteErrorSeen, &rdev->flags); |
425 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
426 | set_bit(MD_RECOVERY_NEEDED, & |
427 | conf->mddev->recovery); | |
428 | ||
212e7eb7 N |
429 | if (test_bit(FailFast, &rdev->flags) && |
430 | (bio->bi_opf & MD_FAILFAST) && | |
431 | /* We never try FailFast to WriteMostly devices */ | |
432 | !test_bit(WriteMostly, &rdev->flags)) { | |
433 | md_error(r1_bio->mddev, rdev); | |
434 | if (!test_bit(Faulty, &rdev->flags)) | |
435 | /* This is the only remaining device, | |
436 | * We need to retry the write without | |
437 | * FailFast | |
438 | */ | |
439 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
440 | else { | |
441 | /* Finished with this branch */ | |
442 | r1_bio->bios[mirror] = NULL; | |
443 | to_put = bio; | |
444 | } | |
445 | } else | |
446 | set_bit(R1BIO_WriteError, &r1_bio->state); | |
4367af55 | 447 | } else { |
1da177e4 | 448 | /* |
e9c7469b TH |
449 | * Set R1BIO_Uptodate in our master bio, so that we |
450 | * will return a good error code for to the higher | |
451 | * levels even if IO on some other mirrored buffer | |
452 | * fails. | |
453 | * | |
454 | * The 'master' represents the composite IO operation | |
455 | * to user-side. So if something waits for IO, then it | |
456 | * will wait for the 'master' bio. | |
1da177e4 | 457 | */ |
4367af55 N |
458 | sector_t first_bad; |
459 | int bad_sectors; | |
460 | ||
cd5ff9a1 N |
461 | r1_bio->bios[mirror] = NULL; |
462 | to_put = bio; | |
3056e3ae AL |
463 | /* |
464 | * Do not set R1BIO_Uptodate if the current device is | |
465 | * rebuilding or Faulty. This is because we cannot use | |
466 | * such device for properly reading the data back (we could | |
467 | * potentially use it, if the current write would have felt | |
468 | * before rdev->recovery_offset, but for simplicity we don't | |
469 | * check this here. | |
470 | */ | |
e5872d58 N |
471 | if (test_bit(In_sync, &rdev->flags) && |
472 | !test_bit(Faulty, &rdev->flags)) | |
3056e3ae | 473 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
e9c7469b | 474 | |
4367af55 | 475 | /* Maybe we can clear some bad blocks. */ |
e5872d58 | 476 | if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
e3f948cd | 477 | &first_bad, &bad_sectors) && !discard_error) { |
4367af55 N |
478 | r1_bio->bios[mirror] = IO_MADE_GOOD; |
479 | set_bit(R1BIO_MadeGood, &r1_bio->state); | |
480 | } | |
481 | } | |
482 | ||
e9c7469b | 483 | if (behind) { |
e5872d58 | 484 | if (test_bit(WriteMostly, &rdev->flags)) |
e9c7469b TH |
485 | atomic_dec(&r1_bio->behind_remaining); |
486 | ||
487 | /* | |
488 | * In behind mode, we ACK the master bio once the I/O | |
489 | * has safely reached all non-writemostly | |
490 | * disks. Setting the Returned bit ensures that this | |
491 | * gets done only once -- we don't ever want to return | |
492 | * -EIO here, instead we'll wait | |
493 | */ | |
494 | if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && | |
495 | test_bit(R1BIO_Uptodate, &r1_bio->state)) { | |
496 | /* Maybe we can return now */ | |
497 | if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { | |
498 | struct bio *mbio = r1_bio->master_bio; | |
36a4e1fe N |
499 | pr_debug("raid1: behind end write sectors" |
500 | " %llu-%llu\n", | |
4f024f37 KO |
501 | (unsigned long long) mbio->bi_iter.bi_sector, |
502 | (unsigned long long) bio_end_sector(mbio) - 1); | |
d2eb35ac | 503 | call_bio_endio(r1_bio); |
4b6d287f N |
504 | } |
505 | } | |
506 | } | |
4367af55 | 507 | if (r1_bio->bios[mirror] == NULL) |
e5872d58 | 508 | rdev_dec_pending(rdev, conf->mddev); |
e9c7469b | 509 | |
1da177e4 | 510 | /* |
1da177e4 LT |
511 | * Let's see if all mirrored write operations have finished |
512 | * already. | |
513 | */ | |
af6d7b76 | 514 | r1_bio_write_done(r1_bio); |
c70810b3 | 515 | |
04b857f7 N |
516 | if (to_put) |
517 | bio_put(to_put); | |
1da177e4 LT |
518 | } |
519 | ||
1da177e4 LT |
520 | /* |
521 | * This routine returns the disk from which the requested read should | |
522 | * be done. There is a per-array 'next expected sequential IO' sector | |
523 | * number - if this matches on the next IO then we use the last disk. | |
524 | * There is also a per-disk 'last know head position' sector that is | |
525 | * maintained from IRQ contexts, both the normal and the resync IO | |
526 | * completion handlers update this position correctly. If there is no | |
527 | * perfect sequential match then we pick the disk whose head is closest. | |
528 | * | |
529 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
530 | * because position is mirror, not device based. | |
531 | * | |
532 | * The rdev for the device selected will have nr_pending incremented. | |
533 | */ | |
e8096360 | 534 | static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors) |
1da177e4 | 535 | { |
af3a2cd6 | 536 | const sector_t this_sector = r1_bio->sector; |
d2eb35ac N |
537 | int sectors; |
538 | int best_good_sectors; | |
9dedf603 SL |
539 | int best_disk, best_dist_disk, best_pending_disk; |
540 | int has_nonrot_disk; | |
be4d3280 | 541 | int disk; |
76073054 | 542 | sector_t best_dist; |
9dedf603 | 543 | unsigned int min_pending; |
3cb03002 | 544 | struct md_rdev *rdev; |
f3ac8bf7 | 545 | int choose_first; |
12cee5a8 | 546 | int choose_next_idle; |
1da177e4 LT |
547 | |
548 | rcu_read_lock(); | |
549 | /* | |
8ddf9efe | 550 | * Check if we can balance. We can balance on the whole |
1da177e4 LT |
551 | * device if no resync is going on, or below the resync window. |
552 | * We take the first readable disk when above the resync window. | |
553 | */ | |
554 | retry: | |
d2eb35ac | 555 | sectors = r1_bio->sectors; |
76073054 | 556 | best_disk = -1; |
9dedf603 | 557 | best_dist_disk = -1; |
76073054 | 558 | best_dist = MaxSector; |
9dedf603 SL |
559 | best_pending_disk = -1; |
560 | min_pending = UINT_MAX; | |
d2eb35ac | 561 | best_good_sectors = 0; |
9dedf603 | 562 | has_nonrot_disk = 0; |
12cee5a8 | 563 | choose_next_idle = 0; |
2e52d449 | 564 | clear_bit(R1BIO_FailFast, &r1_bio->state); |
d2eb35ac | 565 | |
7d49ffcf GR |
566 | if ((conf->mddev->recovery_cp < this_sector + sectors) || |
567 | (mddev_is_clustered(conf->mddev) && | |
90382ed9 | 568 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, |
7d49ffcf GR |
569 | this_sector + sectors))) |
570 | choose_first = 1; | |
571 | else | |
572 | choose_first = 0; | |
1da177e4 | 573 | |
be4d3280 | 574 | for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { |
76073054 | 575 | sector_t dist; |
d2eb35ac N |
576 | sector_t first_bad; |
577 | int bad_sectors; | |
9dedf603 | 578 | unsigned int pending; |
12cee5a8 | 579 | bool nonrot; |
d2eb35ac | 580 | |
f3ac8bf7 N |
581 | rdev = rcu_dereference(conf->mirrors[disk].rdev); |
582 | if (r1_bio->bios[disk] == IO_BLOCKED | |
583 | || rdev == NULL | |
76073054 | 584 | || test_bit(Faulty, &rdev->flags)) |
f3ac8bf7 | 585 | continue; |
76073054 N |
586 | if (!test_bit(In_sync, &rdev->flags) && |
587 | rdev->recovery_offset < this_sector + sectors) | |
1da177e4 | 588 | continue; |
76073054 N |
589 | if (test_bit(WriteMostly, &rdev->flags)) { |
590 | /* Don't balance among write-mostly, just | |
591 | * use the first as a last resort */ | |
d1901ef0 | 592 | if (best_dist_disk < 0) { |
307729c8 N |
593 | if (is_badblock(rdev, this_sector, sectors, |
594 | &first_bad, &bad_sectors)) { | |
816b0acf | 595 | if (first_bad <= this_sector) |
307729c8 N |
596 | /* Cannot use this */ |
597 | continue; | |
598 | best_good_sectors = first_bad - this_sector; | |
599 | } else | |
600 | best_good_sectors = sectors; | |
d1901ef0 TH |
601 | best_dist_disk = disk; |
602 | best_pending_disk = disk; | |
307729c8 | 603 | } |
76073054 N |
604 | continue; |
605 | } | |
606 | /* This is a reasonable device to use. It might | |
607 | * even be best. | |
608 | */ | |
d2eb35ac N |
609 | if (is_badblock(rdev, this_sector, sectors, |
610 | &first_bad, &bad_sectors)) { | |
611 | if (best_dist < MaxSector) | |
612 | /* already have a better device */ | |
613 | continue; | |
614 | if (first_bad <= this_sector) { | |
615 | /* cannot read here. If this is the 'primary' | |
616 | * device, then we must not read beyond | |
617 | * bad_sectors from another device.. | |
618 | */ | |
619 | bad_sectors -= (this_sector - first_bad); | |
620 | if (choose_first && sectors > bad_sectors) | |
621 | sectors = bad_sectors; | |
622 | if (best_good_sectors > sectors) | |
623 | best_good_sectors = sectors; | |
624 | ||
625 | } else { | |
626 | sector_t good_sectors = first_bad - this_sector; | |
627 | if (good_sectors > best_good_sectors) { | |
628 | best_good_sectors = good_sectors; | |
629 | best_disk = disk; | |
630 | } | |
631 | if (choose_first) | |
632 | break; | |
633 | } | |
634 | continue; | |
635 | } else | |
636 | best_good_sectors = sectors; | |
637 | ||
2e52d449 N |
638 | if (best_disk >= 0) |
639 | /* At least two disks to choose from so failfast is OK */ | |
640 | set_bit(R1BIO_FailFast, &r1_bio->state); | |
641 | ||
12cee5a8 SL |
642 | nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); |
643 | has_nonrot_disk |= nonrot; | |
9dedf603 | 644 | pending = atomic_read(&rdev->nr_pending); |
76073054 | 645 | dist = abs(this_sector - conf->mirrors[disk].head_position); |
12cee5a8 | 646 | if (choose_first) { |
76073054 | 647 | best_disk = disk; |
1da177e4 LT |
648 | break; |
649 | } | |
12cee5a8 SL |
650 | /* Don't change to another disk for sequential reads */ |
651 | if (conf->mirrors[disk].next_seq_sect == this_sector | |
652 | || dist == 0) { | |
653 | int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; | |
654 | struct raid1_info *mirror = &conf->mirrors[disk]; | |
655 | ||
656 | best_disk = disk; | |
657 | /* | |
658 | * If buffered sequential IO size exceeds optimal | |
659 | * iosize, check if there is idle disk. If yes, choose | |
660 | * the idle disk. read_balance could already choose an | |
661 | * idle disk before noticing it's a sequential IO in | |
662 | * this disk. This doesn't matter because this disk | |
663 | * will idle, next time it will be utilized after the | |
664 | * first disk has IO size exceeds optimal iosize. In | |
665 | * this way, iosize of the first disk will be optimal | |
666 | * iosize at least. iosize of the second disk might be | |
667 | * small, but not a big deal since when the second disk | |
668 | * starts IO, the first disk is likely still busy. | |
669 | */ | |
670 | if (nonrot && opt_iosize > 0 && | |
671 | mirror->seq_start != MaxSector && | |
672 | mirror->next_seq_sect > opt_iosize && | |
673 | mirror->next_seq_sect - opt_iosize >= | |
674 | mirror->seq_start) { | |
675 | choose_next_idle = 1; | |
676 | continue; | |
677 | } | |
678 | break; | |
679 | } | |
12cee5a8 SL |
680 | |
681 | if (choose_next_idle) | |
682 | continue; | |
9dedf603 SL |
683 | |
684 | if (min_pending > pending) { | |
685 | min_pending = pending; | |
686 | best_pending_disk = disk; | |
687 | } | |
688 | ||
76073054 N |
689 | if (dist < best_dist) { |
690 | best_dist = dist; | |
9dedf603 | 691 | best_dist_disk = disk; |
1da177e4 | 692 | } |
f3ac8bf7 | 693 | } |
1da177e4 | 694 | |
9dedf603 SL |
695 | /* |
696 | * If all disks are rotational, choose the closest disk. If any disk is | |
697 | * non-rotational, choose the disk with less pending request even the | |
698 | * disk is rotational, which might/might not be optimal for raids with | |
699 | * mixed ratation/non-rotational disks depending on workload. | |
700 | */ | |
701 | if (best_disk == -1) { | |
2e52d449 | 702 | if (has_nonrot_disk || min_pending == 0) |
9dedf603 SL |
703 | best_disk = best_pending_disk; |
704 | else | |
705 | best_disk = best_dist_disk; | |
706 | } | |
707 | ||
76073054 N |
708 | if (best_disk >= 0) { |
709 | rdev = rcu_dereference(conf->mirrors[best_disk].rdev); | |
8ddf9efe N |
710 | if (!rdev) |
711 | goto retry; | |
712 | atomic_inc(&rdev->nr_pending); | |
d2eb35ac | 713 | sectors = best_good_sectors; |
12cee5a8 SL |
714 | |
715 | if (conf->mirrors[best_disk].next_seq_sect != this_sector) | |
716 | conf->mirrors[best_disk].seq_start = this_sector; | |
717 | ||
be4d3280 | 718 | conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; |
1da177e4 LT |
719 | } |
720 | rcu_read_unlock(); | |
d2eb35ac | 721 | *max_sectors = sectors; |
1da177e4 | 722 | |
76073054 | 723 | return best_disk; |
1da177e4 LT |
724 | } |
725 | ||
5c675f83 | 726 | static int raid1_congested(struct mddev *mddev, int bits) |
0d129228 | 727 | { |
e8096360 | 728 | struct r1conf *conf = mddev->private; |
0d129228 N |
729 | int i, ret = 0; |
730 | ||
4452226e | 731 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
732 | conf->pending_count >= max_queued_requests) |
733 | return 1; | |
734 | ||
0d129228 | 735 | rcu_read_lock(); |
f53e29fc | 736 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 737 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 738 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 739 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 740 | |
1ed7242e JB |
741 | BUG_ON(!q); |
742 | ||
0d129228 N |
743 | /* Note the '|| 1' - when read_balance prefers |
744 | * non-congested targets, it can be removed | |
745 | */ | |
4452226e | 746 | if ((bits & (1 << WB_async_congested)) || 1) |
0d129228 N |
747 | ret |= bdi_congested(&q->backing_dev_info, bits); |
748 | else | |
749 | ret &= bdi_congested(&q->backing_dev_info, bits); | |
750 | } | |
751 | } | |
752 | rcu_read_unlock(); | |
753 | return ret; | |
754 | } | |
0d129228 | 755 | |
e8096360 | 756 | static void flush_pending_writes(struct r1conf *conf) |
a35e63ef N |
757 | { |
758 | /* Any writes that have been queued but are awaiting | |
759 | * bitmap updates get flushed here. | |
a35e63ef | 760 | */ |
a35e63ef N |
761 | spin_lock_irq(&conf->device_lock); |
762 | ||
763 | if (conf->pending_bio_list.head) { | |
764 | struct bio *bio; | |
765 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 766 | conf->pending_count = 0; |
a35e63ef N |
767 | spin_unlock_irq(&conf->device_lock); |
768 | /* flush any pending bitmap writes to | |
769 | * disk before proceeding w/ I/O */ | |
770 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 771 | wake_up(&conf->wait_barrier); |
a35e63ef N |
772 | |
773 | while (bio) { /* submit pending writes */ | |
774 | struct bio *next = bio->bi_next; | |
5e2c7a36 | 775 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 776 | bio->bi_next = NULL; |
5e2c7a36 N |
777 | bio->bi_bdev = rdev->bdev; |
778 | if (test_bit(Faulty, &rdev->flags)) { | |
779 | bio->bi_error = -EIO; | |
780 | bio_endio(bio); | |
781 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
782 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
2ff8cc2c | 783 | /* Just ignore it */ |
4246a0b6 | 784 | bio_endio(bio); |
2ff8cc2c SL |
785 | else |
786 | generic_make_request(bio); | |
a35e63ef N |
787 | bio = next; |
788 | } | |
a35e63ef N |
789 | } else |
790 | spin_unlock_irq(&conf->device_lock); | |
7eaceacc JA |
791 | } |
792 | ||
17999be4 N |
793 | /* Barriers.... |
794 | * Sometimes we need to suspend IO while we do something else, | |
795 | * either some resync/recovery, or reconfigure the array. | |
796 | * To do this we raise a 'barrier'. | |
797 | * The 'barrier' is a counter that can be raised multiple times | |
798 | * to count how many activities are happening which preclude | |
799 | * normal IO. | |
800 | * We can only raise the barrier if there is no pending IO. | |
801 | * i.e. if nr_pending == 0. | |
802 | * We choose only to raise the barrier if no-one is waiting for the | |
803 | * barrier to go down. This means that as soon as an IO request | |
804 | * is ready, no other operations which require a barrier will start | |
805 | * until the IO request has had a chance. | |
806 | * | |
807 | * So: regular IO calls 'wait_barrier'. When that returns there | |
808 | * is no backgroup IO happening, It must arrange to call | |
809 | * allow_barrier when it has finished its IO. | |
810 | * backgroup IO calls must call raise_barrier. Once that returns | |
811 | * there is no normal IO happeing. It must arrange to call | |
812 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 813 | */ |
c2fd4c94 | 814 | static void raise_barrier(struct r1conf *conf, sector_t sector_nr) |
1da177e4 LT |
815 | { |
816 | spin_lock_irq(&conf->resync_lock); | |
17999be4 N |
817 | |
818 | /* Wait until no block IO is waiting */ | |
819 | wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting, | |
eed8c02e | 820 | conf->resync_lock); |
17999be4 N |
821 | |
822 | /* block any new IO from starting */ | |
823 | conf->barrier++; | |
c2fd4c94 | 824 | conf->next_resync = sector_nr; |
17999be4 | 825 | |
79ef3a8a | 826 | /* For these conditions we must wait: |
827 | * A: while the array is in frozen state | |
828 | * B: while barrier >= RESYNC_DEPTH, meaning resync reach | |
829 | * the max count which allowed. | |
830 | * C: next_resync + RESYNC_SECTORS > start_next_window, meaning | |
831 | * next resync will reach to the window which normal bios are | |
832 | * handling. | |
2f73d3c5 | 833 | * D: while there are any active requests in the current window. |
79ef3a8a | 834 | */ |
17999be4 | 835 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 836 | !conf->array_frozen && |
79ef3a8a | 837 | conf->barrier < RESYNC_DEPTH && |
2f73d3c5 | 838 | conf->current_window_requests == 0 && |
79ef3a8a | 839 | (conf->start_next_window >= |
840 | conf->next_resync + RESYNC_SECTORS), | |
eed8c02e | 841 | conf->resync_lock); |
17999be4 | 842 | |
34e97f17 | 843 | conf->nr_pending++; |
17999be4 N |
844 | spin_unlock_irq(&conf->resync_lock); |
845 | } | |
846 | ||
e8096360 | 847 | static void lower_barrier(struct r1conf *conf) |
17999be4 N |
848 | { |
849 | unsigned long flags; | |
709ae487 | 850 | BUG_ON(conf->barrier <= 0); |
17999be4 N |
851 | spin_lock_irqsave(&conf->resync_lock, flags); |
852 | conf->barrier--; | |
34e97f17 | 853 | conf->nr_pending--; |
17999be4 N |
854 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
855 | wake_up(&conf->wait_barrier); | |
856 | } | |
857 | ||
79ef3a8a | 858 | static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio) |
17999be4 | 859 | { |
79ef3a8a | 860 | bool wait = false; |
861 | ||
862 | if (conf->array_frozen || !bio) | |
863 | wait = true; | |
864 | else if (conf->barrier && bio_data_dir(bio) == WRITE) { | |
23554960 N |
865 | if ((conf->mddev->curr_resync_completed |
866 | >= bio_end_sector(bio)) || | |
f2c771a6 | 867 | (conf->start_next_window + NEXT_NORMALIO_DISTANCE |
23554960 | 868 | <= bio->bi_iter.bi_sector)) |
79ef3a8a | 869 | wait = false; |
870 | else | |
871 | wait = true; | |
872 | } | |
873 | ||
874 | return wait; | |
875 | } | |
876 | ||
877 | static sector_t wait_barrier(struct r1conf *conf, struct bio *bio) | |
878 | { | |
879 | sector_t sector = 0; | |
880 | ||
17999be4 | 881 | spin_lock_irq(&conf->resync_lock); |
79ef3a8a | 882 | if (need_to_wait_for_sync(conf, bio)) { |
17999be4 | 883 | conf->nr_waiting++; |
d6b42dcb N |
884 | /* Wait for the barrier to drop. |
885 | * However if there are already pending | |
886 | * requests (preventing the barrier from | |
887 | * rising completely), and the | |
5965b642 | 888 | * per-process bio queue isn't empty, |
d6b42dcb | 889 | * then don't wait, as we need to empty |
5965b642 N |
890 | * that queue to allow conf->start_next_window |
891 | * to increase. | |
d6b42dcb | 892 | */ |
578b54ad | 893 | raid1_log(conf->mddev, "wait barrier"); |
d6b42dcb | 894 | wait_event_lock_irq(conf->wait_barrier, |
b364e3d0 | 895 | !conf->array_frozen && |
896 | (!conf->barrier || | |
5965b642 N |
897 | ((conf->start_next_window < |
898 | conf->next_resync + RESYNC_SECTORS) && | |
899 | current->bio_list && | |
900 | !bio_list_empty(current->bio_list))), | |
eed8c02e | 901 | conf->resync_lock); |
17999be4 | 902 | conf->nr_waiting--; |
1da177e4 | 903 | } |
79ef3a8a | 904 | |
905 | if (bio && bio_data_dir(bio) == WRITE) { | |
e8ff8bf0 | 906 | if (bio->bi_iter.bi_sector >= conf->next_resync) { |
79ef3a8a | 907 | if (conf->start_next_window == MaxSector) |
908 | conf->start_next_window = | |
909 | conf->next_resync + | |
910 | NEXT_NORMALIO_DISTANCE; | |
911 | ||
912 | if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE) | |
4f024f37 | 913 | <= bio->bi_iter.bi_sector) |
79ef3a8a | 914 | conf->next_window_requests++; |
915 | else | |
916 | conf->current_window_requests++; | |
79ef3a8a | 917 | sector = conf->start_next_window; |
41a336e0 | 918 | } |
79ef3a8a | 919 | } |
920 | ||
17999be4 | 921 | conf->nr_pending++; |
1da177e4 | 922 | spin_unlock_irq(&conf->resync_lock); |
79ef3a8a | 923 | return sector; |
1da177e4 LT |
924 | } |
925 | ||
79ef3a8a | 926 | static void allow_barrier(struct r1conf *conf, sector_t start_next_window, |
927 | sector_t bi_sector) | |
17999be4 N |
928 | { |
929 | unsigned long flags; | |
79ef3a8a | 930 | |
17999be4 N |
931 | spin_lock_irqsave(&conf->resync_lock, flags); |
932 | conf->nr_pending--; | |
79ef3a8a | 933 | if (start_next_window) { |
934 | if (start_next_window == conf->start_next_window) { | |
935 | if (conf->start_next_window + NEXT_NORMALIO_DISTANCE | |
936 | <= bi_sector) | |
937 | conf->next_window_requests--; | |
938 | else | |
939 | conf->current_window_requests--; | |
940 | } else | |
941 | conf->current_window_requests--; | |
942 | ||
943 | if (!conf->current_window_requests) { | |
944 | if (conf->next_window_requests) { | |
945 | conf->current_window_requests = | |
946 | conf->next_window_requests; | |
947 | conf->next_window_requests = 0; | |
948 | conf->start_next_window += | |
949 | NEXT_NORMALIO_DISTANCE; | |
950 | } else | |
951 | conf->start_next_window = MaxSector; | |
952 | } | |
953 | } | |
17999be4 N |
954 | spin_unlock_irqrestore(&conf->resync_lock, flags); |
955 | wake_up(&conf->wait_barrier); | |
956 | } | |
957 | ||
e2d59925 | 958 | static void freeze_array(struct r1conf *conf, int extra) |
ddaf22ab N |
959 | { |
960 | /* stop syncio and normal IO and wait for everything to | |
961 | * go quite. | |
b364e3d0 | 962 | * We wait until nr_pending match nr_queued+extra |
1c830532 N |
963 | * This is called in the context of one normal IO request |
964 | * that has failed. Thus any sync request that might be pending | |
965 | * will be blocked by nr_pending, and we need to wait for | |
966 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 967 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
968 | * must match the number of pending IOs (nr_pending) before |
969 | * we continue. | |
ddaf22ab N |
970 | */ |
971 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 972 | conf->array_frozen = 1; |
578b54ad | 973 | raid1_log(conf->mddev, "wait freeze"); |
eed8c02e | 974 | wait_event_lock_irq_cmd(conf->wait_barrier, |
e2d59925 | 975 | conf->nr_pending == conf->nr_queued+extra, |
eed8c02e LC |
976 | conf->resync_lock, |
977 | flush_pending_writes(conf)); | |
ddaf22ab N |
978 | spin_unlock_irq(&conf->resync_lock); |
979 | } | |
e8096360 | 980 | static void unfreeze_array(struct r1conf *conf) |
ddaf22ab N |
981 | { |
982 | /* reverse the effect of the freeze */ | |
983 | spin_lock_irq(&conf->resync_lock); | |
b364e3d0 | 984 | conf->array_frozen = 0; |
ddaf22ab N |
985 | wake_up(&conf->wait_barrier); |
986 | spin_unlock_irq(&conf->resync_lock); | |
987 | } | |
988 | ||
f72ffdd6 | 989 | /* duplicate the data pages for behind I/O |
4e78064f | 990 | */ |
9f2c9d12 | 991 | static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio) |
4b6d287f N |
992 | { |
993 | int i; | |
994 | struct bio_vec *bvec; | |
2ca68f5e | 995 | struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec), |
4b6d287f | 996 | GFP_NOIO); |
2ca68f5e | 997 | if (unlikely(!bvecs)) |
af6d7b76 | 998 | return; |
4b6d287f | 999 | |
cb34e057 | 1000 | bio_for_each_segment_all(bvec, bio, i) { |
2ca68f5e N |
1001 | bvecs[i] = *bvec; |
1002 | bvecs[i].bv_page = alloc_page(GFP_NOIO); | |
1003 | if (unlikely(!bvecs[i].bv_page)) | |
4b6d287f | 1004 | goto do_sync_io; |
2ca68f5e N |
1005 | memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset, |
1006 | kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len); | |
1007 | kunmap(bvecs[i].bv_page); | |
4b6d287f N |
1008 | kunmap(bvec->bv_page); |
1009 | } | |
2ca68f5e | 1010 | r1_bio->behind_bvecs = bvecs; |
af6d7b76 N |
1011 | r1_bio->behind_page_count = bio->bi_vcnt; |
1012 | set_bit(R1BIO_BehindIO, &r1_bio->state); | |
1013 | return; | |
4b6d287f N |
1014 | |
1015 | do_sync_io: | |
af6d7b76 | 1016 | for (i = 0; i < bio->bi_vcnt; i++) |
2ca68f5e N |
1017 | if (bvecs[i].bv_page) |
1018 | put_page(bvecs[i].bv_page); | |
1019 | kfree(bvecs); | |
4f024f37 KO |
1020 | pr_debug("%dB behind alloc failed, doing sync I/O\n", |
1021 | bio->bi_iter.bi_size); | |
4b6d287f N |
1022 | } |
1023 | ||
f54a9d0e N |
1024 | struct raid1_plug_cb { |
1025 | struct blk_plug_cb cb; | |
1026 | struct bio_list pending; | |
1027 | int pending_cnt; | |
1028 | }; | |
1029 | ||
1030 | static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1031 | { | |
1032 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, | |
1033 | cb); | |
1034 | struct mddev *mddev = plug->cb.data; | |
1035 | struct r1conf *conf = mddev->private; | |
1036 | struct bio *bio; | |
1037 | ||
874807a8 | 1038 | if (from_schedule || current->bio_list) { |
f54a9d0e N |
1039 | spin_lock_irq(&conf->device_lock); |
1040 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1041 | conf->pending_count += plug->pending_cnt; | |
1042 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1043 | wake_up(&conf->wait_barrier); |
f54a9d0e N |
1044 | md_wakeup_thread(mddev->thread); |
1045 | kfree(plug); | |
1046 | return; | |
1047 | } | |
1048 | ||
1049 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1050 | bio = bio_list_get(&plug->pending); | |
1051 | bitmap_unplug(mddev->bitmap); | |
1052 | wake_up(&conf->wait_barrier); | |
1053 | ||
1054 | while (bio) { /* submit pending writes */ | |
1055 | struct bio *next = bio->bi_next; | |
5e2c7a36 | 1056 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
f54a9d0e | 1057 | bio->bi_next = NULL; |
5e2c7a36 N |
1058 | bio->bi_bdev = rdev->bdev; |
1059 | if (test_bit(Faulty, &rdev->flags)) { | |
1060 | bio->bi_error = -EIO; | |
1061 | bio_endio(bio); | |
1062 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
1063 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
32f9f570 | 1064 | /* Just ignore it */ |
4246a0b6 | 1065 | bio_endio(bio); |
32f9f570 SL |
1066 | else |
1067 | generic_make_request(bio); | |
f54a9d0e N |
1068 | bio = next; |
1069 | } | |
1070 | kfree(plug); | |
1071 | } | |
1072 | ||
3b046a97 RL |
1073 | static void raid1_read_request(struct mddev *mddev, struct bio *bio, |
1074 | struct r1bio *r1_bio) | |
1da177e4 | 1075 | { |
e8096360 | 1076 | struct r1conf *conf = mddev->private; |
0eaf822c | 1077 | struct raid1_info *mirror; |
1da177e4 | 1078 | struct bio *read_bio; |
3b046a97 RL |
1079 | struct bitmap *bitmap = mddev->bitmap; |
1080 | const int op = bio_op(bio); | |
1081 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
1082 | int sectors_handled; | |
1083 | int max_sectors; | |
1084 | int rdisk; | |
1085 | ||
1086 | wait_barrier(conf, bio); | |
1087 | ||
1088 | read_again: | |
1089 | rdisk = read_balance(conf, r1_bio, &max_sectors); | |
1090 | ||
1091 | if (rdisk < 0) { | |
1092 | /* couldn't find anywhere to read from */ | |
1093 | raid_end_bio_io(r1_bio); | |
1094 | return; | |
1095 | } | |
1096 | mirror = conf->mirrors + rdisk; | |
1097 | ||
1098 | if (test_bit(WriteMostly, &mirror->rdev->flags) && | |
1099 | bitmap) { | |
1100 | /* | |
1101 | * Reading from a write-mostly device must take care not to | |
1102 | * over-take any writes that are 'behind' | |
1103 | */ | |
1104 | raid1_log(mddev, "wait behind writes"); | |
1105 | wait_event(bitmap->behind_wait, | |
1106 | atomic_read(&bitmap->behind_writes) == 0); | |
1107 | } | |
1108 | r1_bio->read_disk = rdisk; | |
1109 | r1_bio->start_next_window = 0; | |
1110 | ||
1111 | read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev); | |
1112 | bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector, | |
1113 | max_sectors); | |
1114 | ||
1115 | r1_bio->bios[rdisk] = read_bio; | |
1116 | ||
1117 | read_bio->bi_iter.bi_sector = r1_bio->sector + | |
1118 | mirror->rdev->data_offset; | |
1119 | read_bio->bi_bdev = mirror->rdev->bdev; | |
1120 | read_bio->bi_end_io = raid1_end_read_request; | |
1121 | bio_set_op_attrs(read_bio, op, do_sync); | |
1122 | if (test_bit(FailFast, &mirror->rdev->flags) && | |
1123 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
1124 | read_bio->bi_opf |= MD_FAILFAST; | |
1125 | read_bio->bi_private = r1_bio; | |
1126 | ||
1127 | if (mddev->gendisk) | |
1128 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1129 | read_bio, disk_devt(mddev->gendisk), | |
1130 | r1_bio->sector); | |
1131 | ||
1132 | if (max_sectors < r1_bio->sectors) { | |
1133 | /* | |
1134 | * could not read all from this device, so we will need another | |
1135 | * r1_bio. | |
1136 | */ | |
1137 | sectors_handled = (r1_bio->sector + max_sectors | |
1138 | - bio->bi_iter.bi_sector); | |
1139 | r1_bio->sectors = max_sectors; | |
1140 | spin_lock_irq(&conf->device_lock); | |
1141 | if (bio->bi_phys_segments == 0) | |
1142 | bio->bi_phys_segments = 2; | |
1143 | else | |
1144 | bio->bi_phys_segments++; | |
1145 | spin_unlock_irq(&conf->device_lock); | |
1146 | ||
1147 | /* | |
1148 | * Cannot call generic_make_request directly as that will be | |
1149 | * queued in __make_request and subsequent mempool_alloc might | |
1150 | * block waiting for it. So hand bio over to raid1d. | |
1151 | */ | |
1152 | reschedule_retry(r1_bio); | |
1153 | ||
1154 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1155 | ||
1156 | r1_bio->master_bio = bio; | |
1157 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; | |
1158 | r1_bio->state = 0; | |
1159 | r1_bio->mddev = mddev; | |
1160 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; | |
1161 | goto read_again; | |
1162 | } else | |
1163 | generic_make_request(read_bio); | |
1164 | } | |
1165 | ||
1166 | static void raid1_write_request(struct mddev *mddev, struct bio *bio, | |
1167 | struct r1bio *r1_bio) | |
1168 | { | |
1169 | struct r1conf *conf = mddev->private; | |
1f68f0c4 | 1170 | int i, disks; |
3b046a97 | 1171 | struct bitmap *bitmap = mddev->bitmap; |
191ea9b2 | 1172 | unsigned long flags; |
796a5cf0 | 1173 | const int op = bio_op(bio); |
1eff9d32 JA |
1174 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1175 | const unsigned long do_flush_fua = (bio->bi_opf & | |
28a8f0d3 | 1176 | (REQ_PREFLUSH | REQ_FUA)); |
3cb03002 | 1177 | struct md_rdev *blocked_rdev; |
f54a9d0e N |
1178 | struct blk_plug_cb *cb; |
1179 | struct raid1_plug_cb *plug = NULL; | |
1f68f0c4 N |
1180 | int first_clone; |
1181 | int sectors_handled; | |
1182 | int max_sectors; | |
79ef3a8a | 1183 | sector_t start_next_window; |
191ea9b2 | 1184 | |
1da177e4 LT |
1185 | /* |
1186 | * Register the new request and wait if the reconstruction | |
1187 | * thread has put up a bar for new requests. | |
1188 | * Continue immediately if no resync is active currently. | |
1189 | */ | |
62de608d | 1190 | |
3d310eb7 N |
1191 | md_write_start(mddev, bio); /* wait on superblock update early */ |
1192 | ||
3b046a97 | 1193 | if ((bio_end_sector(bio) > mddev->suspend_lo && |
589a1c49 GR |
1194 | bio->bi_iter.bi_sector < mddev->suspend_hi) || |
1195 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1196 | md_cluster_ops->area_resyncing(mddev, WRITE, |
3b046a97 RL |
1197 | bio->bi_iter.bi_sector, bio_end_sector(bio)))) { |
1198 | ||
1199 | /* | |
1200 | * As the suspend_* range is controlled by userspace, we want | |
1201 | * an interruptible wait. | |
6eef4b21 N |
1202 | */ |
1203 | DEFINE_WAIT(w); | |
1204 | for (;;) { | |
1205 | flush_signals(current); | |
1206 | prepare_to_wait(&conf->wait_barrier, | |
1207 | &w, TASK_INTERRUPTIBLE); | |
f73a1c7d | 1208 | if (bio_end_sector(bio) <= mddev->suspend_lo || |
589a1c49 GR |
1209 | bio->bi_iter.bi_sector >= mddev->suspend_hi || |
1210 | (mddev_is_clustered(mddev) && | |
90382ed9 | 1211 | !md_cluster_ops->area_resyncing(mddev, WRITE, |
3b046a97 RL |
1212 | bio->bi_iter.bi_sector, |
1213 | bio_end_sector(bio)))) | |
6eef4b21 N |
1214 | break; |
1215 | schedule(); | |
1216 | } | |
1217 | finish_wait(&conf->wait_barrier, &w); | |
1218 | } | |
79ef3a8a | 1219 | start_next_window = wait_barrier(conf, bio); |
1da177e4 | 1220 | |
34db0cd6 N |
1221 | if (conf->pending_count >= max_queued_requests) { |
1222 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1223 | raid1_log(mddev, "wait queued"); |
34db0cd6 N |
1224 | wait_event(conf->wait_barrier, |
1225 | conf->pending_count < max_queued_requests); | |
1226 | } | |
1f68f0c4 | 1227 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1228 | * inc refcount on their rdev. Record them by setting |
1229 | * bios[x] to bio | |
1f68f0c4 N |
1230 | * If there are known/acknowledged bad blocks on any device on |
1231 | * which we have seen a write error, we want to avoid writing those | |
1232 | * blocks. | |
1233 | * This potentially requires several writes to write around | |
1234 | * the bad blocks. Each set of writes gets it's own r1bio | |
1235 | * with a set of bios attached. | |
1da177e4 | 1236 | */ |
c3b328ac | 1237 | |
8f19ccb2 | 1238 | disks = conf->raid_disks * 2; |
6bfe0b49 | 1239 | retry_write: |
79ef3a8a | 1240 | r1_bio->start_next_window = start_next_window; |
6bfe0b49 | 1241 | blocked_rdev = NULL; |
1da177e4 | 1242 | rcu_read_lock(); |
1f68f0c4 | 1243 | max_sectors = r1_bio->sectors; |
1da177e4 | 1244 | for (i = 0; i < disks; i++) { |
3cb03002 | 1245 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
6bfe0b49 DW |
1246 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1247 | atomic_inc(&rdev->nr_pending); | |
1248 | blocked_rdev = rdev; | |
1249 | break; | |
1250 | } | |
1f68f0c4 | 1251 | r1_bio->bios[i] = NULL; |
8ae12666 | 1252 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
1253 | if (i < conf->raid_disks) |
1254 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
1f68f0c4 N |
1255 | continue; |
1256 | } | |
1257 | ||
1258 | atomic_inc(&rdev->nr_pending); | |
1259 | if (test_bit(WriteErrorSeen, &rdev->flags)) { | |
1260 | sector_t first_bad; | |
1261 | int bad_sectors; | |
1262 | int is_bad; | |
1263 | ||
3b046a97 | 1264 | is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, |
1f68f0c4 N |
1265 | &first_bad, &bad_sectors); |
1266 | if (is_bad < 0) { | |
1267 | /* mustn't write here until the bad block is | |
1268 | * acknowledged*/ | |
1269 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1270 | blocked_rdev = rdev; | |
1271 | break; | |
1272 | } | |
1273 | if (is_bad && first_bad <= r1_bio->sector) { | |
1274 | /* Cannot write here at all */ | |
1275 | bad_sectors -= (r1_bio->sector - first_bad); | |
1276 | if (bad_sectors < max_sectors) | |
1277 | /* mustn't write more than bad_sectors | |
1278 | * to other devices yet | |
1279 | */ | |
1280 | max_sectors = bad_sectors; | |
03c902e1 | 1281 | rdev_dec_pending(rdev, mddev); |
1f68f0c4 N |
1282 | /* We don't set R1BIO_Degraded as that |
1283 | * only applies if the disk is | |
1284 | * missing, so it might be re-added, | |
1285 | * and we want to know to recover this | |
1286 | * chunk. | |
1287 | * In this case the device is here, | |
1288 | * and the fact that this chunk is not | |
1289 | * in-sync is recorded in the bad | |
1290 | * block log | |
1291 | */ | |
1292 | continue; | |
964147d5 | 1293 | } |
1f68f0c4 N |
1294 | if (is_bad) { |
1295 | int good_sectors = first_bad - r1_bio->sector; | |
1296 | if (good_sectors < max_sectors) | |
1297 | max_sectors = good_sectors; | |
1298 | } | |
1299 | } | |
1300 | r1_bio->bios[i] = bio; | |
1da177e4 LT |
1301 | } |
1302 | rcu_read_unlock(); | |
1303 | ||
6bfe0b49 DW |
1304 | if (unlikely(blocked_rdev)) { |
1305 | /* Wait for this device to become unblocked */ | |
1306 | int j; | |
79ef3a8a | 1307 | sector_t old = start_next_window; |
6bfe0b49 DW |
1308 | |
1309 | for (j = 0; j < i; j++) | |
1310 | if (r1_bio->bios[j]) | |
1311 | rdev_dec_pending(conf->mirrors[j].rdev, mddev); | |
1f68f0c4 | 1312 | r1_bio->state = 0; |
4f024f37 | 1313 | allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector); |
578b54ad | 1314 | raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 | 1315 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
79ef3a8a | 1316 | start_next_window = wait_barrier(conf, bio); |
1317 | /* | |
1318 | * We must make sure the multi r1bios of bio have | |
1319 | * the same value of bi_phys_segments | |
1320 | */ | |
1321 | if (bio->bi_phys_segments && old && | |
1322 | old != start_next_window) | |
1323 | /* Wait for the former r1bio(s) to complete */ | |
1324 | wait_event(conf->wait_barrier, | |
1325 | bio->bi_phys_segments == 1); | |
6bfe0b49 DW |
1326 | goto retry_write; |
1327 | } | |
1328 | ||
1f68f0c4 N |
1329 | if (max_sectors < r1_bio->sectors) { |
1330 | /* We are splitting this write into multiple parts, so | |
1331 | * we need to prepare for allocating another r1_bio. | |
1332 | */ | |
1333 | r1_bio->sectors = max_sectors; | |
1334 | spin_lock_irq(&conf->device_lock); | |
1335 | if (bio->bi_phys_segments == 0) | |
1336 | bio->bi_phys_segments = 2; | |
1337 | else | |
1338 | bio->bi_phys_segments++; | |
1339 | spin_unlock_irq(&conf->device_lock); | |
191ea9b2 | 1340 | } |
4f024f37 | 1341 | sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector; |
4b6d287f | 1342 | |
4e78064f | 1343 | atomic_set(&r1_bio->remaining, 1); |
4b6d287f | 1344 | atomic_set(&r1_bio->behind_remaining, 0); |
06d91a5f | 1345 | |
1f68f0c4 | 1346 | first_clone = 1; |
1da177e4 LT |
1347 | for (i = 0; i < disks; i++) { |
1348 | struct bio *mbio; | |
1349 | if (!r1_bio->bios[i]) | |
1350 | continue; | |
1351 | ||
a167f663 | 1352 | mbio = bio_clone_mddev(bio, GFP_NOIO, mddev); |
3b046a97 RL |
1353 | bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, |
1354 | max_sectors); | |
1f68f0c4 N |
1355 | |
1356 | if (first_clone) { | |
1357 | /* do behind I/O ? | |
1358 | * Not if there are too many, or cannot | |
1359 | * allocate memory, or a reader on WriteMostly | |
1360 | * is waiting for behind writes to flush */ | |
1361 | if (bitmap && | |
1362 | (atomic_read(&bitmap->behind_writes) | |
1363 | < mddev->bitmap_info.max_write_behind) && | |
1364 | !waitqueue_active(&bitmap->behind_wait)) | |
1365 | alloc_behind_pages(mbio, r1_bio); | |
1366 | ||
1367 | bitmap_startwrite(bitmap, r1_bio->sector, | |
1368 | r1_bio->sectors, | |
1369 | test_bit(R1BIO_BehindIO, | |
1370 | &r1_bio->state)); | |
1371 | first_clone = 0; | |
1372 | } | |
2ca68f5e | 1373 | if (r1_bio->behind_bvecs) { |
4b6d287f N |
1374 | struct bio_vec *bvec; |
1375 | int j; | |
1376 | ||
cb34e057 KO |
1377 | /* |
1378 | * We trimmed the bio, so _all is legit | |
4b6d287f | 1379 | */ |
d74c6d51 | 1380 | bio_for_each_segment_all(bvec, mbio, j) |
2ca68f5e | 1381 | bvec->bv_page = r1_bio->behind_bvecs[j].bv_page; |
4b6d287f N |
1382 | if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags)) |
1383 | atomic_inc(&r1_bio->behind_remaining); | |
1384 | } | |
1385 | ||
1f68f0c4 N |
1386 | r1_bio->bios[i] = mbio; |
1387 | ||
4f024f37 | 1388 | mbio->bi_iter.bi_sector = (r1_bio->sector + |
1f68f0c4 | 1389 | conf->mirrors[i].rdev->data_offset); |
109e3765 | 1390 | mbio->bi_bdev = conf->mirrors[i].rdev->bdev; |
1f68f0c4 | 1391 | mbio->bi_end_io = raid1_end_write_request; |
288dab8a | 1392 | bio_set_op_attrs(mbio, op, do_flush_fua | do_sync); |
212e7eb7 N |
1393 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && |
1394 | !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && | |
1395 | conf->raid_disks - mddev->degraded > 1) | |
1396 | mbio->bi_opf |= MD_FAILFAST; | |
1f68f0c4 N |
1397 | mbio->bi_private = r1_bio; |
1398 | ||
1da177e4 | 1399 | atomic_inc(&r1_bio->remaining); |
f54a9d0e | 1400 | |
109e3765 N |
1401 | if (mddev->gendisk) |
1402 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1403 | mbio, disk_devt(mddev->gendisk), | |
1404 | r1_bio->sector); | |
1405 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1406 | mbio->bi_bdev = (void*)conf->mirrors[i].rdev; | |
1407 | ||
f54a9d0e N |
1408 | cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug)); |
1409 | if (cb) | |
1410 | plug = container_of(cb, struct raid1_plug_cb, cb); | |
1411 | else | |
1412 | plug = NULL; | |
4e78064f | 1413 | spin_lock_irqsave(&conf->device_lock, flags); |
f54a9d0e N |
1414 | if (plug) { |
1415 | bio_list_add(&plug->pending, mbio); | |
1416 | plug->pending_cnt++; | |
1417 | } else { | |
1418 | bio_list_add(&conf->pending_bio_list, mbio); | |
1419 | conf->pending_count++; | |
1420 | } | |
4e78064f | 1421 | spin_unlock_irqrestore(&conf->device_lock, flags); |
f54a9d0e | 1422 | if (!plug) |
b357f04a | 1423 | md_wakeup_thread(mddev->thread); |
1da177e4 | 1424 | } |
079fa166 N |
1425 | /* Mustn't call r1_bio_write_done before this next test, |
1426 | * as it could result in the bio being freed. | |
1427 | */ | |
aa8b57aa | 1428 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1429 | r1_bio_write_done(r1_bio); |
1f68f0c4 N |
1430 | /* We need another r1_bio. It has already been counted |
1431 | * in bio->bi_phys_segments | |
1432 | */ | |
1433 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1434 | r1_bio->master_bio = bio; | |
aa8b57aa | 1435 | r1_bio->sectors = bio_sectors(bio) - sectors_handled; |
1f68f0c4 N |
1436 | r1_bio->state = 0; |
1437 | r1_bio->mddev = mddev; | |
4f024f37 | 1438 | r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
1f68f0c4 N |
1439 | goto retry_write; |
1440 | } | |
1441 | ||
079fa166 N |
1442 | r1_bio_write_done(r1_bio); |
1443 | ||
1444 | /* In case raid1d snuck in to freeze_array */ | |
1445 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1446 | } |
1447 | ||
3b046a97 RL |
1448 | static void raid1_make_request(struct mddev *mddev, struct bio *bio) |
1449 | { | |
1450 | struct r1conf *conf = mddev->private; | |
1451 | struct r1bio *r1_bio; | |
1452 | ||
1453 | /* | |
1454 | * make_request() can abort the operation when read-ahead is being | |
1455 | * used and no empty request is available. | |
1456 | * | |
1457 | */ | |
1458 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
1459 | ||
1460 | r1_bio->master_bio = bio; | |
1461 | r1_bio->sectors = bio_sectors(bio); | |
1462 | r1_bio->state = 0; | |
1463 | r1_bio->mddev = mddev; | |
1464 | r1_bio->sector = bio->bi_iter.bi_sector; | |
1465 | ||
1466 | /* | |
1467 | * We might need to issue multiple reads to different devices if there | |
1468 | * are bad blocks around, so we keep track of the number of reads in | |
1469 | * bio->bi_phys_segments. If this is 0, there is only one r1_bio and | |
1470 | * no locking will be needed when requests complete. If it is | |
1471 | * non-zero, then it is the number of not-completed requests. | |
1472 | */ | |
1473 | bio->bi_phys_segments = 0; | |
1474 | bio_clear_flag(bio, BIO_SEG_VALID); | |
1475 | ||
1476 | if (bio_data_dir(bio) == READ) | |
1477 | raid1_read_request(mddev, bio, r1_bio); | |
1478 | else | |
1479 | raid1_write_request(mddev, bio, r1_bio); | |
1480 | } | |
1481 | ||
849674e4 | 1482 | static void raid1_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1483 | { |
e8096360 | 1484 | struct r1conf *conf = mddev->private; |
1da177e4 LT |
1485 | int i; |
1486 | ||
1487 | seq_printf(seq, " [%d/%d] [", conf->raid_disks, | |
11ce99e6 | 1488 | conf->raid_disks - mddev->degraded); |
ddac7c7e N |
1489 | rcu_read_lock(); |
1490 | for (i = 0; i < conf->raid_disks; i++) { | |
3cb03002 | 1491 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
1da177e4 | 1492 | seq_printf(seq, "%s", |
ddac7c7e N |
1493 | rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1494 | } | |
1495 | rcu_read_unlock(); | |
1da177e4 LT |
1496 | seq_printf(seq, "]"); |
1497 | } | |
1498 | ||
849674e4 | 1499 | static void raid1_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1500 | { |
1501 | char b[BDEVNAME_SIZE]; | |
e8096360 | 1502 | struct r1conf *conf = mddev->private; |
423f04d6 | 1503 | unsigned long flags; |
1da177e4 LT |
1504 | |
1505 | /* | |
1506 | * If it is not operational, then we have already marked it as dead | |
1507 | * else if it is the last working disks, ignore the error, let the | |
1508 | * next level up know. | |
1509 | * else mark the drive as failed | |
1510 | */ | |
2e52d449 | 1511 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1512 | if (test_bit(In_sync, &rdev->flags) |
4044ba58 | 1513 | && (conf->raid_disks - mddev->degraded) == 1) { |
1da177e4 LT |
1514 | /* |
1515 | * Don't fail the drive, act as though we were just a | |
4044ba58 N |
1516 | * normal single drive. |
1517 | * However don't try a recovery from this drive as | |
1518 | * it is very likely to fail. | |
1da177e4 | 1519 | */ |
5389042f | 1520 | conf->recovery_disabled = mddev->recovery_disabled; |
2e52d449 | 1521 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1522 | return; |
4044ba58 | 1523 | } |
de393cde | 1524 | set_bit(Blocked, &rdev->flags); |
c04be0aa | 1525 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
1da177e4 | 1526 | mddev->degraded++; |
dd00a99e | 1527 | set_bit(Faulty, &rdev->flags); |
dd00a99e N |
1528 | } else |
1529 | set_bit(Faulty, &rdev->flags); | |
423f04d6 | 1530 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2446dba0 N |
1531 | /* |
1532 | * if recovery is running, make sure it aborts. | |
1533 | */ | |
1534 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2953079c SL |
1535 | set_mask_bits(&mddev->sb_flags, 0, |
1536 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
1d41c216 N |
1537 | pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n" |
1538 | "md/raid1:%s: Operation continuing on %d devices.\n", | |
1539 | mdname(mddev), bdevname(rdev->bdev, b), | |
1540 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
1da177e4 LT |
1541 | } |
1542 | ||
e8096360 | 1543 | static void print_conf(struct r1conf *conf) |
1da177e4 LT |
1544 | { |
1545 | int i; | |
1da177e4 | 1546 | |
1d41c216 | 1547 | pr_debug("RAID1 conf printout:\n"); |
1da177e4 | 1548 | if (!conf) { |
1d41c216 | 1549 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1550 | return; |
1551 | } | |
1d41c216 N |
1552 | pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, |
1553 | conf->raid_disks); | |
1da177e4 | 1554 | |
ddac7c7e | 1555 | rcu_read_lock(); |
1da177e4 LT |
1556 | for (i = 0; i < conf->raid_disks; i++) { |
1557 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 1558 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
ddac7c7e | 1559 | if (rdev) |
1d41c216 N |
1560 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1561 | i, !test_bit(In_sync, &rdev->flags), | |
1562 | !test_bit(Faulty, &rdev->flags), | |
1563 | bdevname(rdev->bdev,b)); | |
1da177e4 | 1564 | } |
ddac7c7e | 1565 | rcu_read_unlock(); |
1da177e4 LT |
1566 | } |
1567 | ||
e8096360 | 1568 | static void close_sync(struct r1conf *conf) |
1da177e4 | 1569 | { |
79ef3a8a | 1570 | wait_barrier(conf, NULL); |
1571 | allow_barrier(conf, 0, 0); | |
1da177e4 LT |
1572 | |
1573 | mempool_destroy(conf->r1buf_pool); | |
1574 | conf->r1buf_pool = NULL; | |
79ef3a8a | 1575 | |
669cc7ba | 1576 | spin_lock_irq(&conf->resync_lock); |
e8ff8bf0 | 1577 | conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE; |
79ef3a8a | 1578 | conf->start_next_window = MaxSector; |
669cc7ba N |
1579 | conf->current_window_requests += |
1580 | conf->next_window_requests; | |
1581 | conf->next_window_requests = 0; | |
1582 | spin_unlock_irq(&conf->resync_lock); | |
1da177e4 LT |
1583 | } |
1584 | ||
fd01b88c | 1585 | static int raid1_spare_active(struct mddev *mddev) |
1da177e4 LT |
1586 | { |
1587 | int i; | |
e8096360 | 1588 | struct r1conf *conf = mddev->private; |
6b965620 N |
1589 | int count = 0; |
1590 | unsigned long flags; | |
1da177e4 LT |
1591 | |
1592 | /* | |
f72ffdd6 | 1593 | * Find all failed disks within the RAID1 configuration |
ddac7c7e N |
1594 | * and mark them readable. |
1595 | * Called under mddev lock, so rcu protection not needed. | |
423f04d6 N |
1596 | * device_lock used to avoid races with raid1_end_read_request |
1597 | * which expects 'In_sync' flags and ->degraded to be consistent. | |
1da177e4 | 1598 | */ |
423f04d6 | 1599 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 1600 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 1601 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
8c7a2c2b N |
1602 | struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; |
1603 | if (repl | |
1aee41f6 | 1604 | && !test_bit(Candidate, &repl->flags) |
8c7a2c2b N |
1605 | && repl->recovery_offset == MaxSector |
1606 | && !test_bit(Faulty, &repl->flags) | |
1607 | && !test_and_set_bit(In_sync, &repl->flags)) { | |
1608 | /* replacement has just become active */ | |
1609 | if (!rdev || | |
1610 | !test_and_clear_bit(In_sync, &rdev->flags)) | |
1611 | count++; | |
1612 | if (rdev) { | |
1613 | /* Replaced device not technically | |
1614 | * faulty, but we need to be sure | |
1615 | * it gets removed and never re-added | |
1616 | */ | |
1617 | set_bit(Faulty, &rdev->flags); | |
1618 | sysfs_notify_dirent_safe( | |
1619 | rdev->sysfs_state); | |
1620 | } | |
1621 | } | |
ddac7c7e | 1622 | if (rdev |
61e4947c | 1623 | && rdev->recovery_offset == MaxSector |
ddac7c7e | 1624 | && !test_bit(Faulty, &rdev->flags) |
c04be0aa | 1625 | && !test_and_set_bit(In_sync, &rdev->flags)) { |
6b965620 | 1626 | count++; |
654e8b5a | 1627 | sysfs_notify_dirent_safe(rdev->sysfs_state); |
1da177e4 LT |
1628 | } |
1629 | } | |
6b965620 N |
1630 | mddev->degraded -= count; |
1631 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1632 | |
1633 | print_conf(conf); | |
6b965620 | 1634 | return count; |
1da177e4 LT |
1635 | } |
1636 | ||
fd01b88c | 1637 | static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1638 | { |
e8096360 | 1639 | struct r1conf *conf = mddev->private; |
199050ea | 1640 | int err = -EEXIST; |
41158c7e | 1641 | int mirror = 0; |
0eaf822c | 1642 | struct raid1_info *p; |
6c2fce2e | 1643 | int first = 0; |
30194636 | 1644 | int last = conf->raid_disks - 1; |
1da177e4 | 1645 | |
5389042f N |
1646 | if (mddev->recovery_disabled == conf->recovery_disabled) |
1647 | return -EBUSY; | |
1648 | ||
1501efad DW |
1649 | if (md_integrity_add_rdev(rdev, mddev)) |
1650 | return -ENXIO; | |
1651 | ||
6c2fce2e NB |
1652 | if (rdev->raid_disk >= 0) |
1653 | first = last = rdev->raid_disk; | |
1654 | ||
70bcecdb GR |
1655 | /* |
1656 | * find the disk ... but prefer rdev->saved_raid_disk | |
1657 | * if possible. | |
1658 | */ | |
1659 | if (rdev->saved_raid_disk >= 0 && | |
1660 | rdev->saved_raid_disk >= first && | |
1661 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) | |
1662 | first = last = rdev->saved_raid_disk; | |
1663 | ||
7ef449d1 N |
1664 | for (mirror = first; mirror <= last; mirror++) { |
1665 | p = conf->mirrors+mirror; | |
1666 | if (!p->rdev) { | |
1da177e4 | 1667 | |
9092c02d JB |
1668 | if (mddev->gendisk) |
1669 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1670 | rdev->data_offset << 9); | |
1da177e4 LT |
1671 | |
1672 | p->head_position = 0; | |
1673 | rdev->raid_disk = mirror; | |
199050ea | 1674 | err = 0; |
6aea114a N |
1675 | /* As all devices are equivalent, we don't need a full recovery |
1676 | * if this was recently any drive of the array | |
1677 | */ | |
1678 | if (rdev->saved_raid_disk < 0) | |
41158c7e | 1679 | conf->fullsync = 1; |
d6065f7b | 1680 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
1681 | break; |
1682 | } | |
7ef449d1 N |
1683 | if (test_bit(WantReplacement, &p->rdev->flags) && |
1684 | p[conf->raid_disks].rdev == NULL) { | |
1685 | /* Add this device as a replacement */ | |
1686 | clear_bit(In_sync, &rdev->flags); | |
1687 | set_bit(Replacement, &rdev->flags); | |
1688 | rdev->raid_disk = mirror; | |
1689 | err = 0; | |
1690 | conf->fullsync = 1; | |
1691 | rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); | |
1692 | break; | |
1693 | } | |
1694 | } | |
9092c02d | 1695 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
2ff8cc2c | 1696 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1da177e4 | 1697 | print_conf(conf); |
199050ea | 1698 | return err; |
1da177e4 LT |
1699 | } |
1700 | ||
b8321b68 | 1701 | static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1702 | { |
e8096360 | 1703 | struct r1conf *conf = mddev->private; |
1da177e4 | 1704 | int err = 0; |
b8321b68 | 1705 | int number = rdev->raid_disk; |
0eaf822c | 1706 | struct raid1_info *p = conf->mirrors + number; |
1da177e4 | 1707 | |
b014f14c N |
1708 | if (rdev != p->rdev) |
1709 | p = conf->mirrors + conf->raid_disks + number; | |
1710 | ||
1da177e4 | 1711 | print_conf(conf); |
b8321b68 | 1712 | if (rdev == p->rdev) { |
b2d444d7 | 1713 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
1714 | atomic_read(&rdev->nr_pending)) { |
1715 | err = -EBUSY; | |
1716 | goto abort; | |
1717 | } | |
046abeed | 1718 | /* Only remove non-faulty devices if recovery |
dfc70645 N |
1719 | * is not possible. |
1720 | */ | |
1721 | if (!test_bit(Faulty, &rdev->flags) && | |
5389042f | 1722 | mddev->recovery_disabled != conf->recovery_disabled && |
dfc70645 N |
1723 | mddev->degraded < conf->raid_disks) { |
1724 | err = -EBUSY; | |
1725 | goto abort; | |
1726 | } | |
1da177e4 | 1727 | p->rdev = NULL; |
d787be40 N |
1728 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1729 | synchronize_rcu(); | |
1730 | if (atomic_read(&rdev->nr_pending)) { | |
1731 | /* lost the race, try later */ | |
1732 | err = -EBUSY; | |
1733 | p->rdev = rdev; | |
1734 | goto abort; | |
1735 | } | |
1736 | } | |
1737 | if (conf->mirrors[conf->raid_disks + number].rdev) { | |
8c7a2c2b N |
1738 | /* We just removed a device that is being replaced. |
1739 | * Move down the replacement. We drain all IO before | |
1740 | * doing this to avoid confusion. | |
1741 | */ | |
1742 | struct md_rdev *repl = | |
1743 | conf->mirrors[conf->raid_disks + number].rdev; | |
e2d59925 | 1744 | freeze_array(conf, 0); |
8c7a2c2b N |
1745 | clear_bit(Replacement, &repl->flags); |
1746 | p->rdev = repl; | |
1747 | conf->mirrors[conf->raid_disks + number].rdev = NULL; | |
e2d59925 | 1748 | unfreeze_array(conf); |
8c7a2c2b N |
1749 | clear_bit(WantReplacement, &rdev->flags); |
1750 | } else | |
b014f14c | 1751 | clear_bit(WantReplacement, &rdev->flags); |
a91a2785 | 1752 | err = md_integrity_register(mddev); |
1da177e4 LT |
1753 | } |
1754 | abort: | |
1755 | ||
1756 | print_conf(conf); | |
1757 | return err; | |
1758 | } | |
1759 | ||
4246a0b6 | 1760 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1761 | { |
9f2c9d12 | 1762 | struct r1bio *r1_bio = bio->bi_private; |
1da177e4 | 1763 | |
0fc280f6 | 1764 | update_head_pos(r1_bio->read_disk, r1_bio); |
ba3ae3be | 1765 | |
1da177e4 LT |
1766 | /* |
1767 | * we have read a block, now it needs to be re-written, | |
1768 | * or re-read if the read failed. | |
1769 | * We don't do much here, just schedule handling by raid1d | |
1770 | */ | |
4246a0b6 | 1771 | if (!bio->bi_error) |
1da177e4 | 1772 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
d11c171e N |
1773 | |
1774 | if (atomic_dec_and_test(&r1_bio->remaining)) | |
1775 | reschedule_retry(r1_bio); | |
1da177e4 LT |
1776 | } |
1777 | ||
4246a0b6 | 1778 | static void end_sync_write(struct bio *bio) |
1da177e4 | 1779 | { |
4246a0b6 | 1780 | int uptodate = !bio->bi_error; |
9f2c9d12 | 1781 | struct r1bio *r1_bio = bio->bi_private; |
fd01b88c | 1782 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1783 | struct r1conf *conf = mddev->private; |
4367af55 N |
1784 | sector_t first_bad; |
1785 | int bad_sectors; | |
854abd75 | 1786 | struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; |
ba3ae3be | 1787 | |
6b1117d5 | 1788 | if (!uptodate) { |
57dab0bd | 1789 | sector_t sync_blocks = 0; |
6b1117d5 N |
1790 | sector_t s = r1_bio->sector; |
1791 | long sectors_to_go = r1_bio->sectors; | |
1792 | /* make sure these bits doesn't get cleared. */ | |
1793 | do { | |
5e3db645 | 1794 | bitmap_end_sync(mddev->bitmap, s, |
6b1117d5 N |
1795 | &sync_blocks, 1); |
1796 | s += sync_blocks; | |
1797 | sectors_to_go -= sync_blocks; | |
1798 | } while (sectors_to_go > 0); | |
854abd75 N |
1799 | set_bit(WriteErrorSeen, &rdev->flags); |
1800 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) | |
19d67169 N |
1801 | set_bit(MD_RECOVERY_NEEDED, & |
1802 | mddev->recovery); | |
d8f05d29 | 1803 | set_bit(R1BIO_WriteError, &r1_bio->state); |
854abd75 | 1804 | } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, |
3a9f28a5 N |
1805 | &first_bad, &bad_sectors) && |
1806 | !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, | |
1807 | r1_bio->sector, | |
1808 | r1_bio->sectors, | |
1809 | &first_bad, &bad_sectors) | |
1810 | ) | |
4367af55 | 1811 | set_bit(R1BIO_MadeGood, &r1_bio->state); |
e3b9703e | 1812 | |
1da177e4 | 1813 | if (atomic_dec_and_test(&r1_bio->remaining)) { |
4367af55 | 1814 | int s = r1_bio->sectors; |
d8f05d29 N |
1815 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
1816 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
4367af55 N |
1817 | reschedule_retry(r1_bio); |
1818 | else { | |
1819 | put_buf(r1_bio); | |
1820 | md_done_sync(mddev, s, uptodate); | |
1821 | } | |
1da177e4 | 1822 | } |
1da177e4 LT |
1823 | } |
1824 | ||
3cb03002 | 1825 | static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector, |
d8f05d29 N |
1826 | int sectors, struct page *page, int rw) |
1827 | { | |
796a5cf0 | 1828 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
d8f05d29 N |
1829 | /* success */ |
1830 | return 1; | |
19d67169 | 1831 | if (rw == WRITE) { |
d8f05d29 | 1832 | set_bit(WriteErrorSeen, &rdev->flags); |
19d67169 N |
1833 | if (!test_and_set_bit(WantReplacement, |
1834 | &rdev->flags)) | |
1835 | set_bit(MD_RECOVERY_NEEDED, & | |
1836 | rdev->mddev->recovery); | |
1837 | } | |
d8f05d29 N |
1838 | /* need to record an error - either for the block or the device */ |
1839 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
1840 | md_error(rdev->mddev, rdev); | |
1841 | return 0; | |
1842 | } | |
1843 | ||
9f2c9d12 | 1844 | static int fix_sync_read_error(struct r1bio *r1_bio) |
1da177e4 | 1845 | { |
a68e5870 N |
1846 | /* Try some synchronous reads of other devices to get |
1847 | * good data, much like with normal read errors. Only | |
1848 | * read into the pages we already have so we don't | |
1849 | * need to re-issue the read request. | |
1850 | * We don't need to freeze the array, because being in an | |
1851 | * active sync request, there is no normal IO, and | |
1852 | * no overlapping syncs. | |
06f60385 N |
1853 | * We don't need to check is_badblock() again as we |
1854 | * made sure that anything with a bad block in range | |
1855 | * will have bi_end_io clear. | |
a68e5870 | 1856 | */ |
fd01b88c | 1857 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1858 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1859 | struct bio *bio = r1_bio->bios[r1_bio->read_disk]; |
1860 | sector_t sect = r1_bio->sector; | |
1861 | int sectors = r1_bio->sectors; | |
1862 | int idx = 0; | |
2e52d449 N |
1863 | struct md_rdev *rdev; |
1864 | ||
1865 | rdev = conf->mirrors[r1_bio->read_disk].rdev; | |
1866 | if (test_bit(FailFast, &rdev->flags)) { | |
1867 | /* Don't try recovering from here - just fail it | |
1868 | * ... unless it is the last working device of course */ | |
1869 | md_error(mddev, rdev); | |
1870 | if (test_bit(Faulty, &rdev->flags)) | |
1871 | /* Don't try to read from here, but make sure | |
1872 | * put_buf does it's thing | |
1873 | */ | |
1874 | bio->bi_end_io = end_sync_write; | |
1875 | } | |
a68e5870 N |
1876 | |
1877 | while(sectors) { | |
1878 | int s = sectors; | |
1879 | int d = r1_bio->read_disk; | |
1880 | int success = 0; | |
78d7f5f7 | 1881 | int start; |
a68e5870 N |
1882 | |
1883 | if (s > (PAGE_SIZE>>9)) | |
1884 | s = PAGE_SIZE >> 9; | |
1885 | do { | |
1886 | if (r1_bio->bios[d]->bi_end_io == end_sync_read) { | |
1887 | /* No rcu protection needed here devices | |
1888 | * can only be removed when no resync is | |
1889 | * active, and resync is currently active | |
1890 | */ | |
1891 | rdev = conf->mirrors[d].rdev; | |
9d3d8011 | 1892 | if (sync_page_io(rdev, sect, s<<9, |
a68e5870 | 1893 | bio->bi_io_vec[idx].bv_page, |
796a5cf0 | 1894 | REQ_OP_READ, 0, false)) { |
a68e5870 N |
1895 | success = 1; |
1896 | break; | |
1897 | } | |
1898 | } | |
1899 | d++; | |
8f19ccb2 | 1900 | if (d == conf->raid_disks * 2) |
a68e5870 N |
1901 | d = 0; |
1902 | } while (!success && d != r1_bio->read_disk); | |
1903 | ||
78d7f5f7 | 1904 | if (!success) { |
a68e5870 | 1905 | char b[BDEVNAME_SIZE]; |
3a9f28a5 N |
1906 | int abort = 0; |
1907 | /* Cannot read from anywhere, this block is lost. | |
1908 | * Record a bad block on each device. If that doesn't | |
1909 | * work just disable and interrupt the recovery. | |
1910 | * Don't fail devices as that won't really help. | |
1911 | */ | |
1d41c216 N |
1912 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
1913 | mdname(mddev), | |
1914 | bdevname(bio->bi_bdev, b), | |
1915 | (unsigned long long)r1_bio->sector); | |
8f19ccb2 | 1916 | for (d = 0; d < conf->raid_disks * 2; d++) { |
3a9f28a5 N |
1917 | rdev = conf->mirrors[d].rdev; |
1918 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
1919 | continue; | |
1920 | if (!rdev_set_badblocks(rdev, sect, s, 0)) | |
1921 | abort = 1; | |
1922 | } | |
1923 | if (abort) { | |
d890fa2b N |
1924 | conf->recovery_disabled = |
1925 | mddev->recovery_disabled; | |
3a9f28a5 N |
1926 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
1927 | md_done_sync(mddev, r1_bio->sectors, 0); | |
1928 | put_buf(r1_bio); | |
1929 | return 0; | |
1930 | } | |
1931 | /* Try next page */ | |
1932 | sectors -= s; | |
1933 | sect += s; | |
1934 | idx++; | |
1935 | continue; | |
d11c171e | 1936 | } |
78d7f5f7 N |
1937 | |
1938 | start = d; | |
1939 | /* write it back and re-read */ | |
1940 | while (d != r1_bio->read_disk) { | |
1941 | if (d == 0) | |
8f19ccb2 | 1942 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1943 | d--; |
1944 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1945 | continue; | |
1946 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1947 | if (r1_sync_page_io(rdev, sect, s, |
1948 | bio->bi_io_vec[idx].bv_page, | |
1949 | WRITE) == 0) { | |
78d7f5f7 N |
1950 | r1_bio->bios[d]->bi_end_io = NULL; |
1951 | rdev_dec_pending(rdev, mddev); | |
9d3d8011 | 1952 | } |
78d7f5f7 N |
1953 | } |
1954 | d = start; | |
1955 | while (d != r1_bio->read_disk) { | |
1956 | if (d == 0) | |
8f19ccb2 | 1957 | d = conf->raid_disks * 2; |
78d7f5f7 N |
1958 | d--; |
1959 | if (r1_bio->bios[d]->bi_end_io != end_sync_read) | |
1960 | continue; | |
1961 | rdev = conf->mirrors[d].rdev; | |
d8f05d29 N |
1962 | if (r1_sync_page_io(rdev, sect, s, |
1963 | bio->bi_io_vec[idx].bv_page, | |
1964 | READ) != 0) | |
9d3d8011 | 1965 | atomic_add(s, &rdev->corrected_errors); |
78d7f5f7 | 1966 | } |
a68e5870 N |
1967 | sectors -= s; |
1968 | sect += s; | |
1969 | idx ++; | |
1970 | } | |
78d7f5f7 | 1971 | set_bit(R1BIO_Uptodate, &r1_bio->state); |
4246a0b6 | 1972 | bio->bi_error = 0; |
a68e5870 N |
1973 | return 1; |
1974 | } | |
1975 | ||
c95e6385 | 1976 | static void process_checks(struct r1bio *r1_bio) |
a68e5870 N |
1977 | { |
1978 | /* We have read all readable devices. If we haven't | |
1979 | * got the block, then there is no hope left. | |
1980 | * If we have, then we want to do a comparison | |
1981 | * and skip the write if everything is the same. | |
1982 | * If any blocks failed to read, then we need to | |
1983 | * attempt an over-write | |
1984 | */ | |
fd01b88c | 1985 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 1986 | struct r1conf *conf = mddev->private; |
a68e5870 N |
1987 | int primary; |
1988 | int i; | |
f4380a91 | 1989 | int vcnt; |
a68e5870 | 1990 | |
30bc9b53 N |
1991 | /* Fix variable parts of all bios */ |
1992 | vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); | |
1993 | for (i = 0; i < conf->raid_disks * 2; i++) { | |
1994 | int j; | |
1995 | int size; | |
4246a0b6 | 1996 | int error; |
30bc9b53 N |
1997 | struct bio *b = r1_bio->bios[i]; |
1998 | if (b->bi_end_io != end_sync_read) | |
1999 | continue; | |
4246a0b6 CH |
2000 | /* fixup the bio for reuse, but preserve errno */ |
2001 | error = b->bi_error; | |
30bc9b53 | 2002 | bio_reset(b); |
4246a0b6 | 2003 | b->bi_error = error; |
30bc9b53 | 2004 | b->bi_vcnt = vcnt; |
4f024f37 KO |
2005 | b->bi_iter.bi_size = r1_bio->sectors << 9; |
2006 | b->bi_iter.bi_sector = r1_bio->sector + | |
30bc9b53 N |
2007 | conf->mirrors[i].rdev->data_offset; |
2008 | b->bi_bdev = conf->mirrors[i].rdev->bdev; | |
2009 | b->bi_end_io = end_sync_read; | |
2010 | b->bi_private = r1_bio; | |
2011 | ||
4f024f37 | 2012 | size = b->bi_iter.bi_size; |
30bc9b53 N |
2013 | for (j = 0; j < vcnt ; j++) { |
2014 | struct bio_vec *bi; | |
2015 | bi = &b->bi_io_vec[j]; | |
2016 | bi->bv_offset = 0; | |
2017 | if (size > PAGE_SIZE) | |
2018 | bi->bv_len = PAGE_SIZE; | |
2019 | else | |
2020 | bi->bv_len = size; | |
2021 | size -= PAGE_SIZE; | |
2022 | } | |
2023 | } | |
8f19ccb2 | 2024 | for (primary = 0; primary < conf->raid_disks * 2; primary++) |
a68e5870 | 2025 | if (r1_bio->bios[primary]->bi_end_io == end_sync_read && |
4246a0b6 | 2026 | !r1_bio->bios[primary]->bi_error) { |
a68e5870 N |
2027 | r1_bio->bios[primary]->bi_end_io = NULL; |
2028 | rdev_dec_pending(conf->mirrors[primary].rdev, mddev); | |
2029 | break; | |
2030 | } | |
2031 | r1_bio->read_disk = primary; | |
8f19ccb2 | 2032 | for (i = 0; i < conf->raid_disks * 2; i++) { |
78d7f5f7 | 2033 | int j; |
78d7f5f7 N |
2034 | struct bio *pbio = r1_bio->bios[primary]; |
2035 | struct bio *sbio = r1_bio->bios[i]; | |
4246a0b6 | 2036 | int error = sbio->bi_error; |
a68e5870 | 2037 | |
2aabaa65 | 2038 | if (sbio->bi_end_io != end_sync_read) |
78d7f5f7 | 2039 | continue; |
4246a0b6 CH |
2040 | /* Now we can 'fixup' the error value */ |
2041 | sbio->bi_error = 0; | |
78d7f5f7 | 2042 | |
4246a0b6 | 2043 | if (!error) { |
78d7f5f7 N |
2044 | for (j = vcnt; j-- ; ) { |
2045 | struct page *p, *s; | |
2046 | p = pbio->bi_io_vec[j].bv_page; | |
2047 | s = sbio->bi_io_vec[j].bv_page; | |
2048 | if (memcmp(page_address(p), | |
2049 | page_address(s), | |
5020ad7d | 2050 | sbio->bi_io_vec[j].bv_len)) |
78d7f5f7 | 2051 | break; |
69382e85 | 2052 | } |
78d7f5f7 N |
2053 | } else |
2054 | j = 0; | |
2055 | if (j >= 0) | |
7f7583d4 | 2056 | atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); |
78d7f5f7 | 2057 | if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) |
4246a0b6 | 2058 | && !error)) { |
78d7f5f7 N |
2059 | /* No need to write to this device. */ |
2060 | sbio->bi_end_io = NULL; | |
2061 | rdev_dec_pending(conf->mirrors[i].rdev, mddev); | |
2062 | continue; | |
2063 | } | |
d3b45c2a KO |
2064 | |
2065 | bio_copy_data(sbio, pbio); | |
78d7f5f7 | 2066 | } |
a68e5870 N |
2067 | } |
2068 | ||
9f2c9d12 | 2069 | static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio) |
a68e5870 | 2070 | { |
e8096360 | 2071 | struct r1conf *conf = mddev->private; |
a68e5870 | 2072 | int i; |
8f19ccb2 | 2073 | int disks = conf->raid_disks * 2; |
a68e5870 N |
2074 | struct bio *bio, *wbio; |
2075 | ||
2076 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2077 | ||
a68e5870 N |
2078 | if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) |
2079 | /* ouch - failed to read all of that. */ | |
2080 | if (!fix_sync_read_error(r1_bio)) | |
2081 | return; | |
7ca78d57 N |
2082 | |
2083 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
c95e6385 N |
2084 | process_checks(r1_bio); |
2085 | ||
d11c171e N |
2086 | /* |
2087 | * schedule writes | |
2088 | */ | |
1da177e4 LT |
2089 | atomic_set(&r1_bio->remaining, 1); |
2090 | for (i = 0; i < disks ; i++) { | |
2091 | wbio = r1_bio->bios[i]; | |
3e198f78 N |
2092 | if (wbio->bi_end_io == NULL || |
2093 | (wbio->bi_end_io == end_sync_read && | |
2094 | (i == r1_bio->read_disk || | |
2095 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) | |
1da177e4 | 2096 | continue; |
6e05fc2a N |
2097 | if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) |
2098 | continue; | |
1da177e4 | 2099 | |
796a5cf0 | 2100 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
212e7eb7 N |
2101 | if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) |
2102 | wbio->bi_opf |= MD_FAILFAST; | |
2103 | ||
3e198f78 | 2104 | wbio->bi_end_io = end_sync_write; |
1da177e4 | 2105 | atomic_inc(&r1_bio->remaining); |
aa8b57aa | 2106 | md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); |
191ea9b2 | 2107 | |
1da177e4 LT |
2108 | generic_make_request(wbio); |
2109 | } | |
2110 | ||
2111 | if (atomic_dec_and_test(&r1_bio->remaining)) { | |
191ea9b2 | 2112 | /* if we're here, all write(s) have completed, so clean up */ |
58e94ae1 N |
2113 | int s = r1_bio->sectors; |
2114 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || | |
2115 | test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2116 | reschedule_retry(r1_bio); | |
2117 | else { | |
2118 | put_buf(r1_bio); | |
2119 | md_done_sync(mddev, s, 1); | |
2120 | } | |
1da177e4 LT |
2121 | } |
2122 | } | |
2123 | ||
2124 | /* | |
2125 | * This is a kernel thread which: | |
2126 | * | |
2127 | * 1. Retries failed read operations on working mirrors. | |
2128 | * 2. Updates the raid superblock when problems encounter. | |
d2eb35ac | 2129 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2130 | */ |
2131 | ||
e8096360 | 2132 | static void fix_read_error(struct r1conf *conf, int read_disk, |
867868fb N |
2133 | sector_t sect, int sectors) |
2134 | { | |
fd01b88c | 2135 | struct mddev *mddev = conf->mddev; |
867868fb N |
2136 | while(sectors) { |
2137 | int s = sectors; | |
2138 | int d = read_disk; | |
2139 | int success = 0; | |
2140 | int start; | |
3cb03002 | 2141 | struct md_rdev *rdev; |
867868fb N |
2142 | |
2143 | if (s > (PAGE_SIZE>>9)) | |
2144 | s = PAGE_SIZE >> 9; | |
2145 | ||
2146 | do { | |
d2eb35ac N |
2147 | sector_t first_bad; |
2148 | int bad_sectors; | |
2149 | ||
707a6a42 N |
2150 | rcu_read_lock(); |
2151 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2152 | if (rdev && |
da8840a7 | 2153 | (test_bit(In_sync, &rdev->flags) || |
2154 | (!test_bit(Faulty, &rdev->flags) && | |
2155 | rdev->recovery_offset >= sect + s)) && | |
d2eb35ac | 2156 | is_badblock(rdev, sect, s, |
707a6a42 N |
2157 | &first_bad, &bad_sectors) == 0) { |
2158 | atomic_inc(&rdev->nr_pending); | |
2159 | rcu_read_unlock(); | |
2160 | if (sync_page_io(rdev, sect, s<<9, | |
796a5cf0 | 2161 | conf->tmppage, REQ_OP_READ, 0, false)) |
707a6a42 N |
2162 | success = 1; |
2163 | rdev_dec_pending(rdev, mddev); | |
2164 | if (success) | |
2165 | break; | |
2166 | } else | |
2167 | rcu_read_unlock(); | |
2168 | d++; | |
2169 | if (d == conf->raid_disks * 2) | |
2170 | d = 0; | |
867868fb N |
2171 | } while (!success && d != read_disk); |
2172 | ||
2173 | if (!success) { | |
d8f05d29 | 2174 | /* Cannot read from anywhere - mark it bad */ |
3cb03002 | 2175 | struct md_rdev *rdev = conf->mirrors[read_disk].rdev; |
d8f05d29 N |
2176 | if (!rdev_set_badblocks(rdev, sect, s, 0)) |
2177 | md_error(mddev, rdev); | |
867868fb N |
2178 | break; |
2179 | } | |
2180 | /* write it back and re-read */ | |
2181 | start = d; | |
2182 | while (d != read_disk) { | |
2183 | if (d==0) | |
8f19ccb2 | 2184 | d = conf->raid_disks * 2; |
867868fb | 2185 | d--; |
707a6a42 N |
2186 | rcu_read_lock(); |
2187 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2188 | if (rdev && |
707a6a42 N |
2189 | !test_bit(Faulty, &rdev->flags)) { |
2190 | atomic_inc(&rdev->nr_pending); | |
2191 | rcu_read_unlock(); | |
d8f05d29 N |
2192 | r1_sync_page_io(rdev, sect, s, |
2193 | conf->tmppage, WRITE); | |
707a6a42 N |
2194 | rdev_dec_pending(rdev, mddev); |
2195 | } else | |
2196 | rcu_read_unlock(); | |
867868fb N |
2197 | } |
2198 | d = start; | |
2199 | while (d != read_disk) { | |
2200 | char b[BDEVNAME_SIZE]; | |
2201 | if (d==0) | |
8f19ccb2 | 2202 | d = conf->raid_disks * 2; |
867868fb | 2203 | d--; |
707a6a42 N |
2204 | rcu_read_lock(); |
2205 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
867868fb | 2206 | if (rdev && |
b8cb6b4c | 2207 | !test_bit(Faulty, &rdev->flags)) { |
707a6a42 N |
2208 | atomic_inc(&rdev->nr_pending); |
2209 | rcu_read_unlock(); | |
d8f05d29 N |
2210 | if (r1_sync_page_io(rdev, sect, s, |
2211 | conf->tmppage, READ)) { | |
867868fb | 2212 | atomic_add(s, &rdev->corrected_errors); |
1d41c216 N |
2213 | pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n", |
2214 | mdname(mddev), s, | |
2215 | (unsigned long long)(sect + | |
2216 | rdev->data_offset), | |
2217 | bdevname(rdev->bdev, b)); | |
867868fb | 2218 | } |
707a6a42 N |
2219 | rdev_dec_pending(rdev, mddev); |
2220 | } else | |
2221 | rcu_read_unlock(); | |
867868fb N |
2222 | } |
2223 | sectors -= s; | |
2224 | sect += s; | |
2225 | } | |
2226 | } | |
2227 | ||
9f2c9d12 | 2228 | static int narrow_write_error(struct r1bio *r1_bio, int i) |
cd5ff9a1 | 2229 | { |
fd01b88c | 2230 | struct mddev *mddev = r1_bio->mddev; |
e8096360 | 2231 | struct r1conf *conf = mddev->private; |
3cb03002 | 2232 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
cd5ff9a1 N |
2233 | |
2234 | /* bio has the data to be written to device 'i' where | |
2235 | * we just recently had a write error. | |
2236 | * We repeatedly clone the bio and trim down to one block, | |
2237 | * then try the write. Where the write fails we record | |
2238 | * a bad block. | |
2239 | * It is conceivable that the bio doesn't exactly align with | |
2240 | * blocks. We must handle this somehow. | |
2241 | * | |
2242 | * We currently own a reference on the rdev. | |
2243 | */ | |
2244 | ||
2245 | int block_sectors; | |
2246 | sector_t sector; | |
2247 | int sectors; | |
2248 | int sect_to_write = r1_bio->sectors; | |
2249 | int ok = 1; | |
2250 | ||
2251 | if (rdev->badblocks.shift < 0) | |
2252 | return 0; | |
2253 | ||
ab713cdc ND |
2254 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2255 | bdev_logical_block_size(rdev->bdev) >> 9); | |
cd5ff9a1 N |
2256 | sector = r1_bio->sector; |
2257 | sectors = ((sector + block_sectors) | |
2258 | & ~(sector_t)(block_sectors - 1)) | |
2259 | - sector; | |
2260 | ||
cd5ff9a1 N |
2261 | while (sect_to_write) { |
2262 | struct bio *wbio; | |
2263 | if (sectors > sect_to_write) | |
2264 | sectors = sect_to_write; | |
2265 | /* Write at 'sector' for 'sectors'*/ | |
2266 | ||
b783863f KO |
2267 | if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { |
2268 | unsigned vcnt = r1_bio->behind_page_count; | |
2269 | struct bio_vec *vec = r1_bio->behind_bvecs; | |
2270 | ||
2271 | while (!vec->bv_page) { | |
2272 | vec++; | |
2273 | vcnt--; | |
2274 | } | |
2275 | ||
2276 | wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev); | |
2277 | memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec)); | |
2278 | ||
2279 | wbio->bi_vcnt = vcnt; | |
2280 | } else { | |
2281 | wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
2282 | } | |
2283 | ||
796a5cf0 | 2284 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4f024f37 KO |
2285 | wbio->bi_iter.bi_sector = r1_bio->sector; |
2286 | wbio->bi_iter.bi_size = r1_bio->sectors << 9; | |
cd5ff9a1 | 2287 | |
6678d83f | 2288 | bio_trim(wbio, sector - r1_bio->sector, sectors); |
4f024f37 | 2289 | wbio->bi_iter.bi_sector += rdev->data_offset; |
cd5ff9a1 | 2290 | wbio->bi_bdev = rdev->bdev; |
4e49ea4a MC |
2291 | |
2292 | if (submit_bio_wait(wbio) < 0) | |
cd5ff9a1 N |
2293 | /* failure! */ |
2294 | ok = rdev_set_badblocks(rdev, sector, | |
2295 | sectors, 0) | |
2296 | && ok; | |
2297 | ||
2298 | bio_put(wbio); | |
2299 | sect_to_write -= sectors; | |
2300 | sector += sectors; | |
2301 | sectors = block_sectors; | |
2302 | } | |
2303 | return ok; | |
2304 | } | |
2305 | ||
e8096360 | 2306 | static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2307 | { |
2308 | int m; | |
2309 | int s = r1_bio->sectors; | |
8f19ccb2 | 2310 | for (m = 0; m < conf->raid_disks * 2 ; m++) { |
3cb03002 | 2311 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2312 | struct bio *bio = r1_bio->bios[m]; |
2313 | if (bio->bi_end_io == NULL) | |
2314 | continue; | |
4246a0b6 | 2315 | if (!bio->bi_error && |
62096bce | 2316 | test_bit(R1BIO_MadeGood, &r1_bio->state)) { |
c6563a8c | 2317 | rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); |
62096bce | 2318 | } |
4246a0b6 | 2319 | if (bio->bi_error && |
62096bce N |
2320 | test_bit(R1BIO_WriteError, &r1_bio->state)) { |
2321 | if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) | |
2322 | md_error(conf->mddev, rdev); | |
2323 | } | |
2324 | } | |
2325 | put_buf(r1_bio); | |
2326 | md_done_sync(conf->mddev, s, 1); | |
2327 | } | |
2328 | ||
e8096360 | 2329 | static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2330 | { |
2331 | int m; | |
55ce74d4 | 2332 | bool fail = false; |
8f19ccb2 | 2333 | for (m = 0; m < conf->raid_disks * 2 ; m++) |
62096bce | 2334 | if (r1_bio->bios[m] == IO_MADE_GOOD) { |
3cb03002 | 2335 | struct md_rdev *rdev = conf->mirrors[m].rdev; |
62096bce N |
2336 | rdev_clear_badblocks(rdev, |
2337 | r1_bio->sector, | |
c6563a8c | 2338 | r1_bio->sectors, 0); |
62096bce N |
2339 | rdev_dec_pending(rdev, conf->mddev); |
2340 | } else if (r1_bio->bios[m] != NULL) { | |
2341 | /* This drive got a write error. We need to | |
2342 | * narrow down and record precise write | |
2343 | * errors. | |
2344 | */ | |
55ce74d4 | 2345 | fail = true; |
62096bce N |
2346 | if (!narrow_write_error(r1_bio, m)) { |
2347 | md_error(conf->mddev, | |
2348 | conf->mirrors[m].rdev); | |
2349 | /* an I/O failed, we can't clear the bitmap */ | |
2350 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2351 | } | |
2352 | rdev_dec_pending(conf->mirrors[m].rdev, | |
2353 | conf->mddev); | |
2354 | } | |
55ce74d4 N |
2355 | if (fail) { |
2356 | spin_lock_irq(&conf->device_lock); | |
2357 | list_add(&r1_bio->retry_list, &conf->bio_end_io_list); | |
ccfc7bf1 | 2358 | conf->nr_queued++; |
55ce74d4 N |
2359 | spin_unlock_irq(&conf->device_lock); |
2360 | md_wakeup_thread(conf->mddev->thread); | |
bd8688a1 N |
2361 | } else { |
2362 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2363 | close_write(r1_bio); | |
55ce74d4 | 2364 | raid_end_bio_io(r1_bio); |
bd8688a1 | 2365 | } |
62096bce N |
2366 | } |
2367 | ||
e8096360 | 2368 | static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio) |
62096bce N |
2369 | { |
2370 | int disk; | |
2371 | int max_sectors; | |
fd01b88c | 2372 | struct mddev *mddev = conf->mddev; |
62096bce N |
2373 | struct bio *bio; |
2374 | char b[BDEVNAME_SIZE]; | |
3cb03002 | 2375 | struct md_rdev *rdev; |
109e3765 N |
2376 | dev_t bio_dev; |
2377 | sector_t bio_sector; | |
62096bce N |
2378 | |
2379 | clear_bit(R1BIO_ReadError, &r1_bio->state); | |
2380 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2381 | * the block and we can fix it. | |
2382 | * We freeze all other IO, and try reading the block from | |
2383 | * other devices. When we find one, we re-write | |
2384 | * and check it that fixes the read error. | |
2385 | * This is all done synchronously while the array is | |
2386 | * frozen | |
2387 | */ | |
7449f699 TM |
2388 | |
2389 | bio = r1_bio->bios[r1_bio->read_disk]; | |
2390 | bdevname(bio->bi_bdev, b); | |
109e3765 N |
2391 | bio_dev = bio->bi_bdev->bd_dev; |
2392 | bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector; | |
7449f699 TM |
2393 | bio_put(bio); |
2394 | r1_bio->bios[r1_bio->read_disk] = NULL; | |
2395 | ||
2e52d449 N |
2396 | rdev = conf->mirrors[r1_bio->read_disk].rdev; |
2397 | if (mddev->ro == 0 | |
2398 | && !test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2399 | freeze_array(conf, 1); |
62096bce N |
2400 | fix_read_error(conf, r1_bio->read_disk, |
2401 | r1_bio->sector, r1_bio->sectors); | |
2402 | unfreeze_array(conf); | |
7449f699 TM |
2403 | } else { |
2404 | r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; | |
2405 | } | |
2406 | ||
2e52d449 | 2407 | rdev_dec_pending(rdev, conf->mddev); |
62096bce | 2408 | |
62096bce N |
2409 | read_more: |
2410 | disk = read_balance(conf, r1_bio, &max_sectors); | |
2411 | if (disk == -1) { | |
1d41c216 N |
2412 | pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n", |
2413 | mdname(mddev), b, (unsigned long long)r1_bio->sector); | |
62096bce N |
2414 | raid_end_bio_io(r1_bio); |
2415 | } else { | |
2416 | const unsigned long do_sync | |
1eff9d32 | 2417 | = r1_bio->master_bio->bi_opf & REQ_SYNC; |
62096bce N |
2418 | r1_bio->read_disk = disk; |
2419 | bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev); | |
4f024f37 KO |
2420 | bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector, |
2421 | max_sectors); | |
62096bce N |
2422 | r1_bio->bios[r1_bio->read_disk] = bio; |
2423 | rdev = conf->mirrors[disk].rdev; | |
1d41c216 N |
2424 | pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", |
2425 | mdname(mddev), | |
2426 | (unsigned long long)r1_bio->sector, | |
2427 | bdevname(rdev->bdev, b)); | |
4f024f37 | 2428 | bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset; |
62096bce N |
2429 | bio->bi_bdev = rdev->bdev; |
2430 | bio->bi_end_io = raid1_end_read_request; | |
796a5cf0 | 2431 | bio_set_op_attrs(bio, REQ_OP_READ, do_sync); |
2e52d449 N |
2432 | if (test_bit(FailFast, &rdev->flags) && |
2433 | test_bit(R1BIO_FailFast, &r1_bio->state)) | |
2434 | bio->bi_opf |= MD_FAILFAST; | |
62096bce N |
2435 | bio->bi_private = r1_bio; |
2436 | if (max_sectors < r1_bio->sectors) { | |
2437 | /* Drat - have to split this up more */ | |
2438 | struct bio *mbio = r1_bio->master_bio; | |
2439 | int sectors_handled = (r1_bio->sector + max_sectors | |
4f024f37 | 2440 | - mbio->bi_iter.bi_sector); |
62096bce N |
2441 | r1_bio->sectors = max_sectors; |
2442 | spin_lock_irq(&conf->device_lock); | |
2443 | if (mbio->bi_phys_segments == 0) | |
2444 | mbio->bi_phys_segments = 2; | |
2445 | else | |
2446 | mbio->bi_phys_segments++; | |
2447 | spin_unlock_irq(&conf->device_lock); | |
109e3765 N |
2448 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), |
2449 | bio, bio_dev, bio_sector); | |
62096bce N |
2450 | generic_make_request(bio); |
2451 | bio = NULL; | |
2452 | ||
2453 | r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); | |
2454 | ||
2455 | r1_bio->master_bio = mbio; | |
aa8b57aa | 2456 | r1_bio->sectors = bio_sectors(mbio) - sectors_handled; |
62096bce N |
2457 | r1_bio->state = 0; |
2458 | set_bit(R1BIO_ReadError, &r1_bio->state); | |
2459 | r1_bio->mddev = mddev; | |
4f024f37 KO |
2460 | r1_bio->sector = mbio->bi_iter.bi_sector + |
2461 | sectors_handled; | |
62096bce N |
2462 | |
2463 | goto read_more; | |
109e3765 N |
2464 | } else { |
2465 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), | |
2466 | bio, bio_dev, bio_sector); | |
62096bce | 2467 | generic_make_request(bio); |
109e3765 | 2468 | } |
62096bce N |
2469 | } |
2470 | } | |
2471 | ||
4ed8731d | 2472 | static void raid1d(struct md_thread *thread) |
1da177e4 | 2473 | { |
4ed8731d | 2474 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2475 | struct r1bio *r1_bio; |
1da177e4 | 2476 | unsigned long flags; |
e8096360 | 2477 | struct r1conf *conf = mddev->private; |
1da177e4 | 2478 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2479 | struct blk_plug plug; |
1da177e4 LT |
2480 | |
2481 | md_check_recovery(mddev); | |
e1dfa0a2 | 2482 | |
55ce74d4 | 2483 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2484 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
55ce74d4 N |
2485 | LIST_HEAD(tmp); |
2486 | spin_lock_irqsave(&conf->device_lock, flags); | |
2953079c | 2487 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
ccfc7bf1 ND |
2488 | while (!list_empty(&conf->bio_end_io_list)) { |
2489 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2490 | conf->nr_queued--; | |
2491 | } | |
55ce74d4 N |
2492 | } |
2493 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2494 | while (!list_empty(&tmp)) { | |
a452744b MP |
2495 | r1_bio = list_first_entry(&tmp, struct r1bio, |
2496 | retry_list); | |
55ce74d4 | 2497 | list_del(&r1_bio->retry_list); |
bd8688a1 N |
2498 | if (mddev->degraded) |
2499 | set_bit(R1BIO_Degraded, &r1_bio->state); | |
2500 | if (test_bit(R1BIO_WriteError, &r1_bio->state)) | |
2501 | close_write(r1_bio); | |
55ce74d4 N |
2502 | raid_end_bio_io(r1_bio); |
2503 | } | |
2504 | } | |
2505 | ||
e1dfa0a2 | 2506 | blk_start_plug(&plug); |
1da177e4 | 2507 | for (;;) { |
191ea9b2 | 2508 | |
0021b7bc | 2509 | flush_pending_writes(conf); |
191ea9b2 | 2510 | |
a35e63ef N |
2511 | spin_lock_irqsave(&conf->device_lock, flags); |
2512 | if (list_empty(head)) { | |
2513 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2514 | break; |
a35e63ef | 2515 | } |
9f2c9d12 | 2516 | r1_bio = list_entry(head->prev, struct r1bio, retry_list); |
1da177e4 | 2517 | list_del(head->prev); |
ddaf22ab | 2518 | conf->nr_queued--; |
1da177e4 LT |
2519 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2520 | ||
2521 | mddev = r1_bio->mddev; | |
070ec55d | 2522 | conf = mddev->private; |
4367af55 | 2523 | if (test_bit(R1BIO_IsSync, &r1_bio->state)) { |
d8f05d29 | 2524 | if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2525 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2526 | handle_sync_write_finished(conf, r1_bio); | |
2527 | else | |
4367af55 | 2528 | sync_request_write(mddev, r1_bio); |
cd5ff9a1 | 2529 | } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || |
62096bce N |
2530 | test_bit(R1BIO_WriteError, &r1_bio->state)) |
2531 | handle_write_finished(conf, r1_bio); | |
2532 | else if (test_bit(R1BIO_ReadError, &r1_bio->state)) | |
2533 | handle_read_error(conf, r1_bio); | |
2534 | else | |
d2eb35ac N |
2535 | /* just a partial read to be scheduled from separate |
2536 | * context | |
2537 | */ | |
2538 | generic_make_request(r1_bio->bios[r1_bio->read_disk]); | |
62096bce | 2539 | |
1d9d5241 | 2540 | cond_resched(); |
2953079c | 2541 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2542 | md_check_recovery(mddev); |
1da177e4 | 2543 | } |
e1dfa0a2 | 2544 | blk_finish_plug(&plug); |
1da177e4 LT |
2545 | } |
2546 | ||
e8096360 | 2547 | static int init_resync(struct r1conf *conf) |
1da177e4 LT |
2548 | { |
2549 | int buffs; | |
2550 | ||
2551 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
9e77c485 | 2552 | BUG_ON(conf->r1buf_pool); |
1da177e4 LT |
2553 | conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, |
2554 | conf->poolinfo); | |
2555 | if (!conf->r1buf_pool) | |
2556 | return -ENOMEM; | |
2557 | conf->next_resync = 0; | |
2558 | return 0; | |
2559 | } | |
2560 | ||
2561 | /* | |
2562 | * perform a "sync" on one "block" | |
2563 | * | |
2564 | * We need to make sure that no normal I/O request - particularly write | |
2565 | * requests - conflict with active sync requests. | |
2566 | * | |
2567 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2568 | * that can be installed to exclude normal IO requests. | |
2569 | */ | |
2570 | ||
849674e4 SL |
2571 | static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr, |
2572 | int *skipped) | |
1da177e4 | 2573 | { |
e8096360 | 2574 | struct r1conf *conf = mddev->private; |
9f2c9d12 | 2575 | struct r1bio *r1_bio; |
1da177e4 LT |
2576 | struct bio *bio; |
2577 | sector_t max_sector, nr_sectors; | |
3e198f78 | 2578 | int disk = -1; |
1da177e4 | 2579 | int i; |
3e198f78 N |
2580 | int wonly = -1; |
2581 | int write_targets = 0, read_targets = 0; | |
57dab0bd | 2582 | sector_t sync_blocks; |
e3b9703e | 2583 | int still_degraded = 0; |
06f60385 N |
2584 | int good_sectors = RESYNC_SECTORS; |
2585 | int min_bad = 0; /* number of sectors that are bad in all devices */ | |
1da177e4 LT |
2586 | |
2587 | if (!conf->r1buf_pool) | |
2588 | if (init_resync(conf)) | |
57afd89f | 2589 | return 0; |
1da177e4 | 2590 | |
58c0fed4 | 2591 | max_sector = mddev->dev_sectors; |
1da177e4 | 2592 | if (sector_nr >= max_sector) { |
191ea9b2 N |
2593 | /* If we aborted, we need to abort the |
2594 | * sync on the 'current' bitmap chunk (there will | |
2595 | * only be one in raid1 resync. | |
2596 | * We can find the current addess in mddev->curr_resync | |
2597 | */ | |
6a806c51 N |
2598 | if (mddev->curr_resync < max_sector) /* aborted */ |
2599 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
191ea9b2 | 2600 | &sync_blocks, 1); |
6a806c51 | 2601 | else /* completed sync */ |
191ea9b2 | 2602 | conf->fullsync = 0; |
6a806c51 N |
2603 | |
2604 | bitmap_close_sync(mddev->bitmap); | |
1da177e4 | 2605 | close_sync(conf); |
c40f341f GR |
2606 | |
2607 | if (mddev_is_clustered(mddev)) { | |
2608 | conf->cluster_sync_low = 0; | |
2609 | conf->cluster_sync_high = 0; | |
c40f341f | 2610 | } |
1da177e4 LT |
2611 | return 0; |
2612 | } | |
2613 | ||
07d84d10 N |
2614 | if (mddev->bitmap == NULL && |
2615 | mddev->recovery_cp == MaxSector && | |
6394cca5 | 2616 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
07d84d10 N |
2617 | conf->fullsync == 0) { |
2618 | *skipped = 1; | |
2619 | return max_sector - sector_nr; | |
2620 | } | |
6394cca5 N |
2621 | /* before building a request, check if we can skip these blocks.. |
2622 | * This call the bitmap_start_sync doesn't actually record anything | |
2623 | */ | |
e3b9703e | 2624 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
e5de485f | 2625 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { |
191ea9b2 N |
2626 | /* We can skip this block, and probably several more */ |
2627 | *skipped = 1; | |
2628 | return sync_blocks; | |
2629 | } | |
17999be4 | 2630 | |
7ac50447 TM |
2631 | /* |
2632 | * If there is non-resync activity waiting for a turn, then let it | |
2633 | * though before starting on this new sync request. | |
2634 | */ | |
2635 | if (conf->nr_waiting) | |
2636 | schedule_timeout_uninterruptible(1); | |
2637 | ||
c40f341f GR |
2638 | /* we are incrementing sector_nr below. To be safe, we check against |
2639 | * sector_nr + two times RESYNC_SECTORS | |
2640 | */ | |
2641 | ||
2642 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
2643 | mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
1c4588e9 | 2644 | r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); |
17999be4 | 2645 | |
c2fd4c94 | 2646 | raise_barrier(conf, sector_nr); |
1da177e4 | 2647 | |
3e198f78 | 2648 | rcu_read_lock(); |
1da177e4 | 2649 | /* |
3e198f78 N |
2650 | * If we get a correctably read error during resync or recovery, |
2651 | * we might want to read from a different device. So we | |
2652 | * flag all drives that could conceivably be read from for READ, | |
2653 | * and any others (which will be non-In_sync devices) for WRITE. | |
2654 | * If a read fails, we try reading from something else for which READ | |
2655 | * is OK. | |
1da177e4 | 2656 | */ |
1da177e4 | 2657 | |
1da177e4 LT |
2658 | r1_bio->mddev = mddev; |
2659 | r1_bio->sector = sector_nr; | |
191ea9b2 | 2660 | r1_bio->state = 0; |
1da177e4 | 2661 | set_bit(R1BIO_IsSync, &r1_bio->state); |
1da177e4 | 2662 | |
8f19ccb2 | 2663 | for (i = 0; i < conf->raid_disks * 2; i++) { |
3cb03002 | 2664 | struct md_rdev *rdev; |
1da177e4 | 2665 | bio = r1_bio->bios[i]; |
2aabaa65 | 2666 | bio_reset(bio); |
1da177e4 | 2667 | |
3e198f78 N |
2668 | rdev = rcu_dereference(conf->mirrors[i].rdev); |
2669 | if (rdev == NULL || | |
06f60385 | 2670 | test_bit(Faulty, &rdev->flags)) { |
8f19ccb2 N |
2671 | if (i < conf->raid_disks) |
2672 | still_degraded = 1; | |
3e198f78 | 2673 | } else if (!test_bit(In_sync, &rdev->flags)) { |
796a5cf0 | 2674 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1da177e4 LT |
2675 | bio->bi_end_io = end_sync_write; |
2676 | write_targets ++; | |
3e198f78 N |
2677 | } else { |
2678 | /* may need to read from here */ | |
06f60385 N |
2679 | sector_t first_bad = MaxSector; |
2680 | int bad_sectors; | |
2681 | ||
2682 | if (is_badblock(rdev, sector_nr, good_sectors, | |
2683 | &first_bad, &bad_sectors)) { | |
2684 | if (first_bad > sector_nr) | |
2685 | good_sectors = first_bad - sector_nr; | |
2686 | else { | |
2687 | bad_sectors -= (sector_nr - first_bad); | |
2688 | if (min_bad == 0 || | |
2689 | min_bad > bad_sectors) | |
2690 | min_bad = bad_sectors; | |
2691 | } | |
2692 | } | |
2693 | if (sector_nr < first_bad) { | |
2694 | if (test_bit(WriteMostly, &rdev->flags)) { | |
2695 | if (wonly < 0) | |
2696 | wonly = i; | |
2697 | } else { | |
2698 | if (disk < 0) | |
2699 | disk = i; | |
2700 | } | |
796a5cf0 | 2701 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
06f60385 N |
2702 | bio->bi_end_io = end_sync_read; |
2703 | read_targets++; | |
d57368af AL |
2704 | } else if (!test_bit(WriteErrorSeen, &rdev->flags) && |
2705 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
2706 | !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { | |
2707 | /* | |
2708 | * The device is suitable for reading (InSync), | |
2709 | * but has bad block(s) here. Let's try to correct them, | |
2710 | * if we are doing resync or repair. Otherwise, leave | |
2711 | * this device alone for this sync request. | |
2712 | */ | |
796a5cf0 | 2713 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
d57368af AL |
2714 | bio->bi_end_io = end_sync_write; |
2715 | write_targets++; | |
3e198f78 | 2716 | } |
3e198f78 | 2717 | } |
06f60385 N |
2718 | if (bio->bi_end_io) { |
2719 | atomic_inc(&rdev->nr_pending); | |
4f024f37 | 2720 | bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; |
06f60385 N |
2721 | bio->bi_bdev = rdev->bdev; |
2722 | bio->bi_private = r1_bio; | |
2e52d449 N |
2723 | if (test_bit(FailFast, &rdev->flags)) |
2724 | bio->bi_opf |= MD_FAILFAST; | |
06f60385 | 2725 | } |
1da177e4 | 2726 | } |
3e198f78 N |
2727 | rcu_read_unlock(); |
2728 | if (disk < 0) | |
2729 | disk = wonly; | |
2730 | r1_bio->read_disk = disk; | |
191ea9b2 | 2731 | |
06f60385 N |
2732 | if (read_targets == 0 && min_bad > 0) { |
2733 | /* These sectors are bad on all InSync devices, so we | |
2734 | * need to mark them bad on all write targets | |
2735 | */ | |
2736 | int ok = 1; | |
8f19ccb2 | 2737 | for (i = 0 ; i < conf->raid_disks * 2 ; i++) |
06f60385 | 2738 | if (r1_bio->bios[i]->bi_end_io == end_sync_write) { |
a42f9d83 | 2739 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
06f60385 N |
2740 | ok = rdev_set_badblocks(rdev, sector_nr, |
2741 | min_bad, 0 | |
2742 | ) && ok; | |
2743 | } | |
2953079c | 2744 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
06f60385 N |
2745 | *skipped = 1; |
2746 | put_buf(r1_bio); | |
2747 | ||
2748 | if (!ok) { | |
2749 | /* Cannot record the badblocks, so need to | |
2750 | * abort the resync. | |
2751 | * If there are multiple read targets, could just | |
2752 | * fail the really bad ones ??? | |
2753 | */ | |
2754 | conf->recovery_disabled = mddev->recovery_disabled; | |
2755 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2756 | return 0; | |
2757 | } else | |
2758 | return min_bad; | |
2759 | ||
2760 | } | |
2761 | if (min_bad > 0 && min_bad < good_sectors) { | |
2762 | /* only resync enough to reach the next bad->good | |
2763 | * transition */ | |
2764 | good_sectors = min_bad; | |
2765 | } | |
2766 | ||
3e198f78 N |
2767 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) |
2768 | /* extra read targets are also write targets */ | |
2769 | write_targets += read_targets-1; | |
2770 | ||
2771 | if (write_targets == 0 || read_targets == 0) { | |
1da177e4 LT |
2772 | /* There is nowhere to write, so all non-sync |
2773 | * drives must be failed - so we are finished | |
2774 | */ | |
b7219ccb N |
2775 | sector_t rv; |
2776 | if (min_bad > 0) | |
2777 | max_sector = sector_nr + min_bad; | |
2778 | rv = max_sector - sector_nr; | |
57afd89f | 2779 | *skipped = 1; |
1da177e4 | 2780 | put_buf(r1_bio); |
1da177e4 LT |
2781 | return rv; |
2782 | } | |
2783 | ||
c6207277 N |
2784 | if (max_sector > mddev->resync_max) |
2785 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
06f60385 N |
2786 | if (max_sector > sector_nr + good_sectors) |
2787 | max_sector = sector_nr + good_sectors; | |
1da177e4 | 2788 | nr_sectors = 0; |
289e99e8 | 2789 | sync_blocks = 0; |
1da177e4 LT |
2790 | do { |
2791 | struct page *page; | |
2792 | int len = PAGE_SIZE; | |
2793 | if (sector_nr + (len>>9) > max_sector) | |
2794 | len = (max_sector - sector_nr) << 9; | |
2795 | if (len == 0) | |
2796 | break; | |
6a806c51 N |
2797 | if (sync_blocks == 0) { |
2798 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, | |
e5de485f N |
2799 | &sync_blocks, still_degraded) && |
2800 | !conf->fullsync && | |
2801 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) | |
6a806c51 | 2802 | break; |
7571ae88 | 2803 | if ((len >> 9) > sync_blocks) |
6a806c51 | 2804 | len = sync_blocks<<9; |
ab7a30c7 | 2805 | } |
191ea9b2 | 2806 | |
8f19ccb2 | 2807 | for (i = 0 ; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2808 | bio = r1_bio->bios[i]; |
2809 | if (bio->bi_end_io) { | |
d11c171e | 2810 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
1da177e4 LT |
2811 | if (bio_add_page(bio, page, len, 0) == 0) { |
2812 | /* stop here */ | |
d11c171e | 2813 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; |
1da177e4 LT |
2814 | while (i > 0) { |
2815 | i--; | |
2816 | bio = r1_bio->bios[i]; | |
6a806c51 N |
2817 | if (bio->bi_end_io==NULL) |
2818 | continue; | |
1da177e4 LT |
2819 | /* remove last page from this bio */ |
2820 | bio->bi_vcnt--; | |
4f024f37 | 2821 | bio->bi_iter.bi_size -= len; |
b7c44ed9 | 2822 | bio_clear_flag(bio, BIO_SEG_VALID); |
1da177e4 LT |
2823 | } |
2824 | goto bio_full; | |
2825 | } | |
2826 | } | |
2827 | } | |
2828 | nr_sectors += len>>9; | |
2829 | sector_nr += len>>9; | |
191ea9b2 | 2830 | sync_blocks -= (len>>9); |
1da177e4 LT |
2831 | } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); |
2832 | bio_full: | |
1da177e4 LT |
2833 | r1_bio->sectors = nr_sectors; |
2834 | ||
c40f341f GR |
2835 | if (mddev_is_clustered(mddev) && |
2836 | conf->cluster_sync_high < sector_nr + nr_sectors) { | |
2837 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
2838 | conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; | |
2839 | /* Send resync message */ | |
2840 | md_cluster_ops->resync_info_update(mddev, | |
2841 | conf->cluster_sync_low, | |
2842 | conf->cluster_sync_high); | |
2843 | } | |
2844 | ||
d11c171e N |
2845 | /* For a user-requested sync, we read all readable devices and do a |
2846 | * compare | |
2847 | */ | |
2848 | if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { | |
2849 | atomic_set(&r1_bio->remaining, read_targets); | |
2d4f4f33 | 2850 | for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { |
d11c171e N |
2851 | bio = r1_bio->bios[i]; |
2852 | if (bio->bi_end_io == end_sync_read) { | |
2d4f4f33 | 2853 | read_targets--; |
ddac7c7e | 2854 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2855 | if (read_targets == 1) |
2856 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e N |
2857 | generic_make_request(bio); |
2858 | } | |
2859 | } | |
2860 | } else { | |
2861 | atomic_set(&r1_bio->remaining, 1); | |
2862 | bio = r1_bio->bios[r1_bio->read_disk]; | |
ddac7c7e | 2863 | md_sync_acct(bio->bi_bdev, nr_sectors); |
2e52d449 N |
2864 | if (read_targets == 1) |
2865 | bio->bi_opf &= ~MD_FAILFAST; | |
d11c171e | 2866 | generic_make_request(bio); |
1da177e4 | 2867 | |
d11c171e | 2868 | } |
1da177e4 LT |
2869 | return nr_sectors; |
2870 | } | |
2871 | ||
fd01b88c | 2872 | static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
2873 | { |
2874 | if (sectors) | |
2875 | return sectors; | |
2876 | ||
2877 | return mddev->dev_sectors; | |
2878 | } | |
2879 | ||
e8096360 | 2880 | static struct r1conf *setup_conf(struct mddev *mddev) |
1da177e4 | 2881 | { |
e8096360 | 2882 | struct r1conf *conf; |
709ae487 | 2883 | int i; |
0eaf822c | 2884 | struct raid1_info *disk; |
3cb03002 | 2885 | struct md_rdev *rdev; |
709ae487 | 2886 | int err = -ENOMEM; |
1da177e4 | 2887 | |
e8096360 | 2888 | conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL); |
1da177e4 | 2889 | if (!conf) |
709ae487 | 2890 | goto abort; |
1da177e4 | 2891 | |
0eaf822c | 2892 | conf->mirrors = kzalloc(sizeof(struct raid1_info) |
8f19ccb2 | 2893 | * mddev->raid_disks * 2, |
1da177e4 LT |
2894 | GFP_KERNEL); |
2895 | if (!conf->mirrors) | |
709ae487 | 2896 | goto abort; |
1da177e4 | 2897 | |
ddaf22ab N |
2898 | conf->tmppage = alloc_page(GFP_KERNEL); |
2899 | if (!conf->tmppage) | |
709ae487 | 2900 | goto abort; |
ddaf22ab | 2901 | |
709ae487 | 2902 | conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); |
1da177e4 | 2903 | if (!conf->poolinfo) |
709ae487 | 2904 | goto abort; |
8f19ccb2 | 2905 | conf->poolinfo->raid_disks = mddev->raid_disks * 2; |
1da177e4 LT |
2906 | conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, |
2907 | r1bio_pool_free, | |
2908 | conf->poolinfo); | |
2909 | if (!conf->r1bio_pool) | |
709ae487 N |
2910 | goto abort; |
2911 | ||
ed9bfdf1 | 2912 | conf->poolinfo->mddev = mddev; |
1da177e4 | 2913 | |
c19d5798 | 2914 | err = -EINVAL; |
e7e72bf6 | 2915 | spin_lock_init(&conf->device_lock); |
dafb20fa | 2916 | rdev_for_each(rdev, mddev) { |
aba336bd | 2917 | struct request_queue *q; |
709ae487 | 2918 | int disk_idx = rdev->raid_disk; |
1da177e4 LT |
2919 | if (disk_idx >= mddev->raid_disks |
2920 | || disk_idx < 0) | |
2921 | continue; | |
c19d5798 | 2922 | if (test_bit(Replacement, &rdev->flags)) |
02b898f2 | 2923 | disk = conf->mirrors + mddev->raid_disks + disk_idx; |
c19d5798 N |
2924 | else |
2925 | disk = conf->mirrors + disk_idx; | |
1da177e4 | 2926 | |
c19d5798 N |
2927 | if (disk->rdev) |
2928 | goto abort; | |
1da177e4 | 2929 | disk->rdev = rdev; |
aba336bd | 2930 | q = bdev_get_queue(rdev->bdev); |
1da177e4 LT |
2931 | |
2932 | disk->head_position = 0; | |
12cee5a8 | 2933 | disk->seq_start = MaxSector; |
1da177e4 LT |
2934 | } |
2935 | conf->raid_disks = mddev->raid_disks; | |
2936 | conf->mddev = mddev; | |
1da177e4 | 2937 | INIT_LIST_HEAD(&conf->retry_list); |
55ce74d4 | 2938 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
1da177e4 LT |
2939 | |
2940 | spin_lock_init(&conf->resync_lock); | |
17999be4 | 2941 | init_waitqueue_head(&conf->wait_barrier); |
1da177e4 | 2942 | |
191ea9b2 | 2943 | bio_list_init(&conf->pending_bio_list); |
34db0cd6 | 2944 | conf->pending_count = 0; |
d890fa2b | 2945 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
191ea9b2 | 2946 | |
79ef3a8a | 2947 | conf->start_next_window = MaxSector; |
2948 | conf->current_window_requests = conf->next_window_requests = 0; | |
2949 | ||
c19d5798 | 2950 | err = -EIO; |
8f19ccb2 | 2951 | for (i = 0; i < conf->raid_disks * 2; i++) { |
1da177e4 LT |
2952 | |
2953 | disk = conf->mirrors + i; | |
2954 | ||
c19d5798 N |
2955 | if (i < conf->raid_disks && |
2956 | disk[conf->raid_disks].rdev) { | |
2957 | /* This slot has a replacement. */ | |
2958 | if (!disk->rdev) { | |
2959 | /* No original, just make the replacement | |
2960 | * a recovering spare | |
2961 | */ | |
2962 | disk->rdev = | |
2963 | disk[conf->raid_disks].rdev; | |
2964 | disk[conf->raid_disks].rdev = NULL; | |
2965 | } else if (!test_bit(In_sync, &disk->rdev->flags)) | |
2966 | /* Original is not in_sync - bad */ | |
2967 | goto abort; | |
2968 | } | |
2969 | ||
5fd6c1dc N |
2970 | if (!disk->rdev || |
2971 | !test_bit(In_sync, &disk->rdev->flags)) { | |
1da177e4 | 2972 | disk->head_position = 0; |
4f0a5e01 JB |
2973 | if (disk->rdev && |
2974 | (disk->rdev->saved_raid_disk < 0)) | |
918f0238 | 2975 | conf->fullsync = 1; |
be4d3280 | 2976 | } |
1da177e4 | 2977 | } |
709ae487 | 2978 | |
709ae487 | 2979 | err = -ENOMEM; |
0232605d | 2980 | conf->thread = md_register_thread(raid1d, mddev, "raid1"); |
1d41c216 | 2981 | if (!conf->thread) |
709ae487 | 2982 | goto abort; |
1da177e4 | 2983 | |
709ae487 N |
2984 | return conf; |
2985 | ||
2986 | abort: | |
2987 | if (conf) { | |
644df1a8 | 2988 | mempool_destroy(conf->r1bio_pool); |
709ae487 N |
2989 | kfree(conf->mirrors); |
2990 | safe_put_page(conf->tmppage); | |
2991 | kfree(conf->poolinfo); | |
2992 | kfree(conf); | |
2993 | } | |
2994 | return ERR_PTR(err); | |
2995 | } | |
2996 | ||
afa0f557 | 2997 | static void raid1_free(struct mddev *mddev, void *priv); |
849674e4 | 2998 | static int raid1_run(struct mddev *mddev) |
709ae487 | 2999 | { |
e8096360 | 3000 | struct r1conf *conf; |
709ae487 | 3001 | int i; |
3cb03002 | 3002 | struct md_rdev *rdev; |
5220ea1e | 3003 | int ret; |
2ff8cc2c | 3004 | bool discard_supported = false; |
709ae487 N |
3005 | |
3006 | if (mddev->level != 1) { | |
1d41c216 N |
3007 | pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n", |
3008 | mdname(mddev), mddev->level); | |
709ae487 N |
3009 | return -EIO; |
3010 | } | |
3011 | if (mddev->reshape_position != MaxSector) { | |
1d41c216 N |
3012 | pr_warn("md/raid1:%s: reshape_position set but not supported\n", |
3013 | mdname(mddev)); | |
709ae487 N |
3014 | return -EIO; |
3015 | } | |
1da177e4 | 3016 | /* |
709ae487 N |
3017 | * copy the already verified devices into our private RAID1 |
3018 | * bookkeeping area. [whatever we allocate in run(), | |
afa0f557 | 3019 | * should be freed in raid1_free()] |
1da177e4 | 3020 | */ |
709ae487 N |
3021 | if (mddev->private == NULL) |
3022 | conf = setup_conf(mddev); | |
3023 | else | |
3024 | conf = mddev->private; | |
1da177e4 | 3025 | |
709ae487 N |
3026 | if (IS_ERR(conf)) |
3027 | return PTR_ERR(conf); | |
1da177e4 | 3028 | |
c8dc9c65 | 3029 | if (mddev->queue) |
5026d7a9 PA |
3030 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3031 | ||
dafb20fa | 3032 | rdev_for_each(rdev, mddev) { |
1ed7242e JB |
3033 | if (!mddev->gendisk) |
3034 | continue; | |
709ae487 N |
3035 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
3036 | rdev->data_offset << 9); | |
2ff8cc2c SL |
3037 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
3038 | discard_supported = true; | |
1da177e4 | 3039 | } |
191ea9b2 | 3040 | |
709ae487 N |
3041 | mddev->degraded = 0; |
3042 | for (i=0; i < conf->raid_disks; i++) | |
3043 | if (conf->mirrors[i].rdev == NULL || | |
3044 | !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || | |
3045 | test_bit(Faulty, &conf->mirrors[i].rdev->flags)) | |
3046 | mddev->degraded++; | |
3047 | ||
3048 | if (conf->raid_disks - mddev->degraded == 1) | |
3049 | mddev->recovery_cp = MaxSector; | |
3050 | ||
8c6ac868 | 3051 | if (mddev->recovery_cp != MaxSector) |
1d41c216 N |
3052 | pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", |
3053 | mdname(mddev)); | |
3054 | pr_info("md/raid1:%s: active with %d out of %d mirrors\n", | |
f72ffdd6 | 3055 | mdname(mddev), mddev->raid_disks - mddev->degraded, |
1da177e4 | 3056 | mddev->raid_disks); |
709ae487 | 3057 | |
1da177e4 LT |
3058 | /* |
3059 | * Ok, everything is just fine now | |
3060 | */ | |
709ae487 N |
3061 | mddev->thread = conf->thread; |
3062 | conf->thread = NULL; | |
3063 | mddev->private = conf; | |
46533ff7 | 3064 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
709ae487 | 3065 | |
1f403624 | 3066 | md_set_array_sectors(mddev, raid1_size(mddev, 0, 0)); |
1da177e4 | 3067 | |
1ed7242e | 3068 | if (mddev->queue) { |
2ff8cc2c SL |
3069 | if (discard_supported) |
3070 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3071 | mddev->queue); | |
3072 | else | |
3073 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3074 | mddev->queue); | |
1ed7242e | 3075 | } |
5220ea1e | 3076 | |
3077 | ret = md_integrity_register(mddev); | |
5aa61f42 N |
3078 | if (ret) { |
3079 | md_unregister_thread(&mddev->thread); | |
afa0f557 | 3080 | raid1_free(mddev, conf); |
5aa61f42 | 3081 | } |
5220ea1e | 3082 | return ret; |
1da177e4 LT |
3083 | } |
3084 | ||
afa0f557 | 3085 | static void raid1_free(struct mddev *mddev, void *priv) |
1da177e4 | 3086 | { |
afa0f557 | 3087 | struct r1conf *conf = priv; |
409c57f3 | 3088 | |
644df1a8 | 3089 | mempool_destroy(conf->r1bio_pool); |
990a8baf | 3090 | kfree(conf->mirrors); |
0fea7ed8 | 3091 | safe_put_page(conf->tmppage); |
990a8baf | 3092 | kfree(conf->poolinfo); |
1da177e4 | 3093 | kfree(conf); |
1da177e4 LT |
3094 | } |
3095 | ||
fd01b88c | 3096 | static int raid1_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
3097 | { |
3098 | /* no resync is happening, and there is enough space | |
3099 | * on all devices, so we can resize. | |
3100 | * We need to make sure resync covers any new space. | |
3101 | * If the array is shrinking we should possibly wait until | |
3102 | * any io in the removed space completes, but it hardly seems | |
3103 | * worth it. | |
3104 | */ | |
a4a6125a N |
3105 | sector_t newsize = raid1_size(mddev, sectors, 0); |
3106 | if (mddev->external_size && | |
3107 | mddev->array_sectors > newsize) | |
b522adcd | 3108 | return -EINVAL; |
a4a6125a N |
3109 | if (mddev->bitmap) { |
3110 | int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
3111 | if (ret) | |
3112 | return ret; | |
3113 | } | |
3114 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 3115 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 3116 | revalidate_disk(mddev->gendisk); |
b522adcd | 3117 | if (sectors > mddev->dev_sectors && |
b098636c | 3118 | mddev->recovery_cp > mddev->dev_sectors) { |
58c0fed4 | 3119 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
3120 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3121 | } | |
b522adcd | 3122 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 3123 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3124 | return 0; |
3125 | } | |
3126 | ||
fd01b88c | 3127 | static int raid1_reshape(struct mddev *mddev) |
1da177e4 LT |
3128 | { |
3129 | /* We need to: | |
3130 | * 1/ resize the r1bio_pool | |
3131 | * 2/ resize conf->mirrors | |
3132 | * | |
3133 | * We allocate a new r1bio_pool if we can. | |
3134 | * Then raise a device barrier and wait until all IO stops. | |
3135 | * Then resize conf->mirrors and swap in the new r1bio pool. | |
6ea9c07c N |
3136 | * |
3137 | * At the same time, we "pack" the devices so that all the missing | |
3138 | * devices have the higher raid_disk numbers. | |
1da177e4 LT |
3139 | */ |
3140 | mempool_t *newpool, *oldpool; | |
3141 | struct pool_info *newpoolinfo; | |
0eaf822c | 3142 | struct raid1_info *newmirrors; |
e8096360 | 3143 | struct r1conf *conf = mddev->private; |
63c70c4f | 3144 | int cnt, raid_disks; |
c04be0aa | 3145 | unsigned long flags; |
b5470dc5 | 3146 | int d, d2, err; |
1da177e4 | 3147 | |
63c70c4f | 3148 | /* Cannot change chunk_size, layout, or level */ |
664e7c41 | 3149 | if (mddev->chunk_sectors != mddev->new_chunk_sectors || |
63c70c4f N |
3150 | mddev->layout != mddev->new_layout || |
3151 | mddev->level != mddev->new_level) { | |
664e7c41 | 3152 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
63c70c4f N |
3153 | mddev->new_layout = mddev->layout; |
3154 | mddev->new_level = mddev->level; | |
3155 | return -EINVAL; | |
3156 | } | |
3157 | ||
28c1b9fd GR |
3158 | if (!mddev_is_clustered(mddev)) { |
3159 | err = md_allow_write(mddev); | |
3160 | if (err) | |
3161 | return err; | |
3162 | } | |
2a2275d6 | 3163 | |
63c70c4f N |
3164 | raid_disks = mddev->raid_disks + mddev->delta_disks; |
3165 | ||
6ea9c07c N |
3166 | if (raid_disks < conf->raid_disks) { |
3167 | cnt=0; | |
3168 | for (d= 0; d < conf->raid_disks; d++) | |
3169 | if (conf->mirrors[d].rdev) | |
3170 | cnt++; | |
3171 | if (cnt > raid_disks) | |
1da177e4 | 3172 | return -EBUSY; |
6ea9c07c | 3173 | } |
1da177e4 LT |
3174 | |
3175 | newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); | |
3176 | if (!newpoolinfo) | |
3177 | return -ENOMEM; | |
3178 | newpoolinfo->mddev = mddev; | |
8f19ccb2 | 3179 | newpoolinfo->raid_disks = raid_disks * 2; |
1da177e4 LT |
3180 | |
3181 | newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, | |
3182 | r1bio_pool_free, newpoolinfo); | |
3183 | if (!newpool) { | |
3184 | kfree(newpoolinfo); | |
3185 | return -ENOMEM; | |
3186 | } | |
0eaf822c | 3187 | newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2, |
8f19ccb2 | 3188 | GFP_KERNEL); |
1da177e4 LT |
3189 | if (!newmirrors) { |
3190 | kfree(newpoolinfo); | |
3191 | mempool_destroy(newpool); | |
3192 | return -ENOMEM; | |
3193 | } | |
1da177e4 | 3194 | |
e2d59925 | 3195 | freeze_array(conf, 0); |
1da177e4 LT |
3196 | |
3197 | /* ok, everything is stopped */ | |
3198 | oldpool = conf->r1bio_pool; | |
3199 | conf->r1bio_pool = newpool; | |
6ea9c07c | 3200 | |
a88aa786 | 3201 | for (d = d2 = 0; d < conf->raid_disks; d++) { |
3cb03002 | 3202 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
a88aa786 | 3203 | if (rdev && rdev->raid_disk != d2) { |
36fad858 | 3204 | sysfs_unlink_rdev(mddev, rdev); |
a88aa786 | 3205 | rdev->raid_disk = d2; |
36fad858 NK |
3206 | sysfs_unlink_rdev(mddev, rdev); |
3207 | if (sysfs_link_rdev(mddev, rdev)) | |
1d41c216 N |
3208 | pr_warn("md/raid1:%s: cannot register rd%d\n", |
3209 | mdname(mddev), rdev->raid_disk); | |
6ea9c07c | 3210 | } |
a88aa786 N |
3211 | if (rdev) |
3212 | newmirrors[d2++].rdev = rdev; | |
3213 | } | |
1da177e4 LT |
3214 | kfree(conf->mirrors); |
3215 | conf->mirrors = newmirrors; | |
3216 | kfree(conf->poolinfo); | |
3217 | conf->poolinfo = newpoolinfo; | |
3218 | ||
c04be0aa | 3219 | spin_lock_irqsave(&conf->device_lock, flags); |
1da177e4 | 3220 | mddev->degraded += (raid_disks - conf->raid_disks); |
c04be0aa | 3221 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 3222 | conf->raid_disks = mddev->raid_disks = raid_disks; |
63c70c4f | 3223 | mddev->delta_disks = 0; |
1da177e4 | 3224 | |
e2d59925 | 3225 | unfreeze_array(conf); |
1da177e4 | 3226 | |
985ca973 | 3227 | set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); |
1da177e4 LT |
3228 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3229 | md_wakeup_thread(mddev->thread); | |
3230 | ||
3231 | mempool_destroy(oldpool); | |
3232 | return 0; | |
3233 | } | |
3234 | ||
fd01b88c | 3235 | static void raid1_quiesce(struct mddev *mddev, int state) |
36fa3063 | 3236 | { |
e8096360 | 3237 | struct r1conf *conf = mddev->private; |
36fa3063 N |
3238 | |
3239 | switch(state) { | |
6eef4b21 N |
3240 | case 2: /* wake for suspend */ |
3241 | wake_up(&conf->wait_barrier); | |
3242 | break; | |
9e6603da | 3243 | case 1: |
07169fd4 | 3244 | freeze_array(conf, 0); |
36fa3063 | 3245 | break; |
9e6603da | 3246 | case 0: |
07169fd4 | 3247 | unfreeze_array(conf); |
36fa3063 N |
3248 | break; |
3249 | } | |
36fa3063 N |
3250 | } |
3251 | ||
fd01b88c | 3252 | static void *raid1_takeover(struct mddev *mddev) |
709ae487 N |
3253 | { |
3254 | /* raid1 can take over: | |
3255 | * raid5 with 2 devices, any layout or chunk size | |
3256 | */ | |
3257 | if (mddev->level == 5 && mddev->raid_disks == 2) { | |
e8096360 | 3258 | struct r1conf *conf; |
709ae487 N |
3259 | mddev->new_level = 1; |
3260 | mddev->new_layout = 0; | |
3261 | mddev->new_chunk_sectors = 0; | |
3262 | conf = setup_conf(mddev); | |
6995f0b2 | 3263 | if (!IS_ERR(conf)) { |
07169fd4 | 3264 | /* Array must appear to be quiesced */ |
3265 | conf->array_frozen = 1; | |
394ed8e4 SL |
3266 | mddev_clear_unsupported_flags(mddev, |
3267 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 3268 | } |
709ae487 N |
3269 | return conf; |
3270 | } | |
3271 | return ERR_PTR(-EINVAL); | |
3272 | } | |
1da177e4 | 3273 | |
84fc4b56 | 3274 | static struct md_personality raid1_personality = |
1da177e4 LT |
3275 | { |
3276 | .name = "raid1", | |
2604b703 | 3277 | .level = 1, |
1da177e4 | 3278 | .owner = THIS_MODULE, |
849674e4 SL |
3279 | .make_request = raid1_make_request, |
3280 | .run = raid1_run, | |
afa0f557 | 3281 | .free = raid1_free, |
849674e4 SL |
3282 | .status = raid1_status, |
3283 | .error_handler = raid1_error, | |
1da177e4 LT |
3284 | .hot_add_disk = raid1_add_disk, |
3285 | .hot_remove_disk= raid1_remove_disk, | |
3286 | .spare_active = raid1_spare_active, | |
849674e4 | 3287 | .sync_request = raid1_sync_request, |
1da177e4 | 3288 | .resize = raid1_resize, |
80c3a6ce | 3289 | .size = raid1_size, |
63c70c4f | 3290 | .check_reshape = raid1_reshape, |
36fa3063 | 3291 | .quiesce = raid1_quiesce, |
709ae487 | 3292 | .takeover = raid1_takeover, |
5c675f83 | 3293 | .congested = raid1_congested, |
1da177e4 LT |
3294 | }; |
3295 | ||
3296 | static int __init raid_init(void) | |
3297 | { | |
2604b703 | 3298 | return register_md_personality(&raid1_personality); |
1da177e4 LT |
3299 | } |
3300 | ||
3301 | static void raid_exit(void) | |
3302 | { | |
2604b703 | 3303 | unregister_md_personality(&raid1_personality); |
1da177e4 LT |
3304 | } |
3305 | ||
3306 | module_init(raid_init); | |
3307 | module_exit(raid_exit); | |
3308 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 3309 | MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD"); |
1da177e4 | 3310 | MODULE_ALIAS("md-personality-3"); /* RAID1 */ |
d9d166c2 | 3311 | MODULE_ALIAS("md-raid1"); |
2604b703 | 3312 | MODULE_ALIAS("md-level-1"); |
34db0cd6 N |
3313 | |
3314 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |