]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid10.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 2000-2004 Neil Brown | |
5 | * | |
6 | * RAID-10 support for md. | |
7 | * | |
25985edc | 8 | * Base on code in raid1.c. See raid1.c for further copyright information. |
1da177e4 LT |
9 | * |
10 | * | |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
5a0e3ad6 | 21 | #include <linux/slab.h> |
25570727 | 22 | #include <linux/delay.h> |
bff61975 | 23 | #include <linux/blkdev.h> |
056075c7 | 24 | #include <linux/module.h> |
bff61975 | 25 | #include <linux/seq_file.h> |
8bda470e | 26 | #include <linux/ratelimit.h> |
3ea7daa5 | 27 | #include <linux/kthread.h> |
109e3765 | 28 | #include <trace/events/block.h> |
43b2e5d8 | 29 | #include "md.h" |
ef740c37 | 30 | #include "raid10.h" |
dab8b292 | 31 | #include "raid0.h" |
935fe098 | 32 | #include "md-bitmap.h" |
1da177e4 LT |
33 | |
34 | /* | |
35 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
36 | * The layout of data is defined by | |
37 | * chunk_size | |
38 | * raid_disks | |
39 | * near_copies (stored in low byte of layout) | |
40 | * far_copies (stored in second byte of layout) | |
c93983bf | 41 | * far_offset (stored in bit 16 of layout ) |
475901af | 42 | * use_far_sets (stored in bit 17 of layout ) |
8bce6d35 | 43 | * use_far_sets_bugfixed (stored in bit 18 of layout ) |
1da177e4 | 44 | * |
475901af JB |
45 | * The data to be stored is divided into chunks using chunksize. Each device |
46 | * is divided into far_copies sections. In each section, chunks are laid out | |
47 | * in a style similar to raid0, but near_copies copies of each chunk is stored | |
48 | * (each on a different drive). The starting device for each section is offset | |
49 | * near_copies from the starting device of the previous section. Thus there | |
50 | * are (near_copies * far_copies) of each chunk, and each is on a different | |
51 | * drive. near_copies and far_copies must be at least one, and their product | |
52 | * is at most raid_disks. | |
c93983bf N |
53 | * |
54 | * If far_offset is true, then the far_copies are handled a bit differently. | |
475901af JB |
55 | * The copies are still in different stripes, but instead of being very far |
56 | * apart on disk, there are adjacent stripes. | |
57 | * | |
58 | * The far and offset algorithms are handled slightly differently if | |
59 | * 'use_far_sets' is true. In this case, the array's devices are grouped into | |
60 | * sets that are (near_copies * far_copies) in size. The far copied stripes | |
61 | * are still shifted by 'near_copies' devices, but this shifting stays confined | |
62 | * to the set rather than the entire array. This is done to improve the number | |
63 | * of device combinations that can fail without causing the array to fail. | |
64 | * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk | |
65 | * on a device): | |
66 | * A B C D A B C D E | |
67 | * ... ... | |
68 | * D A B C E A B C D | |
69 | * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s): | |
70 | * [A B] [C D] [A B] [C D E] | |
71 | * |...| |...| |...| | ... | | |
72 | * [B A] [D C] [B A] [E C D] | |
1da177e4 LT |
73 | */ |
74 | ||
75 | /* | |
76 | * Number of guaranteed r10bios in case of extreme VM load: | |
77 | */ | |
78 | #define NR_RAID10_BIOS 256 | |
79 | ||
473e87ce JB |
80 | /* when we get a read error on a read-only array, we redirect to another |
81 | * device without failing the first device, or trying to over-write to | |
82 | * correct the read error. To keep track of bad blocks on a per-bio | |
83 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
84 | */ | |
85 | #define IO_BLOCKED ((struct bio *)1) | |
86 | /* When we successfully write to a known bad-block, we need to remove the | |
87 | * bad-block marking which must be done from process context. So we record | |
88 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
89 | */ | |
90 | #define IO_MADE_GOOD ((struct bio *)2) | |
91 | ||
92 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
93 | ||
94 | /* When there are this many requests queued to be written by | |
34db0cd6 N |
95 | * the raid10 thread, we become 'congested' to provide back-pressure |
96 | * for writeback. | |
97 | */ | |
98 | static int max_queued_requests = 1024; | |
99 | ||
e879a879 N |
100 | static void allow_barrier(struct r10conf *conf); |
101 | static void lower_barrier(struct r10conf *conf); | |
635f6416 | 102 | static int _enough(struct r10conf *conf, int previous, int ignore); |
1919cbb2 | 103 | static int enough(struct r10conf *conf, int ignore); |
3ea7daa5 N |
104 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, |
105 | int *skipped); | |
106 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio); | |
4246a0b6 | 107 | static void end_reshape_write(struct bio *bio); |
3ea7daa5 | 108 | static void end_reshape(struct r10conf *conf); |
0a27ec96 | 109 | |
578b54ad N |
110 | #define raid10_log(md, fmt, args...) \ |
111 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid10 " fmt, ##args); } while (0) | |
112 | ||
fb0eb5df ML |
113 | #include "raid1-10.c" |
114 | ||
f0250618 ML |
115 | /* |
116 | * for resync bio, r10bio pointer can be retrieved from the per-bio | |
117 | * 'struct resync_pages'. | |
118 | */ | |
119 | static inline struct r10bio *get_resync_r10bio(struct bio *bio) | |
120 | { | |
121 | return get_resync_pages(bio)->raid_bio; | |
122 | } | |
123 | ||
dd0fc66f | 124 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 125 | { |
e879a879 | 126 | struct r10conf *conf = data; |
9f2c9d12 | 127 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 128 | |
69335ef3 N |
129 | /* allocate a r10bio with room for raid_disks entries in the |
130 | * bios array */ | |
7eaceacc | 131 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
132 | } |
133 | ||
134 | static void r10bio_pool_free(void *r10_bio, void *data) | |
135 | { | |
136 | kfree(r10_bio); | |
137 | } | |
138 | ||
8db87912 | 139 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
0310fa21 N |
140 | /* amount of memory to reserve for resync requests */ |
141 | #define RESYNC_WINDOW (1024*1024) | |
142 | /* maximum number of concurrent requests, memory permitting */ | |
143 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
8db87912 GJ |
144 | #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW) |
145 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) | |
1da177e4 LT |
146 | |
147 | /* | |
148 | * When performing a resync, we need to read and compare, so | |
149 | * we need as many pages are there are copies. | |
150 | * When performing a recovery, we need 2 bios, one for read, | |
151 | * one for write (we recover only one drive per r10buf) | |
152 | * | |
153 | */ | |
dd0fc66f | 154 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 155 | { |
e879a879 | 156 | struct r10conf *conf = data; |
9f2c9d12 | 157 | struct r10bio *r10_bio; |
1da177e4 | 158 | struct bio *bio; |
f0250618 ML |
159 | int j; |
160 | int nalloc, nalloc_rp; | |
161 | struct resync_pages *rps; | |
1da177e4 LT |
162 | |
163 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 164 | if (!r10_bio) |
1da177e4 | 165 | return NULL; |
1da177e4 | 166 | |
3ea7daa5 N |
167 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || |
168 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
1da177e4 LT |
169 | nalloc = conf->copies; /* resync */ |
170 | else | |
171 | nalloc = 2; /* recovery */ | |
172 | ||
f0250618 ML |
173 | /* allocate once for all bios */ |
174 | if (!conf->have_replacement) | |
175 | nalloc_rp = nalloc; | |
176 | else | |
177 | nalloc_rp = nalloc * 2; | |
178 | rps = kmalloc(sizeof(struct resync_pages) * nalloc_rp, gfp_flags); | |
179 | if (!rps) | |
180 | goto out_free_r10bio; | |
181 | ||
1da177e4 LT |
182 | /* |
183 | * Allocate bios. | |
184 | */ | |
185 | for (j = nalloc ; j-- ; ) { | |
6746557f | 186 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
187 | if (!bio) |
188 | goto out_free_bio; | |
189 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
190 | if (!conf->have_replacement) |
191 | continue; | |
192 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
193 | if (!bio) | |
194 | goto out_free_bio; | |
195 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
196 | } |
197 | /* | |
198 | * Allocate RESYNC_PAGES data pages and attach them | |
199 | * where needed. | |
200 | */ | |
f0250618 | 201 | for (j = 0; j < nalloc; j++) { |
69335ef3 | 202 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
f0250618 ML |
203 | struct resync_pages *rp, *rp_repl; |
204 | ||
205 | rp = &rps[j]; | |
206 | if (rbio) | |
207 | rp_repl = &rps[nalloc + j]; | |
208 | ||
1da177e4 | 209 | bio = r10_bio->devs[j].bio; |
f0250618 ML |
210 | |
211 | if (!j || test_bit(MD_RECOVERY_SYNC, | |
212 | &conf->mddev->recovery)) { | |
213 | if (resync_alloc_pages(rp, gfp_flags)) | |
1da177e4 | 214 | goto out_free_pages; |
f0250618 ML |
215 | } else { |
216 | memcpy(rp, &rps[0], sizeof(*rp)); | |
217 | resync_get_all_pages(rp); | |
218 | } | |
1da177e4 | 219 | |
f0250618 ML |
220 | rp->raid_bio = r10_bio; |
221 | bio->bi_private = rp; | |
222 | if (rbio) { | |
223 | memcpy(rp_repl, rp, sizeof(*rp)); | |
224 | rbio->bi_private = rp_repl; | |
1da177e4 LT |
225 | } |
226 | } | |
227 | ||
228 | return r10_bio; | |
229 | ||
230 | out_free_pages: | |
f0250618 ML |
231 | while (--j >= 0) |
232 | resync_free_pages(&rps[j * 2]); | |
233 | ||
5fdd2cf8 | 234 | j = 0; |
1da177e4 | 235 | out_free_bio: |
5fdd2cf8 | 236 | for ( ; j < nalloc; j++) { |
237 | if (r10_bio->devs[j].bio) | |
238 | bio_put(r10_bio->devs[j].bio); | |
69335ef3 N |
239 | if (r10_bio->devs[j].repl_bio) |
240 | bio_put(r10_bio->devs[j].repl_bio); | |
241 | } | |
f0250618 ML |
242 | kfree(rps); |
243 | out_free_r10bio: | |
1da177e4 LT |
244 | r10bio_pool_free(r10_bio, conf); |
245 | return NULL; | |
246 | } | |
247 | ||
248 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
249 | { | |
e879a879 | 250 | struct r10conf *conf = data; |
9f2c9d12 | 251 | struct r10bio *r10bio = __r10_bio; |
1da177e4 | 252 | int j; |
f0250618 | 253 | struct resync_pages *rp = NULL; |
1da177e4 | 254 | |
f0250618 | 255 | for (j = conf->copies; j--; ) { |
1da177e4 | 256 | struct bio *bio = r10bio->devs[j].bio; |
f0250618 ML |
257 | |
258 | rp = get_resync_pages(bio); | |
259 | resync_free_pages(rp); | |
260 | bio_put(bio); | |
261 | ||
69335ef3 N |
262 | bio = r10bio->devs[j].repl_bio; |
263 | if (bio) | |
264 | bio_put(bio); | |
1da177e4 | 265 | } |
f0250618 ML |
266 | |
267 | /* resync pages array stored in the 1st bio's .bi_private */ | |
268 | kfree(rp); | |
269 | ||
1da177e4 LT |
270 | r10bio_pool_free(r10bio, conf); |
271 | } | |
272 | ||
e879a879 | 273 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
274 | { |
275 | int i; | |
276 | ||
277 | for (i = 0; i < conf->copies; i++) { | |
278 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 279 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
280 | bio_put(*bio); |
281 | *bio = NULL; | |
69335ef3 N |
282 | bio = &r10_bio->devs[i].repl_bio; |
283 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
284 | bio_put(*bio); | |
285 | *bio = NULL; | |
1da177e4 LT |
286 | } |
287 | } | |
288 | ||
9f2c9d12 | 289 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 290 | { |
e879a879 | 291 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 292 | |
1da177e4 LT |
293 | put_all_bios(conf, r10_bio); |
294 | mempool_free(r10_bio, conf->r10bio_pool); | |
295 | } | |
296 | ||
9f2c9d12 | 297 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 298 | { |
e879a879 | 299 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
300 | |
301 | mempool_free(r10_bio, conf->r10buf_pool); | |
302 | ||
0a27ec96 | 303 | lower_barrier(conf); |
1da177e4 LT |
304 | } |
305 | ||
9f2c9d12 | 306 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
307 | { |
308 | unsigned long flags; | |
fd01b88c | 309 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 310 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
311 | |
312 | spin_lock_irqsave(&conf->device_lock, flags); | |
313 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 314 | conf->nr_queued ++; |
1da177e4 LT |
315 | spin_unlock_irqrestore(&conf->device_lock, flags); |
316 | ||
388667be AJ |
317 | /* wake up frozen array... */ |
318 | wake_up(&conf->wait_barrier); | |
319 | ||
1da177e4 LT |
320 | md_wakeup_thread(mddev->thread); |
321 | } | |
322 | ||
323 | /* | |
324 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
325 | * operation and are ready to return a success/failure code to the buffer | |
326 | * cache layer. | |
327 | */ | |
9f2c9d12 | 328 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
329 | { |
330 | struct bio *bio = r10_bio->master_bio; | |
e879a879 | 331 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 332 | |
856e08e2 | 333 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) |
4e4cbee9 | 334 | bio->bi_status = BLK_STS_IOERR; |
fd16f2e8 N |
335 | |
336 | bio_endio(bio); | |
337 | /* | |
338 | * Wake up any possible resync thread that waits for the device | |
339 | * to go idle. | |
340 | */ | |
341 | allow_barrier(conf); | |
342 | ||
1da177e4 LT |
343 | free_r10bio(r10_bio); |
344 | } | |
345 | ||
346 | /* | |
347 | * Update disk head position estimator based on IRQ completion info. | |
348 | */ | |
9f2c9d12 | 349 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 350 | { |
e879a879 | 351 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
352 | |
353 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
354 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
355 | } | |
356 | ||
778ca018 NK |
357 | /* |
358 | * Find the disk number which triggered given bio | |
359 | */ | |
e879a879 | 360 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 361 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
362 | { |
363 | int slot; | |
69335ef3 | 364 | int repl = 0; |
778ca018 | 365 | |
69335ef3 | 366 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
367 | if (r10_bio->devs[slot].bio == bio) |
368 | break; | |
69335ef3 N |
369 | if (r10_bio->devs[slot].repl_bio == bio) { |
370 | repl = 1; | |
371 | break; | |
372 | } | |
373 | } | |
778ca018 NK |
374 | |
375 | BUG_ON(slot == conf->copies); | |
376 | update_head_pos(slot, r10_bio); | |
377 | ||
749c55e9 N |
378 | if (slotp) |
379 | *slotp = slot; | |
69335ef3 N |
380 | if (replp) |
381 | *replp = repl; | |
778ca018 NK |
382 | return r10_bio->devs[slot].devnum; |
383 | } | |
384 | ||
4246a0b6 | 385 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 386 | { |
4e4cbee9 | 387 | int uptodate = !bio->bi_status; |
9f2c9d12 | 388 | struct r10bio *r10_bio = bio->bi_private; |
a0e764c5 | 389 | int slot; |
abbf098e | 390 | struct md_rdev *rdev; |
e879a879 | 391 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 392 | |
1da177e4 | 393 | slot = r10_bio->read_slot; |
abbf098e | 394 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
395 | /* |
396 | * this branch is our 'one mirror IO has finished' event handler: | |
397 | */ | |
4443ae10 N |
398 | update_head_pos(slot, r10_bio); |
399 | ||
400 | if (uptodate) { | |
1da177e4 LT |
401 | /* |
402 | * Set R10BIO_Uptodate in our master bio, so that | |
403 | * we will return a good error code to the higher | |
404 | * levels even if IO on some other mirrored buffer fails. | |
405 | * | |
406 | * The 'master' represents the composite IO operation to | |
407 | * user-side. So if something waits for IO, then it will | |
408 | * wait for the 'master' bio. | |
409 | */ | |
410 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
411 | } else { |
412 | /* If all other devices that store this block have | |
413 | * failed, we want to return the error upwards rather | |
414 | * than fail the last device. Here we redefine | |
415 | * "uptodate" to mean "Don't want to retry" | |
416 | */ | |
635f6416 N |
417 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
418 | rdev->raid_disk)) | |
fae8cc5e | 419 | uptodate = 1; |
fae8cc5e N |
420 | } |
421 | if (uptodate) { | |
1da177e4 | 422 | raid_end_bio_io(r10_bio); |
abbf098e | 423 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 424 | } else { |
1da177e4 | 425 | /* |
7c4e06ff | 426 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
427 | */ |
428 | char b[BDEVNAME_SIZE]; | |
08464e09 | 429 | pr_err_ratelimited("md/raid10:%s: %s: rescheduling sector %llu\n", |
8bda470e | 430 | mdname(conf->mddev), |
abbf098e | 431 | bdevname(rdev->bdev, b), |
8bda470e | 432 | (unsigned long long)r10_bio->sector); |
856e08e2 | 433 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
434 | reschedule_retry(r10_bio); |
435 | } | |
1da177e4 LT |
436 | } |
437 | ||
9f2c9d12 | 438 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
439 | { |
440 | /* clear the bitmap if all writes complete successfully */ | |
441 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
442 | r10_bio->sectors, | |
443 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
444 | 0); | |
445 | md_write_end(r10_bio->mddev); | |
446 | } | |
447 | ||
9f2c9d12 | 448 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
449 | { |
450 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
451 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
452 | reschedule_retry(r10_bio); | |
453 | else { | |
454 | close_write(r10_bio); | |
455 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
456 | reschedule_retry(r10_bio); | |
457 | else | |
458 | raid_end_bio_io(r10_bio); | |
459 | } | |
460 | } | |
461 | } | |
462 | ||
4246a0b6 | 463 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 464 | { |
9f2c9d12 | 465 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 466 | int dev; |
749c55e9 | 467 | int dec_rdev = 1; |
e879a879 | 468 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 469 | int slot, repl; |
4ca40c2c | 470 | struct md_rdev *rdev = NULL; |
1919cbb2 | 471 | struct bio *to_put = NULL; |
579ed34f SL |
472 | bool discard_error; |
473 | ||
4e4cbee9 | 474 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 475 | |
475b0321 | 476 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 477 | |
475b0321 N |
478 | if (repl) |
479 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
480 | if (!rdev) { |
481 | smp_rmb(); | |
482 | repl = 0; | |
475b0321 | 483 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 484 | } |
1da177e4 LT |
485 | /* |
486 | * this branch is our 'one mirror IO has finished' event handler: | |
487 | */ | |
4e4cbee9 | 488 | if (bio->bi_status && !discard_error) { |
475b0321 N |
489 | if (repl) |
490 | /* Never record new bad blocks to replacement, | |
491 | * just fail it. | |
492 | */ | |
493 | md_error(rdev->mddev, rdev); | |
494 | else { | |
495 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
496 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
497 | set_bit(MD_RECOVERY_NEEDED, | |
498 | &rdev->mddev->recovery); | |
1919cbb2 | 499 | |
475b0321 | 500 | dec_rdev = 0; |
1919cbb2 N |
501 | if (test_bit(FailFast, &rdev->flags) && |
502 | (bio->bi_opf & MD_FAILFAST)) { | |
503 | md_error(rdev->mddev, rdev); | |
504 | if (!test_bit(Faulty, &rdev->flags)) | |
505 | /* This is the only remaining device, | |
506 | * We need to retry the write without | |
507 | * FailFast | |
508 | */ | |
509 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
510 | else { | |
511 | r10_bio->devs[slot].bio = NULL; | |
512 | to_put = bio; | |
513 | dec_rdev = 1; | |
514 | } | |
515 | } else | |
516 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
475b0321 | 517 | } |
749c55e9 | 518 | } else { |
1da177e4 LT |
519 | /* |
520 | * Set R10BIO_Uptodate in our master bio, so that | |
521 | * we will return a good error code for to the higher | |
522 | * levels even if IO on some other mirrored buffer fails. | |
523 | * | |
524 | * The 'master' represents the composite IO operation to | |
525 | * user-side. So if something waits for IO, then it will | |
526 | * wait for the 'master' bio. | |
527 | */ | |
749c55e9 N |
528 | sector_t first_bad; |
529 | int bad_sectors; | |
530 | ||
3056e3ae AL |
531 | /* |
532 | * Do not set R10BIO_Uptodate if the current device is | |
533 | * rebuilding or Faulty. This is because we cannot use | |
534 | * such device for properly reading the data back (we could | |
535 | * potentially use it, if the current write would have felt | |
536 | * before rdev->recovery_offset, but for simplicity we don't | |
537 | * check this here. | |
538 | */ | |
539 | if (test_bit(In_sync, &rdev->flags) && | |
540 | !test_bit(Faulty, &rdev->flags)) | |
541 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 542 | |
749c55e9 | 543 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 544 | if (is_badblock(rdev, |
749c55e9 N |
545 | r10_bio->devs[slot].addr, |
546 | r10_bio->sectors, | |
579ed34f | 547 | &first_bad, &bad_sectors) && !discard_error) { |
749c55e9 | 548 | bio_put(bio); |
475b0321 N |
549 | if (repl) |
550 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
551 | else | |
552 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
553 | dec_rdev = 0; |
554 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
555 | } | |
556 | } | |
557 | ||
1da177e4 LT |
558 | /* |
559 | * | |
560 | * Let's see if all mirrored write operations have finished | |
561 | * already. | |
562 | */ | |
19d5f834 | 563 | one_write_done(r10_bio); |
749c55e9 | 564 | if (dec_rdev) |
884162df | 565 | rdev_dec_pending(rdev, conf->mddev); |
1919cbb2 N |
566 | if (to_put) |
567 | bio_put(to_put); | |
1da177e4 LT |
568 | } |
569 | ||
1da177e4 LT |
570 | /* |
571 | * RAID10 layout manager | |
25985edc | 572 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
573 | * parameters: near_copies and far_copies. |
574 | * near_copies * far_copies must be <= raid_disks. | |
575 | * Normally one of these will be 1. | |
576 | * If both are 1, we get raid0. | |
577 | * If near_copies == raid_disks, we get raid1. | |
578 | * | |
25985edc | 579 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
580 | * first chunk, followed by near_copies copies of the next chunk and |
581 | * so on. | |
582 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
583 | * as described above, we start again with a device offset of near_copies. | |
584 | * So we effectively have another copy of the whole array further down all | |
585 | * the drives, but with blocks on different drives. | |
586 | * With this layout, and block is never stored twice on the one device. | |
587 | * | |
588 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 589 | * on each device that it is on. |
1da177e4 LT |
590 | * |
591 | * raid10_find_virt does the reverse mapping, from a device and a | |
592 | * sector offset to a virtual address | |
593 | */ | |
594 | ||
f8c9e74f | 595 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
596 | { |
597 | int n,f; | |
598 | sector_t sector; | |
599 | sector_t chunk; | |
600 | sector_t stripe; | |
601 | int dev; | |
1da177e4 | 602 | int slot = 0; |
9a3152ab JB |
603 | int last_far_set_start, last_far_set_size; |
604 | ||
605 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
606 | last_far_set_start *= geo->far_set_size; | |
607 | ||
608 | last_far_set_size = geo->far_set_size; | |
609 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
610 | |
611 | /* now calculate first sector/dev */ | |
5cf00fcd N |
612 | chunk = r10bio->sector >> geo->chunk_shift; |
613 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 614 | |
5cf00fcd | 615 | chunk *= geo->near_copies; |
1da177e4 | 616 | stripe = chunk; |
5cf00fcd N |
617 | dev = sector_div(stripe, geo->raid_disks); |
618 | if (geo->far_offset) | |
619 | stripe *= geo->far_copies; | |
1da177e4 | 620 | |
5cf00fcd | 621 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
622 | |
623 | /* and calculate all the others */ | |
5cf00fcd | 624 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 625 | int d = dev; |
475901af | 626 | int set; |
1da177e4 | 627 | sector_t s = sector; |
1da177e4 | 628 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 629 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
630 | slot++; |
631 | ||
5cf00fcd | 632 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 633 | set = d / geo->far_set_size; |
5cf00fcd | 634 | d += geo->near_copies; |
475901af | 635 | |
9a3152ab JB |
636 | if ((geo->raid_disks % geo->far_set_size) && |
637 | (d > last_far_set_start)) { | |
638 | d -= last_far_set_start; | |
639 | d %= last_far_set_size; | |
640 | d += last_far_set_start; | |
641 | } else { | |
642 | d %= geo->far_set_size; | |
643 | d += geo->far_set_size * set; | |
644 | } | |
5cf00fcd | 645 | s += geo->stride; |
1da177e4 LT |
646 | r10bio->devs[slot].devnum = d; |
647 | r10bio->devs[slot].addr = s; | |
648 | slot++; | |
649 | } | |
650 | dev++; | |
5cf00fcd | 651 | if (dev >= geo->raid_disks) { |
1da177e4 | 652 | dev = 0; |
5cf00fcd | 653 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
654 | } |
655 | } | |
f8c9e74f N |
656 | } |
657 | ||
658 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
659 | { | |
660 | struct geom *geo = &conf->geo; | |
661 | ||
662 | if (conf->reshape_progress != MaxSector && | |
663 | ((r10bio->sector >= conf->reshape_progress) != | |
664 | conf->mddev->reshape_backwards)) { | |
665 | set_bit(R10BIO_Previous, &r10bio->state); | |
666 | geo = &conf->prev; | |
667 | } else | |
668 | clear_bit(R10BIO_Previous, &r10bio->state); | |
669 | ||
670 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
671 | } |
672 | ||
e879a879 | 673 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
674 | { |
675 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
676 | /* Never use conf->prev as this is only called during resync |
677 | * or recovery, so reshape isn't happening | |
678 | */ | |
5cf00fcd | 679 | struct geom *geo = &conf->geo; |
475901af JB |
680 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
681 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
682 | int last_far_set_start; |
683 | ||
684 | if (geo->raid_disks % geo->far_set_size) { | |
685 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
686 | last_far_set_start *= geo->far_set_size; | |
687 | ||
688 | if (dev >= last_far_set_start) { | |
689 | far_set_size = geo->far_set_size; | |
690 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
691 | far_set_start = last_far_set_start; | |
692 | } | |
693 | } | |
1da177e4 | 694 | |
5cf00fcd N |
695 | offset = sector & geo->chunk_mask; |
696 | if (geo->far_offset) { | |
c93983bf | 697 | int fc; |
5cf00fcd N |
698 | chunk = sector >> geo->chunk_shift; |
699 | fc = sector_div(chunk, geo->far_copies); | |
700 | dev -= fc * geo->near_copies; | |
475901af JB |
701 | if (dev < far_set_start) |
702 | dev += far_set_size; | |
c93983bf | 703 | } else { |
5cf00fcd N |
704 | while (sector >= geo->stride) { |
705 | sector -= geo->stride; | |
475901af JB |
706 | if (dev < (geo->near_copies + far_set_start)) |
707 | dev += far_set_size - geo->near_copies; | |
c93983bf | 708 | else |
5cf00fcd | 709 | dev -= geo->near_copies; |
c93983bf | 710 | } |
5cf00fcd | 711 | chunk = sector >> geo->chunk_shift; |
c93983bf | 712 | } |
5cf00fcd N |
713 | vchunk = chunk * geo->raid_disks + dev; |
714 | sector_div(vchunk, geo->near_copies); | |
715 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
716 | } |
717 | ||
1da177e4 LT |
718 | /* |
719 | * This routine returns the disk from which the requested read should | |
720 | * be done. There is a per-array 'next expected sequential IO' sector | |
721 | * number - if this matches on the next IO then we use the last disk. | |
722 | * There is also a per-disk 'last know head position' sector that is | |
723 | * maintained from IRQ contexts, both the normal and the resync IO | |
724 | * completion handlers update this position correctly. If there is no | |
725 | * perfect sequential match then we pick the disk whose head is closest. | |
726 | * | |
727 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
728 | * because position is mirror, not device based. | |
729 | * | |
730 | * The rdev for the device selected will have nr_pending incremented. | |
731 | */ | |
732 | ||
733 | /* | |
734 | * FIXME: possibly should rethink readbalancing and do it differently | |
735 | * depending on near_copies / far_copies geometry. | |
736 | */ | |
96c3fd1f N |
737 | static struct md_rdev *read_balance(struct r10conf *conf, |
738 | struct r10bio *r10_bio, | |
739 | int *max_sectors) | |
1da177e4 | 740 | { |
af3a2cd6 | 741 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 742 | int disk, slot; |
856e08e2 N |
743 | int sectors = r10_bio->sectors; |
744 | int best_good_sectors; | |
56d99121 | 745 | sector_t new_distance, best_dist; |
3bbae04b | 746 | struct md_rdev *best_rdev, *rdev = NULL; |
56d99121 N |
747 | int do_balance; |
748 | int best_slot; | |
5cf00fcd | 749 | struct geom *geo = &conf->geo; |
1da177e4 LT |
750 | |
751 | raid10_find_phys(conf, r10_bio); | |
752 | rcu_read_lock(); | |
56d99121 | 753 | best_slot = -1; |
abbf098e | 754 | best_rdev = NULL; |
56d99121 | 755 | best_dist = MaxSector; |
856e08e2 | 756 | best_good_sectors = 0; |
56d99121 | 757 | do_balance = 1; |
8d3ca83d | 758 | clear_bit(R10BIO_FailFast, &r10_bio->state); |
1da177e4 LT |
759 | /* |
760 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
761 | * device if no resync is going on (recovery is ok), or below |
762 | * the resync window. We take the first readable disk when | |
763 | * above the resync window. | |
1da177e4 | 764 | */ |
d4098c72 GJ |
765 | if ((conf->mddev->recovery_cp < MaxSector |
766 | && (this_sector + sectors >= conf->next_resync)) || | |
767 | (mddev_is_clustered(conf->mddev) && | |
768 | md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, | |
769 | this_sector + sectors))) | |
56d99121 | 770 | do_balance = 0; |
1da177e4 | 771 | |
56d99121 | 772 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
773 | sector_t first_bad; |
774 | int bad_sectors; | |
775 | sector_t dev_sector; | |
776 | ||
56d99121 N |
777 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
778 | continue; | |
1da177e4 | 779 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
780 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
781 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
782 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
783 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 | 784 | if (rdev == NULL || |
8ae12666 | 785 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
786 | continue; |
787 | if (!test_bit(In_sync, &rdev->flags) && | |
788 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
789 | continue; |
790 | ||
856e08e2 N |
791 | dev_sector = r10_bio->devs[slot].addr; |
792 | if (is_badblock(rdev, dev_sector, sectors, | |
793 | &first_bad, &bad_sectors)) { | |
794 | if (best_dist < MaxSector) | |
795 | /* Already have a better slot */ | |
796 | continue; | |
797 | if (first_bad <= dev_sector) { | |
798 | /* Cannot read here. If this is the | |
799 | * 'primary' device, then we must not read | |
800 | * beyond 'bad_sectors' from another device. | |
801 | */ | |
802 | bad_sectors -= (dev_sector - first_bad); | |
803 | if (!do_balance && sectors > bad_sectors) | |
804 | sectors = bad_sectors; | |
805 | if (best_good_sectors > sectors) | |
806 | best_good_sectors = sectors; | |
807 | } else { | |
808 | sector_t good_sectors = | |
809 | first_bad - dev_sector; | |
810 | if (good_sectors > best_good_sectors) { | |
811 | best_good_sectors = good_sectors; | |
812 | best_slot = slot; | |
abbf098e | 813 | best_rdev = rdev; |
856e08e2 N |
814 | } |
815 | if (!do_balance) | |
816 | /* Must read from here */ | |
817 | break; | |
818 | } | |
819 | continue; | |
820 | } else | |
821 | best_good_sectors = sectors; | |
822 | ||
56d99121 N |
823 | if (!do_balance) |
824 | break; | |
1da177e4 | 825 | |
8d3ca83d N |
826 | if (best_slot >= 0) |
827 | /* At least 2 disks to choose from so failfast is OK */ | |
828 | set_bit(R10BIO_FailFast, &r10_bio->state); | |
22dfdf52 N |
829 | /* This optimisation is debatable, and completely destroys |
830 | * sequential read speed for 'far copies' arrays. So only | |
831 | * keep it for 'near' arrays, and review those later. | |
832 | */ | |
5cf00fcd | 833 | if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
8d3ca83d | 834 | new_distance = 0; |
8ed3a195 KS |
835 | |
836 | /* for far > 1 always use the lowest address */ | |
8d3ca83d | 837 | else if (geo->far_copies > 1) |
56d99121 | 838 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 839 | else |
56d99121 N |
840 | new_distance = abs(r10_bio->devs[slot].addr - |
841 | conf->mirrors[disk].head_position); | |
842 | if (new_distance < best_dist) { | |
843 | best_dist = new_distance; | |
844 | best_slot = slot; | |
abbf098e | 845 | best_rdev = rdev; |
1da177e4 LT |
846 | } |
847 | } | |
abbf098e | 848 | if (slot >= conf->copies) { |
56d99121 | 849 | slot = best_slot; |
abbf098e N |
850 | rdev = best_rdev; |
851 | } | |
1da177e4 | 852 | |
56d99121 | 853 | if (slot >= 0) { |
56d99121 | 854 | atomic_inc(&rdev->nr_pending); |
56d99121 N |
855 | r10_bio->read_slot = slot; |
856 | } else | |
96c3fd1f | 857 | rdev = NULL; |
1da177e4 | 858 | rcu_read_unlock(); |
856e08e2 | 859 | *max_sectors = best_good_sectors; |
1da177e4 | 860 | |
96c3fd1f | 861 | return rdev; |
1da177e4 LT |
862 | } |
863 | ||
5c675f83 | 864 | static int raid10_congested(struct mddev *mddev, int bits) |
0d129228 | 865 | { |
e879a879 | 866 | struct r10conf *conf = mddev->private; |
0d129228 N |
867 | int i, ret = 0; |
868 | ||
4452226e | 869 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
870 | conf->pending_count >= max_queued_requests) |
871 | return 1; | |
872 | ||
0d129228 | 873 | rcu_read_lock(); |
f8c9e74f N |
874 | for (i = 0; |
875 | (i < conf->geo.raid_disks || i < conf->prev.raid_disks) | |
876 | && ret == 0; | |
877 | i++) { | |
3cb03002 | 878 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 879 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 880 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 881 | |
dc3b17cc | 882 | ret |= bdi_congested(q->backing_dev_info, bits); |
0d129228 N |
883 | } |
884 | } | |
885 | rcu_read_unlock(); | |
886 | return ret; | |
887 | } | |
888 | ||
e879a879 | 889 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
890 | { |
891 | /* Any writes that have been queued but are awaiting | |
892 | * bitmap updates get flushed here. | |
a35e63ef | 893 | */ |
a35e63ef N |
894 | spin_lock_irq(&conf->device_lock); |
895 | ||
896 | if (conf->pending_bio_list.head) { | |
18022a1b | 897 | struct blk_plug plug; |
a35e63ef | 898 | struct bio *bio; |
18022a1b | 899 | |
a35e63ef | 900 | bio = bio_list_get(&conf->pending_bio_list); |
34db0cd6 | 901 | conf->pending_count = 0; |
a35e63ef | 902 | spin_unlock_irq(&conf->device_lock); |
18022a1b | 903 | blk_start_plug(&plug); |
a35e63ef N |
904 | /* flush any pending bitmap writes to disk |
905 | * before proceeding w/ I/O */ | |
906 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 907 | wake_up(&conf->wait_barrier); |
a35e63ef N |
908 | |
909 | while (bio) { /* submit pending writes */ | |
910 | struct bio *next = bio->bi_next; | |
74d46992 | 911 | struct md_rdev *rdev = (void*)bio->bi_disk; |
a35e63ef | 912 | bio->bi_next = NULL; |
74d46992 | 913 | bio_set_dev(bio, rdev->bdev); |
a9ae93c8 | 914 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 915 | bio_io_error(bio); |
a9ae93c8 | 916 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
74d46992 | 917 | !blk_queue_discard(bio->bi_disk->queue))) |
532a2a3f | 918 | /* Just ignore it */ |
4246a0b6 | 919 | bio_endio(bio); |
532a2a3f SL |
920 | else |
921 | generic_make_request(bio); | |
a35e63ef N |
922 | bio = next; |
923 | } | |
18022a1b | 924 | blk_finish_plug(&plug); |
a35e63ef N |
925 | } else |
926 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 927 | } |
7eaceacc | 928 | |
0a27ec96 N |
929 | /* Barriers.... |
930 | * Sometimes we need to suspend IO while we do something else, | |
931 | * either some resync/recovery, or reconfigure the array. | |
932 | * To do this we raise a 'barrier'. | |
933 | * The 'barrier' is a counter that can be raised multiple times | |
934 | * to count how many activities are happening which preclude | |
935 | * normal IO. | |
936 | * We can only raise the barrier if there is no pending IO. | |
937 | * i.e. if nr_pending == 0. | |
938 | * We choose only to raise the barrier if no-one is waiting for the | |
939 | * barrier to go down. This means that as soon as an IO request | |
940 | * is ready, no other operations which require a barrier will start | |
941 | * until the IO request has had a chance. | |
942 | * | |
943 | * So: regular IO calls 'wait_barrier'. When that returns there | |
944 | * is no backgroup IO happening, It must arrange to call | |
945 | * allow_barrier when it has finished its IO. | |
946 | * backgroup IO calls must call raise_barrier. Once that returns | |
947 | * there is no normal IO happeing. It must arrange to call | |
948 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 949 | */ |
1da177e4 | 950 | |
e879a879 | 951 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 952 | { |
6cce3b23 | 953 | BUG_ON(force && !conf->barrier); |
1da177e4 | 954 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 955 | |
6cce3b23 N |
956 | /* Wait until no block IO is waiting (unless 'force') */ |
957 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
eed8c02e | 958 | conf->resync_lock); |
0a27ec96 N |
959 | |
960 | /* block any new IO from starting */ | |
961 | conf->barrier++; | |
962 | ||
c3b328ac | 963 | /* Now wait for all pending IO to complete */ |
0a27ec96 | 964 | wait_event_lock_irq(conf->wait_barrier, |
0e5313e2 | 965 | !atomic_read(&conf->nr_pending) && conf->barrier < RESYNC_DEPTH, |
eed8c02e | 966 | conf->resync_lock); |
0a27ec96 N |
967 | |
968 | spin_unlock_irq(&conf->resync_lock); | |
969 | } | |
970 | ||
e879a879 | 971 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
972 | { |
973 | unsigned long flags; | |
974 | spin_lock_irqsave(&conf->resync_lock, flags); | |
975 | conf->barrier--; | |
976 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
977 | wake_up(&conf->wait_barrier); | |
978 | } | |
979 | ||
e879a879 | 980 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
981 | { |
982 | spin_lock_irq(&conf->resync_lock); | |
983 | if (conf->barrier) { | |
984 | conf->nr_waiting++; | |
d6b42dcb N |
985 | /* Wait for the barrier to drop. |
986 | * However if there are already pending | |
987 | * requests (preventing the barrier from | |
988 | * rising completely), and the | |
989 | * pre-process bio queue isn't empty, | |
990 | * then don't wait, as we need to empty | |
991 | * that queue to get the nr_pending | |
992 | * count down. | |
993 | */ | |
578b54ad | 994 | raid10_log(conf->mddev, "wait barrier"); |
d6b42dcb N |
995 | wait_event_lock_irq(conf->wait_barrier, |
996 | !conf->barrier || | |
0e5313e2 | 997 | (atomic_read(&conf->nr_pending) && |
d6b42dcb | 998 | current->bio_list && |
f5fe1b51 N |
999 | (!bio_list_empty(¤t->bio_list[0]) || |
1000 | !bio_list_empty(¤t->bio_list[1]))), | |
eed8c02e | 1001 | conf->resync_lock); |
0a27ec96 | 1002 | conf->nr_waiting--; |
0e5313e2 TM |
1003 | if (!conf->nr_waiting) |
1004 | wake_up(&conf->wait_barrier); | |
1da177e4 | 1005 | } |
0e5313e2 | 1006 | atomic_inc(&conf->nr_pending); |
1da177e4 LT |
1007 | spin_unlock_irq(&conf->resync_lock); |
1008 | } | |
1009 | ||
e879a879 | 1010 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 | 1011 | { |
0e5313e2 TM |
1012 | if ((atomic_dec_and_test(&conf->nr_pending)) || |
1013 | (conf->array_freeze_pending)) | |
1014 | wake_up(&conf->wait_barrier); | |
0a27ec96 N |
1015 | } |
1016 | ||
e2d59925 | 1017 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
1018 | { |
1019 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 1020 | * go quiet. |
4443ae10 | 1021 | * We increment barrier and nr_waiting, and then |
e2d59925 | 1022 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
1023 | * This is called in the context of one normal IO request |
1024 | * that has failed. Thus any sync request that might be pending | |
1025 | * will be blocked by nr_pending, and we need to wait for | |
1026 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 1027 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
1028 | * must match the number of pending IOs (nr_pending) before |
1029 | * we continue. | |
4443ae10 N |
1030 | */ |
1031 | spin_lock_irq(&conf->resync_lock); | |
0e5313e2 | 1032 | conf->array_freeze_pending++; |
4443ae10 N |
1033 | conf->barrier++; |
1034 | conf->nr_waiting++; | |
eed8c02e | 1035 | wait_event_lock_irq_cmd(conf->wait_barrier, |
0e5313e2 | 1036 | atomic_read(&conf->nr_pending) == conf->nr_queued+extra, |
eed8c02e LC |
1037 | conf->resync_lock, |
1038 | flush_pending_writes(conf)); | |
c3b328ac | 1039 | |
0e5313e2 | 1040 | conf->array_freeze_pending--; |
4443ae10 N |
1041 | spin_unlock_irq(&conf->resync_lock); |
1042 | } | |
1043 | ||
e879a879 | 1044 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1045 | { |
1046 | /* reverse the effect of the freeze */ | |
1047 | spin_lock_irq(&conf->resync_lock); | |
1048 | conf->barrier--; | |
1049 | conf->nr_waiting--; | |
1050 | wake_up(&conf->wait_barrier); | |
1051 | spin_unlock_irq(&conf->resync_lock); | |
1052 | } | |
1053 | ||
f8c9e74f N |
1054 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1055 | struct md_rdev *rdev) | |
1056 | { | |
1057 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1058 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1059 | return rdev->data_offset; | |
1060 | else | |
1061 | return rdev->new_data_offset; | |
1062 | } | |
1063 | ||
57c67df4 N |
1064 | struct raid10_plug_cb { |
1065 | struct blk_plug_cb cb; | |
1066 | struct bio_list pending; | |
1067 | int pending_cnt; | |
1068 | }; | |
1069 | ||
1070 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1071 | { | |
1072 | struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb, | |
1073 | cb); | |
1074 | struct mddev *mddev = plug->cb.data; | |
1075 | struct r10conf *conf = mddev->private; | |
1076 | struct bio *bio; | |
1077 | ||
874807a8 | 1078 | if (from_schedule || current->bio_list) { |
57c67df4 N |
1079 | spin_lock_irq(&conf->device_lock); |
1080 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1081 | conf->pending_count += plug->pending_cnt; | |
1082 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1083 | wake_up(&conf->wait_barrier); |
57c67df4 N |
1084 | md_wakeup_thread(mddev->thread); |
1085 | kfree(plug); | |
1086 | return; | |
1087 | } | |
1088 | ||
1089 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1090 | bio = bio_list_get(&plug->pending); | |
1091 | bitmap_unplug(mddev->bitmap); | |
1092 | wake_up(&conf->wait_barrier); | |
1093 | ||
1094 | while (bio) { /* submit pending writes */ | |
1095 | struct bio *next = bio->bi_next; | |
74d46992 | 1096 | struct md_rdev *rdev = (void*)bio->bi_disk; |
57c67df4 | 1097 | bio->bi_next = NULL; |
74d46992 | 1098 | bio_set_dev(bio, rdev->bdev); |
a9ae93c8 | 1099 | if (test_bit(Faulty, &rdev->flags)) { |
6308d8e3 | 1100 | bio_io_error(bio); |
a9ae93c8 | 1101 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && |
74d46992 | 1102 | !blk_queue_discard(bio->bi_disk->queue))) |
32f9f570 | 1103 | /* Just ignore it */ |
4246a0b6 | 1104 | bio_endio(bio); |
32f9f570 SL |
1105 | else |
1106 | generic_make_request(bio); | |
57c67df4 N |
1107 | bio = next; |
1108 | } | |
1109 | kfree(plug); | |
1110 | } | |
1111 | ||
bb5f1ed7 RL |
1112 | static void raid10_read_request(struct mddev *mddev, struct bio *bio, |
1113 | struct r10bio *r10_bio) | |
1da177e4 | 1114 | { |
e879a879 | 1115 | struct r10conf *conf = mddev->private; |
1da177e4 | 1116 | struct bio *read_bio; |
bb5f1ed7 RL |
1117 | const int op = bio_op(bio); |
1118 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
bb5f1ed7 RL |
1119 | int max_sectors; |
1120 | sector_t sectors; | |
1121 | struct md_rdev *rdev; | |
545250f2 N |
1122 | char b[BDEVNAME_SIZE]; |
1123 | int slot = r10_bio->read_slot; | |
1124 | struct md_rdev *err_rdev = NULL; | |
1125 | gfp_t gfp = GFP_NOIO; | |
bb5f1ed7 | 1126 | |
545250f2 N |
1127 | if (r10_bio->devs[slot].rdev) { |
1128 | /* | |
1129 | * This is an error retry, but we cannot | |
1130 | * safely dereference the rdev in the r10_bio, | |
1131 | * we must use the one in conf. | |
1132 | * If it has already been disconnected (unlikely) | |
1133 | * we lose the device name in error messages. | |
1134 | */ | |
1135 | int disk; | |
1136 | /* | |
1137 | * As we are blocking raid10, it is a little safer to | |
1138 | * use __GFP_HIGH. | |
1139 | */ | |
1140 | gfp = GFP_NOIO | __GFP_HIGH; | |
1141 | ||
1142 | rcu_read_lock(); | |
1143 | disk = r10_bio->devs[slot].devnum; | |
1144 | err_rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
1145 | if (err_rdev) | |
1146 | bdevname(err_rdev->bdev, b); | |
1147 | else { | |
1148 | strcpy(b, "???"); | |
1149 | /* This never gets dereferenced */ | |
1150 | err_rdev = r10_bio->devs[slot].rdev; | |
1151 | } | |
1152 | rcu_read_unlock(); | |
1153 | } | |
bb5f1ed7 RL |
1154 | /* |
1155 | * Register the new request and wait if the reconstruction | |
1156 | * thread has put up a bar for new requests. | |
1157 | * Continue immediately if no resync is active currently. | |
1158 | */ | |
1159 | wait_barrier(conf); | |
1160 | ||
fc9977dd | 1161 | sectors = r10_bio->sectors; |
bb5f1ed7 RL |
1162 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
1163 | bio->bi_iter.bi_sector < conf->reshape_progress && | |
1164 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
1165 | /* | |
1166 | * IO spans the reshape position. Need to wait for reshape to | |
1167 | * pass | |
1168 | */ | |
1169 | raid10_log(conf->mddev, "wait reshape"); | |
1170 | allow_barrier(conf); | |
1171 | wait_event(conf->wait_barrier, | |
1172 | conf->reshape_progress <= bio->bi_iter.bi_sector || | |
1173 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1174 | sectors); | |
1175 | wait_barrier(conf); | |
1176 | } | |
1177 | ||
bb5f1ed7 RL |
1178 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1179 | if (!rdev) { | |
545250f2 N |
1180 | if (err_rdev) { |
1181 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1182 | mdname(mddev), b, | |
1183 | (unsigned long long)r10_bio->sector); | |
1184 | } | |
bb5f1ed7 RL |
1185 | raid_end_bio_io(r10_bio); |
1186 | return; | |
1187 | } | |
545250f2 N |
1188 | if (err_rdev) |
1189 | pr_err_ratelimited("md/raid10:%s: %s: redirecting sector %llu to another mirror\n", | |
1190 | mdname(mddev), | |
1191 | bdevname(rdev->bdev, b), | |
1192 | (unsigned long long)r10_bio->sector); | |
fc9977dd N |
1193 | if (max_sectors < bio_sectors(bio)) { |
1194 | struct bio *split = bio_split(bio, max_sectors, | |
545250f2 | 1195 | gfp, conf->bio_split); |
fc9977dd | 1196 | bio_chain(split, bio); |
d7eed9cb | 1197 | allow_barrier(conf); |
fc9977dd | 1198 | generic_make_request(bio); |
d7eed9cb | 1199 | wait_barrier(conf); |
fc9977dd N |
1200 | bio = split; |
1201 | r10_bio->master_bio = bio; | |
1202 | r10_bio->sectors = max_sectors; | |
1203 | } | |
bb5f1ed7 RL |
1204 | slot = r10_bio->read_slot; |
1205 | ||
545250f2 | 1206 | read_bio = bio_clone_fast(bio, gfp, mddev->bio_set); |
bb5f1ed7 RL |
1207 | |
1208 | r10_bio->devs[slot].bio = read_bio; | |
1209 | r10_bio->devs[slot].rdev = rdev; | |
1210 | ||
1211 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + | |
1212 | choose_data_offset(r10_bio, rdev); | |
74d46992 | 1213 | bio_set_dev(read_bio, rdev->bdev); |
bb5f1ed7 RL |
1214 | read_bio->bi_end_io = raid10_end_read_request; |
1215 | bio_set_op_attrs(read_bio, op, do_sync); | |
1216 | if (test_bit(FailFast, &rdev->flags) && | |
1217 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
1218 | read_bio->bi_opf |= MD_FAILFAST; | |
1219 | read_bio->bi_private = r10_bio; | |
1220 | ||
1221 | if (mddev->gendisk) | |
74d46992 | 1222 | trace_block_bio_remap(read_bio->bi_disk->queue, |
bb5f1ed7 RL |
1223 | read_bio, disk_devt(mddev->gendisk), |
1224 | r10_bio->sector); | |
fc9977dd | 1225 | generic_make_request(read_bio); |
bb5f1ed7 RL |
1226 | return; |
1227 | } | |
1228 | ||
27f26a0f GJ |
1229 | static void raid10_write_one_disk(struct mddev *mddev, struct r10bio *r10_bio, |
1230 | struct bio *bio, bool replacement, | |
fc9977dd | 1231 | int n_copy) |
bb5f1ed7 | 1232 | { |
796a5cf0 | 1233 | const int op = bio_op(bio); |
1eff9d32 JA |
1234 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1235 | const unsigned long do_fua = (bio->bi_opf & REQ_FUA); | |
6cce3b23 | 1236 | unsigned long flags; |
57c67df4 N |
1237 | struct blk_plug_cb *cb; |
1238 | struct raid10_plug_cb *plug = NULL; | |
27f26a0f GJ |
1239 | struct r10conf *conf = mddev->private; |
1240 | struct md_rdev *rdev; | |
1241 | int devnum = r10_bio->devs[n_copy].devnum; | |
1242 | struct bio *mbio; | |
1243 | ||
1244 | if (replacement) { | |
1245 | rdev = conf->mirrors[devnum].replacement; | |
1246 | if (rdev == NULL) { | |
1247 | /* Replacement just got moved to main 'rdev' */ | |
1248 | smp_mb(); | |
1249 | rdev = conf->mirrors[devnum].rdev; | |
1250 | } | |
1251 | } else | |
1252 | rdev = conf->mirrors[devnum].rdev; | |
1253 | ||
1254 | mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); | |
27f26a0f GJ |
1255 | if (replacement) |
1256 | r10_bio->devs[n_copy].repl_bio = mbio; | |
1257 | else | |
1258 | r10_bio->devs[n_copy].bio = mbio; | |
1259 | ||
1260 | mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + | |
1261 | choose_data_offset(r10_bio, rdev)); | |
74d46992 | 1262 | bio_set_dev(mbio, rdev->bdev); |
27f26a0f GJ |
1263 | mbio->bi_end_io = raid10_end_write_request; |
1264 | bio_set_op_attrs(mbio, op, do_sync | do_fua); | |
1265 | if (!replacement && test_bit(FailFast, | |
1266 | &conf->mirrors[devnum].rdev->flags) | |
1267 | && enough(conf, devnum)) | |
1268 | mbio->bi_opf |= MD_FAILFAST; | |
1269 | mbio->bi_private = r10_bio; | |
1270 | ||
1271 | if (conf->mddev->gendisk) | |
74d46992 | 1272 | trace_block_bio_remap(mbio->bi_disk->queue, |
27f26a0f GJ |
1273 | mbio, disk_devt(conf->mddev->gendisk), |
1274 | r10_bio->sector); | |
1275 | /* flush_pending_writes() needs access to the rdev so...*/ | |
74d46992 | 1276 | mbio->bi_disk = (void *)rdev; |
27f26a0f GJ |
1277 | |
1278 | atomic_inc(&r10_bio->remaining); | |
1279 | ||
1280 | cb = blk_check_plugged(raid10_unplug, mddev, sizeof(*plug)); | |
1281 | if (cb) | |
1282 | plug = container_of(cb, struct raid10_plug_cb, cb); | |
1283 | else | |
1284 | plug = NULL; | |
27f26a0f GJ |
1285 | if (plug) { |
1286 | bio_list_add(&plug->pending, mbio); | |
1287 | plug->pending_cnt++; | |
1288 | } else { | |
23b245c0 | 1289 | spin_lock_irqsave(&conf->device_lock, flags); |
27f26a0f GJ |
1290 | bio_list_add(&conf->pending_bio_list, mbio); |
1291 | conf->pending_count++; | |
23b245c0 | 1292 | spin_unlock_irqrestore(&conf->device_lock, flags); |
27f26a0f | 1293 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1294 | } |
27f26a0f GJ |
1295 | } |
1296 | ||
1297 | static void raid10_write_request(struct mddev *mddev, struct bio *bio, | |
1298 | struct r10bio *r10_bio) | |
1299 | { | |
1300 | struct r10conf *conf = mddev->private; | |
1301 | int i; | |
1302 | struct md_rdev *blocked_rdev; | |
bb5f1ed7 | 1303 | sector_t sectors; |
d4432c23 | 1304 | int max_sectors; |
1da177e4 | 1305 | |
cb8a7a7e GJ |
1306 | if ((mddev_is_clustered(mddev) && |
1307 | md_cluster_ops->area_resyncing(mddev, WRITE, | |
1308 | bio->bi_iter.bi_sector, | |
1309 | bio_end_sector(bio)))) { | |
1310 | DEFINE_WAIT(w); | |
1311 | for (;;) { | |
1312 | prepare_to_wait(&conf->wait_barrier, | |
1313 | &w, TASK_IDLE); | |
1314 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, | |
1315 | bio->bi_iter.bi_sector, bio_end_sector(bio))) | |
1316 | break; | |
1317 | schedule(); | |
1318 | } | |
1319 | finish_wait(&conf->wait_barrier, &w); | |
1320 | } | |
1321 | ||
cc13b1d1 N |
1322 | /* |
1323 | * Register the new request and wait if the reconstruction | |
1324 | * thread has put up a bar for new requests. | |
1325 | * Continue immediately if no resync is active currently. | |
1326 | */ | |
1327 | wait_barrier(conf); | |
1328 | ||
fc9977dd | 1329 | sectors = r10_bio->sectors; |
3ea7daa5 | 1330 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
4f024f37 KO |
1331 | bio->bi_iter.bi_sector < conf->reshape_progress && |
1332 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
bb5f1ed7 RL |
1333 | /* |
1334 | * IO spans the reshape position. Need to wait for reshape to | |
1335 | * pass | |
3ea7daa5 | 1336 | */ |
578b54ad | 1337 | raid10_log(conf->mddev, "wait reshape"); |
3ea7daa5 N |
1338 | allow_barrier(conf); |
1339 | wait_event(conf->wait_barrier, | |
4f024f37 KO |
1340 | conf->reshape_progress <= bio->bi_iter.bi_sector || |
1341 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1342 | sectors); | |
3ea7daa5 N |
1343 | wait_barrier(conf); |
1344 | } | |
bb5f1ed7 | 1345 | |
3ea7daa5 | 1346 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
3ea7daa5 | 1347 | (mddev->reshape_backwards |
4f024f37 KO |
1348 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1349 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1350 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1351 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1352 | /* Need to update reshape_position in metadata */ |
1353 | mddev->reshape_position = conf->reshape_progress; | |
2953079c SL |
1354 | set_mask_bits(&mddev->sb_flags, 0, |
1355 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
3ea7daa5 | 1356 | md_wakeup_thread(mddev->thread); |
578b54ad | 1357 | raid10_log(conf->mddev, "wait reshape metadata"); |
3ea7daa5 | 1358 | wait_event(mddev->sb_wait, |
2953079c | 1359 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
3ea7daa5 N |
1360 | |
1361 | conf->reshape_safe = mddev->reshape_position; | |
1362 | } | |
1363 | ||
34db0cd6 N |
1364 | if (conf->pending_count >= max_queued_requests) { |
1365 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1366 | raid10_log(mddev, "wait queued"); |
34db0cd6 N |
1367 | wait_event(conf->wait_barrier, |
1368 | conf->pending_count < max_queued_requests); | |
1369 | } | |
6bfe0b49 | 1370 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1371 | * inc refcount on their rdev. Record them by setting |
1372 | * bios[x] to bio | |
d4432c23 N |
1373 | * If there are known/acknowledged bad blocks on any device |
1374 | * on which we have seen a write error, we want to avoid | |
1375 | * writing to those blocks. This potentially requires several | |
1376 | * writes to write around the bad blocks. Each set of writes | |
fd16f2e8 | 1377 | * gets its own r10_bio with a set of bios attached. |
1da177e4 | 1378 | */ |
c3b328ac | 1379 | |
69335ef3 | 1380 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1381 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1382 | retry_write: |
cb6969e8 | 1383 | blocked_rdev = NULL; |
1da177e4 | 1384 | rcu_read_lock(); |
d4432c23 N |
1385 | max_sectors = r10_bio->sectors; |
1386 | ||
1da177e4 LT |
1387 | for (i = 0; i < conf->copies; i++) { |
1388 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1389 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1390 | struct md_rdev *rrdev = rcu_dereference( |
1391 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1392 | if (rdev == rrdev) |
1393 | rrdev = NULL; | |
6bfe0b49 DW |
1394 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1395 | atomic_inc(&rdev->nr_pending); | |
1396 | blocked_rdev = rdev; | |
1397 | break; | |
1398 | } | |
475b0321 N |
1399 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { |
1400 | atomic_inc(&rrdev->nr_pending); | |
1401 | blocked_rdev = rrdev; | |
1402 | break; | |
1403 | } | |
8ae12666 | 1404 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1405 | rdev = NULL; |
8ae12666 | 1406 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1407 | rrdev = NULL; |
1408 | ||
d4432c23 | 1409 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1410 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1411 | |
1412 | if (!rdev && !rrdev) { | |
6cce3b23 | 1413 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1414 | continue; |
1415 | } | |
e7c0c3fa | 1416 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1417 | sector_t first_bad; |
1418 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1419 | int bad_sectors; | |
1420 | int is_bad; | |
1421 | ||
bb5f1ed7 | 1422 | is_bad = is_badblock(rdev, dev_sector, max_sectors, |
d4432c23 N |
1423 | &first_bad, &bad_sectors); |
1424 | if (is_bad < 0) { | |
1425 | /* Mustn't write here until the bad block | |
1426 | * is acknowledged | |
1427 | */ | |
1428 | atomic_inc(&rdev->nr_pending); | |
1429 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1430 | blocked_rdev = rdev; | |
1431 | break; | |
1432 | } | |
1433 | if (is_bad && first_bad <= dev_sector) { | |
1434 | /* Cannot write here at all */ | |
1435 | bad_sectors -= (dev_sector - first_bad); | |
1436 | if (bad_sectors < max_sectors) | |
1437 | /* Mustn't write more than bad_sectors | |
1438 | * to other devices yet | |
1439 | */ | |
1440 | max_sectors = bad_sectors; | |
1441 | /* We don't set R10BIO_Degraded as that | |
1442 | * only applies if the disk is missing, | |
1443 | * so it might be re-added, and we want to | |
1444 | * know to recover this chunk. | |
1445 | * In this case the device is here, and the | |
1446 | * fact that this chunk is not in-sync is | |
1447 | * recorded in the bad block log. | |
1448 | */ | |
1449 | continue; | |
1450 | } | |
1451 | if (is_bad) { | |
1452 | int good_sectors = first_bad - dev_sector; | |
1453 | if (good_sectors < max_sectors) | |
1454 | max_sectors = good_sectors; | |
1455 | } | |
6cce3b23 | 1456 | } |
e7c0c3fa N |
1457 | if (rdev) { |
1458 | r10_bio->devs[i].bio = bio; | |
1459 | atomic_inc(&rdev->nr_pending); | |
1460 | } | |
475b0321 N |
1461 | if (rrdev) { |
1462 | r10_bio->devs[i].repl_bio = bio; | |
1463 | atomic_inc(&rrdev->nr_pending); | |
1464 | } | |
1da177e4 LT |
1465 | } |
1466 | rcu_read_unlock(); | |
1467 | ||
6bfe0b49 DW |
1468 | if (unlikely(blocked_rdev)) { |
1469 | /* Have to wait for this device to get unblocked, then retry */ | |
1470 | int j; | |
1471 | int d; | |
1472 | ||
475b0321 | 1473 | for (j = 0; j < i; j++) { |
6bfe0b49 DW |
1474 | if (r10_bio->devs[j].bio) { |
1475 | d = r10_bio->devs[j].devnum; | |
1476 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1477 | } | |
475b0321 | 1478 | if (r10_bio->devs[j].repl_bio) { |
4ca40c2c | 1479 | struct md_rdev *rdev; |
475b0321 | 1480 | d = r10_bio->devs[j].devnum; |
4ca40c2c N |
1481 | rdev = conf->mirrors[d].replacement; |
1482 | if (!rdev) { | |
1483 | /* Race with remove_disk */ | |
1484 | smp_mb(); | |
1485 | rdev = conf->mirrors[d].rdev; | |
1486 | } | |
1487 | rdev_dec_pending(rdev, mddev); | |
475b0321 N |
1488 | } |
1489 | } | |
6bfe0b49 | 1490 | allow_barrier(conf); |
578b54ad | 1491 | raid10_log(conf->mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 DW |
1492 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
1493 | wait_barrier(conf); | |
1494 | goto retry_write; | |
1495 | } | |
1496 | ||
6b6c8110 | 1497 | if (max_sectors < r10_bio->sectors) |
d4432c23 | 1498 | r10_bio->sectors = max_sectors; |
fc9977dd N |
1499 | |
1500 | if (r10_bio->sectors < bio_sectors(bio)) { | |
1501 | struct bio *split = bio_split(bio, r10_bio->sectors, | |
1502 | GFP_NOIO, conf->bio_split); | |
1503 | bio_chain(split, bio); | |
d7eed9cb | 1504 | allow_barrier(conf); |
fc9977dd | 1505 | generic_make_request(bio); |
d7eed9cb | 1506 | wait_barrier(conf); |
fc9977dd N |
1507 | bio = split; |
1508 | r10_bio->master_bio = bio; | |
d4432c23 | 1509 | } |
d4432c23 | 1510 | |
4e78064f | 1511 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1512 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1513 | |
1da177e4 | 1514 | for (i = 0; i < conf->copies; i++) { |
27f26a0f | 1515 | if (r10_bio->devs[i].bio) |
fc9977dd | 1516 | raid10_write_one_disk(mddev, r10_bio, bio, false, i); |
27f26a0f | 1517 | if (r10_bio->devs[i].repl_bio) |
fc9977dd | 1518 | raid10_write_one_disk(mddev, r10_bio, bio, true, i); |
d4432c23 | 1519 | } |
079fa166 | 1520 | one_write_done(r10_bio); |
20d0189b KO |
1521 | } |
1522 | ||
fc9977dd | 1523 | static void __make_request(struct mddev *mddev, struct bio *bio, int sectors) |
bb5f1ed7 RL |
1524 | { |
1525 | struct r10conf *conf = mddev->private; | |
1526 | struct r10bio *r10_bio; | |
1527 | ||
1528 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1529 | ||
1530 | r10_bio->master_bio = bio; | |
fc9977dd | 1531 | r10_bio->sectors = sectors; |
bb5f1ed7 RL |
1532 | |
1533 | r10_bio->mddev = mddev; | |
1534 | r10_bio->sector = bio->bi_iter.bi_sector; | |
1535 | r10_bio->state = 0; | |
545250f2 | 1536 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * conf->copies); |
bb5f1ed7 RL |
1537 | |
1538 | if (bio_data_dir(bio) == READ) | |
1539 | raid10_read_request(mddev, bio, r10_bio); | |
1540 | else | |
1541 | raid10_write_request(mddev, bio, r10_bio); | |
1542 | } | |
1543 | ||
cc27b0c7 | 1544 | static bool raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1545 | { |
1546 | struct r10conf *conf = mddev->private; | |
1547 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1548 | int chunk_sects = chunk_mask + 1; | |
fc9977dd | 1549 | int sectors = bio_sectors(bio); |
20d0189b | 1550 | |
1eff9d32 | 1551 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
20d0189b | 1552 | md_flush_request(mddev, bio); |
cc27b0c7 | 1553 | return true; |
20d0189b KO |
1554 | } |
1555 | ||
cc27b0c7 N |
1556 | if (!md_write_start(mddev, bio)) |
1557 | return false; | |
1558 | ||
fc9977dd N |
1559 | /* |
1560 | * If this request crosses a chunk boundary, we need to split | |
1561 | * it. | |
1562 | */ | |
1563 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1564 | sectors > chunk_sects | |
1565 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1566 | || conf->prev.near_copies < | |
1567 | conf->prev.raid_disks))) | |
1568 | sectors = chunk_sects - | |
1569 | (bio->bi_iter.bi_sector & | |
1570 | (chunk_sects - 1)); | |
1571 | __make_request(mddev, bio, sectors); | |
079fa166 N |
1572 | |
1573 | /* In case raid10d snuck in to freeze_array */ | |
1574 | wake_up(&conf->wait_barrier); | |
cc27b0c7 | 1575 | return true; |
1da177e4 LT |
1576 | } |
1577 | ||
849674e4 | 1578 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1579 | { |
e879a879 | 1580 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1581 | int i; |
1582 | ||
5cf00fcd | 1583 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1584 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1585 | if (conf->geo.near_copies > 1) |
1586 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1587 | if (conf->geo.far_copies > 1) { | |
1588 | if (conf->geo.far_offset) | |
1589 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1590 | else |
5cf00fcd | 1591 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1592 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1593 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1594 | } |
5cf00fcd N |
1595 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1596 | conf->geo.raid_disks - mddev->degraded); | |
d44b0a92 N |
1597 | rcu_read_lock(); |
1598 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
1599 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1600 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
1601 | } | |
1602 | rcu_read_unlock(); | |
1da177e4 LT |
1603 | seq_printf(seq, "]"); |
1604 | } | |
1605 | ||
700c7213 N |
1606 | /* check if there are enough drives for |
1607 | * every block to appear on atleast one. | |
1608 | * Don't consider the device numbered 'ignore' | |
1609 | * as we might be about to remove it. | |
1610 | */ | |
635f6416 | 1611 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1612 | { |
1613 | int first = 0; | |
725d6e57 | 1614 | int has_enough = 0; |
635f6416 N |
1615 | int disks, ncopies; |
1616 | if (previous) { | |
1617 | disks = conf->prev.raid_disks; | |
1618 | ncopies = conf->prev.near_copies; | |
1619 | } else { | |
1620 | disks = conf->geo.raid_disks; | |
1621 | ncopies = conf->geo.near_copies; | |
1622 | } | |
700c7213 | 1623 | |
725d6e57 | 1624 | rcu_read_lock(); |
700c7213 N |
1625 | do { |
1626 | int n = conf->copies; | |
1627 | int cnt = 0; | |
80b48124 | 1628 | int this = first; |
700c7213 | 1629 | while (n--) { |
725d6e57 N |
1630 | struct md_rdev *rdev; |
1631 | if (this != ignore && | |
1632 | (rdev = rcu_dereference(conf->mirrors[this].rdev)) && | |
1633 | test_bit(In_sync, &rdev->flags)) | |
700c7213 | 1634 | cnt++; |
635f6416 | 1635 | this = (this+1) % disks; |
700c7213 N |
1636 | } |
1637 | if (cnt == 0) | |
725d6e57 | 1638 | goto out; |
635f6416 | 1639 | first = (first + ncopies) % disks; |
700c7213 | 1640 | } while (first != 0); |
725d6e57 N |
1641 | has_enough = 1; |
1642 | out: | |
1643 | rcu_read_unlock(); | |
1644 | return has_enough; | |
700c7213 N |
1645 | } |
1646 | ||
f8c9e74f N |
1647 | static int enough(struct r10conf *conf, int ignore) |
1648 | { | |
635f6416 N |
1649 | /* when calling 'enough', both 'prev' and 'geo' must |
1650 | * be stable. | |
1651 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1652 | * is held. | |
1653 | */ | |
1654 | return _enough(conf, 0, ignore) && | |
1655 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1656 | } |
1657 | ||
849674e4 | 1658 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1659 | { |
1660 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1661 | struct r10conf *conf = mddev->private; |
635f6416 | 1662 | unsigned long flags; |
1da177e4 LT |
1663 | |
1664 | /* | |
1665 | * If it is not operational, then we have already marked it as dead | |
1666 | * else if it is the last working disks, ignore the error, let the | |
1667 | * next level up know. | |
1668 | * else mark the drive as failed | |
1669 | */ | |
635f6416 | 1670 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1671 | if (test_bit(In_sync, &rdev->flags) |
635f6416 | 1672 | && !enough(conf, rdev->raid_disk)) { |
1da177e4 LT |
1673 | /* |
1674 | * Don't fail the drive, just return an IO error. | |
1da177e4 | 1675 | */ |
635f6416 | 1676 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1677 | return; |
635f6416 | 1678 | } |
2446dba0 | 1679 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1680 | mddev->degraded++; |
2446dba0 N |
1681 | /* |
1682 | * If recovery is running, make sure it aborts. | |
1683 | */ | |
1684 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
de393cde | 1685 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1686 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
1687 | set_mask_bits(&mddev->sb_flags, 0, |
1688 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
635f6416 | 1689 | spin_unlock_irqrestore(&conf->device_lock, flags); |
08464e09 N |
1690 | pr_crit("md/raid10:%s: Disk failure on %s, disabling device.\n" |
1691 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
1692 | mdname(mddev), bdevname(rdev->bdev, b), | |
1693 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); | |
1da177e4 LT |
1694 | } |
1695 | ||
e879a879 | 1696 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1697 | { |
1698 | int i; | |
4056ca51 | 1699 | struct md_rdev *rdev; |
1da177e4 | 1700 | |
08464e09 | 1701 | pr_debug("RAID10 conf printout:\n"); |
1da177e4 | 1702 | if (!conf) { |
08464e09 | 1703 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1704 | return; |
1705 | } | |
08464e09 N |
1706 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1707 | conf->geo.raid_disks); | |
1da177e4 | 1708 | |
4056ca51 N |
1709 | /* This is only called with ->reconfix_mutex held, so |
1710 | * rcu protection of rdev is not needed */ | |
5cf00fcd | 1711 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1712 | char b[BDEVNAME_SIZE]; |
4056ca51 N |
1713 | rdev = conf->mirrors[i].rdev; |
1714 | if (rdev) | |
08464e09 N |
1715 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1716 | i, !test_bit(In_sync, &rdev->flags), | |
1717 | !test_bit(Faulty, &rdev->flags), | |
1718 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1719 | } |
1720 | } | |
1721 | ||
e879a879 | 1722 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1723 | { |
0a27ec96 N |
1724 | wait_barrier(conf); |
1725 | allow_barrier(conf); | |
1da177e4 LT |
1726 | |
1727 | mempool_destroy(conf->r10buf_pool); | |
1728 | conf->r10buf_pool = NULL; | |
1729 | } | |
1730 | ||
fd01b88c | 1731 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1732 | { |
1733 | int i; | |
e879a879 | 1734 | struct r10conf *conf = mddev->private; |
dc280d98 | 1735 | struct raid10_info *tmp; |
6b965620 N |
1736 | int count = 0; |
1737 | unsigned long flags; | |
1da177e4 LT |
1738 | |
1739 | /* | |
1740 | * Find all non-in_sync disks within the RAID10 configuration | |
1741 | * and mark them in_sync | |
1742 | */ | |
5cf00fcd | 1743 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1744 | tmp = conf->mirrors + i; |
4ca40c2c N |
1745 | if (tmp->replacement |
1746 | && tmp->replacement->recovery_offset == MaxSector | |
1747 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1748 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1749 | /* Replacement has just become active */ | |
1750 | if (!tmp->rdev | |
1751 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1752 | count++; | |
1753 | if (tmp->rdev) { | |
1754 | /* Replaced device not technically faulty, | |
1755 | * but we need to be sure it gets removed | |
1756 | * and never re-added. | |
1757 | */ | |
1758 | set_bit(Faulty, &tmp->rdev->flags); | |
1759 | sysfs_notify_dirent_safe( | |
1760 | tmp->rdev->sysfs_state); | |
1761 | } | |
1762 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1763 | } else if (tmp->rdev | |
61e4947c | 1764 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
1765 | && !test_bit(Faulty, &tmp->rdev->flags) |
1766 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1767 | count++; |
2863b9eb | 1768 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
1769 | } |
1770 | } | |
6b965620 N |
1771 | spin_lock_irqsave(&conf->device_lock, flags); |
1772 | mddev->degraded -= count; | |
1773 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1774 | |
1775 | print_conf(conf); | |
6b965620 | 1776 | return count; |
1da177e4 LT |
1777 | } |
1778 | ||
fd01b88c | 1779 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1780 | { |
e879a879 | 1781 | struct r10conf *conf = mddev->private; |
199050ea | 1782 | int err = -EEXIST; |
1da177e4 | 1783 | int mirror; |
6c2fce2e | 1784 | int first = 0; |
5cf00fcd | 1785 | int last = conf->geo.raid_disks - 1; |
1da177e4 LT |
1786 | |
1787 | if (mddev->recovery_cp < MaxSector) | |
1788 | /* only hot-add to in-sync arrays, as recovery is | |
1789 | * very different from resync | |
1790 | */ | |
199050ea | 1791 | return -EBUSY; |
635f6416 | 1792 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 1793 | return -EINVAL; |
1da177e4 | 1794 | |
1501efad DW |
1795 | if (md_integrity_add_rdev(rdev, mddev)) |
1796 | return -ENXIO; | |
1797 | ||
a53a6c85 | 1798 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1799 | first = last = rdev->raid_disk; |
1da177e4 | 1800 | |
2c4193df | 1801 | if (rdev->saved_raid_disk >= first && |
edf57bb0 | 1802 | rdev->saved_raid_disk < conf->geo.raid_disks && |
6cce3b23 N |
1803 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1804 | mirror = rdev->saved_raid_disk; | |
1805 | else | |
6c2fce2e | 1806 | mirror = first; |
2bb77736 | 1807 | for ( ; mirror <= last ; mirror++) { |
dc280d98 | 1808 | struct raid10_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1809 | if (p->recovery_disabled == mddev->recovery_disabled) |
1810 | continue; | |
b7044d41 N |
1811 | if (p->rdev) { |
1812 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1813 | p->replacement != NULL) | |
1814 | continue; | |
1815 | clear_bit(In_sync, &rdev->flags); | |
1816 | set_bit(Replacement, &rdev->flags); | |
1817 | rdev->raid_disk = mirror; | |
1818 | err = 0; | |
9092c02d JB |
1819 | if (mddev->gendisk) |
1820 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1821 | rdev->data_offset << 9); | |
b7044d41 N |
1822 | conf->fullsync = 1; |
1823 | rcu_assign_pointer(p->replacement, rdev); | |
1824 | break; | |
1825 | } | |
1da177e4 | 1826 | |
9092c02d JB |
1827 | if (mddev->gendisk) |
1828 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1829 | rdev->data_offset << 9); | |
1da177e4 | 1830 | |
2bb77736 | 1831 | p->head_position = 0; |
d890fa2b | 1832 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1833 | rdev->raid_disk = mirror; |
1834 | err = 0; | |
1835 | if (rdev->saved_raid_disk != mirror) | |
1836 | conf->fullsync = 1; | |
1837 | rcu_assign_pointer(p->rdev, rdev); | |
1838 | break; | |
1839 | } | |
ed30be07 | 1840 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
532a2a3f SL |
1841 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1842 | ||
1da177e4 | 1843 | print_conf(conf); |
199050ea | 1844 | return err; |
1da177e4 LT |
1845 | } |
1846 | ||
b8321b68 | 1847 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1848 | { |
e879a879 | 1849 | struct r10conf *conf = mddev->private; |
1da177e4 | 1850 | int err = 0; |
b8321b68 | 1851 | int number = rdev->raid_disk; |
c8ab903e | 1852 | struct md_rdev **rdevp; |
dc280d98 | 1853 | struct raid10_info *p = conf->mirrors + number; |
1da177e4 LT |
1854 | |
1855 | print_conf(conf); | |
c8ab903e N |
1856 | if (rdev == p->rdev) |
1857 | rdevp = &p->rdev; | |
1858 | else if (rdev == p->replacement) | |
1859 | rdevp = &p->replacement; | |
1860 | else | |
1861 | return 0; | |
1862 | ||
1863 | if (test_bit(In_sync, &rdev->flags) || | |
1864 | atomic_read(&rdev->nr_pending)) { | |
1865 | err = -EBUSY; | |
1866 | goto abort; | |
1867 | } | |
d787be40 | 1868 | /* Only remove non-faulty devices if recovery |
c8ab903e N |
1869 | * is not possible. |
1870 | */ | |
1871 | if (!test_bit(Faulty, &rdev->flags) && | |
1872 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1873 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 1874 | number < conf->geo.raid_disks && |
c8ab903e N |
1875 | enough(conf, -1)) { |
1876 | err = -EBUSY; | |
1877 | goto abort; | |
1da177e4 | 1878 | } |
c8ab903e | 1879 | *rdevp = NULL; |
d787be40 N |
1880 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1881 | synchronize_rcu(); | |
1882 | if (atomic_read(&rdev->nr_pending)) { | |
1883 | /* lost the race, try later */ | |
1884 | err = -EBUSY; | |
1885 | *rdevp = rdev; | |
1886 | goto abort; | |
1887 | } | |
1888 | } | |
1889 | if (p->replacement) { | |
4ca40c2c N |
1890 | /* We must have just cleared 'rdev' */ |
1891 | p->rdev = p->replacement; | |
1892 | clear_bit(Replacement, &p->replacement->flags); | |
1893 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1894 | * but will never see neither -- if they are careful. | |
1895 | */ | |
1896 | p->replacement = NULL; | |
e5bc9c3c | 1897 | } |
4ca40c2c | 1898 | |
e5bc9c3c | 1899 | clear_bit(WantReplacement, &rdev->flags); |
c8ab903e N |
1900 | err = md_integrity_register(mddev); |
1901 | ||
1da177e4 LT |
1902 | abort: |
1903 | ||
1904 | print_conf(conf); | |
1905 | return err; | |
1906 | } | |
1907 | ||
81fa1520 | 1908 | static void __end_sync_read(struct r10bio *r10_bio, struct bio *bio, int d) |
1da177e4 | 1909 | { |
e879a879 | 1910 | struct r10conf *conf = r10_bio->mddev->private; |
0eb3ff12 | 1911 | |
4e4cbee9 | 1912 | if (!bio->bi_status) |
0eb3ff12 | 1913 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
1914 | else |
1915 | /* The write handler will notice the lack of | |
1916 | * R10BIO_Uptodate and record any errors etc | |
1917 | */ | |
4dbcdc75 N |
1918 | atomic_add(r10_bio->sectors, |
1919 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1920 | |
1921 | /* for reconstruct, we always reschedule after a read. | |
1922 | * for resync, only after all reads | |
1923 | */ | |
73d5c38a | 1924 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1925 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1926 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1927 | /* we have read all the blocks, | |
1928 | * do the comparison in process context in raid10d | |
1929 | */ | |
1930 | reschedule_retry(r10_bio); | |
1931 | } | |
1da177e4 LT |
1932 | } |
1933 | ||
81fa1520 ML |
1934 | static void end_sync_read(struct bio *bio) |
1935 | { | |
f0250618 | 1936 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
81fa1520 ML |
1937 | struct r10conf *conf = r10_bio->mddev->private; |
1938 | int d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
1939 | ||
1940 | __end_sync_read(r10_bio, bio, d); | |
1941 | } | |
1942 | ||
1943 | static void end_reshape_read(struct bio *bio) | |
1944 | { | |
f0250618 | 1945 | /* reshape read bio isn't allocated from r10buf_pool */ |
81fa1520 ML |
1946 | struct r10bio *r10_bio = bio->bi_private; |
1947 | ||
1948 | __end_sync_read(r10_bio, bio, r10_bio->read_slot); | |
1949 | } | |
1950 | ||
9f2c9d12 | 1951 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1952 | { |
fd01b88c | 1953 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1954 | |
1da177e4 LT |
1955 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1956 | if (r10_bio->master_bio == NULL) { | |
1957 | /* the primary of several recovery bios */ | |
73d5c38a | 1958 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1959 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1960 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1961 | reschedule_retry(r10_bio); |
1962 | else | |
1963 | put_buf(r10_bio); | |
73d5c38a | 1964 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1965 | break; |
1966 | } else { | |
9f2c9d12 | 1967 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1968 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1969 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1970 | reschedule_retry(r10_bio); |
1971 | else | |
1972 | put_buf(r10_bio); | |
1da177e4 LT |
1973 | r10_bio = r10_bio2; |
1974 | } | |
1975 | } | |
1da177e4 LT |
1976 | } |
1977 | ||
4246a0b6 | 1978 | static void end_sync_write(struct bio *bio) |
5e570289 | 1979 | { |
f0250618 | 1980 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
fd01b88c | 1981 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 1982 | struct r10conf *conf = mddev->private; |
5e570289 N |
1983 | int d; |
1984 | sector_t first_bad; | |
1985 | int bad_sectors; | |
1986 | int slot; | |
9ad1aefc | 1987 | int repl; |
4ca40c2c | 1988 | struct md_rdev *rdev = NULL; |
5e570289 | 1989 | |
9ad1aefc N |
1990 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1991 | if (repl) | |
1992 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 1993 | else |
9ad1aefc | 1994 | rdev = conf->mirrors[d].rdev; |
5e570289 | 1995 | |
4e4cbee9 | 1996 | if (bio->bi_status) { |
9ad1aefc N |
1997 | if (repl) |
1998 | md_error(mddev, rdev); | |
1999 | else { | |
2000 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
2001 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2002 | set_bit(MD_RECOVERY_NEEDED, | |
2003 | &rdev->mddev->recovery); | |
9ad1aefc N |
2004 | set_bit(R10BIO_WriteError, &r10_bio->state); |
2005 | } | |
2006 | } else if (is_badblock(rdev, | |
5e570289 N |
2007 | r10_bio->devs[slot].addr, |
2008 | r10_bio->sectors, | |
2009 | &first_bad, &bad_sectors)) | |
2010 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
2011 | ||
9ad1aefc | 2012 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
2013 | |
2014 | end_sync_request(r10_bio); | |
2015 | } | |
2016 | ||
1da177e4 LT |
2017 | /* |
2018 | * Note: sync and recover and handled very differently for raid10 | |
2019 | * This code is for resync. | |
2020 | * For resync, we read through virtual addresses and read all blocks. | |
2021 | * If there is any error, we schedule a write. The lowest numbered | |
2022 | * drive is authoritative. | |
2023 | * However requests come for physical address, so we need to map. | |
2024 | * For every physical address there are raid_disks/copies virtual addresses, | |
2025 | * which is always are least one, but is not necessarly an integer. | |
2026 | * This means that a physical address can span multiple chunks, so we may | |
2027 | * have to submit multiple io requests for a single sync request. | |
2028 | */ | |
2029 | /* | |
2030 | * We check if all blocks are in-sync and only write to blocks that | |
2031 | * aren't in sync | |
2032 | */ | |
9f2c9d12 | 2033 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2034 | { |
e879a879 | 2035 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
2036 | int i, first; |
2037 | struct bio *tbio, *fbio; | |
f4380a91 | 2038 | int vcnt; |
cdb76be3 | 2039 | struct page **tpages, **fpages; |
1da177e4 LT |
2040 | |
2041 | atomic_set(&r10_bio->remaining, 1); | |
2042 | ||
2043 | /* find the first device with a block */ | |
2044 | for (i=0; i<conf->copies; i++) | |
4e4cbee9 | 2045 | if (!r10_bio->devs[i].bio->bi_status) |
1da177e4 LT |
2046 | break; |
2047 | ||
2048 | if (i == conf->copies) | |
2049 | goto done; | |
2050 | ||
2051 | first = i; | |
2052 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
2053 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
2054 | fbio->bi_iter.bi_idx = 0; | |
cdb76be3 | 2055 | fpages = get_resync_pages(fbio)->pages; |
1da177e4 | 2056 | |
f4380a91 | 2057 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 2058 | /* now find blocks with errors */ |
0eb3ff12 N |
2059 | for (i=0 ; i < conf->copies ; i++) { |
2060 | int j, d; | |
8d3ca83d | 2061 | struct md_rdev *rdev; |
f0250618 | 2062 | struct resync_pages *rp; |
1da177e4 | 2063 | |
1da177e4 | 2064 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
2065 | |
2066 | if (tbio->bi_end_io != end_sync_read) | |
2067 | continue; | |
2068 | if (i == first) | |
1da177e4 | 2069 | continue; |
cdb76be3 ML |
2070 | |
2071 | tpages = get_resync_pages(tbio)->pages; | |
8d3ca83d N |
2072 | d = r10_bio->devs[i].devnum; |
2073 | rdev = conf->mirrors[d].rdev; | |
4e4cbee9 | 2074 | if (!r10_bio->devs[i].bio->bi_status) { |
0eb3ff12 N |
2075 | /* We know that the bi_io_vec layout is the same for |
2076 | * both 'first' and 'i', so we just compare them. | |
2077 | * All vec entries are PAGE_SIZE; | |
2078 | */ | |
7bb23c49 N |
2079 | int sectors = r10_bio->sectors; |
2080 | for (j = 0; j < vcnt; j++) { | |
2081 | int len = PAGE_SIZE; | |
2082 | if (sectors < (len / 512)) | |
2083 | len = sectors * 512; | |
cdb76be3 ML |
2084 | if (memcmp(page_address(fpages[j]), |
2085 | page_address(tpages[j]), | |
7bb23c49 | 2086 | len)) |
0eb3ff12 | 2087 | break; |
7bb23c49 N |
2088 | sectors -= len/512; |
2089 | } | |
0eb3ff12 N |
2090 | if (j == vcnt) |
2091 | continue; | |
7f7583d4 | 2092 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
2093 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
2094 | /* Don't fix anything. */ | |
2095 | continue; | |
8d3ca83d N |
2096 | } else if (test_bit(FailFast, &rdev->flags)) { |
2097 | /* Just give up on this device */ | |
2098 | md_error(rdev->mddev, rdev); | |
2099 | continue; | |
0eb3ff12 | 2100 | } |
f84ee364 N |
2101 | /* Ok, we need to write this bio, either to correct an |
2102 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
2103 | * First we need to fixup bv_offset, bv_len and |
2104 | * bi_vecs, as the read request might have corrupted these | |
2105 | */ | |
f0250618 | 2106 | rp = get_resync_pages(tbio); |
8be185f2 KO |
2107 | bio_reset(tbio); |
2108 | ||
fb0eb5df ML |
2109 | md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size); |
2110 | ||
f0250618 ML |
2111 | rp->raid_bio = r10_bio; |
2112 | tbio->bi_private = rp; | |
4f024f37 | 2113 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 | 2114 | tbio->bi_end_io = end_sync_write; |
796a5cf0 | 2115 | bio_set_op_attrs(tbio, REQ_OP_WRITE, 0); |
1da177e4 | 2116 | |
c31df25f KO |
2117 | bio_copy_data(tbio, fbio); |
2118 | ||
1da177e4 LT |
2119 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2120 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2121 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2122 | |
1919cbb2 N |
2123 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
2124 | tbio->bi_opf |= MD_FAILFAST; | |
4f024f37 | 2125 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
74d46992 | 2126 | bio_set_dev(tbio, conf->mirrors[d].rdev->bdev); |
1da177e4 LT |
2127 | generic_make_request(tbio); |
2128 | } | |
2129 | ||
9ad1aefc N |
2130 | /* Now write out to any replacement devices |
2131 | * that are active | |
2132 | */ | |
2133 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2134 | int d; |
9ad1aefc N |
2135 | |
2136 | tbio = r10_bio->devs[i].repl_bio; | |
2137 | if (!tbio || !tbio->bi_end_io) | |
2138 | continue; | |
2139 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2140 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2141 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2142 | d = r10_bio->devs[i].devnum; |
2143 | atomic_inc(&r10_bio->remaining); | |
2144 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2145 | bio_sectors(tbio)); |
9ad1aefc N |
2146 | generic_make_request(tbio); |
2147 | } | |
2148 | ||
1da177e4 LT |
2149 | done: |
2150 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2151 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2152 | put_buf(r10_bio); | |
2153 | } | |
2154 | } | |
2155 | ||
2156 | /* | |
2157 | * Now for the recovery code. | |
2158 | * Recovery happens across physical sectors. | |
2159 | * We recover all non-is_sync drives by finding the virtual address of | |
2160 | * each, and then choose a working drive that also has that virt address. | |
2161 | * There is a separate r10_bio for each non-in_sync drive. | |
2162 | * Only the first two slots are in use. The first for reading, | |
2163 | * The second for writing. | |
2164 | * | |
2165 | */ | |
9f2c9d12 | 2166 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2167 | { |
2168 | /* We got a read error during recovery. | |
2169 | * We repeat the read in smaller page-sized sections. | |
2170 | * If a read succeeds, write it to the new device or record | |
2171 | * a bad block if we cannot. | |
2172 | * If a read fails, record a bad block on both old and | |
2173 | * new devices. | |
2174 | */ | |
fd01b88c | 2175 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2176 | struct r10conf *conf = mddev->private; |
5e570289 N |
2177 | struct bio *bio = r10_bio->devs[0].bio; |
2178 | sector_t sect = 0; | |
2179 | int sectors = r10_bio->sectors; | |
2180 | int idx = 0; | |
2181 | int dr = r10_bio->devs[0].devnum; | |
2182 | int dw = r10_bio->devs[1].devnum; | |
cdb76be3 | 2183 | struct page **pages = get_resync_pages(bio)->pages; |
5e570289 N |
2184 | |
2185 | while (sectors) { | |
2186 | int s = sectors; | |
3cb03002 | 2187 | struct md_rdev *rdev; |
5e570289 N |
2188 | sector_t addr; |
2189 | int ok; | |
2190 | ||
2191 | if (s > (PAGE_SIZE>>9)) | |
2192 | s = PAGE_SIZE >> 9; | |
2193 | ||
2194 | rdev = conf->mirrors[dr].rdev; | |
2195 | addr = r10_bio->devs[0].addr + sect, | |
2196 | ok = sync_page_io(rdev, | |
2197 | addr, | |
2198 | s << 9, | |
cdb76be3 | 2199 | pages[idx], |
796a5cf0 | 2200 | REQ_OP_READ, 0, false); |
5e570289 N |
2201 | if (ok) { |
2202 | rdev = conf->mirrors[dw].rdev; | |
2203 | addr = r10_bio->devs[1].addr + sect; | |
2204 | ok = sync_page_io(rdev, | |
2205 | addr, | |
2206 | s << 9, | |
cdb76be3 | 2207 | pages[idx], |
796a5cf0 | 2208 | REQ_OP_WRITE, 0, false); |
b7044d41 | 2209 | if (!ok) { |
5e570289 | 2210 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2211 | if (!test_and_set_bit(WantReplacement, |
2212 | &rdev->flags)) | |
2213 | set_bit(MD_RECOVERY_NEEDED, | |
2214 | &rdev->mddev->recovery); | |
2215 | } | |
5e570289 N |
2216 | } |
2217 | if (!ok) { | |
2218 | /* We don't worry if we cannot set a bad block - | |
2219 | * it really is bad so there is no loss in not | |
2220 | * recording it yet | |
2221 | */ | |
2222 | rdev_set_badblocks(rdev, addr, s, 0); | |
2223 | ||
2224 | if (rdev != conf->mirrors[dw].rdev) { | |
2225 | /* need bad block on destination too */ | |
3cb03002 | 2226 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2227 | addr = r10_bio->devs[1].addr + sect; |
2228 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2229 | if (!ok) { | |
2230 | /* just abort the recovery */ | |
08464e09 N |
2231 | pr_notice("md/raid10:%s: recovery aborted due to read error\n", |
2232 | mdname(mddev)); | |
5e570289 N |
2233 | |
2234 | conf->mirrors[dw].recovery_disabled | |
2235 | = mddev->recovery_disabled; | |
2236 | set_bit(MD_RECOVERY_INTR, | |
2237 | &mddev->recovery); | |
2238 | break; | |
2239 | } | |
2240 | } | |
2241 | } | |
2242 | ||
2243 | sectors -= s; | |
2244 | sect += s; | |
2245 | idx++; | |
2246 | } | |
2247 | } | |
1da177e4 | 2248 | |
9f2c9d12 | 2249 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2250 | { |
e879a879 | 2251 | struct r10conf *conf = mddev->private; |
c65060ad | 2252 | int d; |
24afd80d | 2253 | struct bio *wbio, *wbio2; |
1da177e4 | 2254 | |
5e570289 N |
2255 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2256 | fix_recovery_read_error(r10_bio); | |
2257 | end_sync_request(r10_bio); | |
2258 | return; | |
2259 | } | |
2260 | ||
c65060ad NK |
2261 | /* |
2262 | * share the pages with the first bio | |
1da177e4 LT |
2263 | * and submit the write request |
2264 | */ | |
1da177e4 | 2265 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2266 | wbio = r10_bio->devs[1].bio; |
2267 | wbio2 = r10_bio->devs[1].repl_bio; | |
0eb25bb0 N |
2268 | /* Need to test wbio2->bi_end_io before we call |
2269 | * generic_make_request as if the former is NULL, | |
2270 | * the latter is free to free wbio2. | |
2271 | */ | |
2272 | if (wbio2 && !wbio2->bi_end_io) | |
2273 | wbio2 = NULL; | |
24afd80d N |
2274 | if (wbio->bi_end_io) { |
2275 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2276 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
24afd80d N |
2277 | generic_make_request(wbio); |
2278 | } | |
0eb25bb0 | 2279 | if (wbio2) { |
24afd80d N |
2280 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2281 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2282 | bio_sectors(wbio2)); |
24afd80d N |
2283 | generic_make_request(wbio2); |
2284 | } | |
1da177e4 LT |
2285 | } |
2286 | ||
1e50915f RB |
2287 | /* |
2288 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2289 | * We halve the read error count for every hour that has elapsed | |
2290 | * since the last recorded read error. | |
2291 | * | |
2292 | */ | |
fd01b88c | 2293 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f | 2294 | { |
0e3ef49e | 2295 | long cur_time_mon; |
1e50915f RB |
2296 | unsigned long hours_since_last; |
2297 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2298 | ||
0e3ef49e | 2299 | cur_time_mon = ktime_get_seconds(); |
1e50915f | 2300 | |
0e3ef49e | 2301 | if (rdev->last_read_error == 0) { |
1e50915f RB |
2302 | /* first time we've seen a read error */ |
2303 | rdev->last_read_error = cur_time_mon; | |
2304 | return; | |
2305 | } | |
2306 | ||
0e3ef49e AB |
2307 | hours_since_last = (long)(cur_time_mon - |
2308 | rdev->last_read_error) / 3600; | |
1e50915f RB |
2309 | |
2310 | rdev->last_read_error = cur_time_mon; | |
2311 | ||
2312 | /* | |
2313 | * if hours_since_last is > the number of bits in read_errors | |
2314 | * just set read errors to 0. We do this to avoid | |
2315 | * overflowing the shift of read_errors by hours_since_last. | |
2316 | */ | |
2317 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2318 | atomic_set(&rdev->read_errors, 0); | |
2319 | else | |
2320 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2321 | } | |
2322 | ||
3cb03002 | 2323 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2324 | int sectors, struct page *page, int rw) |
2325 | { | |
2326 | sector_t first_bad; | |
2327 | int bad_sectors; | |
2328 | ||
2329 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2330 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2331 | return -1; | |
796a5cf0 | 2332 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
58c54fcc N |
2333 | /* success */ |
2334 | return 1; | |
b7044d41 | 2335 | if (rw == WRITE) { |
58c54fcc | 2336 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2337 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2338 | set_bit(MD_RECOVERY_NEEDED, | |
2339 | &rdev->mddev->recovery); | |
2340 | } | |
58c54fcc N |
2341 | /* need to record an error - either for the block or the device */ |
2342 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2343 | md_error(rdev->mddev, rdev); | |
2344 | return 0; | |
2345 | } | |
2346 | ||
1da177e4 LT |
2347 | /* |
2348 | * This is a kernel thread which: | |
2349 | * | |
2350 | * 1. Retries failed read operations on working mirrors. | |
2351 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2352 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2353 | */ |
2354 | ||
e879a879 | 2355 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2356 | { |
2357 | int sect = 0; /* Offset from r10_bio->sector */ | |
2358 | int sectors = r10_bio->sectors; | |
3cb03002 | 2359 | struct md_rdev*rdev; |
1e50915f | 2360 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2361 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2362 | |
7c4e06ff N |
2363 | /* still own a reference to this rdev, so it cannot |
2364 | * have been cleared recently. | |
2365 | */ | |
2366 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2367 | |
7c4e06ff N |
2368 | if (test_bit(Faulty, &rdev->flags)) |
2369 | /* drive has already been failed, just ignore any | |
2370 | more fix_read_error() attempts */ | |
2371 | return; | |
1e50915f | 2372 | |
7c4e06ff N |
2373 | check_decay_read_errors(mddev, rdev); |
2374 | atomic_inc(&rdev->read_errors); | |
2375 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2376 | char b[BDEVNAME_SIZE]; | |
2377 | bdevname(rdev->bdev, b); | |
1e50915f | 2378 | |
08464e09 N |
2379 | pr_notice("md/raid10:%s: %s: Raid device exceeded read_error threshold [cur %d:max %d]\n", |
2380 | mdname(mddev), b, | |
2381 | atomic_read(&rdev->read_errors), max_read_errors); | |
2382 | pr_notice("md/raid10:%s: %s: Failing raid device\n", | |
2383 | mdname(mddev), b); | |
d683c8e0 | 2384 | md_error(mddev, rdev); |
fae8cc5e | 2385 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2386 | return; |
1e50915f | 2387 | } |
1e50915f | 2388 | |
6814d536 N |
2389 | while(sectors) { |
2390 | int s = sectors; | |
2391 | int sl = r10_bio->read_slot; | |
2392 | int success = 0; | |
2393 | int start; | |
2394 | ||
2395 | if (s > (PAGE_SIZE>>9)) | |
2396 | s = PAGE_SIZE >> 9; | |
2397 | ||
2398 | rcu_read_lock(); | |
2399 | do { | |
8dbed5ce N |
2400 | sector_t first_bad; |
2401 | int bad_sectors; | |
2402 | ||
0544a21d | 2403 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2404 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2405 | if (rdev && | |
8dbed5ce | 2406 | test_bit(In_sync, &rdev->flags) && |
f5b67ae8 | 2407 | !test_bit(Faulty, &rdev->flags) && |
8dbed5ce N |
2408 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, |
2409 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2410 | atomic_inc(&rdev->nr_pending); |
2411 | rcu_read_unlock(); | |
2b193363 | 2412 | success = sync_page_io(rdev, |
6814d536 | 2413 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2414 | sect, |
6814d536 | 2415 | s<<9, |
796a5cf0 MC |
2416 | conf->tmppage, |
2417 | REQ_OP_READ, 0, false); | |
6814d536 N |
2418 | rdev_dec_pending(rdev, mddev); |
2419 | rcu_read_lock(); | |
2420 | if (success) | |
2421 | break; | |
2422 | } | |
2423 | sl++; | |
2424 | if (sl == conf->copies) | |
2425 | sl = 0; | |
2426 | } while (!success && sl != r10_bio->read_slot); | |
2427 | rcu_read_unlock(); | |
2428 | ||
2429 | if (!success) { | |
58c54fcc N |
2430 | /* Cannot read from anywhere, just mark the block |
2431 | * as bad on the first device to discourage future | |
2432 | * reads. | |
2433 | */ | |
6814d536 | 2434 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2435 | rdev = conf->mirrors[dn].rdev; |
2436 | ||
2437 | if (!rdev_set_badblocks( | |
2438 | rdev, | |
2439 | r10_bio->devs[r10_bio->read_slot].addr | |
2440 | + sect, | |
fae8cc5e | 2441 | s, 0)) { |
58c54fcc | 2442 | md_error(mddev, rdev); |
fae8cc5e N |
2443 | r10_bio->devs[r10_bio->read_slot].bio |
2444 | = IO_BLOCKED; | |
2445 | } | |
6814d536 N |
2446 | break; |
2447 | } | |
2448 | ||
2449 | start = sl; | |
2450 | /* write it back and re-read */ | |
2451 | rcu_read_lock(); | |
2452 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2453 | char b[BDEVNAME_SIZE]; |
0544a21d | 2454 | |
6814d536 N |
2455 | if (sl==0) |
2456 | sl = conf->copies; | |
2457 | sl--; | |
2458 | d = r10_bio->devs[sl].devnum; | |
2459 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2460 | if (!rdev || |
f5b67ae8 | 2461 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2462 | !test_bit(In_sync, &rdev->flags)) |
2463 | continue; | |
2464 | ||
2465 | atomic_inc(&rdev->nr_pending); | |
2466 | rcu_read_unlock(); | |
58c54fcc N |
2467 | if (r10_sync_page_io(rdev, |
2468 | r10_bio->devs[sl].addr + | |
2469 | sect, | |
055d3747 | 2470 | s, conf->tmppage, WRITE) |
1294b9c9 N |
2471 | == 0) { |
2472 | /* Well, this device is dead */ | |
08464e09 N |
2473 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %s)\n", |
2474 | mdname(mddev), s, | |
2475 | (unsigned long long)( | |
2476 | sect + | |
2477 | choose_data_offset(r10_bio, | |
2478 | rdev)), | |
2479 | bdevname(rdev->bdev, b)); | |
2480 | pr_notice("md/raid10:%s: %s: failing drive\n", | |
2481 | mdname(mddev), | |
2482 | bdevname(rdev->bdev, b)); | |
6814d536 | 2483 | } |
1294b9c9 N |
2484 | rdev_dec_pending(rdev, mddev); |
2485 | rcu_read_lock(); | |
6814d536 N |
2486 | } |
2487 | sl = start; | |
2488 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2489 | char b[BDEVNAME_SIZE]; |
0544a21d | 2490 | |
6814d536 N |
2491 | if (sl==0) |
2492 | sl = conf->copies; | |
2493 | sl--; | |
2494 | d = r10_bio->devs[sl].devnum; | |
2495 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2496 | if (!rdev || |
f5b67ae8 | 2497 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2498 | !test_bit(In_sync, &rdev->flags)) |
2499 | continue; | |
6814d536 | 2500 | |
1294b9c9 N |
2501 | atomic_inc(&rdev->nr_pending); |
2502 | rcu_read_unlock(); | |
58c54fcc N |
2503 | switch (r10_sync_page_io(rdev, |
2504 | r10_bio->devs[sl].addr + | |
2505 | sect, | |
055d3747 | 2506 | s, conf->tmppage, |
58c54fcc N |
2507 | READ)) { |
2508 | case 0: | |
1294b9c9 | 2509 | /* Well, this device is dead */ |
08464e09 | 2510 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2511 | mdname(mddev), s, |
2512 | (unsigned long long)( | |
f8c9e74f N |
2513 | sect + |
2514 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 | 2515 | bdevname(rdev->bdev, b)); |
08464e09 | 2516 | pr_notice("md/raid10:%s: %s: failing drive\n", |
1294b9c9 N |
2517 | mdname(mddev), |
2518 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2519 | break; |
2520 | case 1: | |
08464e09 | 2521 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2522 | mdname(mddev), s, |
2523 | (unsigned long long)( | |
f8c9e74f N |
2524 | sect + |
2525 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2526 | bdevname(rdev->bdev, b)); |
2527 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2528 | } |
1294b9c9 N |
2529 | |
2530 | rdev_dec_pending(rdev, mddev); | |
2531 | rcu_read_lock(); | |
6814d536 N |
2532 | } |
2533 | rcu_read_unlock(); | |
2534 | ||
2535 | sectors -= s; | |
2536 | sect += s; | |
2537 | } | |
2538 | } | |
2539 | ||
9f2c9d12 | 2540 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2541 | { |
2542 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2543 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2544 | struct r10conf *conf = mddev->private; |
3cb03002 | 2545 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2546 | /* bio has the data to be written to slot 'i' where |
2547 | * we just recently had a write error. | |
2548 | * We repeatedly clone the bio and trim down to one block, | |
2549 | * then try the write. Where the write fails we record | |
2550 | * a bad block. | |
2551 | * It is conceivable that the bio doesn't exactly align with | |
2552 | * blocks. We must handle this. | |
2553 | * | |
2554 | * We currently own a reference to the rdev. | |
2555 | */ | |
2556 | ||
2557 | int block_sectors; | |
2558 | sector_t sector; | |
2559 | int sectors; | |
2560 | int sect_to_write = r10_bio->sectors; | |
2561 | int ok = 1; | |
2562 | ||
2563 | if (rdev->badblocks.shift < 0) | |
2564 | return 0; | |
2565 | ||
f04ebb0b N |
2566 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2567 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2568 | sector = r10_bio->sector; |
2569 | sectors = ((r10_bio->sector + block_sectors) | |
2570 | & ~(sector_t)(block_sectors - 1)) | |
2571 | - sector; | |
2572 | ||
2573 | while (sect_to_write) { | |
2574 | struct bio *wbio; | |
27028626 | 2575 | sector_t wsector; |
bd870a16 N |
2576 | if (sectors > sect_to_write) |
2577 | sectors = sect_to_write; | |
2578 | /* Write at 'sector' for 'sectors' */ | |
d7a10308 | 2579 | wbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
4f024f37 | 2580 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
27028626 TM |
2581 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); |
2582 | wbio->bi_iter.bi_sector = wsector + | |
2583 | choose_data_offset(r10_bio, rdev); | |
74d46992 | 2584 | bio_set_dev(wbio, rdev->bdev); |
796a5cf0 | 2585 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
2586 | |
2587 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 | 2588 | /* Failure! */ |
27028626 | 2589 | ok = rdev_set_badblocks(rdev, wsector, |
bd870a16 N |
2590 | sectors, 0) |
2591 | && ok; | |
2592 | ||
2593 | bio_put(wbio); | |
2594 | sect_to_write -= sectors; | |
2595 | sector += sectors; | |
2596 | sectors = block_sectors; | |
2597 | } | |
2598 | return ok; | |
2599 | } | |
2600 | ||
9f2c9d12 | 2601 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2602 | { |
2603 | int slot = r10_bio->read_slot; | |
560f8e55 | 2604 | struct bio *bio; |
e879a879 | 2605 | struct r10conf *conf = mddev->private; |
abbf098e | 2606 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2607 | |
2608 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2609 | * the block and we can fix it. | |
2610 | * We freeze all other IO, and try reading the block from | |
2611 | * other devices. When we find one, we re-write | |
2612 | * and check it that fixes the read error. | |
2613 | * This is all done synchronously while the array is | |
2614 | * frozen. | |
2615 | */ | |
fae8cc5e | 2616 | bio = r10_bio->devs[slot].bio; |
fae8cc5e N |
2617 | bio_put(bio); |
2618 | r10_bio->devs[slot].bio = NULL; | |
2619 | ||
8d3ca83d N |
2620 | if (mddev->ro) |
2621 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2622 | else if (!test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2623 | freeze_array(conf, 1); |
560f8e55 N |
2624 | fix_read_error(conf, mddev, r10_bio); |
2625 | unfreeze_array(conf); | |
fae8cc5e | 2626 | } else |
8d3ca83d | 2627 | md_error(mddev, rdev); |
fae8cc5e | 2628 | |
abbf098e | 2629 | rdev_dec_pending(rdev, mddev); |
545250f2 N |
2630 | allow_barrier(conf); |
2631 | r10_bio->state = 0; | |
2632 | raid10_read_request(mddev, r10_bio->master_bio, r10_bio); | |
560f8e55 N |
2633 | } |
2634 | ||
e879a879 | 2635 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2636 | { |
2637 | /* Some sort of write request has finished and it | |
2638 | * succeeded in writing where we thought there was a | |
2639 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2640 | * Or possibly if failed and we need to record |
2641 | * a bad block. | |
749c55e9 N |
2642 | */ |
2643 | int m; | |
3cb03002 | 2644 | struct md_rdev *rdev; |
749c55e9 N |
2645 | |
2646 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2647 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2648 | for (m = 0; m < conf->copies; m++) { |
2649 | int dev = r10_bio->devs[m].devnum; | |
2650 | rdev = conf->mirrors[dev].rdev; | |
a2018905 YY |
2651 | if (r10_bio->devs[m].bio == NULL || |
2652 | r10_bio->devs[m].bio->bi_end_io == NULL) | |
1a0b7cd8 | 2653 | continue; |
4e4cbee9 | 2654 | if (!r10_bio->devs[m].bio->bi_status) { |
749c55e9 N |
2655 | rdev_clear_badblocks( |
2656 | rdev, | |
2657 | r10_bio->devs[m].addr, | |
c6563a8c | 2658 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2659 | } else { |
2660 | if (!rdev_set_badblocks( | |
2661 | rdev, | |
2662 | r10_bio->devs[m].addr, | |
2663 | r10_bio->sectors, 0)) | |
2664 | md_error(conf->mddev, rdev); | |
749c55e9 | 2665 | } |
9ad1aefc | 2666 | rdev = conf->mirrors[dev].replacement; |
a2018905 YY |
2667 | if (r10_bio->devs[m].repl_bio == NULL || |
2668 | r10_bio->devs[m].repl_bio->bi_end_io == NULL) | |
9ad1aefc | 2669 | continue; |
4246a0b6 | 2670 | |
4e4cbee9 | 2671 | if (!r10_bio->devs[m].repl_bio->bi_status) { |
9ad1aefc N |
2672 | rdev_clear_badblocks( |
2673 | rdev, | |
2674 | r10_bio->devs[m].addr, | |
c6563a8c | 2675 | r10_bio->sectors, 0); |
9ad1aefc N |
2676 | } else { |
2677 | if (!rdev_set_badblocks( | |
2678 | rdev, | |
2679 | r10_bio->devs[m].addr, | |
2680 | r10_bio->sectors, 0)) | |
2681 | md_error(conf->mddev, rdev); | |
2682 | } | |
1a0b7cd8 | 2683 | } |
749c55e9 N |
2684 | put_buf(r10_bio); |
2685 | } else { | |
95af587e | 2686 | bool fail = false; |
bd870a16 N |
2687 | for (m = 0; m < conf->copies; m++) { |
2688 | int dev = r10_bio->devs[m].devnum; | |
2689 | struct bio *bio = r10_bio->devs[m].bio; | |
2690 | rdev = conf->mirrors[dev].rdev; | |
2691 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2692 | rdev_clear_badblocks( |
2693 | rdev, | |
2694 | r10_bio->devs[m].addr, | |
c6563a8c | 2695 | r10_bio->sectors, 0); |
749c55e9 | 2696 | rdev_dec_pending(rdev, conf->mddev); |
4e4cbee9 | 2697 | } else if (bio != NULL && bio->bi_status) { |
95af587e | 2698 | fail = true; |
bd870a16 N |
2699 | if (!narrow_write_error(r10_bio, m)) { |
2700 | md_error(conf->mddev, rdev); | |
2701 | set_bit(R10BIO_Degraded, | |
2702 | &r10_bio->state); | |
2703 | } | |
2704 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2705 | } |
475b0321 N |
2706 | bio = r10_bio->devs[m].repl_bio; |
2707 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2708 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2709 | rdev_clear_badblocks( |
2710 | rdev, | |
2711 | r10_bio->devs[m].addr, | |
c6563a8c | 2712 | r10_bio->sectors, 0); |
475b0321 N |
2713 | rdev_dec_pending(rdev, conf->mddev); |
2714 | } | |
bd870a16 | 2715 | } |
95af587e N |
2716 | if (fail) { |
2717 | spin_lock_irq(&conf->device_lock); | |
2718 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 2719 | conf->nr_queued++; |
95af587e | 2720 | spin_unlock_irq(&conf->device_lock); |
cf25ae78 GJ |
2721 | /* |
2722 | * In case freeze_array() is waiting for condition | |
2723 | * nr_pending == nr_queued + extra to be true. | |
2724 | */ | |
2725 | wake_up(&conf->wait_barrier); | |
95af587e | 2726 | md_wakeup_thread(conf->mddev->thread); |
c340702c N |
2727 | } else { |
2728 | if (test_bit(R10BIO_WriteError, | |
2729 | &r10_bio->state)) | |
2730 | close_write(r10_bio); | |
95af587e | 2731 | raid_end_bio_io(r10_bio); |
c340702c | 2732 | } |
749c55e9 N |
2733 | } |
2734 | } | |
2735 | ||
4ed8731d | 2736 | static void raid10d(struct md_thread *thread) |
1da177e4 | 2737 | { |
4ed8731d | 2738 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2739 | struct r10bio *r10_bio; |
1da177e4 | 2740 | unsigned long flags; |
e879a879 | 2741 | struct r10conf *conf = mddev->private; |
1da177e4 | 2742 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2743 | struct blk_plug plug; |
1da177e4 LT |
2744 | |
2745 | md_check_recovery(mddev); | |
1da177e4 | 2746 | |
95af587e | 2747 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2748 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
95af587e N |
2749 | LIST_HEAD(tmp); |
2750 | spin_lock_irqsave(&conf->device_lock, flags); | |
2953079c | 2751 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
23ddba80 SL |
2752 | while (!list_empty(&conf->bio_end_io_list)) { |
2753 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2754 | conf->nr_queued--; | |
2755 | } | |
95af587e N |
2756 | } |
2757 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2758 | while (!list_empty(&tmp)) { | |
a452744b MP |
2759 | r10_bio = list_first_entry(&tmp, struct r10bio, |
2760 | retry_list); | |
95af587e | 2761 | list_del(&r10_bio->retry_list); |
c340702c N |
2762 | if (mddev->degraded) |
2763 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
2764 | ||
2765 | if (test_bit(R10BIO_WriteError, | |
2766 | &r10_bio->state)) | |
2767 | close_write(r10_bio); | |
95af587e N |
2768 | raid_end_bio_io(r10_bio); |
2769 | } | |
2770 | } | |
2771 | ||
e1dfa0a2 | 2772 | blk_start_plug(&plug); |
1da177e4 | 2773 | for (;;) { |
6cce3b23 | 2774 | |
0021b7bc | 2775 | flush_pending_writes(conf); |
6cce3b23 | 2776 | |
a35e63ef N |
2777 | spin_lock_irqsave(&conf->device_lock, flags); |
2778 | if (list_empty(head)) { | |
2779 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2780 | break; |
a35e63ef | 2781 | } |
9f2c9d12 | 2782 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2783 | list_del(head->prev); |
4443ae10 | 2784 | conf->nr_queued--; |
1da177e4 LT |
2785 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2786 | ||
2787 | mddev = r10_bio->mddev; | |
070ec55d | 2788 | conf = mddev->private; |
bd870a16 N |
2789 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2790 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 2791 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
2792 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
2793 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 2794 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 2795 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2796 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2797 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2798 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2799 | handle_read_error(mddev, r10_bio); |
fc9977dd N |
2800 | else |
2801 | WARN_ON_ONCE(1); | |
560f8e55 | 2802 | |
1d9d5241 | 2803 | cond_resched(); |
2953079c | 2804 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2805 | md_check_recovery(mddev); |
1da177e4 | 2806 | } |
e1dfa0a2 | 2807 | blk_finish_plug(&plug); |
1da177e4 LT |
2808 | } |
2809 | ||
e879a879 | 2810 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2811 | { |
2812 | int buffs; | |
69335ef3 | 2813 | int i; |
1da177e4 LT |
2814 | |
2815 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2816 | BUG_ON(conf->r10buf_pool); |
69335ef3 | 2817 | conf->have_replacement = 0; |
5cf00fcd | 2818 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
2819 | if (conf->mirrors[i].replacement) |
2820 | conf->have_replacement = 1; | |
1da177e4 LT |
2821 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2822 | if (!conf->r10buf_pool) | |
2823 | return -ENOMEM; | |
2824 | conf->next_resync = 0; | |
2825 | return 0; | |
2826 | } | |
2827 | ||
208410b5 SL |
2828 | static struct r10bio *raid10_alloc_init_r10buf(struct r10conf *conf) |
2829 | { | |
2830 | struct r10bio *r10bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); | |
2831 | struct rsync_pages *rp; | |
2832 | struct bio *bio; | |
2833 | int nalloc; | |
2834 | int i; | |
2835 | ||
2836 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || | |
2837 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
2838 | nalloc = conf->copies; /* resync */ | |
2839 | else | |
2840 | nalloc = 2; /* recovery */ | |
2841 | ||
2842 | for (i = 0; i < nalloc; i++) { | |
2843 | bio = r10bio->devs[i].bio; | |
2844 | rp = bio->bi_private; | |
2845 | bio_reset(bio); | |
2846 | bio->bi_private = rp; | |
2847 | bio = r10bio->devs[i].repl_bio; | |
2848 | if (bio) { | |
2849 | rp = bio->bi_private; | |
2850 | bio_reset(bio); | |
2851 | bio->bi_private = rp; | |
2852 | } | |
2853 | } | |
2854 | return r10bio; | |
2855 | } | |
2856 | ||
8db87912 GJ |
2857 | /* |
2858 | * Set cluster_sync_high since we need other nodes to add the | |
2859 | * range [cluster_sync_low, cluster_sync_high] to suspend list. | |
2860 | */ | |
2861 | static void raid10_set_cluster_sync_high(struct r10conf *conf) | |
2862 | { | |
2863 | sector_t window_size; | |
2864 | int extra_chunk, chunks; | |
2865 | ||
2866 | /* | |
2867 | * First, here we define "stripe" as a unit which across | |
2868 | * all member devices one time, so we get chunks by use | |
2869 | * raid_disks / near_copies. Otherwise, if near_copies is | |
2870 | * close to raid_disks, then resync window could increases | |
2871 | * linearly with the increase of raid_disks, which means | |
2872 | * we will suspend a really large IO window while it is not | |
2873 | * necessary. If raid_disks is not divisible by near_copies, | |
2874 | * an extra chunk is needed to ensure the whole "stripe" is | |
2875 | * covered. | |
2876 | */ | |
2877 | ||
2878 | chunks = conf->geo.raid_disks / conf->geo.near_copies; | |
2879 | if (conf->geo.raid_disks % conf->geo.near_copies == 0) | |
2880 | extra_chunk = 0; | |
2881 | else | |
2882 | extra_chunk = 1; | |
2883 | window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors; | |
2884 | ||
2885 | /* | |
2886 | * At least use a 32M window to align with raid1's resync window | |
2887 | */ | |
2888 | window_size = (CLUSTER_RESYNC_WINDOW_SECTORS > window_size) ? | |
2889 | CLUSTER_RESYNC_WINDOW_SECTORS : window_size; | |
2890 | ||
2891 | conf->cluster_sync_high = conf->cluster_sync_low + window_size; | |
2892 | } | |
2893 | ||
1da177e4 LT |
2894 | /* |
2895 | * perform a "sync" on one "block" | |
2896 | * | |
2897 | * We need to make sure that no normal I/O request - particularly write | |
2898 | * requests - conflict with active sync requests. | |
2899 | * | |
2900 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2901 | * that can be installed to exclude normal IO requests. | |
2902 | * | |
2903 | * Resync and recovery are handled very differently. | |
2904 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2905 | * | |
2906 | * For resync, we iterate over virtual addresses, read all copies, | |
2907 | * and update if there are differences. If only one copy is live, | |
2908 | * skip it. | |
2909 | * For recovery, we iterate over physical addresses, read a good | |
2910 | * value for each non-in_sync drive, and over-write. | |
2911 | * | |
2912 | * So, for recovery we may have several outstanding complex requests for a | |
2913 | * given address, one for each out-of-sync device. We model this by allocating | |
2914 | * a number of r10_bio structures, one for each out-of-sync device. | |
2915 | * As we setup these structures, we collect all bio's together into a list | |
2916 | * which we then process collectively to add pages, and then process again | |
2917 | * to pass to generic_make_request. | |
2918 | * | |
2919 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2920 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2921 | * has its remaining count decremented to 0, the whole complex operation | |
2922 | * is complete. | |
2923 | * | |
2924 | */ | |
2925 | ||
849674e4 | 2926 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 2927 | int *skipped) |
1da177e4 | 2928 | { |
e879a879 | 2929 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2930 | struct r10bio *r10_bio; |
1da177e4 LT |
2931 | struct bio *biolist = NULL, *bio; |
2932 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2933 | int i; |
6cce3b23 | 2934 | int max_sync; |
57dab0bd | 2935 | sector_t sync_blocks; |
1da177e4 LT |
2936 | sector_t sectors_skipped = 0; |
2937 | int chunks_skipped = 0; | |
5cf00fcd | 2938 | sector_t chunk_mask = conf->geo.chunk_mask; |
022e510f | 2939 | int page_idx = 0; |
1da177e4 LT |
2940 | |
2941 | if (!conf->r10buf_pool) | |
2942 | if (init_resync(conf)) | |
57afd89f | 2943 | return 0; |
1da177e4 | 2944 | |
7e83ccbe MW |
2945 | /* |
2946 | * Allow skipping a full rebuild for incremental assembly | |
2947 | * of a clean array, like RAID1 does. | |
2948 | */ | |
2949 | if (mddev->bitmap == NULL && | |
2950 | mddev->recovery_cp == MaxSector && | |
13765120 N |
2951 | mddev->reshape_position == MaxSector && |
2952 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 2953 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 2954 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
2955 | conf->fullsync == 0) { |
2956 | *skipped = 1; | |
13765120 | 2957 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
2958 | } |
2959 | ||
1da177e4 | 2960 | skipped: |
58c0fed4 | 2961 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
2962 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
2963 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
2964 | max_sector = mddev->resync_max_sectors; |
2965 | if (sector_nr >= max_sector) { | |
8db87912 GJ |
2966 | conf->cluster_sync_low = 0; |
2967 | conf->cluster_sync_high = 0; | |
2968 | ||
6cce3b23 N |
2969 | /* If we aborted, we need to abort the |
2970 | * sync on the 'current' bitmap chucks (there can | |
2971 | * be several when recovering multiple devices). | |
2972 | * as we may have started syncing it but not finished. | |
2973 | * We can find the current address in | |
2974 | * mddev->curr_resync, but for recovery, | |
2975 | * we need to convert that to several | |
2976 | * virtual addresses. | |
2977 | */ | |
3ea7daa5 N |
2978 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
2979 | end_reshape(conf); | |
b3968552 | 2980 | close_sync(conf); |
3ea7daa5 N |
2981 | return 0; |
2982 | } | |
2983 | ||
6cce3b23 N |
2984 | if (mddev->curr_resync < max_sector) { /* aborted */ |
2985 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
2986 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
2987 | &sync_blocks, 1); | |
5cf00fcd | 2988 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
2989 | sector_t sect = |
2990 | raid10_find_virt(conf, mddev->curr_resync, i); | |
2991 | bitmap_end_sync(mddev->bitmap, sect, | |
2992 | &sync_blocks, 1); | |
2993 | } | |
9ad1aefc N |
2994 | } else { |
2995 | /* completed sync */ | |
2996 | if ((!mddev->bitmap || conf->fullsync) | |
2997 | && conf->have_replacement | |
2998 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
2999 | /* Completed a full sync so the replacements | |
3000 | * are now fully recovered. | |
3001 | */ | |
f90145f3 N |
3002 | rcu_read_lock(); |
3003 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3004 | struct md_rdev *rdev = | |
3005 | rcu_dereference(conf->mirrors[i].replacement); | |
3006 | if (rdev) | |
3007 | rdev->recovery_offset = MaxSector; | |
3008 | } | |
3009 | rcu_read_unlock(); | |
9ad1aefc | 3010 | } |
6cce3b23 | 3011 | conf->fullsync = 0; |
9ad1aefc | 3012 | } |
6cce3b23 | 3013 | bitmap_close_sync(mddev->bitmap); |
1da177e4 | 3014 | close_sync(conf); |
57afd89f | 3015 | *skipped = 1; |
1da177e4 LT |
3016 | return sectors_skipped; |
3017 | } | |
3ea7daa5 N |
3018 | |
3019 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
3020 | return reshape_request(mddev, sector_nr, skipped); | |
3021 | ||
5cf00fcd | 3022 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
3023 | /* if there has been nothing to do on any drive, |
3024 | * then there is nothing to do at all.. | |
3025 | */ | |
57afd89f N |
3026 | *skipped = 1; |
3027 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
3028 | } |
3029 | ||
c6207277 N |
3030 | if (max_sector > mddev->resync_max) |
3031 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
3032 | ||
1da177e4 LT |
3033 | /* make sure whole request will fit in a chunk - if chunks |
3034 | * are meaningful | |
3035 | */ | |
5cf00fcd N |
3036 | if (conf->geo.near_copies < conf->geo.raid_disks && |
3037 | max_sector > (sector_nr | chunk_mask)) | |
3038 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 | 3039 | |
7ac50447 TM |
3040 | /* |
3041 | * If there is non-resync activity waiting for a turn, then let it | |
3042 | * though before starting on this new sync request. | |
3043 | */ | |
3044 | if (conf->nr_waiting) | |
3045 | schedule_timeout_uninterruptible(1); | |
3046 | ||
1da177e4 LT |
3047 | /* Again, very different code for resync and recovery. |
3048 | * Both must result in an r10bio with a list of bios that | |
74d46992 | 3049 | * have bi_end_io, bi_sector, bi_disk set, |
1da177e4 LT |
3050 | * and bi_private set to the r10bio. |
3051 | * For recovery, we may actually create several r10bios | |
3052 | * with 2 bios in each, that correspond to the bios in the main one. | |
3053 | * In this case, the subordinate r10bios link back through a | |
3054 | * borrowed master_bio pointer, and the counter in the master | |
3055 | * includes a ref from each subordinate. | |
3056 | */ | |
3057 | /* First, we decide what to do and set ->bi_end_io | |
3058 | * To end_sync_read if we want to read, and | |
3059 | * end_sync_write if we will want to write. | |
3060 | */ | |
3061 | ||
6cce3b23 | 3062 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
3063 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
3064 | /* recovery... the complicated one */ | |
e875ecea | 3065 | int j; |
1da177e4 LT |
3066 | r10_bio = NULL; |
3067 | ||
5cf00fcd | 3068 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 3069 | int still_degraded; |
9f2c9d12 | 3070 | struct r10bio *rb2; |
ab9d47e9 N |
3071 | sector_t sect; |
3072 | int must_sync; | |
e875ecea | 3073 | int any_working; |
dc280d98 | 3074 | struct raid10_info *mirror = &conf->mirrors[i]; |
f90145f3 | 3075 | struct md_rdev *mrdev, *mreplace; |
24afd80d | 3076 | |
f90145f3 N |
3077 | rcu_read_lock(); |
3078 | mrdev = rcu_dereference(mirror->rdev); | |
3079 | mreplace = rcu_dereference(mirror->replacement); | |
3080 | ||
3081 | if ((mrdev == NULL || | |
f5b67ae8 | 3082 | test_bit(Faulty, &mrdev->flags) || |
f90145f3 N |
3083 | test_bit(In_sync, &mrdev->flags)) && |
3084 | (mreplace == NULL || | |
3085 | test_bit(Faulty, &mreplace->flags))) { | |
3086 | rcu_read_unlock(); | |
ab9d47e9 | 3087 | continue; |
f90145f3 | 3088 | } |
1da177e4 | 3089 | |
ab9d47e9 N |
3090 | still_degraded = 0; |
3091 | /* want to reconstruct this device */ | |
3092 | rb2 = r10_bio; | |
3093 | sect = raid10_find_virt(conf, sector_nr, i); | |
fc448a18 N |
3094 | if (sect >= mddev->resync_max_sectors) { |
3095 | /* last stripe is not complete - don't | |
3096 | * try to recover this sector. | |
3097 | */ | |
f90145f3 | 3098 | rcu_read_unlock(); |
fc448a18 N |
3099 | continue; |
3100 | } | |
f5b67ae8 N |
3101 | if (mreplace && test_bit(Faulty, &mreplace->flags)) |
3102 | mreplace = NULL; | |
24afd80d N |
3103 | /* Unless we are doing a full sync, or a replacement |
3104 | * we only need to recover the block if it is set in | |
3105 | * the bitmap | |
ab9d47e9 N |
3106 | */ |
3107 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3108 | &sync_blocks, 1); | |
3109 | if (sync_blocks < max_sync) | |
3110 | max_sync = sync_blocks; | |
3111 | if (!must_sync && | |
f90145f3 | 3112 | mreplace == NULL && |
ab9d47e9 N |
3113 | !conf->fullsync) { |
3114 | /* yep, skip the sync_blocks here, but don't assume | |
3115 | * that there will never be anything to do here | |
3116 | */ | |
3117 | chunks_skipped = -1; | |
f90145f3 | 3118 | rcu_read_unlock(); |
ab9d47e9 N |
3119 | continue; |
3120 | } | |
f90145f3 N |
3121 | atomic_inc(&mrdev->nr_pending); |
3122 | if (mreplace) | |
3123 | atomic_inc(&mreplace->nr_pending); | |
3124 | rcu_read_unlock(); | |
6cce3b23 | 3125 | |
208410b5 | 3126 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3127 | r10_bio->state = 0; |
ab9d47e9 N |
3128 | raise_barrier(conf, rb2 != NULL); |
3129 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 3130 | |
ab9d47e9 N |
3131 | r10_bio->master_bio = (struct bio*)rb2; |
3132 | if (rb2) | |
3133 | atomic_inc(&rb2->remaining); | |
3134 | r10_bio->mddev = mddev; | |
3135 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
3136 | r10_bio->sector = sect; | |
1da177e4 | 3137 | |
ab9d47e9 N |
3138 | raid10_find_phys(conf, r10_bio); |
3139 | ||
3140 | /* Need to check if the array will still be | |
3141 | * degraded | |
3142 | */ | |
f90145f3 N |
3143 | rcu_read_lock(); |
3144 | for (j = 0; j < conf->geo.raid_disks; j++) { | |
3145 | struct md_rdev *rdev = rcu_dereference( | |
3146 | conf->mirrors[j].rdev); | |
3147 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
ab9d47e9 | 3148 | still_degraded = 1; |
87fc767b | 3149 | break; |
1da177e4 | 3150 | } |
f90145f3 | 3151 | } |
ab9d47e9 N |
3152 | |
3153 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3154 | &sync_blocks, still_degraded); | |
3155 | ||
e875ecea | 3156 | any_working = 0; |
ab9d47e9 | 3157 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3158 | int k; |
ab9d47e9 | 3159 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3160 | sector_t from_addr, to_addr; |
f90145f3 N |
3161 | struct md_rdev *rdev = |
3162 | rcu_dereference(conf->mirrors[d].rdev); | |
40c356ce N |
3163 | sector_t sector, first_bad; |
3164 | int bad_sectors; | |
f90145f3 N |
3165 | if (!rdev || |
3166 | !test_bit(In_sync, &rdev->flags)) | |
ab9d47e9 N |
3167 | continue; |
3168 | /* This is where we read from */ | |
e875ecea | 3169 | any_working = 1; |
40c356ce N |
3170 | sector = r10_bio->devs[j].addr; |
3171 | ||
3172 | if (is_badblock(rdev, sector, max_sync, | |
3173 | &first_bad, &bad_sectors)) { | |
3174 | if (first_bad > sector) | |
3175 | max_sync = first_bad - sector; | |
3176 | else { | |
3177 | bad_sectors -= (sector | |
3178 | - first_bad); | |
3179 | if (max_sync > bad_sectors) | |
3180 | max_sync = bad_sectors; | |
3181 | continue; | |
3182 | } | |
3183 | } | |
ab9d47e9 N |
3184 | bio = r10_bio->devs[0].bio; |
3185 | bio->bi_next = biolist; | |
3186 | biolist = bio; | |
ab9d47e9 | 3187 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3188 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3189 | if (test_bit(FailFast, &rdev->flags)) |
3190 | bio->bi_opf |= MD_FAILFAST; | |
5e570289 | 3191 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3192 | bio->bi_iter.bi_sector = from_addr + |
3193 | rdev->data_offset; | |
74d46992 | 3194 | bio_set_dev(bio, rdev->bdev); |
24afd80d N |
3195 | atomic_inc(&rdev->nr_pending); |
3196 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3197 | |
3198 | for (k=0; k<conf->copies; k++) | |
3199 | if (r10_bio->devs[k].devnum == i) | |
3200 | break; | |
3201 | BUG_ON(k == conf->copies); | |
5e570289 | 3202 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3203 | r10_bio->devs[0].devnum = d; |
5e570289 | 3204 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3205 | r10_bio->devs[1].devnum = i; |
5e570289 | 3206 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3207 | |
f90145f3 | 3208 | if (!test_bit(In_sync, &mrdev->flags)) { |
24afd80d N |
3209 | bio = r10_bio->devs[1].bio; |
3210 | bio->bi_next = biolist; | |
3211 | biolist = bio; | |
24afd80d | 3212 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3213 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3214 | bio->bi_iter.bi_sector = to_addr |
f90145f3 | 3215 | + mrdev->data_offset; |
74d46992 | 3216 | bio_set_dev(bio, mrdev->bdev); |
24afd80d N |
3217 | atomic_inc(&r10_bio->remaining); |
3218 | } else | |
3219 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3220 | ||
3221 | /* and maybe write to replacement */ | |
3222 | bio = r10_bio->devs[1].repl_bio; | |
3223 | if (bio) | |
3224 | bio->bi_end_io = NULL; | |
f90145f3 | 3225 | /* Note: if mreplace != NULL, then bio |
24afd80d N |
3226 | * cannot be NULL as r10buf_pool_alloc will |
3227 | * have allocated it. | |
3228 | * So the second test here is pointless. | |
3229 | * But it keeps semantic-checkers happy, and | |
3230 | * this comment keeps human reviewers | |
3231 | * happy. | |
3232 | */ | |
f90145f3 N |
3233 | if (mreplace == NULL || bio == NULL || |
3234 | test_bit(Faulty, &mreplace->flags)) | |
24afd80d N |
3235 | break; |
3236 | bio->bi_next = biolist; | |
3237 | biolist = bio; | |
24afd80d | 3238 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3239 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3240 | bio->bi_iter.bi_sector = to_addr + |
f90145f3 | 3241 | mreplace->data_offset; |
74d46992 | 3242 | bio_set_dev(bio, mreplace->bdev); |
24afd80d | 3243 | atomic_inc(&r10_bio->remaining); |
ab9d47e9 N |
3244 | break; |
3245 | } | |
f90145f3 | 3246 | rcu_read_unlock(); |
ab9d47e9 | 3247 | if (j == conf->copies) { |
e875ecea N |
3248 | /* Cannot recover, so abort the recovery or |
3249 | * record a bad block */ | |
e875ecea N |
3250 | if (any_working) { |
3251 | /* problem is that there are bad blocks | |
3252 | * on other device(s) | |
3253 | */ | |
3254 | int k; | |
3255 | for (k = 0; k < conf->copies; k++) | |
3256 | if (r10_bio->devs[k].devnum == i) | |
3257 | break; | |
24afd80d | 3258 | if (!test_bit(In_sync, |
f90145f3 | 3259 | &mrdev->flags) |
24afd80d | 3260 | && !rdev_set_badblocks( |
f90145f3 | 3261 | mrdev, |
24afd80d N |
3262 | r10_bio->devs[k].addr, |
3263 | max_sync, 0)) | |
3264 | any_working = 0; | |
f90145f3 | 3265 | if (mreplace && |
24afd80d | 3266 | !rdev_set_badblocks( |
f90145f3 | 3267 | mreplace, |
e875ecea N |
3268 | r10_bio->devs[k].addr, |
3269 | max_sync, 0)) | |
3270 | any_working = 0; | |
3271 | } | |
3272 | if (!any_working) { | |
3273 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3274 | &mddev->recovery)) | |
08464e09 | 3275 | pr_warn("md/raid10:%s: insufficient working devices for recovery.\n", |
e875ecea | 3276 | mdname(mddev)); |
24afd80d | 3277 | mirror->recovery_disabled |
e875ecea N |
3278 | = mddev->recovery_disabled; |
3279 | } | |
e8b84915 N |
3280 | put_buf(r10_bio); |
3281 | if (rb2) | |
3282 | atomic_dec(&rb2->remaining); | |
3283 | r10_bio = rb2; | |
f90145f3 N |
3284 | rdev_dec_pending(mrdev, mddev); |
3285 | if (mreplace) | |
3286 | rdev_dec_pending(mreplace, mddev); | |
ab9d47e9 | 3287 | break; |
1da177e4 | 3288 | } |
f90145f3 N |
3289 | rdev_dec_pending(mrdev, mddev); |
3290 | if (mreplace) | |
3291 | rdev_dec_pending(mreplace, mddev); | |
8d3ca83d N |
3292 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { |
3293 | /* Only want this if there is elsewhere to | |
3294 | * read from. 'j' is currently the first | |
3295 | * readable copy. | |
3296 | */ | |
3297 | int targets = 1; | |
3298 | for (; j < conf->copies; j++) { | |
3299 | int d = r10_bio->devs[j].devnum; | |
3300 | if (conf->mirrors[d].rdev && | |
3301 | test_bit(In_sync, | |
3302 | &conf->mirrors[d].rdev->flags)) | |
3303 | targets++; | |
3304 | } | |
3305 | if (targets == 1) | |
3306 | r10_bio->devs[0].bio->bi_opf | |
3307 | &= ~MD_FAILFAST; | |
3308 | } | |
ab9d47e9 | 3309 | } |
1da177e4 LT |
3310 | if (biolist == NULL) { |
3311 | while (r10_bio) { | |
9f2c9d12 N |
3312 | struct r10bio *rb2 = r10_bio; |
3313 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3314 | rb2->master_bio = NULL; |
3315 | put_buf(rb2); | |
3316 | } | |
3317 | goto giveup; | |
3318 | } | |
3319 | } else { | |
3320 | /* resync. Schedule a read for every block at this virt offset */ | |
3321 | int count = 0; | |
6cce3b23 | 3322 | |
8db87912 GJ |
3323 | /* |
3324 | * Since curr_resync_completed could probably not update in | |
3325 | * time, and we will set cluster_sync_low based on it. | |
3326 | * Let's check against "sector_nr + 2 * RESYNC_SECTORS" for | |
3327 | * safety reason, which ensures curr_resync_completed is | |
3328 | * updated in bitmap_cond_end_sync. | |
3329 | */ | |
3330 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, | |
3331 | mddev_is_clustered(mddev) && | |
3332 | (sector_nr + 2 * RESYNC_SECTORS > | |
3333 | conf->cluster_sync_high)); | |
78200d45 | 3334 | |
6cce3b23 N |
3335 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
3336 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3337 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3338 | &mddev->recovery)) { | |
6cce3b23 N |
3339 | /* We can skip this block */ |
3340 | *skipped = 1; | |
3341 | return sync_blocks + sectors_skipped; | |
3342 | } | |
3343 | if (sync_blocks < max_sync) | |
3344 | max_sync = sync_blocks; | |
208410b5 | 3345 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3346 | r10_bio->state = 0; |
1da177e4 | 3347 | |
1da177e4 LT |
3348 | r10_bio->mddev = mddev; |
3349 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3350 | raise_barrier(conf, 0); |
3351 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3352 | |
3353 | r10_bio->master_bio = NULL; | |
3354 | r10_bio->sector = sector_nr; | |
3355 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3356 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3357 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3358 | |
5cf00fcd | 3359 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3360 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3361 | sector_t first_bad, sector; |
3362 | int bad_sectors; | |
f90145f3 | 3363 | struct md_rdev *rdev; |
40c356ce | 3364 | |
9ad1aefc N |
3365 | if (r10_bio->devs[i].repl_bio) |
3366 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3367 | ||
1da177e4 | 3368 | bio = r10_bio->devs[i].bio; |
4e4cbee9 | 3369 | bio->bi_status = BLK_STS_IOERR; |
f90145f3 N |
3370 | rcu_read_lock(); |
3371 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
3372 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3373 | rcu_read_unlock(); | |
1da177e4 | 3374 | continue; |
f90145f3 | 3375 | } |
40c356ce | 3376 | sector = r10_bio->devs[i].addr; |
f90145f3 | 3377 | if (is_badblock(rdev, sector, max_sync, |
40c356ce N |
3378 | &first_bad, &bad_sectors)) { |
3379 | if (first_bad > sector) | |
3380 | max_sync = first_bad - sector; | |
3381 | else { | |
3382 | bad_sectors -= (sector - first_bad); | |
3383 | if (max_sync > bad_sectors) | |
91502f09 | 3384 | max_sync = bad_sectors; |
f90145f3 | 3385 | rcu_read_unlock(); |
40c356ce N |
3386 | continue; |
3387 | } | |
3388 | } | |
f90145f3 | 3389 | atomic_inc(&rdev->nr_pending); |
1da177e4 LT |
3390 | atomic_inc(&r10_bio->remaining); |
3391 | bio->bi_next = biolist; | |
3392 | biolist = bio; | |
1da177e4 | 3393 | bio->bi_end_io = end_sync_read; |
796a5cf0 | 3394 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
1cdd1257 | 3395 | if (test_bit(FailFast, &rdev->flags)) |
8d3ca83d | 3396 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3397 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3398 | bio_set_dev(bio, rdev->bdev); |
1da177e4 | 3399 | count++; |
9ad1aefc | 3400 | |
f90145f3 N |
3401 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3402 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3403 | rcu_read_unlock(); | |
9ad1aefc | 3404 | continue; |
f90145f3 N |
3405 | } |
3406 | atomic_inc(&rdev->nr_pending); | |
9ad1aefc N |
3407 | |
3408 | /* Need to set up for writing to the replacement */ | |
3409 | bio = r10_bio->devs[i].repl_bio; | |
4e4cbee9 | 3410 | bio->bi_status = BLK_STS_IOERR; |
9ad1aefc N |
3411 | |
3412 | sector = r10_bio->devs[i].addr; | |
9ad1aefc N |
3413 | bio->bi_next = biolist; |
3414 | biolist = bio; | |
9ad1aefc | 3415 | bio->bi_end_io = end_sync_write; |
796a5cf0 | 3416 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1cdd1257 | 3417 | if (test_bit(FailFast, &rdev->flags)) |
1919cbb2 | 3418 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3419 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3420 | bio_set_dev(bio, rdev->bdev); |
9ad1aefc | 3421 | count++; |
1cdd1257 | 3422 | rcu_read_unlock(); |
1da177e4 LT |
3423 | } |
3424 | ||
3425 | if (count < 2) { | |
3426 | for (i=0; i<conf->copies; i++) { | |
3427 | int d = r10_bio->devs[i].devnum; | |
3428 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3429 | rdev_dec_pending(conf->mirrors[d].rdev, |
3430 | mddev); | |
9ad1aefc N |
3431 | if (r10_bio->devs[i].repl_bio && |
3432 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3433 | rdev_dec_pending( | |
3434 | conf->mirrors[d].replacement, | |
3435 | mddev); | |
1da177e4 LT |
3436 | } |
3437 | put_buf(r10_bio); | |
3438 | biolist = NULL; | |
3439 | goto giveup; | |
3440 | } | |
3441 | } | |
3442 | ||
1da177e4 | 3443 | nr_sectors = 0; |
6cce3b23 N |
3444 | if (sector_nr + max_sync < max_sector) |
3445 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3446 | do { |
3447 | struct page *page; | |
3448 | int len = PAGE_SIZE; | |
1da177e4 LT |
3449 | if (sector_nr + (len>>9) > max_sector) |
3450 | len = (max_sector - sector_nr) << 9; | |
3451 | if (len == 0) | |
3452 | break; | |
3453 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
f0250618 | 3454 | struct resync_pages *rp = get_resync_pages(bio); |
022e510f | 3455 | page = resync_fetch_page(rp, page_idx); |
c85ba149 ML |
3456 | /* |
3457 | * won't fail because the vec table is big enough | |
3458 | * to hold all these pages | |
3459 | */ | |
3460 | bio_add_page(bio, page, len, 0); | |
1da177e4 LT |
3461 | } |
3462 | nr_sectors += len>>9; | |
3463 | sector_nr += len>>9; | |
022e510f | 3464 | } while (++page_idx < RESYNC_PAGES); |
1da177e4 LT |
3465 | r10_bio->sectors = nr_sectors; |
3466 | ||
8db87912 GJ |
3467 | if (mddev_is_clustered(mddev) && |
3468 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3469 | /* It is resync not recovery */ | |
3470 | if (conf->cluster_sync_high < sector_nr + nr_sectors) { | |
3471 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
3472 | raid10_set_cluster_sync_high(conf); | |
3473 | /* Send resync message */ | |
3474 | md_cluster_ops->resync_info_update(mddev, | |
3475 | conf->cluster_sync_low, | |
3476 | conf->cluster_sync_high); | |
3477 | } | |
3478 | } else if (mddev_is_clustered(mddev)) { | |
3479 | /* This is recovery not resync */ | |
3480 | sector_t sect_va1, sect_va2; | |
3481 | bool broadcast_msg = false; | |
3482 | ||
3483 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3484 | /* | |
3485 | * sector_nr is a device address for recovery, so we | |
3486 | * need translate it to array address before compare | |
3487 | * with cluster_sync_high. | |
3488 | */ | |
3489 | sect_va1 = raid10_find_virt(conf, sector_nr, i); | |
3490 | ||
3491 | if (conf->cluster_sync_high < sect_va1 + nr_sectors) { | |
3492 | broadcast_msg = true; | |
3493 | /* | |
3494 | * curr_resync_completed is similar as | |
3495 | * sector_nr, so make the translation too. | |
3496 | */ | |
3497 | sect_va2 = raid10_find_virt(conf, | |
3498 | mddev->curr_resync_completed, i); | |
3499 | ||
3500 | if (conf->cluster_sync_low == 0 || | |
3501 | conf->cluster_sync_low > sect_va2) | |
3502 | conf->cluster_sync_low = sect_va2; | |
3503 | } | |
3504 | } | |
3505 | if (broadcast_msg) { | |
3506 | raid10_set_cluster_sync_high(conf); | |
3507 | md_cluster_ops->resync_info_update(mddev, | |
3508 | conf->cluster_sync_low, | |
3509 | conf->cluster_sync_high); | |
3510 | } | |
3511 | } | |
3512 | ||
1da177e4 LT |
3513 | while (biolist) { |
3514 | bio = biolist; | |
3515 | biolist = biolist->bi_next; | |
3516 | ||
3517 | bio->bi_next = NULL; | |
f0250618 | 3518 | r10_bio = get_resync_r10bio(bio); |
1da177e4 LT |
3519 | r10_bio->sectors = nr_sectors; |
3520 | ||
3521 | if (bio->bi_end_io == end_sync_read) { | |
74d46992 | 3522 | md_sync_acct_bio(bio, nr_sectors); |
4e4cbee9 | 3523 | bio->bi_status = 0; |
1da177e4 LT |
3524 | generic_make_request(bio); |
3525 | } | |
3526 | } | |
3527 | ||
57afd89f N |
3528 | if (sectors_skipped) |
3529 | /* pretend they weren't skipped, it makes | |
3530 | * no important difference in this case | |
3531 | */ | |
3532 | md_done_sync(mddev, sectors_skipped, 1); | |
3533 | ||
1da177e4 LT |
3534 | return sectors_skipped + nr_sectors; |
3535 | giveup: | |
3536 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3537 | * drives must be failed or in resync, all drives |
3538 | * have a bad block, so try the next chunk... | |
1da177e4 | 3539 | */ |
09b4068a N |
3540 | if (sector_nr + max_sync < max_sector) |
3541 | max_sector = sector_nr + max_sync; | |
3542 | ||
3543 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3544 | chunks_skipped ++; |
3545 | sector_nr = max_sector; | |
1da177e4 | 3546 | goto skipped; |
1da177e4 LT |
3547 | } |
3548 | ||
80c3a6ce | 3549 | static sector_t |
fd01b88c | 3550 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3551 | { |
3552 | sector_t size; | |
e879a879 | 3553 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3554 | |
3555 | if (!raid_disks) | |
3ea7daa5 N |
3556 | raid_disks = min(conf->geo.raid_disks, |
3557 | conf->prev.raid_disks); | |
80c3a6ce | 3558 | if (!sectors) |
dab8b292 | 3559 | sectors = conf->dev_sectors; |
80c3a6ce | 3560 | |
5cf00fcd N |
3561 | size = sectors >> conf->geo.chunk_shift; |
3562 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3563 | size = size * raid_disks; |
5cf00fcd | 3564 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3565 | |
5cf00fcd | 3566 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3567 | } |
3568 | ||
6508fdbf N |
3569 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3570 | { | |
3571 | /* Calculate the number of sectors-per-device that will | |
3572 | * actually be used, and set conf->dev_sectors and | |
3573 | * conf->stride | |
3574 | */ | |
3575 | ||
5cf00fcd N |
3576 | size = size >> conf->geo.chunk_shift; |
3577 | sector_div(size, conf->geo.far_copies); | |
3578 | size = size * conf->geo.raid_disks; | |
3579 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3580 | /* 'size' is now the number of chunks in the array */ |
3581 | /* calculate "used chunks per device" */ | |
3582 | size = size * conf->copies; | |
3583 | ||
3584 | /* We need to round up when dividing by raid_disks to | |
3585 | * get the stride size. | |
3586 | */ | |
5cf00fcd | 3587 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3588 | |
5cf00fcd | 3589 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3590 | |
5cf00fcd N |
3591 | if (conf->geo.far_offset) |
3592 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3593 | else { |
5cf00fcd N |
3594 | sector_div(size, conf->geo.far_copies); |
3595 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3596 | } |
3597 | } | |
dab8b292 | 3598 | |
deb200d0 N |
3599 | enum geo_type {geo_new, geo_old, geo_start}; |
3600 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3601 | { | |
3602 | int nc, fc, fo; | |
3603 | int layout, chunk, disks; | |
3604 | switch (new) { | |
3605 | case geo_old: | |
3606 | layout = mddev->layout; | |
3607 | chunk = mddev->chunk_sectors; | |
3608 | disks = mddev->raid_disks - mddev->delta_disks; | |
3609 | break; | |
3610 | case geo_new: | |
3611 | layout = mddev->new_layout; | |
3612 | chunk = mddev->new_chunk_sectors; | |
3613 | disks = mddev->raid_disks; | |
3614 | break; | |
3615 | default: /* avoid 'may be unused' warnings */ | |
3616 | case geo_start: /* new when starting reshape - raid_disks not | |
3617 | * updated yet. */ | |
3618 | layout = mddev->new_layout; | |
3619 | chunk = mddev->new_chunk_sectors; | |
3620 | disks = mddev->raid_disks + mddev->delta_disks; | |
3621 | break; | |
3622 | } | |
8bce6d35 | 3623 | if (layout >> 19) |
deb200d0 N |
3624 | return -1; |
3625 | if (chunk < (PAGE_SIZE >> 9) || | |
3626 | !is_power_of_2(chunk)) | |
3627 | return -2; | |
3628 | nc = layout & 255; | |
3629 | fc = (layout >> 8) & 255; | |
3630 | fo = layout & (1<<16); | |
3631 | geo->raid_disks = disks; | |
3632 | geo->near_copies = nc; | |
3633 | geo->far_copies = fc; | |
3634 | geo->far_offset = fo; | |
8bce6d35 N |
3635 | switch (layout >> 17) { |
3636 | case 0: /* original layout. simple but not always optimal */ | |
3637 | geo->far_set_size = disks; | |
3638 | break; | |
3639 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3640 | * actually using this, but leave code here just in case.*/ | |
3641 | geo->far_set_size = disks/fc; | |
3642 | WARN(geo->far_set_size < fc, | |
3643 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3644 | break; | |
3645 | case 2: /* "improved" layout fixed to match documentation */ | |
3646 | geo->far_set_size = fc * nc; | |
3647 | break; | |
3648 | default: /* Not a valid layout */ | |
3649 | return -1; | |
3650 | } | |
deb200d0 N |
3651 | geo->chunk_mask = chunk - 1; |
3652 | geo->chunk_shift = ffz(~chunk); | |
3653 | return nc*fc; | |
3654 | } | |
3655 | ||
e879a879 | 3656 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3657 | { |
e879a879 | 3658 | struct r10conf *conf = NULL; |
dab8b292 | 3659 | int err = -EINVAL; |
deb200d0 N |
3660 | struct geom geo; |
3661 | int copies; | |
3662 | ||
3663 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3664 | |
deb200d0 | 3665 | if (copies == -2) { |
08464e09 N |
3666 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n", |
3667 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3668 | goto out; |
1da177e4 | 3669 | } |
2604b703 | 3670 | |
deb200d0 | 3671 | if (copies < 2 || copies > mddev->raid_disks) { |
08464e09 N |
3672 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
3673 | mdname(mddev), mddev->new_layout); | |
1da177e4 LT |
3674 | goto out; |
3675 | } | |
dab8b292 TM |
3676 | |
3677 | err = -ENOMEM; | |
e879a879 | 3678 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3679 | if (!conf) |
1da177e4 | 3680 | goto out; |
dab8b292 | 3681 | |
3ea7daa5 | 3682 | /* FIXME calc properly */ |
dc280d98 | 3683 | conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks + |
78eaa0d4 | 3684 | max(0,-mddev->delta_disks)), |
dab8b292 TM |
3685 | GFP_KERNEL); |
3686 | if (!conf->mirrors) | |
3687 | goto out; | |
4443ae10 N |
3688 | |
3689 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3690 | if (!conf->tmppage) | |
dab8b292 TM |
3691 | goto out; |
3692 | ||
deb200d0 N |
3693 | conf->geo = geo; |
3694 | conf->copies = copies; | |
dab8b292 TM |
3695 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, |
3696 | r10bio_pool_free, conf); | |
3697 | if (!conf->r10bio_pool) | |
3698 | goto out; | |
3699 | ||
011067b0 | 3700 | conf->bio_split = bioset_create(BIO_POOL_SIZE, 0, 0); |
fc9977dd N |
3701 | if (!conf->bio_split) |
3702 | goto out; | |
3703 | ||
6508fdbf | 3704 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3705 | if (mddev->reshape_position == MaxSector) { |
3706 | conf->prev = conf->geo; | |
3707 | conf->reshape_progress = MaxSector; | |
3708 | } else { | |
3709 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3710 | err = -EINVAL; | |
3711 | goto out; | |
3712 | } | |
3713 | conf->reshape_progress = mddev->reshape_position; | |
3714 | if (conf->prev.far_offset) | |
3715 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
3716 | else | |
3717 | /* far_copies must be 1 */ | |
3718 | conf->prev.stride = conf->dev_sectors; | |
3719 | } | |
299b0685 | 3720 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 3721 | spin_lock_init(&conf->device_lock); |
dab8b292 | 3722 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 3723 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 TM |
3724 | |
3725 | spin_lock_init(&conf->resync_lock); | |
3726 | init_waitqueue_head(&conf->wait_barrier); | |
0e5313e2 | 3727 | atomic_set(&conf->nr_pending, 0); |
dab8b292 | 3728 | |
0232605d | 3729 | conf->thread = md_register_thread(raid10d, mddev, "raid10"); |
dab8b292 TM |
3730 | if (!conf->thread) |
3731 | goto out; | |
3732 | ||
dab8b292 TM |
3733 | conf->mddev = mddev; |
3734 | return conf; | |
3735 | ||
3736 | out: | |
dab8b292 | 3737 | if (conf) { |
644df1a8 | 3738 | mempool_destroy(conf->r10bio_pool); |
dab8b292 TM |
3739 | kfree(conf->mirrors); |
3740 | safe_put_page(conf->tmppage); | |
fc9977dd N |
3741 | if (conf->bio_split) |
3742 | bioset_free(conf->bio_split); | |
dab8b292 TM |
3743 | kfree(conf); |
3744 | } | |
3745 | return ERR_PTR(err); | |
3746 | } | |
3747 | ||
849674e4 | 3748 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 3749 | { |
e879a879 | 3750 | struct r10conf *conf; |
dab8b292 | 3751 | int i, disk_idx, chunk_size; |
dc280d98 | 3752 | struct raid10_info *disk; |
3cb03002 | 3753 | struct md_rdev *rdev; |
dab8b292 | 3754 | sector_t size; |
3ea7daa5 N |
3755 | sector_t min_offset_diff = 0; |
3756 | int first = 1; | |
532a2a3f | 3757 | bool discard_supported = false; |
dab8b292 | 3758 | |
a415c0f1 N |
3759 | if (mddev_init_writes_pending(mddev) < 0) |
3760 | return -ENOMEM; | |
3761 | ||
dab8b292 TM |
3762 | if (mddev->private == NULL) { |
3763 | conf = setup_conf(mddev); | |
3764 | if (IS_ERR(conf)) | |
3765 | return PTR_ERR(conf); | |
3766 | mddev->private = conf; | |
3767 | } | |
3768 | conf = mddev->private; | |
3769 | if (!conf) | |
3770 | goto out; | |
3771 | ||
8db87912 GJ |
3772 | if (mddev_is_clustered(conf->mddev)) { |
3773 | int fc, fo; | |
3774 | ||
3775 | fc = (mddev->layout >> 8) & 255; | |
3776 | fo = mddev->layout & (1<<16); | |
3777 | if (fc > 1 || fo > 0) { | |
3778 | pr_err("only near layout is supported by clustered" | |
3779 | " raid10\n"); | |
3780 | goto out; | |
3781 | } | |
3782 | } | |
3783 | ||
dab8b292 TM |
3784 | mddev->thread = conf->thread; |
3785 | conf->thread = NULL; | |
3786 | ||
8f6c2e4b | 3787 | chunk_size = mddev->chunk_sectors << 9; |
cc4d1efd | 3788 | if (mddev->queue) { |
532a2a3f SL |
3789 | blk_queue_max_discard_sectors(mddev->queue, |
3790 | mddev->chunk_sectors); | |
5026d7a9 | 3791 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 3792 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
cc4d1efd JB |
3793 | blk_queue_io_min(mddev->queue, chunk_size); |
3794 | if (conf->geo.raid_disks % conf->geo.near_copies) | |
3795 | blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks); | |
3796 | else | |
3797 | blk_queue_io_opt(mddev->queue, chunk_size * | |
3798 | (conf->geo.raid_disks / conf->geo.near_copies)); | |
3799 | } | |
8f6c2e4b | 3800 | |
dafb20fa | 3801 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 3802 | long long diff; |
34b343cf | 3803 | |
1da177e4 | 3804 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
3805 | if (disk_idx < 0) |
3806 | continue; | |
3807 | if (disk_idx >= conf->geo.raid_disks && | |
3808 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
3809 | continue; |
3810 | disk = conf->mirrors + disk_idx; | |
3811 | ||
56a2559b N |
3812 | if (test_bit(Replacement, &rdev->flags)) { |
3813 | if (disk->replacement) | |
3814 | goto out_free_conf; | |
3815 | disk->replacement = rdev; | |
3816 | } else { | |
3817 | if (disk->rdev) | |
3818 | goto out_free_conf; | |
3819 | disk->rdev = rdev; | |
3820 | } | |
3ea7daa5 N |
3821 | diff = (rdev->new_data_offset - rdev->data_offset); |
3822 | if (!mddev->reshape_backwards) | |
3823 | diff = -diff; | |
3824 | if (diff < 0) | |
3825 | diff = 0; | |
3826 | if (first || diff < min_offset_diff) | |
3827 | min_offset_diff = diff; | |
56a2559b | 3828 | |
cc4d1efd JB |
3829 | if (mddev->gendisk) |
3830 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
3831 | rdev->data_offset << 9); | |
1da177e4 LT |
3832 | |
3833 | disk->head_position = 0; | |
532a2a3f SL |
3834 | |
3835 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) | |
3836 | discard_supported = true; | |
6f287ca6 | 3837 | first = 0; |
1da177e4 | 3838 | } |
3ea7daa5 | 3839 | |
ed30be07 JB |
3840 | if (mddev->queue) { |
3841 | if (discard_supported) | |
3842 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3843 | mddev->queue); | |
3844 | else | |
3845 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3846 | mddev->queue); | |
3847 | } | |
6d508242 | 3848 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3849 | if (!enough(conf, -1)) { |
08464e09 | 3850 | pr_err("md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3851 | mdname(mddev)); |
1da177e4 LT |
3852 | goto out_free_conf; |
3853 | } | |
3854 | ||
3ea7daa5 N |
3855 | if (conf->reshape_progress != MaxSector) { |
3856 | /* must ensure that shape change is supported */ | |
3857 | if (conf->geo.far_copies != 1 && | |
3858 | conf->geo.far_offset == 0) | |
3859 | goto out_free_conf; | |
3860 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 3861 | conf->prev.far_offset == 0) |
3ea7daa5 N |
3862 | goto out_free_conf; |
3863 | } | |
3864 | ||
1da177e4 | 3865 | mddev->degraded = 0; |
f8c9e74f N |
3866 | for (i = 0; |
3867 | i < conf->geo.raid_disks | |
3868 | || i < conf->prev.raid_disks; | |
3869 | i++) { | |
1da177e4 LT |
3870 | |
3871 | disk = conf->mirrors + i; | |
3872 | ||
56a2559b N |
3873 | if (!disk->rdev && disk->replacement) { |
3874 | /* The replacement is all we have - use it */ | |
3875 | disk->rdev = disk->replacement; | |
3876 | disk->replacement = NULL; | |
3877 | clear_bit(Replacement, &disk->rdev->flags); | |
3878 | } | |
3879 | ||
5fd6c1dc | 3880 | if (!disk->rdev || |
2e333e89 | 3881 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3882 | disk->head_position = 0; |
3883 | mddev->degraded++; | |
0b59bb64 N |
3884 | if (disk->rdev && |
3885 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 3886 | conf->fullsync = 1; |
1da177e4 | 3887 | } |
ff9533a5 BC |
3888 | |
3889 | if (disk->replacement && | |
3890 | !test_bit(In_sync, &disk->replacement->flags) && | |
3891 | disk->replacement->saved_raid_disk < 0) { | |
3892 | conf->fullsync = 1; | |
3893 | } | |
3894 | ||
d890fa2b | 3895 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3896 | } |
3897 | ||
8c6ac868 | 3898 | if (mddev->recovery_cp != MaxSector) |
08464e09 N |
3899 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", |
3900 | mdname(mddev)); | |
3901 | pr_info("md/raid10:%s: active with %d out of %d devices\n", | |
5cf00fcd N |
3902 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
3903 | conf->geo.raid_disks); | |
1da177e4 LT |
3904 | /* |
3905 | * Ok, everything is just fine now | |
3906 | */ | |
dab8b292 TM |
3907 | mddev->dev_sectors = conf->dev_sectors; |
3908 | size = raid10_size(mddev, 0, 0); | |
3909 | md_set_array_sectors(mddev, size); | |
3910 | mddev->resync_max_sectors = size; | |
46533ff7 | 3911 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
1da177e4 | 3912 | |
cc4d1efd | 3913 | if (mddev->queue) { |
5cf00fcd | 3914 | int stripe = conf->geo.raid_disks * |
9d8f0363 | 3915 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
cc4d1efd JB |
3916 | |
3917 | /* Calculate max read-ahead size. | |
3918 | * We need to readahead at least twice a whole stripe.... | |
3919 | * maybe... | |
3920 | */ | |
5cf00fcd | 3921 | stripe /= conf->geo.near_copies; |
dc3b17cc JK |
3922 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
3923 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
1da177e4 LT |
3924 | } |
3925 | ||
a91a2785 MP |
3926 | if (md_integrity_register(mddev)) |
3927 | goto out_free_conf; | |
3928 | ||
3ea7daa5 N |
3929 | if (conf->reshape_progress != MaxSector) { |
3930 | unsigned long before_length, after_length; | |
3931 | ||
3932 | before_length = ((1 << conf->prev.chunk_shift) * | |
3933 | conf->prev.far_copies); | |
3934 | after_length = ((1 << conf->geo.chunk_shift) * | |
3935 | conf->geo.far_copies); | |
3936 | ||
3937 | if (max(before_length, after_length) > min_offset_diff) { | |
3938 | /* This cannot work */ | |
08464e09 | 3939 | pr_warn("md/raid10: offset difference not enough to continue reshape\n"); |
3ea7daa5 N |
3940 | goto out_free_conf; |
3941 | } | |
3942 | conf->offset_diff = min_offset_diff; | |
3943 | ||
3ea7daa5 N |
3944 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
3945 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3946 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3947 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3948 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3949 | "reshape"); | |
4b73eea6 AP |
3950 | if (!mddev->sync_thread) |
3951 | goto out_free_conf; | |
3ea7daa5 N |
3952 | } |
3953 | ||
1da177e4 LT |
3954 | return 0; |
3955 | ||
3956 | out_free_conf: | |
01f96c0a | 3957 | md_unregister_thread(&mddev->thread); |
644df1a8 | 3958 | mempool_destroy(conf->r10bio_pool); |
1345b1d8 | 3959 | safe_put_page(conf->tmppage); |
990a8baf | 3960 | kfree(conf->mirrors); |
1da177e4 LT |
3961 | kfree(conf); |
3962 | mddev->private = NULL; | |
3963 | out: | |
3964 | return -EIO; | |
3965 | } | |
3966 | ||
afa0f557 | 3967 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 3968 | { |
afa0f557 | 3969 | struct r10conf *conf = priv; |
1da177e4 | 3970 | |
644df1a8 | 3971 | mempool_destroy(conf->r10bio_pool); |
0fea7ed8 | 3972 | safe_put_page(conf->tmppage); |
990a8baf | 3973 | kfree(conf->mirrors); |
c4796e21 N |
3974 | kfree(conf->mirrors_old); |
3975 | kfree(conf->mirrors_new); | |
fc9977dd N |
3976 | if (conf->bio_split) |
3977 | bioset_free(conf->bio_split); | |
1da177e4 | 3978 | kfree(conf); |
1da177e4 LT |
3979 | } |
3980 | ||
b03e0ccb | 3981 | static void raid10_quiesce(struct mddev *mddev, int quiesce) |
6cce3b23 | 3982 | { |
e879a879 | 3983 | struct r10conf *conf = mddev->private; |
6cce3b23 | 3984 | |
b03e0ccb | 3985 | if (quiesce) |
6cce3b23 | 3986 | raise_barrier(conf, 0); |
b03e0ccb | 3987 | else |
6cce3b23 | 3988 | lower_barrier(conf); |
6cce3b23 | 3989 | } |
1da177e4 | 3990 | |
006a09a0 N |
3991 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3992 | { | |
3993 | /* Resize of 'far' arrays is not supported. | |
3994 | * For 'near' and 'offset' arrays we can set the | |
3995 | * number of sectors used to be an appropriate multiple | |
3996 | * of the chunk size. | |
3997 | * For 'offset', this is far_copies*chunksize. | |
3998 | * For 'near' the multiplier is the LCM of | |
3999 | * near_copies and raid_disks. | |
4000 | * So if far_copies > 1 && !far_offset, fail. | |
4001 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
4002 | * multiply by chunk_size. Then round to this number. | |
4003 | * This is mostly done by raid10_size() | |
4004 | */ | |
4005 | struct r10conf *conf = mddev->private; | |
4006 | sector_t oldsize, size; | |
4007 | ||
f8c9e74f N |
4008 | if (mddev->reshape_position != MaxSector) |
4009 | return -EBUSY; | |
4010 | ||
5cf00fcd | 4011 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
4012 | return -EINVAL; |
4013 | ||
4014 | oldsize = raid10_size(mddev, 0, 0); | |
4015 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
4016 | if (mddev->external_size && |
4017 | mddev->array_sectors > size) | |
006a09a0 | 4018 | return -EINVAL; |
a4a6125a N |
4019 | if (mddev->bitmap) { |
4020 | int ret = bitmap_resize(mddev->bitmap, size, 0, 0); | |
4021 | if (ret) | |
4022 | return ret; | |
4023 | } | |
4024 | md_set_array_sectors(mddev, size); | |
006a09a0 N |
4025 | if (sectors > mddev->dev_sectors && |
4026 | mddev->recovery_cp > oldsize) { | |
4027 | mddev->recovery_cp = oldsize; | |
4028 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4029 | } | |
6508fdbf N |
4030 | calc_sectors(conf, sectors); |
4031 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
4032 | mddev->resync_max_sectors = size; |
4033 | return 0; | |
4034 | } | |
4035 | ||
53a6ab4d | 4036 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 4037 | { |
3cb03002 | 4038 | struct md_rdev *rdev; |
e879a879 | 4039 | struct r10conf *conf; |
dab8b292 TM |
4040 | |
4041 | if (mddev->degraded > 0) { | |
08464e09 N |
4042 | pr_warn("md/raid10:%s: Error: degraded raid0!\n", |
4043 | mdname(mddev)); | |
dab8b292 TM |
4044 | return ERR_PTR(-EINVAL); |
4045 | } | |
53a6ab4d | 4046 | sector_div(size, devs); |
dab8b292 | 4047 | |
dab8b292 TM |
4048 | /* Set new parameters */ |
4049 | mddev->new_level = 10; | |
4050 | /* new layout: far_copies = 1, near_copies = 2 */ | |
4051 | mddev->new_layout = (1<<8) + 2; | |
4052 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
4053 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
4054 | mddev->raid_disks *= 2; |
4055 | /* make sure it will be not marked as dirty */ | |
4056 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 4057 | mddev->dev_sectors = size; |
dab8b292 TM |
4058 | |
4059 | conf = setup_conf(mddev); | |
02214dc5 | 4060 | if (!IS_ERR(conf)) { |
dafb20fa | 4061 | rdev_for_each(rdev, mddev) |
53a6ab4d | 4062 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 4063 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
4064 | rdev->sectors = size; |
4065 | } | |
02214dc5 KW |
4066 | conf->barrier = 1; |
4067 | } | |
4068 | ||
dab8b292 TM |
4069 | return conf; |
4070 | } | |
4071 | ||
fd01b88c | 4072 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 4073 | { |
e373ab10 | 4074 | struct r0conf *raid0_conf; |
dab8b292 TM |
4075 | |
4076 | /* raid10 can take over: | |
4077 | * raid0 - providing it has only two drives | |
4078 | */ | |
4079 | if (mddev->level == 0) { | |
4080 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
4081 | raid0_conf = mddev->private; |
4082 | if (raid0_conf->nr_strip_zones > 1) { | |
08464e09 N |
4083 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n", |
4084 | mdname(mddev)); | |
dab8b292 TM |
4085 | return ERR_PTR(-EINVAL); |
4086 | } | |
53a6ab4d N |
4087 | return raid10_takeover_raid0(mddev, |
4088 | raid0_conf->strip_zone->zone_end, | |
4089 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
4090 | } |
4091 | return ERR_PTR(-EINVAL); | |
4092 | } | |
4093 | ||
3ea7daa5 N |
4094 | static int raid10_check_reshape(struct mddev *mddev) |
4095 | { | |
4096 | /* Called when there is a request to change | |
4097 | * - layout (to ->new_layout) | |
4098 | * - chunk size (to ->new_chunk_sectors) | |
4099 | * - raid_disks (by delta_disks) | |
4100 | * or when trying to restart a reshape that was ongoing. | |
4101 | * | |
4102 | * We need to validate the request and possibly allocate | |
4103 | * space if that might be an issue later. | |
4104 | * | |
4105 | * Currently we reject any reshape of a 'far' mode array, | |
4106 | * allow chunk size to change if new is generally acceptable, | |
4107 | * allow raid_disks to increase, and allow | |
4108 | * a switch between 'near' mode and 'offset' mode. | |
4109 | */ | |
4110 | struct r10conf *conf = mddev->private; | |
4111 | struct geom geo; | |
4112 | ||
4113 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
4114 | return -EINVAL; | |
4115 | ||
4116 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
4117 | /* mustn't change number of copies */ | |
4118 | return -EINVAL; | |
4119 | if (geo.far_copies > 1 && !geo.far_offset) | |
4120 | /* Cannot switch to 'far' mode */ | |
4121 | return -EINVAL; | |
4122 | ||
4123 | if (mddev->array_sectors & geo.chunk_mask) | |
4124 | /* not factor of array size */ | |
4125 | return -EINVAL; | |
4126 | ||
3ea7daa5 N |
4127 | if (!enough(conf, -1)) |
4128 | return -EINVAL; | |
4129 | ||
4130 | kfree(conf->mirrors_new); | |
4131 | conf->mirrors_new = NULL; | |
4132 | if (mddev->delta_disks > 0) { | |
4133 | /* allocate new 'mirrors' list */ | |
4134 | conf->mirrors_new = kzalloc( | |
dc280d98 | 4135 | sizeof(struct raid10_info) |
3ea7daa5 N |
4136 | *(mddev->raid_disks + |
4137 | mddev->delta_disks), | |
4138 | GFP_KERNEL); | |
4139 | if (!conf->mirrors_new) | |
4140 | return -ENOMEM; | |
4141 | } | |
4142 | return 0; | |
4143 | } | |
4144 | ||
4145 | /* | |
4146 | * Need to check if array has failed when deciding whether to: | |
4147 | * - start an array | |
4148 | * - remove non-faulty devices | |
4149 | * - add a spare | |
4150 | * - allow a reshape | |
4151 | * This determination is simple when no reshape is happening. | |
4152 | * However if there is a reshape, we need to carefully check | |
4153 | * both the before and after sections. | |
4154 | * This is because some failed devices may only affect one | |
4155 | * of the two sections, and some non-in_sync devices may | |
4156 | * be insync in the section most affected by failed devices. | |
4157 | */ | |
4158 | static int calc_degraded(struct r10conf *conf) | |
4159 | { | |
4160 | int degraded, degraded2; | |
4161 | int i; | |
4162 | ||
4163 | rcu_read_lock(); | |
4164 | degraded = 0; | |
4165 | /* 'prev' section first */ | |
4166 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
4167 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4168 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4169 | degraded++; | |
4170 | else if (!test_bit(In_sync, &rdev->flags)) | |
4171 | /* When we can reduce the number of devices in | |
4172 | * an array, this might not contribute to | |
4173 | * 'degraded'. It does now. | |
4174 | */ | |
4175 | degraded++; | |
4176 | } | |
4177 | rcu_read_unlock(); | |
4178 | if (conf->geo.raid_disks == conf->prev.raid_disks) | |
4179 | return degraded; | |
4180 | rcu_read_lock(); | |
4181 | degraded2 = 0; | |
4182 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
4183 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4184 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4185 | degraded2++; | |
4186 | else if (!test_bit(In_sync, &rdev->flags)) { | |
4187 | /* If reshape is increasing the number of devices, | |
4188 | * this section has already been recovered, so | |
4189 | * it doesn't contribute to degraded. | |
4190 | * else it does. | |
4191 | */ | |
4192 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
4193 | degraded2++; | |
4194 | } | |
4195 | } | |
4196 | rcu_read_unlock(); | |
4197 | if (degraded2 > degraded) | |
4198 | return degraded2; | |
4199 | return degraded; | |
4200 | } | |
4201 | ||
4202 | static int raid10_start_reshape(struct mddev *mddev) | |
4203 | { | |
4204 | /* A 'reshape' has been requested. This commits | |
4205 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
4206 | * This also checks if there are enough spares and adds them | |
4207 | * to the array. | |
4208 | * We currently require enough spares to make the final | |
4209 | * array non-degraded. We also require that the difference | |
4210 | * between old and new data_offset - on each device - is | |
4211 | * enough that we never risk over-writing. | |
4212 | */ | |
4213 | ||
4214 | unsigned long before_length, after_length; | |
4215 | sector_t min_offset_diff = 0; | |
4216 | int first = 1; | |
4217 | struct geom new; | |
4218 | struct r10conf *conf = mddev->private; | |
4219 | struct md_rdev *rdev; | |
4220 | int spares = 0; | |
bb63a701 | 4221 | int ret; |
3ea7daa5 N |
4222 | |
4223 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4224 | return -EBUSY; | |
4225 | ||
4226 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
4227 | return -EINVAL; | |
4228 | ||
4229 | before_length = ((1 << conf->prev.chunk_shift) * | |
4230 | conf->prev.far_copies); | |
4231 | after_length = ((1 << conf->geo.chunk_shift) * | |
4232 | conf->geo.far_copies); | |
4233 | ||
4234 | rdev_for_each(rdev, mddev) { | |
4235 | if (!test_bit(In_sync, &rdev->flags) | |
4236 | && !test_bit(Faulty, &rdev->flags)) | |
4237 | spares++; | |
4238 | if (rdev->raid_disk >= 0) { | |
4239 | long long diff = (rdev->new_data_offset | |
4240 | - rdev->data_offset); | |
4241 | if (!mddev->reshape_backwards) | |
4242 | diff = -diff; | |
4243 | if (diff < 0) | |
4244 | diff = 0; | |
4245 | if (first || diff < min_offset_diff) | |
4246 | min_offset_diff = diff; | |
b506335e | 4247 | first = 0; |
3ea7daa5 N |
4248 | } |
4249 | } | |
4250 | ||
4251 | if (max(before_length, after_length) > min_offset_diff) | |
4252 | return -EINVAL; | |
4253 | ||
4254 | if (spares < mddev->delta_disks) | |
4255 | return -EINVAL; | |
4256 | ||
4257 | conf->offset_diff = min_offset_diff; | |
4258 | spin_lock_irq(&conf->device_lock); | |
4259 | if (conf->mirrors_new) { | |
4260 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4261 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4262 | smp_mb(); |
c4796e21 | 4263 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4264 | conf->mirrors_old = conf->mirrors; |
4265 | conf->mirrors = conf->mirrors_new; | |
4266 | conf->mirrors_new = NULL; | |
4267 | } | |
4268 | setup_geo(&conf->geo, mddev, geo_start); | |
4269 | smp_mb(); | |
4270 | if (mddev->reshape_backwards) { | |
4271 | sector_t size = raid10_size(mddev, 0, 0); | |
4272 | if (size < mddev->array_sectors) { | |
4273 | spin_unlock_irq(&conf->device_lock); | |
08464e09 N |
4274 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n", |
4275 | mdname(mddev)); | |
3ea7daa5 N |
4276 | return -EINVAL; |
4277 | } | |
4278 | mddev->resync_max_sectors = size; | |
4279 | conf->reshape_progress = size; | |
4280 | } else | |
4281 | conf->reshape_progress = 0; | |
299b0685 | 4282 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4283 | spin_unlock_irq(&conf->device_lock); |
4284 | ||
bb63a701 N |
4285 | if (mddev->delta_disks && mddev->bitmap) { |
4286 | ret = bitmap_resize(mddev->bitmap, | |
4287 | raid10_size(mddev, 0, | |
4288 | conf->geo.raid_disks), | |
4289 | 0, 0); | |
4290 | if (ret) | |
4291 | goto abort; | |
4292 | } | |
3ea7daa5 N |
4293 | if (mddev->delta_disks > 0) { |
4294 | rdev_for_each(rdev, mddev) | |
4295 | if (rdev->raid_disk < 0 && | |
4296 | !test_bit(Faulty, &rdev->flags)) { | |
4297 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4298 | if (rdev->raid_disk >= | |
4299 | conf->prev.raid_disks) | |
4300 | set_bit(In_sync, &rdev->flags); | |
4301 | else | |
4302 | rdev->recovery_offset = 0; | |
4303 | ||
4304 | if (sysfs_link_rdev(mddev, rdev)) | |
4305 | /* Failure here is OK */; | |
4306 | } | |
4307 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4308 | && !test_bit(Faulty, &rdev->flags)) { | |
4309 | /* This is a spare that was manually added */ | |
4310 | set_bit(In_sync, &rdev->flags); | |
4311 | } | |
4312 | } | |
4313 | /* When a reshape changes the number of devices, | |
4314 | * ->degraded is measured against the larger of the | |
4315 | * pre and post numbers. | |
4316 | */ | |
4317 | spin_lock_irq(&conf->device_lock); | |
4318 | mddev->degraded = calc_degraded(conf); | |
4319 | spin_unlock_irq(&conf->device_lock); | |
4320 | mddev->raid_disks = conf->geo.raid_disks; | |
4321 | mddev->reshape_position = conf->reshape_progress; | |
2953079c | 4322 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 N |
4323 | |
4324 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4325 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4326 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 N |
4327 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
4328 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4329 | ||
4330 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4331 | "reshape"); | |
4332 | if (!mddev->sync_thread) { | |
bb63a701 N |
4333 | ret = -EAGAIN; |
4334 | goto abort; | |
3ea7daa5 N |
4335 | } |
4336 | conf->reshape_checkpoint = jiffies; | |
4337 | md_wakeup_thread(mddev->sync_thread); | |
4338 | md_new_event(mddev); | |
4339 | return 0; | |
bb63a701 N |
4340 | |
4341 | abort: | |
4342 | mddev->recovery = 0; | |
4343 | spin_lock_irq(&conf->device_lock); | |
4344 | conf->geo = conf->prev; | |
4345 | mddev->raid_disks = conf->geo.raid_disks; | |
4346 | rdev_for_each(rdev, mddev) | |
4347 | rdev->new_data_offset = rdev->data_offset; | |
4348 | smp_wmb(); | |
4349 | conf->reshape_progress = MaxSector; | |
299b0685 | 4350 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4351 | mddev->reshape_position = MaxSector; |
4352 | spin_unlock_irq(&conf->device_lock); | |
4353 | return ret; | |
3ea7daa5 N |
4354 | } |
4355 | ||
4356 | /* Calculate the last device-address that could contain | |
4357 | * any block from the chunk that includes the array-address 's' | |
4358 | * and report the next address. | |
4359 | * i.e. the address returned will be chunk-aligned and after | |
4360 | * any data that is in the chunk containing 's'. | |
4361 | */ | |
4362 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4363 | { | |
4364 | s = (s | geo->chunk_mask) + 1; | |
4365 | s >>= geo->chunk_shift; | |
4366 | s *= geo->near_copies; | |
4367 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4368 | s *= geo->far_copies; | |
4369 | s <<= geo->chunk_shift; | |
4370 | return s; | |
4371 | } | |
4372 | ||
4373 | /* Calculate the first device-address that could contain | |
4374 | * any block from the chunk that includes the array-address 's'. | |
4375 | * This too will be the start of a chunk | |
4376 | */ | |
4377 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4378 | { | |
4379 | s >>= geo->chunk_shift; | |
4380 | s *= geo->near_copies; | |
4381 | sector_div(s, geo->raid_disks); | |
4382 | s *= geo->far_copies; | |
4383 | s <<= geo->chunk_shift; | |
4384 | return s; | |
4385 | } | |
4386 | ||
4387 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4388 | int *skipped) | |
4389 | { | |
4390 | /* We simply copy at most one chunk (smallest of old and new) | |
4391 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4392 | * or we hit a bad block or something. | |
4393 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4394 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4395 | * can record any location. |
4396 | * | |
4397 | * If we will want to write to a location that isn't | |
4398 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4399 | * we need to flush all reshape requests and update the metadata. | |
4400 | * | |
4401 | * When reshaping forwards (e.g. to more devices), we interpret | |
4402 | * 'safe' as the earliest block which might not have been copied | |
4403 | * down yet. We divide this by previous stripe size and multiply | |
4404 | * by previous stripe length to get lowest device offset that we | |
4405 | * cannot write to yet. | |
4406 | * We interpret 'sector_nr' as an address that we want to write to. | |
4407 | * From this we use last_device_address() to find where we might | |
4408 | * write to, and first_device_address on the 'safe' position. | |
4409 | * If this 'next' write position is after the 'safe' position, | |
4410 | * we must update the metadata to increase the 'safe' position. | |
4411 | * | |
4412 | * When reshaping backwards, we round in the opposite direction | |
4413 | * and perform the reverse test: next write position must not be | |
4414 | * less than current safe position. | |
4415 | * | |
4416 | * In all this the minimum difference in data offsets | |
4417 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4418 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4419 | * |
4420 | * We need to prepare all the bios here before we start any IO | |
4421 | * to ensure the size we choose is acceptable to all devices. | |
4422 | * The means one for each copy for write-out and an extra one for | |
4423 | * read-in. | |
4424 | * We store the read-in bio in ->master_bio and the others in | |
4425 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4426 | */ | |
4427 | struct r10conf *conf = mddev->private; | |
4428 | struct r10bio *r10_bio; | |
4429 | sector_t next, safe, last; | |
4430 | int max_sectors; | |
4431 | int nr_sectors; | |
4432 | int s; | |
4433 | struct md_rdev *rdev; | |
4434 | int need_flush = 0; | |
4435 | struct bio *blist; | |
4436 | struct bio *bio, *read_bio; | |
4437 | int sectors_done = 0; | |
f0250618 | 4438 | struct page **pages; |
3ea7daa5 N |
4439 | |
4440 | if (sector_nr == 0) { | |
4441 | /* If restarting in the middle, skip the initial sectors */ | |
4442 | if (mddev->reshape_backwards && | |
4443 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4444 | sector_nr = (raid10_size(mddev, 0, 0) | |
4445 | - conf->reshape_progress); | |
4446 | } else if (!mddev->reshape_backwards && | |
4447 | conf->reshape_progress > 0) | |
4448 | sector_nr = conf->reshape_progress; | |
4449 | if (sector_nr) { | |
4450 | mddev->curr_resync_completed = sector_nr; | |
4451 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
4452 | *skipped = 1; | |
4453 | return sector_nr; | |
4454 | } | |
4455 | } | |
4456 | ||
4457 | /* We don't use sector_nr to track where we are up to | |
4458 | * as that doesn't work well for ->reshape_backwards. | |
4459 | * So just use ->reshape_progress. | |
4460 | */ | |
4461 | if (mddev->reshape_backwards) { | |
4462 | /* 'next' is the earliest device address that we might | |
4463 | * write to for this chunk in the new layout | |
4464 | */ | |
4465 | next = first_dev_address(conf->reshape_progress - 1, | |
4466 | &conf->geo); | |
4467 | ||
4468 | /* 'safe' is the last device address that we might read from | |
4469 | * in the old layout after a restart | |
4470 | */ | |
4471 | safe = last_dev_address(conf->reshape_safe - 1, | |
4472 | &conf->prev); | |
4473 | ||
4474 | if (next + conf->offset_diff < safe) | |
4475 | need_flush = 1; | |
4476 | ||
4477 | last = conf->reshape_progress - 1; | |
4478 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4479 | & conf->prev.chunk_mask); | |
4480 | if (sector_nr + RESYNC_BLOCK_SIZE/512 < last) | |
4481 | sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512; | |
4482 | } else { | |
4483 | /* 'next' is after the last device address that we | |
4484 | * might write to for this chunk in the new layout | |
4485 | */ | |
4486 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4487 | ||
4488 | /* 'safe' is the earliest device address that we might | |
4489 | * read from in the old layout after a restart | |
4490 | */ | |
4491 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4492 | ||
4493 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4494 | * as that would possibly corrupt data | |
4495 | */ | |
4496 | if (next > safe + conf->offset_diff) | |
4497 | need_flush = 1; | |
4498 | ||
4499 | sector_nr = conf->reshape_progress; | |
4500 | last = sector_nr | (conf->geo.chunk_mask | |
4501 | & conf->prev.chunk_mask); | |
4502 | ||
4503 | if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last) | |
4504 | last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1; | |
4505 | } | |
4506 | ||
4507 | if (need_flush || | |
4508 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4509 | /* Need to update reshape_position in metadata */ | |
4510 | wait_barrier(conf); | |
4511 | mddev->reshape_position = conf->reshape_progress; | |
4512 | if (mddev->reshape_backwards) | |
4513 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4514 | - conf->reshape_progress; | |
4515 | else | |
4516 | mddev->curr_resync_completed = conf->reshape_progress; | |
4517 | conf->reshape_checkpoint = jiffies; | |
2953079c | 4518 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 | 4519 | md_wakeup_thread(mddev->thread); |
2953079c | 4520 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
4521 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4522 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4523 | allow_barrier(conf); | |
4524 | return sectors_done; | |
4525 | } | |
3ea7daa5 N |
4526 | conf->reshape_safe = mddev->reshape_position; |
4527 | allow_barrier(conf); | |
4528 | } | |
4529 | ||
6333b13e | 4530 | raise_barrier(conf, 0); |
3ea7daa5 N |
4531 | read_more: |
4532 | /* Now schedule reads for blocks from sector_nr to last */ | |
208410b5 | 4533 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 4534 | r10_bio->state = 0; |
6333b13e | 4535 | raise_barrier(conf, 1); |
3ea7daa5 N |
4536 | atomic_set(&r10_bio->remaining, 0); |
4537 | r10_bio->mddev = mddev; | |
4538 | r10_bio->sector = sector_nr; | |
4539 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4540 | r10_bio->sectors = last - sector_nr + 1; | |
4541 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4542 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4543 | ||
4544 | if (!rdev) { | |
4545 | /* Cannot read from here, so need to record bad blocks | |
4546 | * on all the target devices. | |
4547 | */ | |
4548 | // FIXME | |
e337aead | 4549 | mempool_free(r10_bio, conf->r10buf_pool); |
3ea7daa5 N |
4550 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4551 | return sectors_done; | |
4552 | } | |
4553 | ||
4554 | read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev); | |
4555 | ||
74d46992 | 4556 | bio_set_dev(read_bio, rdev->bdev); |
4f024f37 | 4557 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4558 | + rdev->data_offset); |
4559 | read_bio->bi_private = r10_bio; | |
81fa1520 | 4560 | read_bio->bi_end_io = end_reshape_read; |
796a5cf0 | 4561 | bio_set_op_attrs(read_bio, REQ_OP_READ, 0); |
ce0b0a46 | 4562 | read_bio->bi_flags &= (~0UL << BIO_RESET_BITS); |
4e4cbee9 | 4563 | read_bio->bi_status = 0; |
3ea7daa5 | 4564 | read_bio->bi_vcnt = 0; |
4f024f37 | 4565 | read_bio->bi_iter.bi_size = 0; |
3ea7daa5 N |
4566 | r10_bio->master_bio = read_bio; |
4567 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4568 | ||
4569 | /* Now find the locations in the new layout */ | |
4570 | __raid10_find_phys(&conf->geo, r10_bio); | |
4571 | ||
4572 | blist = read_bio; | |
4573 | read_bio->bi_next = NULL; | |
4574 | ||
d094d686 | 4575 | rcu_read_lock(); |
3ea7daa5 N |
4576 | for (s = 0; s < conf->copies*2; s++) { |
4577 | struct bio *b; | |
4578 | int d = r10_bio->devs[s/2].devnum; | |
4579 | struct md_rdev *rdev2; | |
4580 | if (s&1) { | |
d094d686 | 4581 | rdev2 = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4582 | b = r10_bio->devs[s/2].repl_bio; |
4583 | } else { | |
d094d686 | 4584 | rdev2 = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4585 | b = r10_bio->devs[s/2].bio; |
4586 | } | |
4587 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4588 | continue; | |
8be185f2 | 4589 | |
74d46992 | 4590 | bio_set_dev(b, rdev2->bdev); |
4f024f37 KO |
4591 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4592 | rdev2->new_data_offset; | |
3ea7daa5 | 4593 | b->bi_end_io = end_reshape_write; |
796a5cf0 | 4594 | bio_set_op_attrs(b, REQ_OP_WRITE, 0); |
3ea7daa5 | 4595 | b->bi_next = blist; |
3ea7daa5 N |
4596 | blist = b; |
4597 | } | |
4598 | ||
4599 | /* Now add as many pages as possible to all of these bios. */ | |
4600 | ||
4601 | nr_sectors = 0; | |
f0250618 | 4602 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; |
3ea7daa5 | 4603 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { |
f0250618 | 4604 | struct page *page = pages[s / (PAGE_SIZE >> 9)]; |
3ea7daa5 N |
4605 | int len = (max_sectors - s) << 9; |
4606 | if (len > PAGE_SIZE) | |
4607 | len = PAGE_SIZE; | |
4608 | for (bio = blist; bio ; bio = bio->bi_next) { | |
c85ba149 ML |
4609 | /* |
4610 | * won't fail because the vec table is big enough | |
4611 | * to hold all these pages | |
4612 | */ | |
4613 | bio_add_page(bio, page, len, 0); | |
3ea7daa5 N |
4614 | } |
4615 | sector_nr += len >> 9; | |
4616 | nr_sectors += len >> 9; | |
4617 | } | |
d094d686 | 4618 | rcu_read_unlock(); |
3ea7daa5 N |
4619 | r10_bio->sectors = nr_sectors; |
4620 | ||
4621 | /* Now submit the read */ | |
74d46992 | 4622 | md_sync_acct_bio(read_bio, r10_bio->sectors); |
3ea7daa5 N |
4623 | atomic_inc(&r10_bio->remaining); |
4624 | read_bio->bi_next = NULL; | |
4625 | generic_make_request(read_bio); | |
3ea7daa5 N |
4626 | sectors_done += nr_sectors; |
4627 | if (sector_nr <= last) | |
4628 | goto read_more; | |
4629 | ||
6333b13e XN |
4630 | lower_barrier(conf); |
4631 | ||
3ea7daa5 N |
4632 | /* Now that we have done the whole section we can |
4633 | * update reshape_progress | |
4634 | */ | |
4635 | if (mddev->reshape_backwards) | |
4636 | conf->reshape_progress -= sectors_done; | |
4637 | else | |
4638 | conf->reshape_progress += sectors_done; | |
4639 | ||
4640 | return sectors_done; | |
4641 | } | |
4642 | ||
4643 | static void end_reshape_request(struct r10bio *r10_bio); | |
4644 | static int handle_reshape_read_error(struct mddev *mddev, | |
4645 | struct r10bio *r10_bio); | |
4646 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4647 | { | |
4648 | /* Reshape read completed. Hopefully we have a block | |
4649 | * to write out. | |
4650 | * If we got a read error then we do sync 1-page reads from | |
4651 | * elsewhere until we find the data - or give up. | |
4652 | */ | |
4653 | struct r10conf *conf = mddev->private; | |
4654 | int s; | |
4655 | ||
4656 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4657 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4658 | /* Reshape has been aborted */ | |
4659 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4660 | return; | |
4661 | } | |
4662 | ||
4663 | /* We definitely have the data in the pages, schedule the | |
4664 | * writes. | |
4665 | */ | |
4666 | atomic_set(&r10_bio->remaining, 1); | |
4667 | for (s = 0; s < conf->copies*2; s++) { | |
4668 | struct bio *b; | |
4669 | int d = r10_bio->devs[s/2].devnum; | |
4670 | struct md_rdev *rdev; | |
d094d686 | 4671 | rcu_read_lock(); |
3ea7daa5 | 4672 | if (s&1) { |
d094d686 | 4673 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4674 | b = r10_bio->devs[s/2].repl_bio; |
4675 | } else { | |
d094d686 | 4676 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4677 | b = r10_bio->devs[s/2].bio; |
4678 | } | |
d094d686 N |
4679 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
4680 | rcu_read_unlock(); | |
3ea7daa5 | 4681 | continue; |
d094d686 | 4682 | } |
3ea7daa5 | 4683 | atomic_inc(&rdev->nr_pending); |
d094d686 | 4684 | rcu_read_unlock(); |
74d46992 | 4685 | md_sync_acct_bio(b, r10_bio->sectors); |
3ea7daa5 N |
4686 | atomic_inc(&r10_bio->remaining); |
4687 | b->bi_next = NULL; | |
4688 | generic_make_request(b); | |
4689 | } | |
4690 | end_reshape_request(r10_bio); | |
4691 | } | |
4692 | ||
4693 | static void end_reshape(struct r10conf *conf) | |
4694 | { | |
4695 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
4696 | return; | |
4697 | ||
4698 | spin_lock_irq(&conf->device_lock); | |
4699 | conf->prev = conf->geo; | |
4700 | md_finish_reshape(conf->mddev); | |
4701 | smp_wmb(); | |
4702 | conf->reshape_progress = MaxSector; | |
299b0685 | 4703 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
4704 | spin_unlock_irq(&conf->device_lock); |
4705 | ||
4706 | /* read-ahead size must cover two whole stripes, which is | |
4707 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4708 | */ | |
4709 | if (conf->mddev->queue) { | |
4710 | int stripe = conf->geo.raid_disks * | |
4711 | ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE); | |
4712 | stripe /= conf->geo.near_copies; | |
dc3b17cc JK |
4713 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
4714 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
3ea7daa5 N |
4715 | } |
4716 | conf->fullsync = 0; | |
4717 | } | |
4718 | ||
3ea7daa5 N |
4719 | static int handle_reshape_read_error(struct mddev *mddev, |
4720 | struct r10bio *r10_bio) | |
4721 | { | |
4722 | /* Use sync reads to get the blocks from somewhere else */ | |
4723 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 4724 | struct r10conf *conf = mddev->private; |
584ed9fa | 4725 | struct r10bio *r10b; |
3ea7daa5 N |
4726 | int slot = 0; |
4727 | int idx = 0; | |
2d06e3b7 ML |
4728 | struct page **pages; |
4729 | ||
584ed9fa MK |
4730 | r10b = kmalloc(sizeof(*r10b) + |
4731 | sizeof(struct r10dev) * conf->copies, GFP_NOIO); | |
4732 | if (!r10b) { | |
4733 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
4734 | return -ENOMEM; | |
4735 | } | |
4736 | ||
2d06e3b7 ML |
4737 | /* reshape IOs share pages from .devs[0].bio */ |
4738 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; | |
3ea7daa5 | 4739 | |
e0ee7785 N |
4740 | r10b->sector = r10_bio->sector; |
4741 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
4742 | |
4743 | while (sectors) { | |
4744 | int s = sectors; | |
4745 | int success = 0; | |
4746 | int first_slot = slot; | |
4747 | ||
4748 | if (s > (PAGE_SIZE >> 9)) | |
4749 | s = PAGE_SIZE >> 9; | |
4750 | ||
d094d686 | 4751 | rcu_read_lock(); |
3ea7daa5 | 4752 | while (!success) { |
e0ee7785 | 4753 | int d = r10b->devs[slot].devnum; |
d094d686 | 4754 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4755 | sector_t addr; |
4756 | if (rdev == NULL || | |
4757 | test_bit(Faulty, &rdev->flags) || | |
4758 | !test_bit(In_sync, &rdev->flags)) | |
4759 | goto failed; | |
4760 | ||
e0ee7785 | 4761 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
d094d686 N |
4762 | atomic_inc(&rdev->nr_pending); |
4763 | rcu_read_unlock(); | |
3ea7daa5 N |
4764 | success = sync_page_io(rdev, |
4765 | addr, | |
4766 | s << 9, | |
2d06e3b7 | 4767 | pages[idx], |
796a5cf0 | 4768 | REQ_OP_READ, 0, false); |
d094d686 N |
4769 | rdev_dec_pending(rdev, mddev); |
4770 | rcu_read_lock(); | |
3ea7daa5 N |
4771 | if (success) |
4772 | break; | |
4773 | failed: | |
4774 | slot++; | |
4775 | if (slot >= conf->copies) | |
4776 | slot = 0; | |
4777 | if (slot == first_slot) | |
4778 | break; | |
4779 | } | |
d094d686 | 4780 | rcu_read_unlock(); |
3ea7daa5 N |
4781 | if (!success) { |
4782 | /* couldn't read this block, must give up */ | |
4783 | set_bit(MD_RECOVERY_INTR, | |
4784 | &mddev->recovery); | |
584ed9fa | 4785 | kfree(r10b); |
3ea7daa5 N |
4786 | return -EIO; |
4787 | } | |
4788 | sectors -= s; | |
4789 | idx++; | |
4790 | } | |
584ed9fa | 4791 | kfree(r10b); |
3ea7daa5 N |
4792 | return 0; |
4793 | } | |
4794 | ||
4246a0b6 | 4795 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 4796 | { |
f0250618 | 4797 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
3ea7daa5 N |
4798 | struct mddev *mddev = r10_bio->mddev; |
4799 | struct r10conf *conf = mddev->private; | |
4800 | int d; | |
4801 | int slot; | |
4802 | int repl; | |
4803 | struct md_rdev *rdev = NULL; | |
4804 | ||
4805 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
4806 | if (repl) | |
4807 | rdev = conf->mirrors[d].replacement; | |
4808 | if (!rdev) { | |
4809 | smp_mb(); | |
4810 | rdev = conf->mirrors[d].rdev; | |
4811 | } | |
4812 | ||
4e4cbee9 | 4813 | if (bio->bi_status) { |
3ea7daa5 N |
4814 | /* FIXME should record badblock */ |
4815 | md_error(mddev, rdev); | |
4816 | } | |
4817 | ||
4818 | rdev_dec_pending(rdev, mddev); | |
4819 | end_reshape_request(r10_bio); | |
4820 | } | |
4821 | ||
4822 | static void end_reshape_request(struct r10bio *r10_bio) | |
4823 | { | |
4824 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
4825 | return; | |
4826 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
4827 | bio_put(r10_bio->master_bio); | |
4828 | put_buf(r10_bio); | |
4829 | } | |
4830 | ||
4831 | static void raid10_finish_reshape(struct mddev *mddev) | |
4832 | { | |
4833 | struct r10conf *conf = mddev->private; | |
4834 | ||
4835 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
4836 | return; | |
4837 | ||
4838 | if (mddev->delta_disks > 0) { | |
3ea7daa5 N |
4839 | if (mddev->recovery_cp > mddev->resync_max_sectors) { |
4840 | mddev->recovery_cp = mddev->resync_max_sectors; | |
4841 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4842 | } | |
e66db6f0 | 4843 | mddev->resync_max_sectors = mddev->array_sectors; |
63aced61 N |
4844 | } else { |
4845 | int d; | |
d094d686 | 4846 | rcu_read_lock(); |
63aced61 N |
4847 | for (d = conf->geo.raid_disks ; |
4848 | d < conf->geo.raid_disks - mddev->delta_disks; | |
4849 | d++) { | |
d094d686 | 4850 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
63aced61 N |
4851 | if (rdev) |
4852 | clear_bit(In_sync, &rdev->flags); | |
d094d686 | 4853 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
63aced61 N |
4854 | if (rdev) |
4855 | clear_bit(In_sync, &rdev->flags); | |
4856 | } | |
d094d686 | 4857 | rcu_read_unlock(); |
3ea7daa5 N |
4858 | } |
4859 | mddev->layout = mddev->new_layout; | |
4860 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
4861 | mddev->reshape_position = MaxSector; | |
4862 | mddev->delta_disks = 0; | |
4863 | mddev->reshape_backwards = 0; | |
4864 | } | |
4865 | ||
84fc4b56 | 4866 | static struct md_personality raid10_personality = |
1da177e4 LT |
4867 | { |
4868 | .name = "raid10", | |
2604b703 | 4869 | .level = 10, |
1da177e4 | 4870 | .owner = THIS_MODULE, |
849674e4 SL |
4871 | .make_request = raid10_make_request, |
4872 | .run = raid10_run, | |
afa0f557 | 4873 | .free = raid10_free, |
849674e4 SL |
4874 | .status = raid10_status, |
4875 | .error_handler = raid10_error, | |
1da177e4 LT |
4876 | .hot_add_disk = raid10_add_disk, |
4877 | .hot_remove_disk= raid10_remove_disk, | |
4878 | .spare_active = raid10_spare_active, | |
849674e4 | 4879 | .sync_request = raid10_sync_request, |
6cce3b23 | 4880 | .quiesce = raid10_quiesce, |
80c3a6ce | 4881 | .size = raid10_size, |
006a09a0 | 4882 | .resize = raid10_resize, |
dab8b292 | 4883 | .takeover = raid10_takeover, |
3ea7daa5 N |
4884 | .check_reshape = raid10_check_reshape, |
4885 | .start_reshape = raid10_start_reshape, | |
4886 | .finish_reshape = raid10_finish_reshape, | |
5c675f83 | 4887 | .congested = raid10_congested, |
1da177e4 LT |
4888 | }; |
4889 | ||
4890 | static int __init raid_init(void) | |
4891 | { | |
2604b703 | 4892 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
4893 | } |
4894 | ||
4895 | static void raid_exit(void) | |
4896 | { | |
2604b703 | 4897 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
4898 | } |
4899 | ||
4900 | module_init(raid_init); | |
4901 | module_exit(raid_exit); | |
4902 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 4903 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 4904 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 4905 | MODULE_ALIAS("md-raid10"); |
2604b703 | 4906 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
4907 | |
4908 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |