1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22#include <linux/blkdev.h>
23#include <linux/module.h>
24#include <linux/raid/md_u.h>
25#include <linux/seq_file.h>
26#include <linux/slab.h>
27#include "md.h"
28#include "md-multipath.h"
29
30#define MAX_WORK_PER_DISK 128
31
32#define NR_RESERVED_BUFS 32
33
34static int multipath_map (struct mpconf *conf)
35{
36 int i, disks = conf->raid_disks;
37
38
39
40
41
42
43 rcu_read_lock();
44 for (i = 0; i < disks; i++) {
45 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
46 if (rdev && test_bit(In_sync, &rdev->flags) &&
47 !test_bit(Faulty, &rdev->flags)) {
48 atomic_inc(&rdev->nr_pending);
49 rcu_read_unlock();
50 return i;
51 }
52 }
53 rcu_read_unlock();
54
55 pr_crit_ratelimited("multipath_map(): no more operational IO paths?\n");
56 return (-1);
57}
58
59static void multipath_reschedule_retry (struct multipath_bh *mp_bh)
60{
61 unsigned long flags;
62 struct mddev *mddev = mp_bh->mddev;
63 struct mpconf *conf = mddev->private;
64
65 spin_lock_irqsave(&conf->device_lock, flags);
66 list_add(&mp_bh->retry_list, &conf->retry_list);
67 spin_unlock_irqrestore(&conf->device_lock, flags);
68 md_wakeup_thread(mddev->thread);
69}
70
71
72
73
74
75
76static void multipath_end_bh_io(struct multipath_bh *mp_bh, blk_status_t status)
77{
78 struct bio *bio = mp_bh->master_bio;
79 struct mpconf *conf = mp_bh->mddev->private;
80
81 bio->bi_status = status;
82 bio_endio(bio);
83 mempool_free(mp_bh, &conf->pool);
84}
85
86static void multipath_end_request(struct bio *bio)
87{
88 struct multipath_bh *mp_bh = bio->bi_private;
89 struct mpconf *conf = mp_bh->mddev->private;
90 struct md_rdev *rdev = conf->multipaths[mp_bh->path].rdev;
91
92 if (!bio->bi_status)
93 multipath_end_bh_io(mp_bh, 0);
94 else if (!(bio->bi_opf & REQ_RAHEAD)) {
95
96
97
98 char b[BDEVNAME_SIZE];
99 md_error (mp_bh->mddev, rdev);
100 pr_info("multipath: %s: rescheduling sector %llu\n",
101 bdevname(rdev->bdev,b),
102 (unsigned long long)bio->bi_iter.bi_sector);
103 multipath_reschedule_retry(mp_bh);
104 } else
105 multipath_end_bh_io(mp_bh, bio->bi_status);
106 rdev_dec_pending(rdev, conf->mddev);
107}
108
109static bool multipath_make_request(struct mddev *mddev, struct bio * bio)
110{
111 struct mpconf *conf = mddev->private;
112 struct multipath_bh * mp_bh;
113 struct multipath_info *multipath;
114
115 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
116 && md_flush_request(mddev, bio))
117 return true;
118
119 mp_bh = mempool_alloc(&conf->pool, GFP_NOIO);
120
121 mp_bh->master_bio = bio;
122 mp_bh->mddev = mddev;
123
124 mp_bh->path = multipath_map(conf);
125 if (mp_bh->path < 0) {
126 bio_io_error(bio);
127 mempool_free(mp_bh, &conf->pool);
128 return true;
129 }
130 multipath = conf->multipaths + mp_bh->path;
131
132 bio_init(&mp_bh->bio, NULL, 0);
133 __bio_clone_fast(&mp_bh->bio, bio);
134
135 mp_bh->bio.bi_iter.bi_sector += multipath->rdev->data_offset;
136 bio_set_dev(&mp_bh->bio, multipath->rdev->bdev);
137 mp_bh->bio.bi_opf |= REQ_FAILFAST_TRANSPORT;
138 mp_bh->bio.bi_end_io = multipath_end_request;
139 mp_bh->bio.bi_private = mp_bh;
140 mddev_check_writesame(mddev, &mp_bh->bio);
141 mddev_check_write_zeroes(mddev, &mp_bh->bio);
142 generic_make_request(&mp_bh->bio);
143 return true;
144}
145
146static void multipath_status(struct seq_file *seq, struct mddev *mddev)
147{
148 struct mpconf *conf = mddev->private;
149 int i;
150
151 seq_printf (seq, " [%d/%d] [", conf->raid_disks,
152 conf->raid_disks - mddev->degraded);
153 rcu_read_lock();
154 for (i = 0; i < conf->raid_disks; i++) {
155 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
156 seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
157 }
158 rcu_read_unlock();
159 seq_putc(seq, ']');
160}
161
162static int multipath_congested(struct mddev *mddev, int bits)
163{
164 struct mpconf *conf = mddev->private;
165 int i, ret = 0;
166
167 rcu_read_lock();
168 for (i = 0; i < mddev->raid_disks ; i++) {
169 struct md_rdev *rdev = rcu_dereference(conf->multipaths[i].rdev);
170 if (rdev && !test_bit(Faulty, &rdev->flags)) {
171 struct request_queue *q = bdev_get_queue(rdev->bdev);
172
173 ret |= bdi_congested(q->backing_dev_info, bits);
174
175
176
177 break;
178 }
179 }
180 rcu_read_unlock();
181 return ret;
182}
183
184
185
186
187static void multipath_error (struct mddev *mddev, struct md_rdev *rdev)
188{
189 struct mpconf *conf = mddev->private;
190 char b[BDEVNAME_SIZE];
191
192 if (conf->raid_disks - mddev->degraded <= 1) {
193
194
195
196
197
198 pr_warn("multipath: only one IO path left and IO error.\n");
199
200 return;
201 }
202
203
204
205 if (test_and_clear_bit(In_sync, &rdev->flags)) {
206 unsigned long flags;
207 spin_lock_irqsave(&conf->device_lock, flags);
208 mddev->degraded++;
209 spin_unlock_irqrestore(&conf->device_lock, flags);
210 }
211 set_bit(Faulty, &rdev->flags);
212 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
213 pr_err("multipath: IO failure on %s, disabling IO path.\n"
214 "multipath: Operation continuing on %d IO paths.\n",
215 bdevname(rdev->bdev, b),
216 conf->raid_disks - mddev->degraded);
217}
218
219static void print_multipath_conf (struct mpconf *conf)
220{
221 int i;
222 struct multipath_info *tmp;
223
224 pr_debug("MULTIPATH conf printout:\n");
225 if (!conf) {
226 pr_debug("(conf==NULL)\n");
227 return;
228 }
229 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
230 conf->raid_disks);
231
232 for (i = 0; i < conf->raid_disks; i++) {
233 char b[BDEVNAME_SIZE];
234 tmp = conf->multipaths + i;
235 if (tmp->rdev)
236 pr_debug(" disk%d, o:%d, dev:%s\n",
237 i,!test_bit(Faulty, &tmp->rdev->flags),
238 bdevname(tmp->rdev->bdev,b));
239 }
240}
241
242static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
243{
244 struct mpconf *conf = mddev->private;
245 int err = -EEXIST;
246 int path;
247 struct multipath_info *p;
248 int first = 0;
249 int last = mddev->raid_disks - 1;
250
251 if (rdev->raid_disk >= 0)
252 first = last = rdev->raid_disk;
253
254 print_multipath_conf(conf);
255
256 for (path = first; path <= last; path++)
257 if ((p=conf->multipaths+path)->rdev == NULL) {
258 disk_stack_limits(mddev->gendisk, rdev->bdev,
259 rdev->data_offset << 9);
260
261 err = md_integrity_add_rdev(rdev, mddev);
262 if (err)
263 break;
264 spin_lock_irq(&conf->device_lock);
265 mddev->degraded--;
266 rdev->raid_disk = path;
267 set_bit(In_sync, &rdev->flags);
268 spin_unlock_irq(&conf->device_lock);
269 rcu_assign_pointer(p->rdev, rdev);
270 err = 0;
271 break;
272 }
273
274 print_multipath_conf(conf);
275
276 return err;
277}
278
279static int multipath_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
280{
281 struct mpconf *conf = mddev->private;
282 int err = 0;
283 int number = rdev->raid_disk;
284 struct multipath_info *p = conf->multipaths + number;
285
286 print_multipath_conf(conf);
287
288 if (rdev == p->rdev) {
289 if (test_bit(In_sync, &rdev->flags) ||
290 atomic_read(&rdev->nr_pending)) {
291 pr_warn("hot-remove-disk, slot %d is identified but is still operational!\n", number);
292 err = -EBUSY;
293 goto abort;
294 }
295 p->rdev = NULL;
296 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
297 synchronize_rcu();
298 if (atomic_read(&rdev->nr_pending)) {
299
300 err = -EBUSY;
301 p->rdev = rdev;
302 goto abort;
303 }
304 }
305 err = md_integrity_register(mddev);
306 }
307abort:
308
309 print_multipath_conf(conf);
310 return err;
311}
312
313
314
315
316
317
318
319
320
321static void multipathd(struct md_thread *thread)
322{
323 struct mddev *mddev = thread->mddev;
324 struct multipath_bh *mp_bh;
325 struct bio *bio;
326 unsigned long flags;
327 struct mpconf *conf = mddev->private;
328 struct list_head *head = &conf->retry_list;
329
330 md_check_recovery(mddev);
331 for (;;) {
332 char b[BDEVNAME_SIZE];
333 spin_lock_irqsave(&conf->device_lock, flags);
334 if (list_empty(head))
335 break;
336 mp_bh = list_entry(head->prev, struct multipath_bh, retry_list);
337 list_del(head->prev);
338 spin_unlock_irqrestore(&conf->device_lock, flags);
339
340 bio = &mp_bh->bio;
341 bio->bi_iter.bi_sector = mp_bh->master_bio->bi_iter.bi_sector;
342
343 if ((mp_bh->path = multipath_map (conf))<0) {
344 pr_err("multipath: %s: unrecoverable IO read error for block %llu\n",
345 bio_devname(bio, b),
346 (unsigned long long)bio->bi_iter.bi_sector);
347 multipath_end_bh_io(mp_bh, BLK_STS_IOERR);
348 } else {
349 pr_err("multipath: %s: redirecting sector %llu to another IO path\n",
350 bio_devname(bio, b),
351 (unsigned long long)bio->bi_iter.bi_sector);
352 *bio = *(mp_bh->master_bio);
353 bio->bi_iter.bi_sector +=
354 conf->multipaths[mp_bh->path].rdev->data_offset;
355 bio_set_dev(bio, conf->multipaths[mp_bh->path].rdev->bdev);
356 bio->bi_opf |= REQ_FAILFAST_TRANSPORT;
357 bio->bi_end_io = multipath_end_request;
358 bio->bi_private = mp_bh;
359 generic_make_request(bio);
360 }
361 }
362 spin_unlock_irqrestore(&conf->device_lock, flags);
363}
364
365static sector_t multipath_size(struct mddev *mddev, sector_t sectors, int raid_disks)
366{
367 WARN_ONCE(sectors || raid_disks,
368 "%s does not support generic reshape\n", __func__);
369
370 return mddev->dev_sectors;
371}
372
373static int multipath_run (struct mddev *mddev)
374{
375 struct mpconf *conf;
376 int disk_idx;
377 struct multipath_info *disk;
378 struct md_rdev *rdev;
379 int working_disks;
380 int ret;
381
382 if (md_check_no_bitmap(mddev))
383 return -EINVAL;
384
385 if (mddev->level != LEVEL_MULTIPATH) {
386 pr_warn("multipath: %s: raid level not set to multipath IO (%d)\n",
387 mdname(mddev), mddev->level);
388 goto out;
389 }
390
391
392
393
394
395
396 conf = kzalloc(sizeof(struct mpconf), GFP_KERNEL);
397 mddev->private = conf;
398 if (!conf)
399 goto out;
400
401 conf->multipaths = kcalloc(mddev->raid_disks,
402 sizeof(struct multipath_info),
403 GFP_KERNEL);
404 if (!conf->multipaths)
405 goto out_free_conf;
406
407 working_disks = 0;
408 rdev_for_each(rdev, mddev) {
409 disk_idx = rdev->raid_disk;
410 if (disk_idx < 0 ||
411 disk_idx >= mddev->raid_disks)
412 continue;
413
414 disk = conf->multipaths + disk_idx;
415 disk->rdev = rdev;
416 disk_stack_limits(mddev->gendisk, rdev->bdev,
417 rdev->data_offset << 9);
418
419 if (!test_bit(Faulty, &rdev->flags))
420 working_disks++;
421 }
422
423 conf->raid_disks = mddev->raid_disks;
424 conf->mddev = mddev;
425 spin_lock_init(&conf->device_lock);
426 INIT_LIST_HEAD(&conf->retry_list);
427
428 if (!working_disks) {
429 pr_warn("multipath: no operational IO paths for %s\n",
430 mdname(mddev));
431 goto out_free_conf;
432 }
433 mddev->degraded = conf->raid_disks - working_disks;
434
435 ret = mempool_init_kmalloc_pool(&conf->pool, NR_RESERVED_BUFS,
436 sizeof(struct multipath_bh));
437 if (ret)
438 goto out_free_conf;
439
440 mddev->thread = md_register_thread(multipathd, mddev,
441 "multipath");
442 if (!mddev->thread)
443 goto out_free_conf;
444
445 pr_info("multipath: array %s active with %d out of %d IO paths\n",
446 mdname(mddev), conf->raid_disks - mddev->degraded,
447 mddev->raid_disks);
448
449
450
451 md_set_array_sectors(mddev, multipath_size(mddev, 0, 0));
452
453 if (md_integrity_register(mddev))
454 goto out_free_conf;
455
456 return 0;
457
458out_free_conf:
459 mempool_exit(&conf->pool);
460 kfree(conf->multipaths);
461 kfree(conf);
462 mddev->private = NULL;
463out:
464 return -EIO;
465}
466
467static void multipath_free(struct mddev *mddev, void *priv)
468{
469 struct mpconf *conf = priv;
470
471 mempool_exit(&conf->pool);
472 kfree(conf->multipaths);
473 kfree(conf);
474}
475
476static struct md_personality multipath_personality =
477{
478 .name = "multipath",
479 .level = LEVEL_MULTIPATH,
480 .owner = THIS_MODULE,
481 .make_request = multipath_make_request,
482 .run = multipath_run,
483 .free = multipath_free,
484 .status = multipath_status,
485 .error_handler = multipath_error,
486 .hot_add_disk = multipath_add_disk,
487 .hot_remove_disk= multipath_remove_disk,
488 .size = multipath_size,
489 .congested = multipath_congested,
490};
491
492static int __init multipath_init (void)
493{
494 return register_md_personality (&multipath_personality);
495}
496
497static void __exit multipath_exit (void)
498{
499 unregister_md_personality (&multipath_personality);
500}
501
502module_init(multipath_init);
503module_exit(multipath_exit);
504MODULE_LICENSE("GPL");
505MODULE_DESCRIPTION("simple multi-path personality for MD");
506MODULE_ALIAS("md-personality-7");
507MODULE_ALIAS("md-multipath");
508MODULE_ALIAS("md-level--4");
509
