blob: b3ccf1e0d4f218d8cb55f5d895e1bf46ac26ac8b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2003 Sistina Software Limited.
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01003 * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * This file is released under the GPL.
6 */
7
Jonathan Brassow06386bb2008-02-08 02:11:37 +00008#include "dm-bio-record.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -07009
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/init.h>
11#include <linux/mempool.h>
12#include <linux/module.h>
13#include <linux/pagemap.h>
14#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/workqueue.h>
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010016#include <linux/device-mapper.h>
Alasdair G Kergona765e202008-04-24 22:02:01 +010017#include <linux/dm-io.h>
18#include <linux/dm-dirty-log.h>
19#include <linux/dm-kcopyd.h>
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010020#include <linux/dm-region-hash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Alasdair G Kergon72d94862006-06-26 00:27:35 -070022#define DM_MSG_PREFIX "raid1"
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010023
24#define MAX_RECOVERY 1 /* Maximum number of regions recovered in parallel. */
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025
Lidong Zhonged632872015-05-13 14:04:10 +080026#define DM_RAID1_HANDLE_ERRORS 0x01
27#define DM_RAID1_KEEP_LOG 0x02
Jonathan Brassowf44db672007-07-12 17:29:04 +010028#define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS)
Lidong Zhonged632872015-05-13 14:04:10 +080029#define keep_log(p) ((p)->features & DM_RAID1_KEEP_LOG)
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -070030
Jonathan E Brassow33184042006-11-08 17:44:44 -080031static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Linus Torvalds1da177e2005-04-16 15:20:36 -070033/*-----------------------------------------------------------------
Neil Browne4c8b3b2006-06-26 00:27:26 -070034 * Mirror set structures.
35 *---------------------------------------------------------------*/
Jonathan Brassow72f4b312008-02-08 02:11:29 +000036enum dm_raid1_error {
37 DM_RAID1_WRITE_ERROR,
Mikulas Patocka64b30c42009-12-10 23:52:02 +000038 DM_RAID1_FLUSH_ERROR,
Jonathan Brassow72f4b312008-02-08 02:11:29 +000039 DM_RAID1_SYNC_ERROR,
40 DM_RAID1_READ_ERROR
41};
42
Neil Browne4c8b3b2006-06-26 00:27:26 -070043struct mirror {
Jonathan Brassowaa5617c2007-10-19 22:47:58 +010044 struct mirror_set *ms;
Neil Browne4c8b3b2006-06-26 00:27:26 -070045 atomic_t error_count;
Al Viro39ed7ad2008-02-13 03:53:00 +000046 unsigned long error_type;
Neil Browne4c8b3b2006-06-26 00:27:26 -070047 struct dm_dev *dev;
48 sector_t offset;
49};
50
51struct mirror_set {
52 struct dm_target *ti;
53 struct list_head list;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010054
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -070055 uint64_t features;
Neil Browne4c8b3b2006-06-26 00:27:26 -070056
Jonathan Brassow72f4b312008-02-08 02:11:29 +000057 spinlock_t lock; /* protects the lists */
Neil Browne4c8b3b2006-06-26 00:27:26 -070058 struct bio_list reads;
59 struct bio_list writes;
Jonathan Brassow72f4b312008-02-08 02:11:29 +000060 struct bio_list failures;
Mikulas Patocka04788502009-12-10 23:52:03 +000061 struct bio_list holds; /* bios are waiting until suspend */
Neil Browne4c8b3b2006-06-26 00:27:26 -070062
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010063 struct dm_region_hash *rh;
64 struct dm_kcopyd_client *kcopyd_client;
Milan Broz88be1632007-05-09 02:33:04 -070065 struct dm_io_client *io_client;
66
Neil Browne4c8b3b2006-06-26 00:27:26 -070067 /* recovery */
68 region_t nr_regions;
69 int in_sync;
Jonathan Brassowfc1ff952007-07-12 17:29:15 +010070 int log_failure;
Mikulas Patocka929be8f2009-12-10 23:52:06 +000071 int leg_failure;
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +000072 atomic_t suspend;
Neil Browne4c8b3b2006-06-26 00:27:26 -070073
Jonathan Brassow72f4b312008-02-08 02:11:29 +000074 atomic_t default_mirror; /* Default mirror */
Neil Browne4c8b3b2006-06-26 00:27:26 -070075
Holger Smolinski6ad36fe2007-05-09 02:32:50 -070076 struct workqueue_struct *kmirrord_wq;
77 struct work_struct kmirrord_work;
Mikulas Patockaa2aebe02008-04-24 22:10:42 +010078 struct timer_list timer;
79 unsigned long timer_pending;
80
Jonathan Brassow72f4b312008-02-08 02:11:29 +000081 struct work_struct trigger_event;
Holger Smolinski6ad36fe2007-05-09 02:32:50 -070082
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010083 unsigned nr_mirrors;
Neil Browne4c8b3b2006-06-26 00:27:26 -070084 struct mirror mirror[0];
85};
86
Mikulas Patockadf5d2e92013-03-01 22:45:49 +000087DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(raid1_resync_throttle,
88 "A percentage of time allocated for raid resynchronization");
89
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010090static void wakeup_mirrord(void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010092 struct mirror_set *ms = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Holger Smolinski6ad36fe2007-05-09 02:32:50 -070094 queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
95}
96
Mikulas Patockaa2aebe02008-04-24 22:10:42 +010097static void delayed_wake_fn(unsigned long data)
98{
99 struct mirror_set *ms = (struct mirror_set *) data;
100
101 clear_bit(0, &ms->timer_pending);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100102 wakeup_mirrord(ms);
Mikulas Patockaa2aebe02008-04-24 22:10:42 +0100103}
104
105static void delayed_wake(struct mirror_set *ms)
106{
107 if (test_and_set_bit(0, &ms->timer_pending))
108 return;
109
110 ms->timer.expires = jiffies + HZ / 5;
111 ms->timer.data = (unsigned long) ms;
112 ms->timer.function = delayed_wake_fn;
113 add_timer(&ms->timer);
114}
115
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100116static void wakeup_all_recovery_waiters(void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100118 wake_up_all(&_kmirrord_recovery_stopped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119}
120
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100121static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
123 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 int should_wake = 0;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100125 struct bio_list *bl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100127 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
128 spin_lock_irqsave(&ms->lock, flags);
129 should_wake = !(bl->head);
130 bio_list_add(bl, bio);
131 spin_unlock_irqrestore(&ms->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133 if (should_wake)
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100134 wakeup_mirrord(ms);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100137static void dispatch_bios(void *context, struct bio_list *bio_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100139 struct mirror_set *ms = context;
140 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100142 while ((bio = bio_list_pop(bio_list)))
143 queue_bio(ms, bio, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
Mikulas Patocka89c7cd82012-12-21 20:23:39 +0000146struct dm_raid1_bio_record {
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000147 struct mirror *m;
Mikulas Patocka0045d612012-12-21 20:23:40 +0000148 /* if details->bi_bdev == NULL, details were not saved */
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000149 struct dm_bio_details details;
Mikulas Patocka0045d612012-12-21 20:23:40 +0000150 region_t write_region;
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000151};
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153/*
154 * Every mirror should look like this one.
155 */
156#define DEFAULT_MIRROR 0
157
158/*
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000159 * This is yucky. We squirrel the mirror struct away inside
160 * bi_next for read/write buffers. This is safe since the bh
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 * doesn't get submitted to the lower levels of block layer.
162 */
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000163static struct mirror *bio_get_m(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000165 return (struct mirror *) bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000168static void bio_set_m(struct bio *bio, struct mirror *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000170 bio->bi_next = (struct bio *) m;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
172
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000173static struct mirror *get_default_mirror(struct mirror_set *ms)
174{
175 return &ms->mirror[atomic_read(&ms->default_mirror)];
176}
177
178static void set_default_mirror(struct mirror *m)
179{
180 struct mirror_set *ms = m->ms;
181 struct mirror *m0 = &(ms->mirror[0]);
182
183 atomic_set(&ms->default_mirror, m - m0);
184}
185
Mikulas Patocka87968dd2009-12-10 23:52:04 +0000186static struct mirror *get_valid_mirror(struct mirror_set *ms)
187{
188 struct mirror *m;
189
190 for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++)
191 if (!atomic_read(&m->error_count))
192 return m;
193
194 return NULL;
195}
196
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000197/* fail_mirror
198 * @m: mirror device to fail
199 * @error_type: one of the enum's, DM_RAID1_*_ERROR
200 *
201 * If errors are being handled, record the type of
202 * error encountered for this device. If this type
203 * of error has already been recorded, we can return;
204 * otherwise, we must signal userspace by triggering
205 * an event. Additionally, if the device is the
206 * primary device, we must choose a new primary, but
207 * only if the mirror is in-sync.
208 *
209 * This function must not block.
210 */
211static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
212{
213 struct mirror_set *ms = m->ms;
214 struct mirror *new;
215
Mikulas Patocka929be8f2009-12-10 23:52:06 +0000216 ms->leg_failure = 1;
217
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000218 /*
219 * error_count is used for nothing more than a
220 * simple way to tell if a device has encountered
221 * errors.
222 */
223 atomic_inc(&m->error_count);
224
225 if (test_and_set_bit(error_type, &m->error_type))
226 return;
227
Jonathan Brassowd460c652009-01-06 03:04:57 +0000228 if (!errors_handled(ms))
229 return;
230
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000231 if (m != get_default_mirror(ms))
232 goto out;
233
Lidong Zhonged632872015-05-13 14:04:10 +0800234 if (!ms->in_sync && !keep_log(ms)) {
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000235 /*
236 * Better to issue requests to same failing device
237 * than to risk returning corrupt data.
238 */
239 DMERR("Primary mirror (%s) failed while out-of-sync: "
240 "Reads may fail.", m->dev->name);
241 goto out;
242 }
243
Mikulas Patocka87968dd2009-12-10 23:52:04 +0000244 new = get_valid_mirror(ms);
245 if (new)
246 set_default_mirror(new);
247 else
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000248 DMWARN("All sides of mirror have failed.");
249
250out:
251 schedule_work(&ms->trigger_event);
252}
253
Mikulas Patockac0da3742009-12-10 23:52:02 +0000254static int mirror_flush(struct dm_target *ti)
255{
256 struct mirror_set *ms = ti->private;
257 unsigned long error_bits;
258
259 unsigned int i;
260 struct dm_io_region io[ms->nr_mirrors];
261 struct mirror *m;
262 struct dm_io_request io_req = {
Tejun Heod87f4c12010-09-03 11:56:19 +0200263 .bi_rw = WRITE_FLUSH,
Mikulas Patockac0da3742009-12-10 23:52:02 +0000264 .mem.type = DM_IO_KMEM,
Mike Snitzer5fc2ffe2011-01-13 19:59:48 +0000265 .mem.ptr.addr = NULL,
Mikulas Patockac0da3742009-12-10 23:52:02 +0000266 .client = ms->io_client,
267 };
268
269 for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) {
270 io[i].bdev = m->dev->bdev;
271 io[i].sector = 0;
272 io[i].count = 0;
273 }
274
275 error_bits = -1;
276 dm_io(&io_req, ms->nr_mirrors, io, &error_bits);
277 if (unlikely(error_bits != 0)) {
278 for (i = 0; i < ms->nr_mirrors; i++)
279 if (test_bit(i, &error_bits))
280 fail_mirror(ms->mirror + i,
Mikulas Patocka64b30c42009-12-10 23:52:02 +0000281 DM_RAID1_FLUSH_ERROR);
Mikulas Patockac0da3742009-12-10 23:52:02 +0000282 return -EIO;
283 }
284
285 return 0;
286}
287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288/*-----------------------------------------------------------------
289 * Recovery.
290 *
291 * When a mirror is first activated we may find that some regions
292 * are in the no-sync state. We have to recover these by
293 * recopying from the default mirror to all the others.
294 *---------------------------------------------------------------*/
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700295static void recovery_complete(int read_err, unsigned long write_err,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 void *context)
297{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100298 struct dm_region *reg = context;
299 struct mirror_set *ms = dm_rh_region_context(reg);
Jonathan Brassow8f0205b2008-02-08 02:11:32 +0000300 int m, bit = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
Jonathan Brassow8f0205b2008-02-08 02:11:32 +0000302 if (read_err) {
Jonathan Brassowf44db672007-07-12 17:29:04 +0100303 /* Read error means the failure of default mirror. */
304 DMERR_LIMIT("Unable to read primary mirror during recovery");
Jonathan Brassow8f0205b2008-02-08 02:11:32 +0000305 fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR);
306 }
Jonathan Brassowf44db672007-07-12 17:29:04 +0100307
Jonathan Brassow8f0205b2008-02-08 02:11:32 +0000308 if (write_err) {
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700309 DMERR_LIMIT("Write error during recovery (error = 0x%lx)",
Jonathan Brassowf44db672007-07-12 17:29:04 +0100310 write_err);
Jonathan Brassow8f0205b2008-02-08 02:11:32 +0000311 /*
312 * Bits correspond to devices (excluding default mirror).
313 * The default mirror cannot change during recovery.
314 */
315 for (m = 0; m < ms->nr_mirrors; m++) {
316 if (&ms->mirror[m] == get_default_mirror(ms))
317 continue;
318 if (test_bit(bit, &write_err))
319 fail_mirror(ms->mirror + m,
320 DM_RAID1_SYNC_ERROR);
321 bit++;
322 }
323 }
Jonathan Brassowf44db672007-07-12 17:29:04 +0100324
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100325 dm_rh_recovery_end(reg, !(read_err || write_err));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100328static int recover(struct mirror_set *ms, struct dm_region *reg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329{
330 int r;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100331 unsigned i;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100332 struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 struct mirror *m;
334 unsigned long flags = 0;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100335 region_t key = dm_rh_get_region_key(reg);
336 sector_t region_size = dm_rh_get_region_size(ms->rh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 /* fill in the source */
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000339 m = get_default_mirror(ms);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 from.bdev = m->dev->bdev;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100341 from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
342 if (key == (ms->nr_regions - 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 /*
344 * The final region may be smaller than
345 * region_size.
346 */
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100347 from.count = ms->ti->len & (region_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 if (!from.count)
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100349 from.count = region_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 } else
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100351 from.count = region_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 /* fill in the destinations */
354 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000355 if (&ms->mirror[i] == get_default_mirror(ms))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 continue;
357
358 m = ms->mirror + i;
359 dest->bdev = m->dev->bdev;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100360 dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 dest->count = from.count;
362 dest++;
363 }
364
365 /* hand to kcopyd */
Jonathan Brassowf7c83e22008-10-10 13:36:59 +0100366 if (!errors_handled(ms))
367 set_bit(DM_KCOPYD_IGNORE_ERROR, &flags);
368
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100369 r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to,
370 flags, recovery_complete, reg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 return r;
373}
374
Lidong Zhonged632872015-05-13 14:04:10 +0800375static void reset_ms_flags(struct mirror_set *ms)
376{
377 unsigned int m;
378
379 ms->leg_failure = 0;
380 for (m = 0; m < ms->nr_mirrors; m++) {
381 atomic_set(&(ms->mirror[m].error_count), 0);
382 ms->mirror[m].error_type = 0;
383 }
384}
385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386static void do_recovery(struct mirror_set *ms)
387{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100388 struct dm_region *reg;
389 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 int r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 /*
393 * Start quiescing some regions.
394 */
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100395 dm_rh_recovery_prepare(ms->rh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397 /*
398 * Copy any already quiesced regions.
399 */
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100400 while ((reg = dm_rh_recovery_start(ms->rh))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 r = recover(ms, reg);
402 if (r)
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100403 dm_rh_recovery_end(reg, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 }
405
406 /*
407 * Update the in sync flag.
408 */
409 if (!ms->in_sync &&
410 (log->type->get_sync_count(log) == ms->nr_regions)) {
411 /* the sync is complete */
412 dm_table_event(ms->ti->table);
413 ms->in_sync = 1;
Lidong Zhonged632872015-05-13 14:04:10 +0800414 reset_ms_flags(ms);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 }
416}
417
418/*-----------------------------------------------------------------
419 * Reads
420 *---------------------------------------------------------------*/
421static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
422{
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000423 struct mirror *m = get_default_mirror(ms);
424
425 do {
426 if (likely(!atomic_read(&m->error_count)))
427 return m;
428
429 if (m-- == ms->mirror)
430 m += ms->nr_mirrors;
431 } while (m != get_default_mirror(ms));
432
433 return NULL;
434}
435
436static int default_ok(struct mirror *m)
437{
438 struct mirror *default_mirror = get_default_mirror(m->ms);
439
440 return !atomic_read(&default_mirror->error_count);
441}
442
443static int mirror_available(struct mirror_set *ms, struct bio *bio)
444{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100445 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
446 region_t region = dm_rh_bio_to_region(ms->rh, bio);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000447
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100448 if (log->type->in_sync(log, region, 0))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700449 return choose_mirror(ms, bio->bi_iter.bi_sector) ? 1 : 0;
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000450
451 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
454/*
455 * remap a buffer to a particular mirror.
456 */
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000457static sector_t map_sector(struct mirror *m, struct bio *bio)
458{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700459 if (unlikely(!bio->bi_iter.bi_size))
Mikulas Patocka41841532009-12-10 23:51:59 +0000460 return 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700461 return m->offset + dm_target_offset(m->ms->ti, bio->bi_iter.bi_sector);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000462}
463
464static void map_bio(struct mirror *m, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 bio->bi_bdev = m->dev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700467 bio->bi_iter.bi_sector = map_sector(m, bio);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000468}
469
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +0100470static void map_region(struct dm_io_region *io, struct mirror *m,
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000471 struct bio *bio)
472{
473 io->bdev = m->dev->bdev;
474 io->sector = map_sector(m, bio);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800475 io->count = bio_sectors(bio);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000476}
477
Mikulas Patocka04788502009-12-10 23:52:03 +0000478static void hold_bio(struct mirror_set *ms, struct bio *bio)
479{
480 /*
Takahiro Yasuif0703042010-03-06 02:32:35 +0000481 * Lock is required to avoid race condition during suspend
482 * process.
Mikulas Patocka04788502009-12-10 23:52:03 +0000483 */
Takahiro Yasuif0703042010-03-06 02:32:35 +0000484 spin_lock_irq(&ms->lock);
485
Mikulas Patocka04788502009-12-10 23:52:03 +0000486 if (atomic_read(&ms->suspend)) {
Takahiro Yasuif0703042010-03-06 02:32:35 +0000487 spin_unlock_irq(&ms->lock);
488
489 /*
490 * If device is suspended, complete the bio.
491 */
Mikulas Patocka04788502009-12-10 23:52:03 +0000492 if (dm_noflush_suspending(ms->ti))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200493 bio->bi_error = DM_ENDIO_REQUEUE;
Mikulas Patocka04788502009-12-10 23:52:03 +0000494 else
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200495 bio->bi_error = -EIO;
496
497 bio_endio(bio);
Mikulas Patocka04788502009-12-10 23:52:03 +0000498 return;
499 }
500
501 /*
502 * Hold bio until the suspend is complete.
503 */
Mikulas Patocka04788502009-12-10 23:52:03 +0000504 bio_list_add(&ms->holds, bio);
505 spin_unlock_irq(&ms->lock);
506}
507
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000508/*-----------------------------------------------------------------
509 * Reads
510 *---------------------------------------------------------------*/
511static void read_callback(unsigned long error, void *context)
512{
513 struct bio *bio = context;
514 struct mirror *m;
515
516 m = bio_get_m(bio);
517 bio_set_m(bio, NULL);
518
519 if (likely(!error)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200520 bio_endio(bio);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000521 return;
522 }
523
524 fail_mirror(m, DM_RAID1_READ_ERROR);
525
526 if (likely(default_ok(m)) || mirror_available(m->ms, bio)) {
527 DMWARN_LIMIT("Read failure on mirror device %s. "
528 "Trying alternative device.",
529 m->dev->name);
530 queue_bio(m->ms, bio, bio_rw(bio));
531 return;
532 }
533
534 DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.",
535 m->dev->name);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200536 bio_io_error(bio);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000537}
538
539/* Asynchronous read. */
540static void read_async_bio(struct mirror *m, struct bio *bio)
541{
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +0100542 struct dm_io_region io;
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000543 struct dm_io_request io_req = {
544 .bi_rw = READ,
Kent Overstreet003b5c52013-10-11 15:45:43 -0700545 .mem.type = DM_IO_BIO,
546 .mem.ptr.bio = bio,
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000547 .notify.fn = read_callback,
548 .notify.context = bio,
549 .client = m->ms->io_client,
550 };
551
552 map_region(&io, m, bio);
553 bio_set_m(bio, m);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100554 BUG_ON(dm_io(&io_req, 1, &io, NULL));
555}
556
557static inline int region_in_sync(struct mirror_set *ms, region_t region,
558 int may_block)
559{
560 int state = dm_rh_get_state(ms->rh, region, may_block);
561 return state == DM_RH_CLEAN || state == DM_RH_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562}
563
564static void do_reads(struct mirror_set *ms, struct bio_list *reads)
565{
566 region_t region;
567 struct bio *bio;
568 struct mirror *m;
569
570 while ((bio = bio_list_pop(reads))) {
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100571 region = dm_rh_bio_to_region(ms->rh, bio);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000572 m = get_default_mirror(ms);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
574 /*
575 * We can only read balance if the region is in sync.
576 */
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100577 if (likely(region_in_sync(ms, region, 1)))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700578 m = choose_mirror(ms, bio->bi_iter.bi_sector);
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000579 else if (m && atomic_read(&m->error_count))
580 m = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000582 if (likely(m))
583 read_async_bio(m, bio);
584 else
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200585 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
587}
588
589/*-----------------------------------------------------------------
590 * Writes.
591 *
592 * We do different things with the write io depending on the
593 * state of the region that it's in:
594 *
595 * SYNC: increment pending, use kcopyd to write to *all* mirrors
596 * RECOVERING: delay the io until recovery completes
597 * NOSYNC: increment pending, just write to the default mirror
598 *---------------------------------------------------------------*/
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000599
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601static void write_callback(unsigned long error, void *context)
602{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200603 unsigned i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 struct bio *bio = (struct bio *) context;
605 struct mirror_set *ms;
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000606 int should_wake = 0;
607 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000609 ms = bio_get_m(bio)->ms;
610 bio_set_m(bio, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 /*
613 * NOTE: We don't decrement the pending count here,
614 * instead it is done by the targets endio function.
615 * This way we handle both writes to SYNC and NOSYNC
616 * regions with the same code.
617 */
Mikulas Patocka60f355e2009-12-10 23:52:05 +0000618 if (likely(!error)) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200619 bio_endio(bio);
Mikulas Patocka60f355e2009-12-10 23:52:05 +0000620 return;
621 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Mikulas Patockaf2ed51a2015-02-12 10:09:20 -0500623 /*
624 * If the bio is discard, return an error, but do not
625 * degrade the array.
626 */
627 if (bio->bi_rw & REQ_DISCARD) {
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200628 bio->bi_error = -EOPNOTSUPP;
629 bio_endio(bio);
Mikulas Patockaf2ed51a2015-02-12 10:09:20 -0500630 return;
631 }
632
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000633 for (i = 0; i < ms->nr_mirrors; i++)
634 if (test_bit(i, &error))
635 fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000636
Mikulas Patocka60f355e2009-12-10 23:52:05 +0000637 /*
638 * Need to raise event. Since raising
639 * events can block, we need to do it in
640 * the main thread.
641 */
642 spin_lock_irqsave(&ms->lock, flags);
643 if (!ms->failures.head)
644 should_wake = 1;
645 bio_list_add(&ms->failures, bio);
646 spin_unlock_irqrestore(&ms->lock, flags);
647 if (should_wake)
648 wakeup_mirrord(ms);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
651static void do_write(struct mirror_set *ms, struct bio *bio)
652{
653 unsigned int i;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +0100654 struct dm_io_region io[ms->nr_mirrors], *dest = io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 struct mirror *m;
Milan Broz88be1632007-05-09 02:33:04 -0700656 struct dm_io_request io_req = {
Tejun Heod87f4c12010-09-03 11:56:19 +0200657 .bi_rw = WRITE | (bio->bi_rw & WRITE_FLUSH_FUA),
Kent Overstreet003b5c52013-10-11 15:45:43 -0700658 .mem.type = DM_IO_BIO,
659 .mem.ptr.bio = bio,
Milan Broz88be1632007-05-09 02:33:04 -0700660 .notify.fn = write_callback,
661 .notify.context = bio,
662 .client = ms->io_client,
663 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Mike Snitzer5fc2ffe2011-01-13 19:59:48 +0000665 if (bio->bi_rw & REQ_DISCARD) {
666 io_req.bi_rw |= REQ_DISCARD;
667 io_req.mem.type = DM_IO_KMEM;
668 io_req.mem.ptr.addr = NULL;
669 }
670
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000671 for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
672 map_region(dest++, m, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Jonathan Brassow06386bb2008-02-08 02:11:37 +0000674 /*
675 * Use default mirror because we only need it to retrieve the reference
676 * to the mirror set in write_callback().
677 */
678 bio_set_m(bio, get_default_mirror(ms));
Milan Broz88be1632007-05-09 02:33:04 -0700679
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100680 BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
683static void do_writes(struct mirror_set *ms, struct bio_list *writes)
684{
685 int state;
686 struct bio *bio;
687 struct bio_list sync, nosync, recover, *this_list = NULL;
Jonathan Brassow7513c2a2009-04-02 19:55:30 +0100688 struct bio_list requeue;
689 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
690 region_t region;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
692 if (!writes->head)
693 return;
694
695 /*
696 * Classify each write.
697 */
698 bio_list_init(&sync);
699 bio_list_init(&nosync);
700 bio_list_init(&recover);
Jonathan Brassow7513c2a2009-04-02 19:55:30 +0100701 bio_list_init(&requeue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
703 while ((bio = bio_list_pop(writes))) {
Mike Snitzer5fc2ffe2011-01-13 19:59:48 +0000704 if ((bio->bi_rw & REQ_FLUSH) ||
705 (bio->bi_rw & REQ_DISCARD)) {
Mikulas Patocka41841532009-12-10 23:51:59 +0000706 bio_list_add(&sync, bio);
707 continue;
708 }
709
Jonathan Brassow7513c2a2009-04-02 19:55:30 +0100710 region = dm_rh_bio_to_region(ms->rh, bio);
711
712 if (log->type->is_remote_recovering &&
713 log->type->is_remote_recovering(log, region)) {
714 bio_list_add(&requeue, bio);
715 continue;
716 }
717
718 state = dm_rh_get_state(ms->rh, region, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 switch (state) {
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100720 case DM_RH_CLEAN:
721 case DM_RH_DIRTY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 this_list = &sync;
723 break;
724
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100725 case DM_RH_NOSYNC:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 this_list = &nosync;
727 break;
728
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100729 case DM_RH_RECOVERING:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 this_list = &recover;
731 break;
732 }
733
734 bio_list_add(this_list, bio);
735 }
736
737 /*
Jonathan Brassow7513c2a2009-04-02 19:55:30 +0100738 * Add bios that are delayed due to remote recovery
739 * back on to the write queue
740 */
741 if (unlikely(requeue.head)) {
742 spin_lock_irq(&ms->lock);
743 bio_list_merge(&ms->writes, &requeue);
744 spin_unlock_irq(&ms->lock);
Mikulas Patocka69885682009-07-23 20:30:37 +0100745 delayed_wake(ms);
Jonathan Brassow7513c2a2009-04-02 19:55:30 +0100746 }
747
748 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 * Increment the pending counts for any regions that will
750 * be written to (writes to recover regions are going to
751 * be delayed).
752 */
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100753 dm_rh_inc_pending(ms->rh, &sync);
754 dm_rh_inc_pending(ms->rh, &nosync);
Jonathan Brassowd2b69862009-09-04 20:40:32 +0100755
756 /*
757 * If the flush fails on a previous call and succeeds here,
758 * we must not reset the log_failure variable. We need
759 * userspace interaction to do that.
760 */
761 ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
763 /*
764 * Dispatch io.
765 */
Mikulas Patocka5528d172010-02-16 18:42:55 +0000766 if (unlikely(ms->log_failure) && errors_handled(ms)) {
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000767 spin_lock_irq(&ms->lock);
768 bio_list_merge(&ms->failures, &sync);
769 spin_unlock_irq(&ms->lock);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100770 wakeup_mirrord(ms);
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000771 } else
Jonathan Brassowfc1ff952007-07-12 17:29:15 +0100772 while ((bio = bio_list_pop(&sync)))
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000773 do_write(ms, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
775 while ((bio = bio_list_pop(&recover)))
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100776 dm_rh_delay(ms->rh, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 while ((bio = bio_list_pop(&nosync))) {
Lidong Zhonged632872015-05-13 14:04:10 +0800779 if (unlikely(ms->leg_failure) && errors_handled(ms) && !keep_log(ms)) {
Mikulas Patockaede5ea02010-03-06 02:32:22 +0000780 spin_lock_irq(&ms->lock);
781 bio_list_add(&ms->failures, bio);
782 spin_unlock_irq(&ms->lock);
783 wakeup_mirrord(ms);
784 } else {
Mikulas Patocka929be8f2009-12-10 23:52:06 +0000785 map_bio(get_default_mirror(ms), bio);
786 generic_make_request(bio);
787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
789}
790
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000791static void do_failures(struct mirror_set *ms, struct bio_list *failures)
792{
793 struct bio *bio;
794
Mikulas Patocka0f398a82009-12-10 23:52:04 +0000795 if (likely(!failures->head))
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000796 return;
797
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000798 /*
799 * If the log has failed, unattempted writes are being
Mikulas Patocka0f398a82009-12-10 23:52:04 +0000800 * put on the holds list. We can't issue those writes
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000801 * until a log has been marked, so we must store them.
802 *
803 * If a 'noflush' suspend is in progress, we can requeue
804 * the I/O's to the core. This give userspace a chance
805 * to reconfigure the mirror, at which point the core
806 * will reissue the writes. If the 'noflush' flag is
807 * not set, we have no choice but to return errors.
808 *
809 * Some writes on the failures list may have been
810 * submitted before the log failure and represent a
811 * failure to write to one of the devices. It is ok
812 * for us to treat them the same and requeue them
813 * as well.
814 */
Mikulas Patocka0f398a82009-12-10 23:52:04 +0000815 while ((bio = bio_list_pop(failures))) {
Mikulas Patocka60f355e2009-12-10 23:52:05 +0000816 if (!ms->log_failure) {
Mikulas Patocka0f398a82009-12-10 23:52:04 +0000817 ms->in_sync = 0;
Mikulas Patockac58098b2009-12-10 23:52:05 +0000818 dm_rh_mark_nosync(ms->rh, bio);
Mikulas Patocka0f398a82009-12-10 23:52:04 +0000819 }
Mikulas Patocka60f355e2009-12-10 23:52:05 +0000820
821 /*
822 * If all the legs are dead, fail the I/O.
Lidong Zhonged632872015-05-13 14:04:10 +0800823 * If the device has failed and keep_log is enabled,
824 * fail the I/O.
825 *
826 * If we have been told to handle errors, and keep_log
827 * isn't enabled, hold the bio and wait for userspace to
828 * deal with the problem.
829 *
Mikulas Patocka60f355e2009-12-10 23:52:05 +0000830 * Otherwise pretend that the I/O succeeded. (This would
831 * be wrong if the failed leg returned after reboot and
832 * got replicated back to the good legs.)
833 */
Lidong Zhonged632872015-05-13 14:04:10 +0800834 if (unlikely(!get_valid_mirror(ms) || (keep_log(ms) && ms->log_failure)))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200835 bio_io_error(bio);
Lidong Zhonged632872015-05-13 14:04:10 +0800836 else if (errors_handled(ms) && !keep_log(ms))
Mikulas Patocka60f355e2009-12-10 23:52:05 +0000837 hold_bio(ms, bio);
838 else
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200839 bio_endio(bio);
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000840 }
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000841}
842
843static void trigger_event(struct work_struct *work)
844{
845 struct mirror_set *ms =
846 container_of(work, struct mirror_set, trigger_event);
847
848 dm_table_event(ms->ti->table);
849}
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851/*-----------------------------------------------------------------
852 * kmirrord
853 *---------------------------------------------------------------*/
Mikulas Patockaa2aebe02008-04-24 22:10:42 +0100854static void do_mirror(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100856 struct mirror_set *ms = container_of(work, struct mirror_set,
857 kmirrord_work);
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000858 struct bio_list reads, writes, failures;
859 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000861 spin_lock_irqsave(&ms->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 reads = ms->reads;
863 writes = ms->writes;
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000864 failures = ms->failures;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 bio_list_init(&ms->reads);
866 bio_list_init(&ms->writes);
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000867 bio_list_init(&ms->failures);
868 spin_unlock_irqrestore(&ms->lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100870 dm_rh_update_states(ms->rh, errors_handled(ms));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 do_recovery(ms);
872 do_reads(ms, &reads);
873 do_writes(ms, &writes);
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000874 do_failures(ms, &failures);
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000875}
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877/*-----------------------------------------------------------------
878 * Target functions
879 *---------------------------------------------------------------*/
880static struct mirror_set *alloc_context(unsigned int nr_mirrors,
881 uint32_t region_size,
882 struct dm_target *ti,
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100883 struct dm_dirty_log *dl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884{
885 size_t len;
886 struct mirror_set *ms = NULL;
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
889
Yoann Padioleaudd00cc42007-07-19 01:49:03 -0700890 ms = kzalloc(len, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 if (!ms) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700892 ti->error = "Cannot allocate mirror context";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 return NULL;
894 }
895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 spin_lock_init(&ms->lock);
Mikulas Patocka5339fc22009-12-10 23:52:06 +0000897 bio_list_init(&ms->reads);
898 bio_list_init(&ms->writes);
899 bio_list_init(&ms->failures);
900 bio_list_init(&ms->holds);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 ms->ti = ti;
903 ms->nr_mirrors = nr_mirrors;
904 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
905 ms->in_sync = 0;
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000906 ms->log_failure = 0;
Mikulas Patocka929be8f2009-12-10 23:52:06 +0000907 ms->leg_failure = 0;
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +0000908 atomic_set(&ms->suspend, 0);
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000909 atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Mikulas Patockabda8efe2011-05-29 13:03:09 +0100911 ms->io_client = dm_io_client_create();
Milan Broz88be1632007-05-09 02:33:04 -0700912 if (IS_ERR(ms->io_client)) {
913 ti->error = "Error creating dm_io client";
914 kfree(ms);
915 return NULL;
916 }
917
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100918 ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord,
919 wakeup_all_recovery_waiters,
920 ms->ti->begin, MAX_RECOVERY,
921 dl, region_size, ms->nr_regions);
922 if (IS_ERR(ms->rh)) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700923 ti->error = "Error creating dirty region hash";
Dmitry Monakhova72cf732007-10-19 22:38:39 +0100924 dm_io_client_destroy(ms->io_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 kfree(ms);
926 return NULL;
927 }
928
929 return ms;
930}
931
932static void free_context(struct mirror_set *ms, struct dm_target *ti,
933 unsigned int m)
934{
935 while (m--)
936 dm_put_device(ti, ms->mirror[m].dev);
937
Milan Broz88be1632007-05-09 02:33:04 -0700938 dm_io_client_destroy(ms->io_client);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100939 dm_region_hash_destroy(ms->rh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 kfree(ms);
941}
942
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
944 unsigned int mirror, char **argv)
945{
Andrew Morton4ee218c2006-03-27 01:17:48 -0800946 unsigned long long offset;
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100947 char dummy;
Vivek Goyale80d1c82015-07-31 09:20:36 -0400948 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100950 if (sscanf(argv[1], "%llu%c", &offset, &dummy) != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700951 ti->error = "Invalid offset";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 return -EINVAL;
953 }
954
Vivek Goyale80d1c82015-07-31 09:20:36 -0400955 ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
956 &ms->mirror[mirror].dev);
957 if (ret) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700958 ti->error = "Device lookup failure";
Vivek Goyale80d1c82015-07-31 09:20:36 -0400959 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
961
Jonathan Brassowaa5617c2007-10-19 22:47:58 +0100962 ms->mirror[mirror].ms = ms;
Jonathan Brassow72f4b312008-02-08 02:11:29 +0000963 atomic_set(&(ms->mirror[mirror].error_count), 0);
964 ms->mirror[mirror].error_type = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 ms->mirror[mirror].offset = offset;
966
967 return 0;
968}
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970/*
971 * Create dirty log: log_type #log_params <log_params>
972 */
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100973static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100974 unsigned argc, char **argv,
975 unsigned *args_used)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100977 unsigned param_count;
Heinz Mauelshagen416cd172008-04-24 21:43:35 +0100978 struct dm_dirty_log *dl;
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100979 char dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 if (argc < 2) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700982 ti->error = "Insufficient mirror log arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 return NULL;
984 }
985
Mikulas Patocka31998ef2012-03-28 18:41:26 +0100986 if (sscanf(argv[1], "%u%c", &param_count, &dummy) != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700987 ti->error = "Invalid mirror log argument count";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 return NULL;
989 }
990
991 *args_used = 2 + param_count;
992
993 if (argc < *args_used) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700994 ti->error = "Insufficient mirror log arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 return NULL;
996 }
997
Mikulas Patockac0da3742009-12-10 23:52:02 +0000998 dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count,
999 argv + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (!dl) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001001 ti->error = "Error creating mirror dirty log";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 return NULL;
1003 }
1004
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 return dl;
1006}
1007
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001008static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
1009 unsigned *args_used)
1010{
1011 unsigned num_features;
1012 struct dm_target *ti = ms->ti;
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001013 char dummy;
Lidong Zhonged632872015-05-13 14:04:10 +08001014 int i;
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001015
1016 *args_used = 0;
1017
1018 if (!argc)
1019 return 0;
1020
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001021 if (sscanf(argv[0], "%u%c", &num_features, &dummy) != 1) {
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001022 ti->error = "Invalid number of features";
1023 return -EINVAL;
1024 }
1025
1026 argc--;
1027 argv++;
1028 (*args_used)++;
1029
1030 if (num_features > argc) {
1031 ti->error = "Not enough arguments to support feature count";
1032 return -EINVAL;
1033 }
1034
Lidong Zhonged632872015-05-13 14:04:10 +08001035 for (i = 0; i < num_features; i++) {
1036 if (!strcmp("handle_errors", argv[0]))
1037 ms->features |= DM_RAID1_HANDLE_ERRORS;
1038 else if (!strcmp("keep_log", argv[0]))
1039 ms->features |= DM_RAID1_KEEP_LOG;
1040 else {
1041 ti->error = "Unrecognised feature requested";
1042 return -EINVAL;
1043 }
1044
1045 argc--;
1046 argv++;
1047 (*args_used)++;
1048 }
1049 if (!errors_handled(ms) && keep_log(ms)) {
1050 ti->error = "keep_log feature requires the handle_errors feature";
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001051 return -EINVAL;
1052 }
1053
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001054 return 0;
1055}
1056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057/*
1058 * Construct a mirror mapping:
1059 *
1060 * log_type #log_params <log_params>
1061 * #mirrors [mirror_path offset]{2,}
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001062 * [#features <features>]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 *
1064 * log_type is "core" or "disk"
1065 * #log_params is between 1 and 3
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001066 *
Lidong Zhonged632872015-05-13 14:04:10 +08001067 * If present, supported features are "handle_errors" and "keep_log".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1070{
1071 int r;
1072 unsigned int nr_mirrors, m, args_used;
1073 struct mirror_set *ms;
Heinz Mauelshagen416cd172008-04-24 21:43:35 +01001074 struct dm_dirty_log *dl;
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001075 char dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077 dl = create_dirty_log(ti, argc, argv, &args_used);
1078 if (!dl)
1079 return -EINVAL;
1080
1081 argv += args_used;
1082 argc -= args_used;
1083
Mikulas Patocka31998ef2012-03-28 18:41:26 +01001084 if (!argc || sscanf(argv[0], "%u%c", &nr_mirrors, &dummy) != 1 ||
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001085 nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001086 ti->error = "Invalid number of mirrors";
Heinz Mauelshagen416cd172008-04-24 21:43:35 +01001087 dm_dirty_log_destroy(dl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 return -EINVAL;
1089 }
1090
1091 argv++, argc--;
1092
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001093 if (argc < nr_mirrors * 2) {
1094 ti->error = "Too few mirror arguments";
Heinz Mauelshagen416cd172008-04-24 21:43:35 +01001095 dm_dirty_log_destroy(dl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 return -EINVAL;
1097 }
1098
1099 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1100 if (!ms) {
Heinz Mauelshagen416cd172008-04-24 21:43:35 +01001101 dm_dirty_log_destroy(dl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return -ENOMEM;
1103 }
1104
1105 /* Get the mirror parameter sets */
1106 for (m = 0; m < nr_mirrors; m++) {
1107 r = get_mirror(ms, ti, m, argv);
1108 if (r) {
1109 free_context(ms, ti, m);
1110 return r;
1111 }
1112 argv += 2;
1113 argc -= 2;
1114 }
1115
1116 ti->private = ms;
Mike Snitzer542f9032012-07-27 15:08:00 +01001117
1118 r = dm_set_target_max_io_len(ti, dm_rh_get_region_size(ms->rh));
1119 if (r)
1120 goto err_free_context;
1121
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00001122 ti->num_flush_bios = 1;
1123 ti->num_discard_bios = 1;
Mike Snitzer30187e12016-01-31 13:28:26 -05001124 ti->per_io_data_size = sizeof(struct dm_raid1_bio_record);
Alasdair G Kergon0ac55482012-07-27 15:08:08 +01001125 ti->discard_zeroes_data_unsupported = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Tejun Heo670368a2013-07-30 08:40:21 -04001127 ms->kmirrord_wq = alloc_workqueue("kmirrord", WQ_MEM_RECLAIM, 0);
Holger Smolinski6ad36fe2007-05-09 02:32:50 -07001128 if (!ms->kmirrord_wq) {
1129 DMERR("couldn't start kmirrord");
Dmitry Monakhova72cf732007-10-19 22:38:39 +01001130 r = -ENOMEM;
1131 goto err_free_context;
Holger Smolinski6ad36fe2007-05-09 02:32:50 -07001132 }
1133 INIT_WORK(&ms->kmirrord_work, do_mirror);
Mikulas Patockaa2aebe02008-04-24 22:10:42 +01001134 init_timer(&ms->timer);
1135 ms->timer_pending = 0;
Jonathan Brassow72f4b312008-02-08 02:11:29 +00001136 INIT_WORK(&ms->trigger_event, trigger_event);
Holger Smolinski6ad36fe2007-05-09 02:32:50 -07001137
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001138 r = parse_features(ms, argc, argv, &args_used);
Dmitry Monakhova72cf732007-10-19 22:38:39 +01001139 if (r)
1140 goto err_destroy_wq;
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001141
1142 argv += args_used;
1143 argc -= args_used;
1144
Jonathan Brassowf44db672007-07-12 17:29:04 +01001145 /*
1146 * Any read-balancing addition depends on the
1147 * DM_RAID1_HANDLE_ERRORS flag being present.
1148 * This is because the decision to balance depends
1149 * on the sync state of a region. If the above
1150 * flag is not present, we ignore errors; and
1151 * the sync state may be inaccurate.
1152 */
1153
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001154 if (argc) {
1155 ti->error = "Too many mirror arguments";
Dmitry Monakhova72cf732007-10-19 22:38:39 +01001156 r = -EINVAL;
1157 goto err_destroy_wq;
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001158 }
1159
Mikulas Patockadf5d2e92013-03-01 22:45:49 +00001160 ms->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
Mikulas Patockafa34ce72011-05-29 13:03:13 +01001161 if (IS_ERR(ms->kcopyd_client)) {
1162 r = PTR_ERR(ms->kcopyd_client);
Dmitry Monakhova72cf732007-10-19 22:38:39 +01001163 goto err_destroy_wq;
Mikulas Patockafa34ce72011-05-29 13:03:13 +01001164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001166 wakeup_mirrord(ms);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 return 0;
Dmitry Monakhova72cf732007-10-19 22:38:39 +01001168
1169err_destroy_wq:
1170 destroy_workqueue(ms->kmirrord_wq);
1171err_free_context:
1172 free_context(ms, ti, ms->nr_mirrors);
1173 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
1176static void mirror_dtr(struct dm_target *ti)
1177{
1178 struct mirror_set *ms = (struct mirror_set *) ti->private;
1179
Mikulas Patockaa2aebe02008-04-24 22:10:42 +01001180 del_timer_sync(&ms->timer);
Holger Smolinski6ad36fe2007-05-09 02:32:50 -07001181 flush_workqueue(ms->kmirrord_wq);
Tejun Heo43829732012-08-20 14:51:24 -07001182 flush_work(&ms->trigger_event);
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +01001183 dm_kcopyd_client_destroy(ms->kcopyd_client);
Holger Smolinski6ad36fe2007-05-09 02:32:50 -07001184 destroy_workqueue(ms->kmirrord_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 free_context(ms, ti, ms->nr_mirrors);
1186}
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188/*
1189 * Mirror mapping function
1190 */
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001191static int mirror_map(struct dm_target *ti, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
1193 int r, rw = bio_rw(bio);
1194 struct mirror *m;
1195 struct mirror_set *ms = ti->private;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001196 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
Mikulas Patocka0045d612012-12-21 20:23:40 +00001197 struct dm_raid1_bio_record *bio_record =
1198 dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
1199
1200 bio_record->details.bi_bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 if (rw == WRITE) {
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001203 /* Save region for mirror_end_io() handler */
Mikulas Patocka0045d612012-12-21 20:23:40 +00001204 bio_record->write_region = dm_rh_bio_to_region(ms->rh, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 queue_bio(ms, bio, rw);
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001206 return DM_MAPIO_SUBMITTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 }
1208
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001209 r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 if (r < 0 && r != -EWOULDBLOCK)
1211 return r;
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 /*
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001214 * If region is not in-sync queue the bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 */
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001216 if (!r || (r == -EWOULDBLOCK)) {
1217 if (rw == READA)
1218 return -EWOULDBLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 queue_bio(ms, bio, rw);
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001221 return DM_MAPIO_SUBMITTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 }
1223
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001224 /*
1225 * The region is in-sync and we can perform reads directly.
1226 * Store enough information so we can retry if it fails.
1227 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07001228 m = choose_mirror(ms, bio->bi_iter.bi_sector);
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001229 if (unlikely(!m))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 return -EIO;
1231
Mikulas Patocka89c7cd82012-12-21 20:23:39 +00001232 dm_bio_record(&bio_record->details, bio);
Mikulas Patocka89c7cd82012-12-21 20:23:39 +00001233 bio_record->m = m;
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001234
1235 map_bio(m, bio);
1236
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001237 return DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238}
1239
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00001240static int mirror_end_io(struct dm_target *ti, struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241{
1242 int rw = bio_rw(bio);
1243 struct mirror_set *ms = (struct mirror_set *) ti->private;
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001244 struct mirror *m = NULL;
1245 struct dm_bio_details *bd = NULL;
Mikulas Patocka0045d612012-12-21 20:23:40 +00001246 struct dm_raid1_bio_record *bio_record =
1247 dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249 /*
1250 * We need to dec pending if this was a write.
1251 */
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001252 if (rw == WRITE) {
Mikulas Patocka751f1882012-07-20 14:25:03 +01001253 if (!(bio->bi_rw & (REQ_FLUSH | REQ_DISCARD)))
Mikulas Patocka0045d612012-12-21 20:23:40 +00001254 dm_rh_dec(ms->rh, bio_record->write_region);
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001255 return error;
1256 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001258 if (error == -EOPNOTSUPP)
1259 goto out;
1260
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001261 if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD))
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001262 goto out;
1263
1264 if (unlikely(error)) {
Mikulas Patocka0045d612012-12-21 20:23:40 +00001265 if (!bio_record->details.bi_bdev) {
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001266 /*
1267 * There wasn't enough memory to record necessary
1268 * information for a retry or there was no other
1269 * mirror in-sync.
1270 */
Adrian Bunke03f1a82008-02-19 19:44:19 +00001271 DMERR_LIMIT("Mirror read failed.");
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001272 return -EIO;
1273 }
Adrian Bunke03f1a82008-02-19 19:44:19 +00001274
Mikulas Patocka89c7cd82012-12-21 20:23:39 +00001275 m = bio_record->m;
Adrian Bunke03f1a82008-02-19 19:44:19 +00001276
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001277 DMERR("Mirror read failed from %s. Trying alternative device.",
1278 m->dev->name);
1279
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001280 fail_mirror(m, DM_RAID1_READ_ERROR);
1281
1282 /*
1283 * A failed read is requeued for another attempt using an intact
1284 * mirror.
1285 */
1286 if (default_ok(m) || mirror_available(ms, bio)) {
Mikulas Patocka89c7cd82012-12-21 20:23:39 +00001287 bd = &bio_record->details;
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001288
1289 dm_bio_restore(bd, bio);
Mikulas Patocka0045d612012-12-21 20:23:40 +00001290 bio_record->details.bi_bdev = NULL;
Mikulas Patockaf3a44fe2014-02-18 09:57:22 -05001291
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001292 queue_bio(ms, bio, rw);
Mikulas Patocka19cbbc62012-12-21 20:23:32 +00001293 return DM_ENDIO_INCOMPLETE;
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001294 }
1295 DMERR("All replicated volumes dead, failing I/O");
1296 }
1297
1298out:
Mikulas Patocka0045d612012-12-21 20:23:40 +00001299 bio_record->details.bi_bdev = NULL;
Jonathan Brassow06386bb2008-02-08 02:11:37 +00001300
1301 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001304static void mirror_presuspend(struct dm_target *ti)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305{
1306 struct mirror_set *ms = (struct mirror_set *) ti->private;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001307 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Mikulas Patocka04788502009-12-10 23:52:03 +00001309 struct bio_list holds;
1310 struct bio *bio;
1311
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001312 atomic_set(&ms->suspend, 1);
1313
1314 /*
Takahiro Yasuif0703042010-03-06 02:32:35 +00001315 * Process bios in the hold list to start recovery waiting
1316 * for bios in the hold list. After the process, no bio has
1317 * a chance to be added in the hold list because ms->suspend
1318 * is set.
1319 */
1320 spin_lock_irq(&ms->lock);
1321 holds = ms->holds;
1322 bio_list_init(&ms->holds);
1323 spin_unlock_irq(&ms->lock);
1324
1325 while ((bio = bio_list_pop(&holds)))
1326 hold_bio(ms, bio);
1327
1328 /*
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001329 * We must finish up all the work that we've
1330 * generated (i.e. recovery work).
1331 */
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001332 dm_rh_stop_recovery(ms->rh);
Jonathan E Brassow33184042006-11-08 17:44:44 -08001333
Jonathan E Brassow33184042006-11-08 17:44:44 -08001334 wait_event(_kmirrord_recovery_stopped,
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001335 !dm_rh_recovery_in_flight(ms->rh));
Jonathan E Brassow33184042006-11-08 17:44:44 -08001336
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001337 if (log->type->presuspend && log->type->presuspend(log))
1338 /* FIXME: need better error handling */
1339 DMWARN("log presuspend failed");
1340
1341 /*
1342 * Now that recovery is complete/stopped and the
1343 * delayed bios are queued, we need to wait for
1344 * the worker thread to complete. This way,
1345 * we know that all of our I/O has been pushed.
1346 */
1347 flush_workqueue(ms->kmirrord_wq);
1348}
1349
1350static void mirror_postsuspend(struct dm_target *ti)
1351{
1352 struct mirror_set *ms = ti->private;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001353 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001354
Jonathan Brassow6b3df0d2007-10-19 22:47:57 +01001355 if (log->type->postsuspend && log->type->postsuspend(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 /* FIXME: need better error handling */
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001357 DMWARN("log postsuspend failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358}
1359
1360static void mirror_resume(struct dm_target *ti)
1361{
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001362 struct mirror_set *ms = ti->private;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001363 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001364
1365 atomic_set(&ms->suspend, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (log->type->resume && log->type->resume(log))
1367 /* FIXME: need better error handling */
1368 DMWARN("log resume failed");
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001369 dm_rh_start_recovery(ms->rh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370}
1371
Jonathan Brassowaf195ac2008-02-08 02:11:39 +00001372/*
1373 * device_status_char
1374 * @m: mirror device/leg we want the status of
1375 *
1376 * We return one character representing the most severe error
1377 * we have encountered.
1378 * A => Alive - No failures
1379 * D => Dead - A write failure occurred leaving mirror out-of-sync
1380 * S => Sync - A sychronization failure occurred, mirror out-of-sync
1381 * R => Read - A read failure occurred, mirror data unaffected
1382 *
1383 * Returns: <char>
1384 */
1385static char device_status_char(struct mirror *m)
1386{
1387 if (!atomic_read(&(m->error_count)))
1388 return 'A';
1389
Mikulas Patocka64b30c42009-12-10 23:52:02 +00001390 return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' :
1391 (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
Jonathan Brassowaf195ac2008-02-08 02:11:39 +00001392 (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
1393 (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
1394}
1395
1396
Mikulas Patockafd7c0922013-03-01 22:45:44 +00001397static void mirror_status(struct dm_target *ti, status_type_t type,
1398 unsigned status_flags, char *result, unsigned maxlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
Jonathan E Brassow315dcc22007-05-09 02:32:58 -07001400 unsigned int m, sz = 0;
Lidong Zhonged632872015-05-13 14:04:10 +08001401 int num_feature_args = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 struct mirror_set *ms = (struct mirror_set *) ti->private;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001403 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
Jonathan Brassowaf195ac2008-02-08 02:11:39 +00001404 char buffer[ms->nr_mirrors + 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 switch (type) {
1407 case STATUSTYPE_INFO:
1408 DMEMIT("%d ", ms->nr_mirrors);
Jonathan Brassowaf195ac2008-02-08 02:11:39 +00001409 for (m = 0; m < ms->nr_mirrors; m++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 DMEMIT("%s ", ms->mirror[m].dev->name);
Jonathan Brassowaf195ac2008-02-08 02:11:39 +00001411 buffer[m] = device_status_char(&(ms->mirror[m]));
1412 }
1413 buffer[m] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Jonathan Brassowaf195ac2008-02-08 02:11:39 +00001415 DMEMIT("%llu/%llu 1 %s ",
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001416 (unsigned long long)log->type->get_sync_count(log),
Jonathan Brassowaf195ac2008-02-08 02:11:39 +00001417 (unsigned long long)ms->nr_regions, buffer);
Jonathan E Brassow315dcc22007-05-09 02:32:58 -07001418
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001419 sz += log->type->status(log, type, result+sz, maxlen-sz);
Jonathan E Brassow315dcc22007-05-09 02:32:58 -07001420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 break;
1422
1423 case STATUSTYPE_TABLE:
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001424 sz = log->type->status(log, type, result, maxlen);
Jonathan E Brassow315dcc22007-05-09 02:32:58 -07001425
Jonathan Brassowe52b8f62006-10-03 01:15:32 -07001426 DMEMIT("%d", ms->nr_mirrors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 for (m = 0; m < ms->nr_mirrors; m++)
Jonathan Brassowe52b8f62006-10-03 01:15:32 -07001428 DMEMIT(" %s %llu", ms->mirror[m].dev->name,
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001429 (unsigned long long)ms->mirror[m].offset);
Jonathan E Brassowa8e6afa2007-05-09 02:32:59 -07001430
Lidong Zhonged632872015-05-13 14:04:10 +08001431 num_feature_args += !!errors_handled(ms);
1432 num_feature_args += !!keep_log(ms);
1433 if (num_feature_args) {
1434 DMEMIT(" %d", num_feature_args);
1435 if (errors_handled(ms))
1436 DMEMIT(" handle_errors");
1437 if (keep_log(ms))
1438 DMEMIT(" keep_log");
1439 }
1440
1441 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001445static int mirror_iterate_devices(struct dm_target *ti,
1446 iterate_devices_callout_fn fn, void *data)
1447{
1448 struct mirror_set *ms = ti->private;
1449 int ret = 0;
1450 unsigned i;
1451
1452 for (i = 0; !ret && i < ms->nr_mirrors; i++)
1453 ret = fn(ti, ms->mirror[i].dev,
Mike Snitzer5dea2712009-07-23 20:30:42 +01001454 ms->mirror[i].offset, ti->len, data);
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001455
1456 return ret;
1457}
1458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459static struct target_type mirror_target = {
1460 .name = "mirror",
Lidong Zhonged632872015-05-13 14:04:10 +08001461 .version = {1, 14, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 .module = THIS_MODULE,
1463 .ctr = mirror_ctr,
1464 .dtr = mirror_dtr,
1465 .map = mirror_map,
1466 .end_io = mirror_end_io,
Jonathan Brassowb80aa7a2008-02-08 02:11:35 +00001467 .presuspend = mirror_presuspend,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 .postsuspend = mirror_postsuspend,
1469 .resume = mirror_resume,
1470 .status = mirror_status,
Mike Snitzeraf4874e2009-06-22 10:12:33 +01001471 .iterate_devices = mirror_iterate_devices,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472};
1473
1474static int __init dm_mirror_init(void)
1475{
1476 int r;
1477
Mikulas Patocka95f8fac2009-04-02 19:55:24 +01001478 r = dm_register_target(&mirror_target);
1479 if (r < 0) {
1480 DMERR("Failed to register mirror target");
1481 goto bad_target;
1482 }
1483
1484 return 0;
1485
1486bad_target:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 return r;
1488}
1489
1490static void __exit dm_mirror_exit(void)
1491{
Mikulas Patocka10d3bd02009-01-06 03:04:58 +00001492 dm_unregister_target(&mirror_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
1494
1495/* Module hooks */
1496module_init(dm_mirror_init);
1497module_exit(dm_mirror_exit);
1498
1499MODULE_DESCRIPTION(DM_NAME " mirror target");
1500MODULE_AUTHOR("Joe Thornber");
1501MODULE_LICENSE("GPL");