Merge commit 'v3.5-rc2' into next
[platform/adaptation/renesas_rcar/renesas_kernel.git] / drivers / md / dm-raid.c
1 /*
2  * Copyright (C) 2010-2011 Neil Brown
3  * Copyright (C) 2010-2011 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7
8 #include <linux/slab.h>
9 #include <linux/module.h>
10
11 #include "md.h"
12 #include "raid1.h"
13 #include "raid5.h"
14 #include "bitmap.h"
15
16 #include <linux/device-mapper.h>
17
18 #define DM_MSG_PREFIX "raid"
19
20 /*
21  * The following flags are used by dm-raid.c to set up the array state.
22  * They must be cleared before md_run is called.
23  */
24 #define FirstUse 10             /* rdev flag */
25
26 struct raid_dev {
27         /*
28          * Two DM devices, one to hold metadata and one to hold the
29          * actual data/parity.  The reason for this is to not confuse
30          * ti->len and give more flexibility in altering size and
31          * characteristics.
32          *
33          * While it is possible for this device to be associated
34          * with a different physical device than the data_dev, it
35          * is intended for it to be the same.
36          *    |--------- Physical Device ---------|
37          *    |- meta_dev -|------ data_dev ------|
38          */
39         struct dm_dev *meta_dev;
40         struct dm_dev *data_dev;
41         struct md_rdev rdev;
42 };
43
44 /*
45  * Flags for rs->print_flags field.
46  */
47 #define DMPF_SYNC              0x1
48 #define DMPF_NOSYNC            0x2
49 #define DMPF_REBUILD           0x4
50 #define DMPF_DAEMON_SLEEP      0x8
51 #define DMPF_MIN_RECOVERY_RATE 0x10
52 #define DMPF_MAX_RECOVERY_RATE 0x20
53 #define DMPF_MAX_WRITE_BEHIND  0x40
54 #define DMPF_STRIPE_CACHE      0x80
55 #define DMPF_REGION_SIZE       0X100
56 struct raid_set {
57         struct dm_target *ti;
58
59         uint32_t bitmap_loaded;
60         uint32_t print_flags;
61
62         struct mddev md;
63         struct raid_type *raid_type;
64         struct dm_target_callbacks callbacks;
65
66         struct raid_dev dev[0];
67 };
68
69 /* Supported raid types and properties. */
70 static struct raid_type {
71         const char *name;               /* RAID algorithm. */
72         const char *descr;              /* Descriptor text for logging. */
73         const unsigned parity_devs;     /* # of parity devices. */
74         const unsigned minimal_devs;    /* minimal # of devices in set. */
75         const unsigned level;           /* RAID level. */
76         const unsigned algorithm;       /* RAID algorithm. */
77 } raid_types[] = {
78         {"raid1",    "RAID1 (mirroring)",               0, 2, 1, 0 /* NONE */},
79         {"raid4",    "RAID4 (dedicated parity disk)",   1, 2, 5, ALGORITHM_PARITY_0},
80         {"raid5_la", "RAID5 (left asymmetric)",         1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
81         {"raid5_ra", "RAID5 (right asymmetric)",        1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
82         {"raid5_ls", "RAID5 (left symmetric)",          1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
83         {"raid5_rs", "RAID5 (right symmetric)",         1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
84         {"raid6_zr", "RAID6 (zero restart)",            2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
85         {"raid6_nr", "RAID6 (N restart)",               2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
86         {"raid6_nc", "RAID6 (N continue)",              2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
87 };
88
89 static struct raid_type *get_raid_type(char *name)
90 {
91         int i;
92
93         for (i = 0; i < ARRAY_SIZE(raid_types); i++)
94                 if (!strcmp(raid_types[i].name, name))
95                         return &raid_types[i];
96
97         return NULL;
98 }
99
100 static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
101 {
102         unsigned i;
103         struct raid_set *rs;
104         sector_t sectors_per_dev;
105
106         if (raid_devs <= raid_type->parity_devs) {
107                 ti->error = "Insufficient number of devices";
108                 return ERR_PTR(-EINVAL);
109         }
110
111         sectors_per_dev = ti->len;
112         if ((raid_type->level > 1) &&
113             sector_div(sectors_per_dev, (raid_devs - raid_type->parity_devs))) {
114                 ti->error = "Target length not divisible by number of data devices";
115                 return ERR_PTR(-EINVAL);
116         }
117
118         rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
119         if (!rs) {
120                 ti->error = "Cannot allocate raid context";
121                 return ERR_PTR(-ENOMEM);
122         }
123
124         mddev_init(&rs->md);
125
126         rs->ti = ti;
127         rs->raid_type = raid_type;
128         rs->md.raid_disks = raid_devs;
129         rs->md.level = raid_type->level;
130         rs->md.new_level = rs->md.level;
131         rs->md.dev_sectors = sectors_per_dev;
132         rs->md.layout = raid_type->algorithm;
133         rs->md.new_layout = rs->md.layout;
134         rs->md.delta_disks = 0;
135         rs->md.recovery_cp = 0;
136
137         for (i = 0; i < raid_devs; i++)
138                 md_rdev_init(&rs->dev[i].rdev);
139
140         /*
141          * Remaining items to be initialized by further RAID params:
142          *  rs->md.persistent
143          *  rs->md.external
144          *  rs->md.chunk_sectors
145          *  rs->md.new_chunk_sectors
146          */
147
148         return rs;
149 }
150
151 static void context_free(struct raid_set *rs)
152 {
153         int i;
154
155         for (i = 0; i < rs->md.raid_disks; i++) {
156                 if (rs->dev[i].meta_dev)
157                         dm_put_device(rs->ti, rs->dev[i].meta_dev);
158                 md_rdev_clear(&rs->dev[i].rdev);
159                 if (rs->dev[i].data_dev)
160                         dm_put_device(rs->ti, rs->dev[i].data_dev);
161         }
162
163         kfree(rs);
164 }
165
166 /*
167  * For every device we have two words
168  *  <meta_dev>: meta device name or '-' if missing
169  *  <data_dev>: data device name or '-' if missing
170  *
171  * The following are permitted:
172  *    - -
173  *    - <data_dev>
174  *    <meta_dev> <data_dev>
175  *
176  * The following is not allowed:
177  *    <meta_dev> -
178  *
179  * This code parses those words.  If there is a failure,
180  * the caller must use context_free to unwind the operations.
181  */
182 static int dev_parms(struct raid_set *rs, char **argv)
183 {
184         int i;
185         int rebuild = 0;
186         int metadata_available = 0;
187         int ret = 0;
188
189         for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
190                 rs->dev[i].rdev.raid_disk = i;
191
192                 rs->dev[i].meta_dev = NULL;
193                 rs->dev[i].data_dev = NULL;
194
195                 /*
196                  * There are no offsets, since there is a separate device
197                  * for data and metadata.
198                  */
199                 rs->dev[i].rdev.data_offset = 0;
200                 rs->dev[i].rdev.mddev = &rs->md;
201
202                 if (strcmp(argv[0], "-")) {
203                         ret = dm_get_device(rs->ti, argv[0],
204                                             dm_table_get_mode(rs->ti->table),
205                                             &rs->dev[i].meta_dev);
206                         rs->ti->error = "RAID metadata device lookup failure";
207                         if (ret)
208                                 return ret;
209
210                         rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
211                         if (!rs->dev[i].rdev.sb_page)
212                                 return -ENOMEM;
213                 }
214
215                 if (!strcmp(argv[1], "-")) {
216                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
217                             (!rs->dev[i].rdev.recovery_offset)) {
218                                 rs->ti->error = "Drive designated for rebuild not specified";
219                                 return -EINVAL;
220                         }
221
222                         rs->ti->error = "No data device supplied with metadata device";
223                         if (rs->dev[i].meta_dev)
224                                 return -EINVAL;
225
226                         continue;
227                 }
228
229                 ret = dm_get_device(rs->ti, argv[1],
230                                     dm_table_get_mode(rs->ti->table),
231                                     &rs->dev[i].data_dev);
232                 if (ret) {
233                         rs->ti->error = "RAID device lookup failure";
234                         return ret;
235                 }
236
237                 if (rs->dev[i].meta_dev) {
238                         metadata_available = 1;
239                         rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
240                 }
241                 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
242                 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
243                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
244                         rebuild++;
245         }
246
247         if (metadata_available) {
248                 rs->md.external = 0;
249                 rs->md.persistent = 1;
250                 rs->md.major_version = 2;
251         } else if (rebuild && !rs->md.recovery_cp) {
252                 /*
253                  * Without metadata, we will not be able to tell if the array
254                  * is in-sync or not - we must assume it is not.  Therefore,
255                  * it is impossible to rebuild a drive.
256                  *
257                  * Even if there is metadata, the on-disk information may
258                  * indicate that the array is not in-sync and it will then
259                  * fail at that time.
260                  *
261                  * User could specify 'nosync' option if desperate.
262                  */
263                 DMERR("Unable to rebuild drive while array is not in-sync");
264                 rs->ti->error = "RAID device lookup failure";
265                 return -EINVAL;
266         }
267
268         return 0;
269 }
270
271 /*
272  * validate_region_size
273  * @rs
274  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
275  *
276  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
277  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
278  *
279  * Returns: 0 on success, -EINVAL on failure.
280  */
281 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
282 {
283         unsigned long min_region_size = rs->ti->len / (1 << 21);
284
285         if (!region_size) {
286                 /*
287                  * Choose a reasonable default.  All figures in sectors.
288                  */
289                 if (min_region_size > (1 << 13)) {
290                         DMINFO("Choosing default region size of %lu sectors",
291                                region_size);
292                         region_size = min_region_size;
293                 } else {
294                         DMINFO("Choosing default region size of 4MiB");
295                         region_size = 1 << 13; /* sectors */
296                 }
297         } else {
298                 /*
299                  * Validate user-supplied value.
300                  */
301                 if (region_size > rs->ti->len) {
302                         rs->ti->error = "Supplied region size is too large";
303                         return -EINVAL;
304                 }
305
306                 if (region_size < min_region_size) {
307                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
308                               region_size, min_region_size);
309                         rs->ti->error = "Supplied region size is too small";
310                         return -EINVAL;
311                 }
312
313                 if (!is_power_of_2(region_size)) {
314                         rs->ti->error = "Region size is not a power of 2";
315                         return -EINVAL;
316                 }
317
318                 if (region_size < rs->md.chunk_sectors) {
319                         rs->ti->error = "Region size is smaller than the chunk size";
320                         return -EINVAL;
321                 }
322         }
323
324         /*
325          * Convert sectors to bytes.
326          */
327         rs->md.bitmap_info.chunksize = (region_size << 9);
328
329         return 0;
330 }
331
332 /*
333  * Possible arguments are...
334  *      <chunk_size> [optional_args]
335  *
336  * Argument definitions
337  *    <chunk_size>                      The number of sectors per disk that
338  *                                      will form the "stripe"
339  *    [[no]sync]                        Force or prevent recovery of the
340  *                                      entire array
341  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
342  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
343  *                                      clear bits
344  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
345  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
346  *    [write_mostly <idx>]              Indicate a write mostly drive via index
347  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
348  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
349  *    [region_size <sectors>]           Defines granularity of bitmap
350  */
351 static int parse_raid_params(struct raid_set *rs, char **argv,
352                              unsigned num_raid_params)
353 {
354         unsigned i, rebuild_cnt = 0;
355         unsigned long value, region_size = 0;
356         char *key;
357
358         /*
359          * First, parse the in-order required arguments
360          * "chunk_size" is the only argument of this type.
361          */
362         if ((strict_strtoul(argv[0], 10, &value) < 0)) {
363                 rs->ti->error = "Bad chunk size";
364                 return -EINVAL;
365         } else if (rs->raid_type->level == 1) {
366                 if (value)
367                         DMERR("Ignoring chunk size parameter for RAID 1");
368                 value = 0;
369         } else if (!is_power_of_2(value)) {
370                 rs->ti->error = "Chunk size must be a power of 2";
371                 return -EINVAL;
372         } else if (value < 8) {
373                 rs->ti->error = "Chunk size value is too small";
374                 return -EINVAL;
375         }
376
377         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
378         argv++;
379         num_raid_params--;
380
381         /*
382          * We set each individual device as In_sync with a completed
383          * 'recovery_offset'.  If there has been a device failure or
384          * replacement then one of the following cases applies:
385          *
386          *   1) User specifies 'rebuild'.
387          *      - Device is reset when param is read.
388          *   2) A new device is supplied.
389          *      - No matching superblock found, resets device.
390          *   3) Device failure was transient and returns on reload.
391          *      - Failure noticed, resets device for bitmap replay.
392          *   4) Device hadn't completed recovery after previous failure.
393          *      - Superblock is read and overrides recovery_offset.
394          *
395          * What is found in the superblocks of the devices is always
396          * authoritative, unless 'rebuild' or '[no]sync' was specified.
397          */
398         for (i = 0; i < rs->md.raid_disks; i++) {
399                 set_bit(In_sync, &rs->dev[i].rdev.flags);
400                 rs->dev[i].rdev.recovery_offset = MaxSector;
401         }
402
403         /*
404          * Second, parse the unordered optional arguments
405          */
406         for (i = 0; i < num_raid_params; i++) {
407                 if (!strcasecmp(argv[i], "nosync")) {
408                         rs->md.recovery_cp = MaxSector;
409                         rs->print_flags |= DMPF_NOSYNC;
410                         continue;
411                 }
412                 if (!strcasecmp(argv[i], "sync")) {
413                         rs->md.recovery_cp = 0;
414                         rs->print_flags |= DMPF_SYNC;
415                         continue;
416                 }
417
418                 /* The rest of the optional arguments come in key/value pairs */
419                 if ((i + 1) >= num_raid_params) {
420                         rs->ti->error = "Wrong number of raid parameters given";
421                         return -EINVAL;
422                 }
423
424                 key = argv[i++];
425                 if (strict_strtoul(argv[i], 10, &value) < 0) {
426                         rs->ti->error = "Bad numerical argument given in raid params";
427                         return -EINVAL;
428                 }
429
430                 if (!strcasecmp(key, "rebuild")) {
431                         rebuild_cnt++;
432                         if (((rs->raid_type->level != 1) &&
433                              (rebuild_cnt > rs->raid_type->parity_devs)) ||
434                             ((rs->raid_type->level == 1) &&
435                              (rebuild_cnt > (rs->md.raid_disks - 1)))) {
436                                 rs->ti->error = "Too many rebuild devices specified for given RAID type";
437                                 return -EINVAL;
438                         }
439                         if (value > rs->md.raid_disks) {
440                                 rs->ti->error = "Invalid rebuild index given";
441                                 return -EINVAL;
442                         }
443                         clear_bit(In_sync, &rs->dev[value].rdev.flags);
444                         rs->dev[value].rdev.recovery_offset = 0;
445                         rs->print_flags |= DMPF_REBUILD;
446                 } else if (!strcasecmp(key, "write_mostly")) {
447                         if (rs->raid_type->level != 1) {
448                                 rs->ti->error = "write_mostly option is only valid for RAID1";
449                                 return -EINVAL;
450                         }
451                         if (value >= rs->md.raid_disks) {
452                                 rs->ti->error = "Invalid write_mostly drive index given";
453                                 return -EINVAL;
454                         }
455                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
456                 } else if (!strcasecmp(key, "max_write_behind")) {
457                         if (rs->raid_type->level != 1) {
458                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
459                                 return -EINVAL;
460                         }
461                         rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
462
463                         /*
464                          * In device-mapper, we specify things in sectors, but
465                          * MD records this value in kB
466                          */
467                         value /= 2;
468                         if (value > COUNTER_MAX) {
469                                 rs->ti->error = "Max write-behind limit out of range";
470                                 return -EINVAL;
471                         }
472                         rs->md.bitmap_info.max_write_behind = value;
473                 } else if (!strcasecmp(key, "daemon_sleep")) {
474                         rs->print_flags |= DMPF_DAEMON_SLEEP;
475                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
476                                 rs->ti->error = "daemon sleep period out of range";
477                                 return -EINVAL;
478                         }
479                         rs->md.bitmap_info.daemon_sleep = value;
480                 } else if (!strcasecmp(key, "stripe_cache")) {
481                         rs->print_flags |= DMPF_STRIPE_CACHE;
482
483                         /*
484                          * In device-mapper, we specify things in sectors, but
485                          * MD records this value in kB
486                          */
487                         value /= 2;
488
489                         if (rs->raid_type->level < 5) {
490                                 rs->ti->error = "Inappropriate argument: stripe_cache";
491                                 return -EINVAL;
492                         }
493                         if (raid5_set_cache_size(&rs->md, (int)value)) {
494                                 rs->ti->error = "Bad stripe_cache size";
495                                 return -EINVAL;
496                         }
497                 } else if (!strcasecmp(key, "min_recovery_rate")) {
498                         rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
499                         if (value > INT_MAX) {
500                                 rs->ti->error = "min_recovery_rate out of range";
501                                 return -EINVAL;
502                         }
503                         rs->md.sync_speed_min = (int)value;
504                 } else if (!strcasecmp(key, "max_recovery_rate")) {
505                         rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
506                         if (value > INT_MAX) {
507                                 rs->ti->error = "max_recovery_rate out of range";
508                                 return -EINVAL;
509                         }
510                         rs->md.sync_speed_max = (int)value;
511                 } else if (!strcasecmp(key, "region_size")) {
512                         rs->print_flags |= DMPF_REGION_SIZE;
513                         region_size = value;
514                 } else {
515                         DMERR("Unable to parse RAID parameter: %s", key);
516                         rs->ti->error = "Unable to parse RAID parameters";
517                         return -EINVAL;
518                 }
519         }
520
521         if (validate_region_size(rs, region_size))
522                 return -EINVAL;
523
524         if (rs->md.chunk_sectors)
525                 rs->ti->split_io = rs->md.chunk_sectors;
526         else
527                 rs->ti->split_io = region_size;
528
529         if (rs->md.chunk_sectors)
530                 rs->ti->split_io = rs->md.chunk_sectors;
531         else
532                 rs->ti->split_io = region_size;
533
534         /* Assume there are no metadata devices until the drives are parsed */
535         rs->md.persistent = 0;
536         rs->md.external = 1;
537
538         return 0;
539 }
540
541 static void do_table_event(struct work_struct *ws)
542 {
543         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
544
545         dm_table_event(rs->ti->table);
546 }
547
548 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
549 {
550         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
551
552         if (rs->raid_type->level == 1)
553                 return md_raid1_congested(&rs->md, bits);
554
555         return md_raid5_congested(&rs->md, bits);
556 }
557
558 /*
559  * This structure is never routinely used by userspace, unlike md superblocks.
560  * Devices with this superblock should only ever be accessed via device-mapper.
561  */
562 #define DM_RAID_MAGIC 0x64526D44
563 struct dm_raid_superblock {
564         __le32 magic;           /* "DmRd" */
565         __le32 features;        /* Used to indicate possible future changes */
566
567         __le32 num_devices;     /* Number of devices in this array. (Max 64) */
568         __le32 array_position;  /* The position of this drive in the array */
569
570         __le64 events;          /* Incremented by md when superblock updated */
571         __le64 failed_devices;  /* Bit field of devices to indicate failures */
572
573         /*
574          * This offset tracks the progress of the repair or replacement of
575          * an individual drive.
576          */
577         __le64 disk_recovery_offset;
578
579         /*
580          * This offset tracks the progress of the initial array
581          * synchronisation/parity calculation.
582          */
583         __le64 array_resync_offset;
584
585         /*
586          * RAID characteristics
587          */
588         __le32 level;
589         __le32 layout;
590         __le32 stripe_sectors;
591
592         __u8 pad[452];          /* Round struct to 512 bytes. */
593                                 /* Always set to 0 when writing. */
594 } __packed;
595
596 static int read_disk_sb(struct md_rdev *rdev, int size)
597 {
598         BUG_ON(!rdev->sb_page);
599
600         if (rdev->sb_loaded)
601                 return 0;
602
603         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
604                 DMERR("Failed to read superblock of device at position %d",
605                       rdev->raid_disk);
606                 md_error(rdev->mddev, rdev);
607                 return -EINVAL;
608         }
609
610         rdev->sb_loaded = 1;
611
612         return 0;
613 }
614
615 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
616 {
617         int i;
618         uint64_t failed_devices;
619         struct dm_raid_superblock *sb;
620         struct raid_set *rs = container_of(mddev, struct raid_set, md);
621
622         sb = page_address(rdev->sb_page);
623         failed_devices = le64_to_cpu(sb->failed_devices);
624
625         for (i = 0; i < mddev->raid_disks; i++)
626                 if (!rs->dev[i].data_dev ||
627                     test_bit(Faulty, &(rs->dev[i].rdev.flags)))
628                         failed_devices |= (1ULL << i);
629
630         memset(sb, 0, sizeof(*sb));
631
632         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
633         sb->features = cpu_to_le32(0);  /* No features yet */
634
635         sb->num_devices = cpu_to_le32(mddev->raid_disks);
636         sb->array_position = cpu_to_le32(rdev->raid_disk);
637
638         sb->events = cpu_to_le64(mddev->events);
639         sb->failed_devices = cpu_to_le64(failed_devices);
640
641         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
642         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
643
644         sb->level = cpu_to_le32(mddev->level);
645         sb->layout = cpu_to_le32(mddev->layout);
646         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
647 }
648
649 /*
650  * super_load
651  *
652  * This function creates a superblock if one is not found on the device
653  * and will decide which superblock to use if there's a choice.
654  *
655  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
656  */
657 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
658 {
659         int ret;
660         struct dm_raid_superblock *sb;
661         struct dm_raid_superblock *refsb;
662         uint64_t events_sb, events_refsb;
663
664         rdev->sb_start = 0;
665         rdev->sb_size = sizeof(*sb);
666
667         ret = read_disk_sb(rdev, rdev->sb_size);
668         if (ret)
669                 return ret;
670
671         sb = page_address(rdev->sb_page);
672
673         /*
674          * Two cases that we want to write new superblocks and rebuild:
675          * 1) New device (no matching magic number)
676          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
677          */
678         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
679             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
680                 super_sync(rdev->mddev, rdev);
681
682                 set_bit(FirstUse, &rdev->flags);
683
684                 /* Force writing of superblocks to disk */
685                 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
686
687                 /* Any superblock is better than none, choose that if given */
688                 return refdev ? 0 : 1;
689         }
690
691         if (!refdev)
692                 return 1;
693
694         events_sb = le64_to_cpu(sb->events);
695
696         refsb = page_address(refdev->sb_page);
697         events_refsb = le64_to_cpu(refsb->events);
698
699         return (events_sb > events_refsb) ? 1 : 0;
700 }
701
702 static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
703 {
704         int role;
705         struct raid_set *rs = container_of(mddev, struct raid_set, md);
706         uint64_t events_sb;
707         uint64_t failed_devices;
708         struct dm_raid_superblock *sb;
709         uint32_t new_devs = 0;
710         uint32_t rebuilds = 0;
711         struct md_rdev *r;
712         struct dm_raid_superblock *sb2;
713
714         sb = page_address(rdev->sb_page);
715         events_sb = le64_to_cpu(sb->events);
716         failed_devices = le64_to_cpu(sb->failed_devices);
717
718         /*
719          * Initialise to 1 if this is a new superblock.
720          */
721         mddev->events = events_sb ? : 1;
722
723         /*
724          * Reshaping is not currently allowed
725          */
726         if ((le32_to_cpu(sb->level) != mddev->level) ||
727             (le32_to_cpu(sb->layout) != mddev->layout) ||
728             (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
729                 DMERR("Reshaping arrays not yet supported.");
730                 return -EINVAL;
731         }
732
733         /* We can only change the number of devices in RAID1 right now */
734         if ((rs->raid_type->level != 1) &&
735             (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
736                 DMERR("Reshaping arrays not yet supported.");
737                 return -EINVAL;
738         }
739
740         if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
741                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
742
743         /*
744          * During load, we set FirstUse if a new superblock was written.
745          * There are two reasons we might not have a superblock:
746          * 1) The array is brand new - in which case, all of the
747          *    devices must have their In_sync bit set.  Also,
748          *    recovery_cp must be 0, unless forced.
749          * 2) This is a new device being added to an old array
750          *    and the new device needs to be rebuilt - in which
751          *    case the In_sync bit will /not/ be set and
752          *    recovery_cp must be MaxSector.
753          */
754         rdev_for_each(r, mddev) {
755                 if (!test_bit(In_sync, &r->flags)) {
756                         DMINFO("Device %d specified for rebuild: "
757                                "Clearing superblock", r->raid_disk);
758                         rebuilds++;
759                 } else if (test_bit(FirstUse, &r->flags))
760                         new_devs++;
761         }
762
763         if (!rebuilds) {
764                 if (new_devs == mddev->raid_disks) {
765                         DMINFO("Superblocks created for new array");
766                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
767                 } else if (new_devs) {
768                         DMERR("New device injected "
769                               "into existing array without 'rebuild' "
770                               "parameter specified");
771                         return -EINVAL;
772                 }
773         } else if (new_devs) {
774                 DMERR("'rebuild' devices cannot be "
775                       "injected into an array with other first-time devices");
776                 return -EINVAL;
777         } else if (mddev->recovery_cp != MaxSector) {
778                 DMERR("'rebuild' specified while array is not in-sync");
779                 return -EINVAL;
780         }
781
782         /*
783          * Now we set the Faulty bit for those devices that are
784          * recorded in the superblock as failed.
785          */
786         rdev_for_each(r, mddev) {
787                 if (!r->sb_page)
788                         continue;
789                 sb2 = page_address(r->sb_page);
790                 sb2->failed_devices = 0;
791
792                 /*
793                  * Check for any device re-ordering.
794                  */
795                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
796                         role = le32_to_cpu(sb2->array_position);
797                         if (role != r->raid_disk) {
798                                 if (rs->raid_type->level != 1) {
799                                         rs->ti->error = "Cannot change device "
800                                                 "positions in RAID array";
801                                         return -EINVAL;
802                                 }
803                                 DMINFO("RAID1 device #%d now at position #%d",
804                                        role, r->raid_disk);
805                         }
806
807                         /*
808                          * Partial recovery is performed on
809                          * returning failed devices.
810                          */
811                         if (failed_devices & (1 << role))
812                                 set_bit(Faulty, &r->flags);
813                 }
814         }
815
816         return 0;
817 }
818
819 static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
820 {
821         struct dm_raid_superblock *sb = page_address(rdev->sb_page);
822
823         /*
824          * If mddev->events is not set, we know we have not yet initialized
825          * the array.
826          */
827         if (!mddev->events && super_init_validation(mddev, rdev))
828                 return -EINVAL;
829
830         mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
831         rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
832         if (!test_bit(FirstUse, &rdev->flags)) {
833                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
834                 if (rdev->recovery_offset != MaxSector)
835                         clear_bit(In_sync, &rdev->flags);
836         }
837
838         /*
839          * If a device comes back, set it as not In_sync and no longer faulty.
840          */
841         if (test_bit(Faulty, &rdev->flags)) {
842                 clear_bit(Faulty, &rdev->flags);
843                 clear_bit(In_sync, &rdev->flags);
844                 rdev->saved_raid_disk = rdev->raid_disk;
845                 rdev->recovery_offset = 0;
846         }
847
848         clear_bit(FirstUse, &rdev->flags);
849
850         return 0;
851 }
852
853 /*
854  * Analyse superblocks and select the freshest.
855  */
856 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
857 {
858         int ret;
859         unsigned redundancy = 0;
860         struct raid_dev *dev;
861         struct md_rdev *rdev, *tmp, *freshest;
862         struct mddev *mddev = &rs->md;
863
864         switch (rs->raid_type->level) {
865         case 1:
866                 redundancy = rs->md.raid_disks - 1;
867                 break;
868         case 4:
869         case 5:
870         case 6:
871                 redundancy = rs->raid_type->parity_devs;
872                 break;
873         default:
874                 ti->error = "Unknown RAID type";
875                 return -EINVAL;
876         }
877
878         freshest = NULL;
879         rdev_for_each_safe(rdev, tmp, mddev) {
880                 if (!rdev->meta_bdev)
881                         continue;
882
883                 ret = super_load(rdev, freshest);
884
885                 switch (ret) {
886                 case 1:
887                         freshest = rdev;
888                         break;
889                 case 0:
890                         break;
891                 default:
892                         dev = container_of(rdev, struct raid_dev, rdev);
893                         if (redundancy--) {
894                                 if (dev->meta_dev)
895                                         dm_put_device(ti, dev->meta_dev);
896
897                                 dev->meta_dev = NULL;
898                                 rdev->meta_bdev = NULL;
899
900                                 if (rdev->sb_page)
901                                         put_page(rdev->sb_page);
902
903                                 rdev->sb_page = NULL;
904
905                                 rdev->sb_loaded = 0;
906
907                                 /*
908                                  * We might be able to salvage the data device
909                                  * even though the meta device has failed.  For
910                                  * now, we behave as though '- -' had been
911                                  * set for this device in the table.
912                                  */
913                                 if (dev->data_dev)
914                                         dm_put_device(ti, dev->data_dev);
915
916                                 dev->data_dev = NULL;
917                                 rdev->bdev = NULL;
918
919                                 list_del(&rdev->same_set);
920
921                                 continue;
922                         }
923                         ti->error = "Failed to load superblock";
924                         return ret;
925                 }
926         }
927
928         if (!freshest)
929                 return 0;
930
931         /*
932          * Validation of the freshest device provides the source of
933          * validation for the remaining devices.
934          */
935         ti->error = "Unable to assemble array: Invalid superblocks";
936         if (super_validate(mddev, freshest))
937                 return -EINVAL;
938
939         rdev_for_each(rdev, mddev)
940                 if ((rdev != freshest) && super_validate(mddev, rdev))
941                         return -EINVAL;
942
943         return 0;
944 }
945
946 /*
947  * Construct a RAID4/5/6 mapping:
948  * Args:
949  *      <raid_type> <#raid_params> <raid_params>                \
950  *      <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
951  *
952  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
953  * details on possible <raid_params>.
954  */
955 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
956 {
957         int ret;
958         struct raid_type *rt;
959         unsigned long num_raid_params, num_raid_devs;
960         struct raid_set *rs = NULL;
961
962         /* Must have at least <raid_type> <#raid_params> */
963         if (argc < 2) {
964                 ti->error = "Too few arguments";
965                 return -EINVAL;
966         }
967
968         /* raid type */
969         rt = get_raid_type(argv[0]);
970         if (!rt) {
971                 ti->error = "Unrecognised raid_type";
972                 return -EINVAL;
973         }
974         argc--;
975         argv++;
976
977         /* number of RAID parameters */
978         if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
979                 ti->error = "Cannot understand number of RAID parameters";
980                 return -EINVAL;
981         }
982         argc--;
983         argv++;
984
985         /* Skip over RAID params for now and find out # of devices */
986         if (num_raid_params + 1 > argc) {
987                 ti->error = "Arguments do not agree with counts given";
988                 return -EINVAL;
989         }
990
991         if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
992             (num_raid_devs >= INT_MAX)) {
993                 ti->error = "Cannot understand number of raid devices";
994                 return -EINVAL;
995         }
996
997         rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
998         if (IS_ERR(rs))
999                 return PTR_ERR(rs);
1000
1001         ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1002         if (ret)
1003                 goto bad;
1004
1005         ret = -EINVAL;
1006
1007         argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1008         argv += num_raid_params + 1;
1009
1010         if (argc != (num_raid_devs * 2)) {
1011                 ti->error = "Supplied RAID devices does not match the count given";
1012                 goto bad;
1013         }
1014
1015         ret = dev_parms(rs, argv);
1016         if (ret)
1017                 goto bad;
1018
1019         rs->md.sync_super = super_sync;
1020         ret = analyse_superblocks(ti, rs);
1021         if (ret)
1022                 goto bad;
1023
1024         INIT_WORK(&rs->md.event_work, do_table_event);
1025         ti->private = rs;
1026         ti->num_flush_requests = 1;
1027
1028         mutex_lock(&rs->md.reconfig_mutex);
1029         ret = md_run(&rs->md);
1030         rs->md.in_sync = 0; /* Assume already marked dirty */
1031         mutex_unlock(&rs->md.reconfig_mutex);
1032
1033         if (ret) {
1034                 ti->error = "Fail to run raid array";
1035                 goto bad;
1036         }
1037
1038         rs->callbacks.congested_fn = raid_is_congested;
1039         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1040
1041         mddev_suspend(&rs->md);
1042         return 0;
1043
1044 bad:
1045         context_free(rs);
1046
1047         return ret;
1048 }
1049
1050 static void raid_dtr(struct dm_target *ti)
1051 {
1052         struct raid_set *rs = ti->private;
1053
1054         list_del_init(&rs->callbacks.list);
1055         md_stop(&rs->md);
1056         context_free(rs);
1057 }
1058
1059 static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context)
1060 {
1061         struct raid_set *rs = ti->private;
1062         struct mddev *mddev = &rs->md;
1063
1064         mddev->pers->make_request(mddev, bio);
1065
1066         return DM_MAPIO_SUBMITTED;
1067 }
1068
1069 static int raid_status(struct dm_target *ti, status_type_t type,
1070                        char *result, unsigned maxlen)
1071 {
1072         struct raid_set *rs = ti->private;
1073         unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1074         unsigned sz = 0;
1075         int i, array_in_sync = 0;
1076         sector_t sync;
1077
1078         switch (type) {
1079         case STATUSTYPE_INFO:
1080                 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1081
1082                 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1083                         sync = rs->md.curr_resync_completed;
1084                 else
1085                         sync = rs->md.recovery_cp;
1086
1087                 if (sync >= rs->md.resync_max_sectors) {
1088                         array_in_sync = 1;
1089                         sync = rs->md.resync_max_sectors;
1090                 } else {
1091                         /*
1092                          * The array may be doing an initial sync, or it may
1093                          * be rebuilding individual components.  If all the
1094                          * devices are In_sync, then it is the array that is
1095                          * being initialized.
1096                          */
1097                         for (i = 0; i < rs->md.raid_disks; i++)
1098                                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1099                                         array_in_sync = 1;
1100                 }
1101                 /*
1102                  * Status characters:
1103                  *  'D' = Dead/Failed device
1104                  *  'a' = Alive but not in-sync
1105                  *  'A' = Alive and in-sync
1106                  */
1107                 for (i = 0; i < rs->md.raid_disks; i++) {
1108                         if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1109                                 DMEMIT("D");
1110                         else if (!array_in_sync ||
1111                                  !test_bit(In_sync, &rs->dev[i].rdev.flags))
1112                                 DMEMIT("a");
1113                         else
1114                                 DMEMIT("A");
1115                 }
1116
1117                 /*
1118                  * In-sync ratio:
1119                  *  The in-sync ratio shows the progress of:
1120                  *   - Initializing the array
1121                  *   - Rebuilding a subset of devices of the array
1122                  *  The user can distinguish between the two by referring
1123                  *  to the status characters.
1124                  */
1125                 DMEMIT(" %llu/%llu",
1126                        (unsigned long long) sync,
1127                        (unsigned long long) rs->md.resync_max_sectors);
1128
1129                 break;
1130         case STATUSTYPE_TABLE:
1131                 /* The string you would use to construct this array */
1132                 for (i = 0; i < rs->md.raid_disks; i++) {
1133                         if ((rs->print_flags & DMPF_REBUILD) &&
1134                             rs->dev[i].data_dev &&
1135                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1136                                 raid_param_cnt += 2; /* for rebuilds */
1137                         if (rs->dev[i].data_dev &&
1138                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1139                                 raid_param_cnt += 2;
1140                 }
1141
1142                 raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1143                 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1144                         raid_param_cnt--;
1145
1146                 DMEMIT("%s %u %u", rs->raid_type->name,
1147                        raid_param_cnt, rs->md.chunk_sectors);
1148
1149                 if ((rs->print_flags & DMPF_SYNC) &&
1150                     (rs->md.recovery_cp == MaxSector))
1151                         DMEMIT(" sync");
1152                 if (rs->print_flags & DMPF_NOSYNC)
1153                         DMEMIT(" nosync");
1154
1155                 for (i = 0; i < rs->md.raid_disks; i++)
1156                         if ((rs->print_flags & DMPF_REBUILD) &&
1157                             rs->dev[i].data_dev &&
1158                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1159                                 DMEMIT(" rebuild %u", i);
1160
1161                 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1162                         DMEMIT(" daemon_sleep %lu",
1163                                rs->md.bitmap_info.daemon_sleep);
1164
1165                 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1166                         DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1167
1168                 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1169                         DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1170
1171                 for (i = 0; i < rs->md.raid_disks; i++)
1172                         if (rs->dev[i].data_dev &&
1173                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1174                                 DMEMIT(" write_mostly %u", i);
1175
1176                 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1177                         DMEMIT(" max_write_behind %lu",
1178                                rs->md.bitmap_info.max_write_behind);
1179
1180                 if (rs->print_flags & DMPF_STRIPE_CACHE) {
1181                         struct r5conf *conf = rs->md.private;
1182
1183                         /* convert from kiB to sectors */
1184                         DMEMIT(" stripe_cache %d",
1185                                conf ? conf->max_nr_stripes * 2 : 0);
1186                 }
1187
1188                 if (rs->print_flags & DMPF_REGION_SIZE)
1189                         DMEMIT(" region_size %lu",
1190                                rs->md.bitmap_info.chunksize >> 9);
1191
1192                 DMEMIT(" %d", rs->md.raid_disks);
1193                 for (i = 0; i < rs->md.raid_disks; i++) {
1194                         if (rs->dev[i].meta_dev)
1195                                 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1196                         else
1197                                 DMEMIT(" -");
1198
1199                         if (rs->dev[i].data_dev)
1200                                 DMEMIT(" %s", rs->dev[i].data_dev->name);
1201                         else
1202                                 DMEMIT(" -");
1203                 }
1204         }
1205
1206         return 0;
1207 }
1208
1209 static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
1210 {
1211         struct raid_set *rs = ti->private;
1212         unsigned i;
1213         int ret = 0;
1214
1215         for (i = 0; !ret && i < rs->md.raid_disks; i++)
1216                 if (rs->dev[i].data_dev)
1217                         ret = fn(ti,
1218                                  rs->dev[i].data_dev,
1219                                  0, /* No offset on data devs */
1220                                  rs->md.dev_sectors,
1221                                  data);
1222
1223         return ret;
1224 }
1225
1226 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1227 {
1228         struct raid_set *rs = ti->private;
1229         unsigned chunk_size = rs->md.chunk_sectors << 9;
1230         struct r5conf *conf = rs->md.private;
1231
1232         blk_limits_io_min(limits, chunk_size);
1233         blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1234 }
1235
1236 static void raid_presuspend(struct dm_target *ti)
1237 {
1238         struct raid_set *rs = ti->private;
1239
1240         md_stop_writes(&rs->md);
1241 }
1242
1243 static void raid_postsuspend(struct dm_target *ti)
1244 {
1245         struct raid_set *rs = ti->private;
1246
1247         mddev_suspend(&rs->md);
1248 }
1249
1250 static void raid_resume(struct dm_target *ti)
1251 {
1252         struct raid_set *rs = ti->private;
1253
1254         set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1255         if (!rs->bitmap_loaded) {
1256                 bitmap_load(&rs->md);
1257                 rs->bitmap_loaded = 1;
1258         }
1259
1260         clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1261         mddev_resume(&rs->md);
1262 }
1263
1264 static struct target_type raid_target = {
1265         .name = "raid",
1266         .version = {1, 2, 0},
1267         .module = THIS_MODULE,
1268         .ctr = raid_ctr,
1269         .dtr = raid_dtr,
1270         .map = raid_map,
1271         .status = raid_status,
1272         .iterate_devices = raid_iterate_devices,
1273         .io_hints = raid_io_hints,
1274         .presuspend = raid_presuspend,
1275         .postsuspend = raid_postsuspend,
1276         .resume = raid_resume,
1277 };
1278
1279 static int __init dm_raid_init(void)
1280 {
1281         return dm_register_target(&raid_target);
1282 }
1283
1284 static void __exit dm_raid_exit(void)
1285 {
1286         dm_unregister_target(&raid_target);
1287 }
1288
1289 module_init(dm_raid_init);
1290 module_exit(dm_raid_exit);
1291
1292 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1293 MODULE_ALIAS("dm-raid4");
1294 MODULE_ALIAS("dm-raid5");
1295 MODULE_ALIAS("dm-raid6");
1296 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1297 MODULE_LICENSE("GPL");