projects/linux·blob master · refs

4182 lines · 120.3 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2010-2011 Neil Brown
4 * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
5 *
6 * This file is released under the GPL.
7 */
8
9#include <linux/slab.h>
10#include <linux/module.h>
11
12#include "md.h"
13#include "raid1.h"
14#include "raid5.h"
15#include "raid10.h"
16#include "md-bitmap.h"
17
18#include <linux/device-mapper.h>
19
20#define DM_MSG_PREFIX "raid"
21#define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
22
23/*
24 * Minimum sectors of free reshape space per raid device
25 */
26#define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
27
28/*
29 * Minimum journal space 4 MiB in sectors.
30 */
31#define MIN_RAID456_JOURNAL_SPACE (4*2048)
32
33static bool devices_handle_discard_safely;
34
35/*
36 * The following flags are used by dm-raid to set up the array state.
37 * They must be cleared before md_run is called.
38 */
39#define FirstUse 10 /* rdev flag */
40
41struct raid_dev {
42 /*
43 * Two DM devices, one to hold metadata and one to hold the
44 * actual data/parity. The reason for this is to not confuse
45 * ti->len and give more flexibility in altering size and
46 * characteristics.
47 *
48 * While it is possible for this device to be associated
49 * with a different physical device than the data_dev, it
50 * is intended for it to be the same.
51 * |--------- Physical Device ---------|
52 * |- meta_dev -|------ data_dev ------|
53 */
54 struct dm_dev *meta_dev;
55 struct dm_dev *data_dev;
56 struct md_rdev rdev;
57};
58
59/*
60 * Bits for establishing rs->ctr_flags
61 *
62 * 1 = no flag value
63 * 2 = flag with value
64 */
65#define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
66#define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
67#define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
68#define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
69#define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
70#define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
71#define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
72#define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
73#define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
74#define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
75#define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
76#define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
77/* New for v1.9.0 */
78#define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
79#define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
80#define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
81
82/* New for v1.10.0 */
83#define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
84
85/* New for v1.11.1 */
86#define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
87
88/*
89 * Flags for rs->ctr_flags field.
90 */
91#define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
92#define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
93#define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
94#define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
95#define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
96#define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
97#define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
98#define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
99#define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
100#define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
101#define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
102#define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
103#define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
104#define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
105#define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
106#define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
107#define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
108
109/*
110 * Definitions of various constructor flags to
111 * be used in checks of valid / invalid flags
112 * per raid level.
113 */
114/* Define all any sync flags */
115#define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
116
117/* Define flags for options without argument (e.g. 'nosync') */
118#define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
119 CTR_FLAG_RAID10_USE_NEAR_SETS)
120
121/* Define flags for options with one argument (e.g. 'delta_disks +2') */
122#define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
123 CTR_FLAG_WRITE_MOSTLY | \
124 CTR_FLAG_DAEMON_SLEEP | \
125 CTR_FLAG_MIN_RECOVERY_RATE | \
126 CTR_FLAG_MAX_RECOVERY_RATE | \
127 CTR_FLAG_MAX_WRITE_BEHIND | \
128 CTR_FLAG_STRIPE_CACHE | \
129 CTR_FLAG_REGION_SIZE | \
130 CTR_FLAG_RAID10_COPIES | \
131 CTR_FLAG_RAID10_FORMAT | \
132 CTR_FLAG_DELTA_DISKS | \
133 CTR_FLAG_DATA_OFFSET | \
134 CTR_FLAG_JOURNAL_DEV | \
135 CTR_FLAG_JOURNAL_MODE)
136
137/* Valid options definitions per raid level... */
138
139/* "raid0" does only accept data offset */
140#define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
141
142/* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
143#define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
144 CTR_FLAG_REBUILD | \
145 CTR_FLAG_WRITE_MOSTLY | \
146 CTR_FLAG_DAEMON_SLEEP | \
147 CTR_FLAG_MIN_RECOVERY_RATE | \
148 CTR_FLAG_MAX_RECOVERY_RATE | \
149 CTR_FLAG_MAX_WRITE_BEHIND | \
150 CTR_FLAG_REGION_SIZE | \
151 CTR_FLAG_DELTA_DISKS | \
152 CTR_FLAG_DATA_OFFSET)
153
154/* "raid10" does not accept any raid1 or stripe cache options */
155#define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
156 CTR_FLAG_REBUILD | \
157 CTR_FLAG_DAEMON_SLEEP | \
158 CTR_FLAG_MIN_RECOVERY_RATE | \
159 CTR_FLAG_MAX_RECOVERY_RATE | \
160 CTR_FLAG_REGION_SIZE | \
161 CTR_FLAG_RAID10_COPIES | \
162 CTR_FLAG_RAID10_FORMAT | \
163 CTR_FLAG_DELTA_DISKS | \
164 CTR_FLAG_DATA_OFFSET | \
165 CTR_FLAG_RAID10_USE_NEAR_SETS)
166
167/*
168 * "raid4/5/6" do not accept any raid1 or raid10 specific options
169 *
170 * "raid6" does not accept "nosync", because it is not guaranteed
171 * that both parity and q-syndrome are being written properly with
172 * any writes
173 */
174#define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
175 CTR_FLAG_REBUILD | \
176 CTR_FLAG_DAEMON_SLEEP | \
177 CTR_FLAG_MIN_RECOVERY_RATE | \
178 CTR_FLAG_MAX_RECOVERY_RATE | \
179 CTR_FLAG_STRIPE_CACHE | \
180 CTR_FLAG_REGION_SIZE | \
181 CTR_FLAG_DELTA_DISKS | \
182 CTR_FLAG_DATA_OFFSET | \
183 CTR_FLAG_JOURNAL_DEV | \
184 CTR_FLAG_JOURNAL_MODE)
185
186#define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
187 CTR_FLAG_REBUILD | \
188 CTR_FLAG_DAEMON_SLEEP | \
189 CTR_FLAG_MIN_RECOVERY_RATE | \
190 CTR_FLAG_MAX_RECOVERY_RATE | \
191 CTR_FLAG_STRIPE_CACHE | \
192 CTR_FLAG_REGION_SIZE | \
193 CTR_FLAG_DELTA_DISKS | \
194 CTR_FLAG_DATA_OFFSET | \
195 CTR_FLAG_JOURNAL_DEV | \
196 CTR_FLAG_JOURNAL_MODE)
197/* ...valid options definitions per raid level */
198
199/*
200 * Flags for rs->runtime_flags field
201 * (RT_FLAG prefix meaning "runtime flag")
202 *
203 * These are all internal and used to define runtime state,
204 * e.g. to prevent another resume from preresume processing
205 * the raid set all over again.
206 */
207#define RT_FLAG_RS_PRERESUMED 0
208#define RT_FLAG_RS_RESUMED 1
209#define RT_FLAG_RS_BITMAP_LOADED 2
210#define RT_FLAG_UPDATE_SBS 3
211#define RT_FLAG_RESHAPE_RS 4
212#define RT_FLAG_RS_SUSPENDED 5
213#define RT_FLAG_RS_IN_SYNC 6
214#define RT_FLAG_RS_RESYNCING 7
215#define RT_FLAG_RS_GROW 8
216#define RT_FLAG_RS_FROZEN 9
217
218/* Array elements of 64 bit needed for rebuild/failed disk bits */
219#define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
220
221/*
222 * raid set level, layout and chunk sectors backup/restore
223 */
224struct rs_layout {
225 int new_level;
226 int new_layout;
227 int new_chunk_sectors;
228};
229
230struct raid_set {
231 struct dm_target *ti;
232
233 uint32_t stripe_cache_entries;
234 unsigned long ctr_flags;
235 unsigned long runtime_flags;
236
237 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
238
239 int raid_disks;
240 int delta_disks;
241 int data_offset;
242 int raid10_copies;
243 int requested_bitmap_chunk_sectors;
244
245 struct mddev md;
246 struct raid_type *raid_type;
247
248 sector_t array_sectors;
249 sector_t dev_sectors;
250
251 /* Optional raid4/5/6 journal device */
252 struct journal_dev {
253 struct dm_dev *dev;
254 struct md_rdev rdev;
255 int mode;
256 } journal_dev;
257
258 struct raid_dev dev[] __counted_by(raid_disks);
259};
260
261static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
262{
263 struct mddev *mddev = &rs->md;
264
265 l->new_level = mddev->new_level;
266 l->new_layout = mddev->new_layout;
267 l->new_chunk_sectors = mddev->new_chunk_sectors;
268}
269
270static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
271{
272 struct mddev *mddev = &rs->md;
273
274 mddev->new_level = l->new_level;
275 mddev->new_layout = l->new_layout;
276 mddev->new_chunk_sectors = l->new_chunk_sectors;
277}
278
279/* raid10 algorithms (i.e. formats) */
280#define ALGORITHM_RAID10_DEFAULT 0
281#define ALGORITHM_RAID10_NEAR 1
282#define ALGORITHM_RAID10_OFFSET 2
283#define ALGORITHM_RAID10_FAR 3
284
285/* Supported raid types and properties. */
286static struct raid_type {
287 const char *name; /* RAID algorithm. */
288 const char *descr; /* Descriptor text for logging. */
289 const unsigned int parity_devs; /* # of parity devices. */
290 const unsigned int minimal_devs;/* minimal # of devices in set. */
291 const unsigned int level; /* RAID level. */
292 const unsigned int algorithm; /* RAID algorithm. */
293} raid_types[] = {
294 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
295 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
296 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
297 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
298 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
299 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
300 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
301 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
302 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
303 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
304 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
305 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
306 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
307 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
308 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
309 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
310 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
311 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
312 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
313 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
314};
315
316/* True, if @v is in inclusive range [@min, @max] */
317static bool __within_range(long v, long min, long max)
318{
319 return v >= min && v <= max;
320}
321
322/* All table line arguments are defined here */
323static struct arg_name_flag {
324 const unsigned long flag;
325 const char *name;
326} __arg_name_flags[] = {
327 { CTR_FLAG_SYNC, "sync"},
328 { CTR_FLAG_NOSYNC, "nosync"},
329 { CTR_FLAG_REBUILD, "rebuild"},
330 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
331 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
332 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
333 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
334 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
335 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
336 { CTR_FLAG_REGION_SIZE, "region_size"},
337 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
338 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
339 { CTR_FLAG_DATA_OFFSET, "data_offset"},
340 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
341 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
342 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
343 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
344};
345
346/* Return argument name string for given @flag */
347static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
348{
349 if (hweight32(flag) == 1) {
350 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
351
352 while (anf-- > __arg_name_flags)
353 if (flag & anf->flag)
354 return anf->name;
355
356 } else
357 DMERR("%s called with more than one flag!", __func__);
358
359 return NULL;
360}
361
362/* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
363static struct {
364 const int mode;
365 const char *param;
366} _raid456_journal_mode[] = {
367 { R5C_JOURNAL_MODE_WRITE_THROUGH, "writethrough" },
368 { R5C_JOURNAL_MODE_WRITE_BACK, "writeback" }
369};
370
371/* Return MD raid4/5/6 journal mode for dm @journal_mode one */
372static int dm_raid_journal_mode_to_md(const char *mode)
373{
374 int m = ARRAY_SIZE(_raid456_journal_mode);
375
376 while (m--)
377 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
378 return _raid456_journal_mode[m].mode;
379
380 return -EINVAL;
381}
382
383/* Return dm-raid raid4/5/6 journal mode string for @mode */
384static const char *md_journal_mode_to_dm_raid(const int mode)
385{
386 int m = ARRAY_SIZE(_raid456_journal_mode);
387
388 while (m--)
389 if (mode == _raid456_journal_mode[m].mode)
390 return _raid456_journal_mode[m].param;
391
392 return "unknown";
393}
394
395/*
396 * Bool helpers to test for various raid levels of a raid set.
397 * It's level as reported by the superblock rather than
398 * the requested raid_type passed to the constructor.
399 */
400/* Return true, if raid set in @rs is raid0 */
401static bool rs_is_raid0(struct raid_set *rs)
402{
403 return !rs->md.level;
404}
405
406/* Return true, if raid set in @rs is raid1 */
407static bool rs_is_raid1(struct raid_set *rs)
408{
409 return rs->md.level == 1;
410}
411
412/* Return true, if raid set in @rs is raid10 */
413static bool rs_is_raid10(struct raid_set *rs)
414{
415 return rs->md.level == 10;
416}
417
418/* Return true, if raid set in @rs is level 6 */
419static bool rs_is_raid6(struct raid_set *rs)
420{
421 return rs->md.level == 6;
422}
423
424/* Return true, if raid set in @rs is level 4, 5 or 6 */
425static bool rs_is_raid456(struct raid_set *rs)
426{
427 return __within_range(rs->md.level, 4, 6);
428}
429
430/* Return true, if raid set in @rs is reshapable */
431static bool __is_raid10_far(int layout);
432static bool rs_is_reshapable(struct raid_set *rs)
433{
434 return rs_is_raid456(rs) ||
435 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
436}
437
438/* Return true, if raid set in @rs is recovering */
439static bool rs_is_recovering(struct raid_set *rs)
440{
441 return rs->md.resync_offset < rs->md.dev_sectors;
442}
443
444/* Return true, if raid set in @rs is reshaping */
445static bool rs_is_reshaping(struct raid_set *rs)
446{
447 return rs->md.reshape_position != MaxSector;
448}
449
450/*
451 * bool helpers to test for various raid levels of a raid type @rt
452 */
453
454/* Return true, if raid type in @rt is raid0 */
455static bool rt_is_raid0(struct raid_type *rt)
456{
457 return !rt->level;
458}
459
460/* Return true, if raid type in @rt is raid1 */
461static bool rt_is_raid1(struct raid_type *rt)
462{
463 return rt->level == 1;
464}
465
466/* Return true, if raid type in @rt is raid10 */
467static bool rt_is_raid10(struct raid_type *rt)
468{
469 return rt->level == 10;
470}
471
472/* Return true, if raid type in @rt is raid4/5 */
473static bool rt_is_raid45(struct raid_type *rt)
474{
475 return __within_range(rt->level, 4, 5);
476}
477
478/* Return true, if raid type in @rt is raid6 */
479static bool rt_is_raid6(struct raid_type *rt)
480{
481 return rt->level == 6;
482}
483
484/* Return true, if raid type in @rt is raid4/5/6 */
485static bool rt_is_raid456(struct raid_type *rt)
486{
487 return __within_range(rt->level, 4, 6);
488}
489/* END: raid level bools */
490
491/* Return valid ctr flags for the raid level of @rs */
492static unsigned long __valid_flags(struct raid_set *rs)
493{
494 if (rt_is_raid0(rs->raid_type))
495 return RAID0_VALID_FLAGS;
496 else if (rt_is_raid1(rs->raid_type))
497 return RAID1_VALID_FLAGS;
498 else if (rt_is_raid10(rs->raid_type))
499 return RAID10_VALID_FLAGS;
500 else if (rt_is_raid45(rs->raid_type))
501 return RAID45_VALID_FLAGS;
502 else if (rt_is_raid6(rs->raid_type))
503 return RAID6_VALID_FLAGS;
504
505 return 0;
506}
507
508/*
509 * Check for valid flags set on @rs
510 *
511 * Has to be called after parsing of the ctr flags!
512 */
513static int rs_check_for_valid_flags(struct raid_set *rs)
514{
515 if (rs->ctr_flags & ~__valid_flags(rs)) {
516 rs->ti->error = "Invalid flags combination";
517 return -EINVAL;
518 }
519
520 return 0;
521}
522
523/* MD raid10 bit definitions and helpers */
524#define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
525#define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
526#define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
527#define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
528
529/* Return md raid10 near copies for @layout */
530static unsigned int __raid10_near_copies(int layout)
531{
532 return layout & 0xFF;
533}
534
535/* Return md raid10 far copies for @layout */
536static unsigned int __raid10_far_copies(int layout)
537{
538 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
539}
540
541/* Return true if md raid10 offset for @layout */
542static bool __is_raid10_offset(int layout)
543{
544 return !!(layout & RAID10_OFFSET);
545}
546
547/* Return true if md raid10 near for @layout */
548static bool __is_raid10_near(int layout)
549{
550 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
551}
552
553/* Return true if md raid10 far for @layout */
554static bool __is_raid10_far(int layout)
555{
556 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
557}
558
559/* Return md raid10 layout string for @layout */
560static const char *raid10_md_layout_to_format(int layout)
561{
562 /*
563 * Bit 16 stands for "offset"
564 * (i.e. adjacent stripes hold copies)
565 *
566 * Refer to MD's raid10.c for details
567 */
568 if (__is_raid10_offset(layout))
569 return "offset";
570
571 if (__raid10_near_copies(layout) > 1)
572 return "near";
573
574 if (__raid10_far_copies(layout) > 1)
575 return "far";
576
577 return "unknown";
578}
579
580/* Return md raid10 algorithm for @name */
581static int raid10_name_to_format(const char *name)
582{
583 if (!strcasecmp(name, "near"))
584 return ALGORITHM_RAID10_NEAR;
585 else if (!strcasecmp(name, "offset"))
586 return ALGORITHM_RAID10_OFFSET;
587 else if (!strcasecmp(name, "far"))
588 return ALGORITHM_RAID10_FAR;
589
590 return -EINVAL;
591}
592
593/* Return md raid10 copies for @layout */
594static unsigned int raid10_md_layout_to_copies(int layout)
595{
596 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
597}
598
599/* Return md raid10 format id for @format string */
600static int raid10_format_to_md_layout(struct raid_set *rs,
601 unsigned int algorithm,
602 unsigned int copies)
603{
604 unsigned int n = 1, f = 1, r = 0;
605
606 /*
607 * MD resilienece flaw:
608 *
609 * enabling use_far_sets for far/offset formats causes copies
610 * to be colocated on the same devs together with their origins!
611 *
612 * -> disable it for now in the definition above
613 */
614 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
615 algorithm == ALGORITHM_RAID10_NEAR)
616 n = copies;
617
618 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
619 f = copies;
620 r = RAID10_OFFSET;
621 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
622 r |= RAID10_USE_FAR_SETS;
623
624 } else if (algorithm == ALGORITHM_RAID10_FAR) {
625 f = copies;
626 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
627 r |= RAID10_USE_FAR_SETS;
628
629 } else
630 return -EINVAL;
631
632 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
633}
634/* END: MD raid10 bit definitions and helpers */
635
636/* Check for any of the raid10 algorithms */
637static bool __got_raid10(struct raid_type *rtp, const int layout)
638{
639 if (rtp->level == 10) {
640 switch (rtp->algorithm) {
641 case ALGORITHM_RAID10_DEFAULT:
642 case ALGORITHM_RAID10_NEAR:
643 return __is_raid10_near(layout);
644 case ALGORITHM_RAID10_OFFSET:
645 return __is_raid10_offset(layout);
646 case ALGORITHM_RAID10_FAR:
647 return __is_raid10_far(layout);
648 default:
649 break;
650 }
651 }
652
653 return false;
654}
655
656/* Return raid_type for @name */
657static struct raid_type *get_raid_type(const char *name)
658{
659 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
660
661 while (rtp-- > raid_types)
662 if (!strcasecmp(rtp->name, name))
663 return rtp;
664
665 return NULL;
666}
667
668/* Return raid_type for @name based derived from @level and @layout */
669static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
670{
671 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
672
673 while (rtp-- > raid_types) {
674 /* RAID10 special checks based on @layout flags/properties */
675 if (rtp->level == level &&
676 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
677 return rtp;
678 }
679
680 return NULL;
681}
682
683/* Adjust rdev sectors */
684static void rs_set_rdev_sectors(struct raid_set *rs)
685{
686 struct mddev *mddev = &rs->md;
687 struct md_rdev *rdev;
688
689 /*
690 * raid10 sets rdev->sector to the device size, which
691 * is unintended in case of out-of-place reshaping
692 */
693 rdev_for_each(rdev, mddev)
694 if (!test_bit(Journal, &rdev->flags))
695 rdev->sectors = mddev->dev_sectors;
696}
697
698/*
699 * Change bdev capacity of @rs in case of a disk add/remove reshape
700 */
701static void rs_set_capacity(struct raid_set *rs)
702{
703 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
704
705 set_capacity_and_notify(gendisk, rs->md.array_sectors);
706}
707
708/*
709 * Set the mddev properties in @rs to the current
710 * ones retrieved from the freshest superblock
711 */
712static void rs_set_cur(struct raid_set *rs)
713{
714 struct mddev *mddev = &rs->md;
715
716 mddev->new_level = mddev->level;
717 mddev->new_layout = mddev->layout;
718 mddev->new_chunk_sectors = mddev->chunk_sectors;
719}
720
721/*
722 * Set the mddev properties in @rs to the new
723 * ones requested by the ctr
724 */
725static void rs_set_new(struct raid_set *rs)
726{
727 struct mddev *mddev = &rs->md;
728
729 mddev->level = mddev->new_level;
730 mddev->layout = mddev->new_layout;
731 mddev->chunk_sectors = mddev->new_chunk_sectors;
732 mddev->raid_disks = rs->raid_disks;
733 mddev->delta_disks = 0;
734}
735
736static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
737 unsigned int raid_devs)
738{
739 unsigned int i;
740 struct raid_set *rs;
741
742 if (raid_devs <= raid_type->parity_devs) {
743 ti->error = "Insufficient number of devices";
744 return ERR_PTR(-EINVAL);
745 }
746
747 rs = kzalloc_flex(*rs, dev, raid_devs);
748 if (!rs) {
749 ti->error = "Cannot allocate raid context";
750 return ERR_PTR(-ENOMEM);
751 }
752
753 if (mddev_init(&rs->md)) {
754 kfree(rs);
755 ti->error = "Cannot initialize raid context";
756 return ERR_PTR(-ENOMEM);
757 }
758
759 rs->raid_disks = raid_devs;
760 rs->delta_disks = 0;
761
762 rs->ti = ti;
763 rs->raid_type = raid_type;
764 rs->stripe_cache_entries = 256;
765 rs->md.raid_disks = raid_devs;
766 rs->md.level = raid_type->level;
767 rs->md.new_level = rs->md.level;
768 rs->md.layout = raid_type->algorithm;
769 rs->md.new_layout = rs->md.layout;
770 rs->md.delta_disks = 0;
771 rs->md.resync_offset = MaxSector;
772
773 for (i = 0; i < raid_devs; i++)
774 md_rdev_init(&rs->dev[i].rdev);
775
776 /*
777 * Remaining items to be initialized by further RAID params:
778 * rs->md.persistent
779 * rs->md.external
780 * rs->md.chunk_sectors
781 * rs->md.new_chunk_sectors
782 * rs->md.dev_sectors
783 */
784
785 return rs;
786}
787
788/* Free all @rs allocations */
789static void raid_set_free(struct raid_set *rs)
790{
791 int i;
792
793 if (rs->journal_dev.dev) {
794 md_rdev_clear(&rs->journal_dev.rdev);
795 dm_put_device(rs->ti, rs->journal_dev.dev);
796 }
797
798 for (i = 0; i < rs->raid_disks; i++) {
799 if (rs->dev[i].meta_dev)
800 dm_put_device(rs->ti, rs->dev[i].meta_dev);
801 md_rdev_clear(&rs->dev[i].rdev);
802 if (rs->dev[i].data_dev)
803 dm_put_device(rs->ti, rs->dev[i].data_dev);
804 }
805
806 mddev_destroy(&rs->md);
807 kfree(rs);
808}
809
810/*
811 * For every device we have two words
812 * <meta_dev>: meta device name or '-' if missing
813 * <data_dev>: data device name or '-' if missing
814 *
815 * The following are permitted:
816 * - -
817 * - <data_dev>
818 * <meta_dev> <data_dev>
819 *
820 * The following is not allowed:
821 * <meta_dev> -
822 *
823 * This code parses those words. If there is a failure,
824 * the caller must use raid_set_free() to unwind the operations.
825 */
826static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
827{
828 int i;
829 int rebuild = 0;
830 int metadata_available = 0;
831 int r = 0;
832 const char *arg;
833
834 /* Put off the number of raid devices argument to get to dev pairs */
835 arg = dm_shift_arg(as);
836 if (!arg)
837 return -EINVAL;
838
839 for (i = 0; i < rs->raid_disks; i++) {
840 rs->dev[i].rdev.raid_disk = i;
841
842 rs->dev[i].meta_dev = NULL;
843 rs->dev[i].data_dev = NULL;
844
845 /*
846 * There are no offsets initially.
847 * Out of place reshape will set them accordingly.
848 */
849 rs->dev[i].rdev.data_offset = 0;
850 rs->dev[i].rdev.new_data_offset = 0;
851 rs->dev[i].rdev.mddev = &rs->md;
852
853 arg = dm_shift_arg(as);
854 if (!arg)
855 return -EINVAL;
856
857 if (strcmp(arg, "-")) {
858 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
859 &rs->dev[i].meta_dev);
860 if (r) {
861 rs->ti->error = "RAID metadata device lookup failure";
862 return r;
863 }
864
865 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
866 if (!rs->dev[i].rdev.sb_page) {
867 rs->ti->error = "Failed to allocate superblock page";
868 return -ENOMEM;
869 }
870 }
871
872 arg = dm_shift_arg(as);
873 if (!arg)
874 return -EINVAL;
875
876 if (!strcmp(arg, "-")) {
877 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
878 (!rs->dev[i].rdev.recovery_offset)) {
879 rs->ti->error = "Drive designated for rebuild not specified";
880 return -EINVAL;
881 }
882
883 if (rs->dev[i].meta_dev) {
884 rs->ti->error = "No data device supplied with metadata device";
885 return -EINVAL;
886 }
887
888 continue;
889 }
890
891 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
892 &rs->dev[i].data_dev);
893 if (r) {
894 rs->ti->error = "RAID device lookup failure";
895 return r;
896 }
897
898 if (rs->dev[i].meta_dev) {
899 metadata_available = 1;
900 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
901 }
902 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
903 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
904 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
905 rebuild++;
906 }
907
908 if (rs->journal_dev.dev)
909 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
910
911 if (metadata_available) {
912 rs->md.external = 0;
913 rs->md.persistent = 1;
914 rs->md.major_version = 2;
915 } else if (rebuild && !rs->md.resync_offset) {
916 /*
917 * Without metadata, we will not be able to tell if the array
918 * is in-sync or not - we must assume it is not. Therefore,
919 * it is impossible to rebuild a drive.
920 *
921 * Even if there is metadata, the on-disk information may
922 * indicate that the array is not in-sync and it will then
923 * fail at that time.
924 *
925 * User could specify 'nosync' option if desperate.
926 */
927 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
928 return -EINVAL;
929 }
930
931 return 0;
932}
933
934/*
935 * validate_region_size
936 * @rs
937 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
938 *
939 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
940 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
941 *
942 * Returns: 0 on success, -EINVAL on failure.
943 */
944static int validate_region_size(struct raid_set *rs, unsigned long region_size)
945{
946 unsigned long min_region_size = rs->ti->len / (1 << 21);
947
948 if (rs_is_raid0(rs))
949 return 0;
950
951 if (!region_size) {
952 /*
953 * Choose a reasonable default. All figures in sectors.
954 */
955 if (min_region_size > (1 << 13)) {
956 /* If not a power of 2, make it the next power of 2 */
957 region_size = roundup_pow_of_two(min_region_size);
958 DMINFO("Choosing default region size of %lu sectors",
959 region_size);
960 } else {
961 DMINFO("Choosing default region size of 4MiB");
962 region_size = 1 << 13; /* sectors */
963 }
964 } else {
965 /*
966 * Validate user-supplied value.
967 */
968 if (region_size > rs->ti->len) {
969 rs->ti->error = "Supplied region size is too large";
970 return -EINVAL;
971 }
972
973 if (region_size < min_region_size) {
974 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
975 region_size, min_region_size);
976 rs->ti->error = "Supplied region size is too small";
977 return -EINVAL;
978 }
979
980 if (!is_power_of_2(region_size)) {
981 rs->ti->error = "Region size is not a power of 2";
982 return -EINVAL;
983 }
984
985 if (region_size < rs->md.chunk_sectors) {
986 rs->ti->error = "Region size is smaller than the chunk size";
987 return -EINVAL;
988 }
989 }
990
991 /*
992 * Convert sectors to bytes.
993 */
994 rs->md.bitmap_info.chunksize = to_bytes(region_size);
995
996 return 0;
997}
998
999/*
1000 * validate_raid_redundancy
1001 * @rs
1002 *
1003 * Determine if there are enough devices in the array that haven't
1004 * failed (or are being rebuilt) to form a usable array.
1005 *
1006 * Returns: 0 on success, -EINVAL on failure.
1007 */
1008static int validate_raid_redundancy(struct raid_set *rs)
1009{
1010 unsigned int i, rebuild_cnt = 0;
1011 unsigned int rebuilds_per_group = 0, copies, raid_disks;
1012 unsigned int group_size, last_group_start;
1013
1014 for (i = 0; i < rs->raid_disks; i++)
1015 if (!test_bit(FirstUse, &rs->dev[i].rdev.flags) &&
1016 ((!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1017 !rs->dev[i].rdev.sb_page)))
1018 rebuild_cnt++;
1019
1020 switch (rs->md.level) {
1021 case 0:
1022 break;
1023 case 1:
1024 if (rebuild_cnt >= rs->md.raid_disks)
1025 goto too_many;
1026 break;
1027 case 4:
1028 case 5:
1029 case 6:
1030 if (rebuild_cnt > rs->raid_type->parity_devs)
1031 goto too_many;
1032 break;
1033 case 10:
1034 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1035 if (copies < 2) {
1036 DMERR("Bogus raid10 data copies < 2!");
1037 return -EINVAL;
1038 }
1039
1040 if (rebuild_cnt < copies)
1041 break;
1042
1043 /*
1044 * It is possible to have a higher rebuild count for RAID10,
1045 * as long as the failed devices occur in different mirror
1046 * groups (i.e. different stripes).
1047 *
1048 * When checking "near" format, make sure no adjacent devices
1049 * have failed beyond what can be handled. In addition to the
1050 * simple case where the number of devices is a multiple of the
1051 * number of copies, we must also handle cases where the number
1052 * of devices is not a multiple of the number of copies.
1053 * E.g. dev1 dev2 dev3 dev4 dev5
1054 * A A B B C
1055 * C D D E E
1056 */
1057 raid_disks = min(rs->raid_disks, rs->md.raid_disks);
1058 if (__is_raid10_near(rs->md.new_layout)) {
1059 for (i = 0; i < raid_disks; i++) {
1060 if (!(i % copies))
1061 rebuilds_per_group = 0;
1062 if ((!rs->dev[i].rdev.sb_page ||
1063 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1064 (++rebuilds_per_group >= copies))
1065 goto too_many;
1066 }
1067 break;
1068 }
1069
1070 /*
1071 * When checking "far" and "offset" formats, we need to ensure
1072 * that the device that holds its copy is not also dead or
1073 * being rebuilt. (Note that "far" and "offset" formats only
1074 * support two copies right now. These formats also only ever
1075 * use the 'use_far_sets' variant.)
1076 *
1077 * This check is somewhat complicated by the need to account
1078 * for arrays that are not a multiple of (far) copies. This
1079 * results in the need to treat the last (potentially larger)
1080 * set differently.
1081 */
1082 group_size = (raid_disks / copies);
1083 last_group_start = (raid_disks / group_size) - 1;
1084 last_group_start *= group_size;
1085 for (i = 0; i < raid_disks; i++) {
1086 if (!(i % copies) && !(i > last_group_start))
1087 rebuilds_per_group = 0;
1088 if ((!rs->dev[i].rdev.sb_page ||
1089 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1090 (++rebuilds_per_group >= copies))
1091 goto too_many;
1092 }
1093 break;
1094 default:
1095 if (rebuild_cnt)
1096 return -EINVAL;
1097 }
1098
1099 return 0;
1100
1101too_many:
1102 return -EINVAL;
1103}
1104
1105/*
1106 * Possible arguments are...
1107 * <chunk_size> [optional_args]
1108 *
1109 * Argument definitions
1110 * <chunk_size> The number of sectors per disk that
1111 * will form the "stripe"
1112 * [[no]sync] Force or prevent recovery of the
1113 * entire array
1114 * [rebuild <idx>] Rebuild the drive indicated by the index
1115 * [daemon_sleep <ms>] Time between bitmap daemon work to
1116 * clear bits
1117 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1118 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1119 * [write_mostly <idx>] Indicate a write mostly drive via index
1120 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1121 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
1122 * [region_size <sectors>] Defines granularity of bitmap
1123 * [journal_dev <dev>] raid4/5/6 journaling deviice
1124 * (i.e. write hole closing log)
1125 *
1126 * RAID10-only options:
1127 * [raid10_copies <# copies>] Number of copies. (Default: 2)
1128 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
1129 */
1130static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1131 unsigned int num_raid_params)
1132{
1133 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1134 unsigned int raid10_copies = 2;
1135 unsigned int i, write_mostly = 0;
1136 unsigned int region_size = 0;
1137 sector_t max_io_len;
1138 const char *arg, *key;
1139 struct raid_dev *rd;
1140 struct raid_type *rt = rs->raid_type;
1141
1142 arg = dm_shift_arg(as);
1143 num_raid_params--; /* Account for chunk_size argument */
1144
1145 if (kstrtoint(arg, 10, &value) < 0) {
1146 rs->ti->error = "Bad numerical argument given for chunk_size";
1147 return -EINVAL;
1148 }
1149
1150 /*
1151 * First, parse the in-order required arguments
1152 * "chunk_size" is the only argument of this type.
1153 */
1154 if (rt_is_raid1(rt)) {
1155 if (value)
1156 DMERR("Ignoring chunk size parameter for RAID 1");
1157 value = 0;
1158 } else if (!is_power_of_2(value)) {
1159 rs->ti->error = "Chunk size must be a power of 2";
1160 return -EINVAL;
1161 } else if (value < 8) {
1162 rs->ti->error = "Chunk size value is too small";
1163 return -EINVAL;
1164 }
1165
1166 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1167
1168 /*
1169 * We set each individual device as In_sync with a completed
1170 * 'recovery_offset'. If there has been a device failure or
1171 * replacement then one of the following cases applies:
1172 *
1173 * 1) User specifies 'rebuild'.
1174 * - Device is reset when param is read.
1175 * 2) A new device is supplied.
1176 * - No matching superblock found, resets device.
1177 * 3) Device failure was transient and returns on reload.
1178 * - Failure noticed, resets device for bitmap replay.
1179 * 4) Device hadn't completed recovery after previous failure.
1180 * - Superblock is read and overrides recovery_offset.
1181 *
1182 * What is found in the superblocks of the devices is always
1183 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1184 */
1185 for (i = 0; i < rs->raid_disks; i++) {
1186 set_bit(In_sync, &rs->dev[i].rdev.flags);
1187 rs->dev[i].rdev.recovery_offset = MaxSector;
1188 }
1189
1190 /*
1191 * Second, parse the unordered optional arguments
1192 */
1193 for (i = 0; i < num_raid_params; i++) {
1194 key = dm_shift_arg(as);
1195 if (!key) {
1196 rs->ti->error = "Not enough raid parameters given";
1197 return -EINVAL;
1198 }
1199
1200 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1201 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1202 rs->ti->error = "Only one 'nosync' argument allowed";
1203 return -EINVAL;
1204 }
1205 continue;
1206 }
1207 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1208 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1209 rs->ti->error = "Only one 'sync' argument allowed";
1210 return -EINVAL;
1211 }
1212 continue;
1213 }
1214 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1215 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1216 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1217 return -EINVAL;
1218 }
1219 continue;
1220 }
1221
1222 arg = dm_shift_arg(as);
1223 i++; /* Account for the argument pairs */
1224 if (!arg) {
1225 rs->ti->error = "Wrong number of raid parameters given";
1226 return -EINVAL;
1227 }
1228
1229 /*
1230 * Parameters that take a string value are checked here.
1231 */
1232 /* "raid10_format {near|offset|far} */
1233 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1234 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1235 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1236 return -EINVAL;
1237 }
1238 if (!rt_is_raid10(rt)) {
1239 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1240 return -EINVAL;
1241 }
1242 raid10_format = raid10_name_to_format(arg);
1243 if (raid10_format < 0) {
1244 rs->ti->error = "Invalid 'raid10_format' value given";
1245 return raid10_format;
1246 }
1247 continue;
1248 }
1249
1250 /* "journal_dev <dev>" */
1251 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1252 int r;
1253 struct md_rdev *jdev;
1254
1255 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1256 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1257 return -EINVAL;
1258 }
1259 if (!rt_is_raid456(rt)) {
1260 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1261 return -EINVAL;
1262 }
1263 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1264 &rs->journal_dev.dev);
1265 if (r) {
1266 rs->ti->error = "raid4/5/6 journal device lookup failure";
1267 return r;
1268 }
1269 jdev = &rs->journal_dev.rdev;
1270 md_rdev_init(jdev);
1271 jdev->mddev = &rs->md;
1272 jdev->bdev = rs->journal_dev.dev->bdev;
1273 jdev->sectors = bdev_nr_sectors(jdev->bdev);
1274 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1275 rs->ti->error = "No space for raid4/5/6 journal";
1276 return -ENOSPC;
1277 }
1278 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1279 set_bit(Journal, &jdev->flags);
1280 continue;
1281 }
1282
1283 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1284 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1285 int r;
1286
1287 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1288 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1289 return -EINVAL;
1290 }
1291 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1292 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1293 return -EINVAL;
1294 }
1295 r = dm_raid_journal_mode_to_md(arg);
1296 if (r < 0) {
1297 rs->ti->error = "Invalid 'journal_mode' argument";
1298 return r;
1299 }
1300 rs->journal_dev.mode = r;
1301 continue;
1302 }
1303
1304 /*
1305 * Parameters with number values from here on.
1306 */
1307 if (kstrtoint(arg, 10, &value) < 0) {
1308 rs->ti->error = "Bad numerical argument given in raid params";
1309 return -EINVAL;
1310 }
1311
1312 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1313 /*
1314 * "rebuild" is being passed in by userspace to provide
1315 * indexes of replaced devices and to set up additional
1316 * devices on raid level takeover.
1317 */
1318 if (!__within_range(value, 0, rs->raid_disks - 1)) {
1319 rs->ti->error = "Invalid rebuild index given";
1320 return -EINVAL;
1321 }
1322
1323 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1324 rs->ti->error = "rebuild for this index already given";
1325 return -EINVAL;
1326 }
1327
1328 rd = rs->dev + value;
1329 clear_bit(In_sync, &rd->rdev.flags);
1330 clear_bit(Faulty, &rd->rdev.flags);
1331 rd->rdev.recovery_offset = 0;
1332 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1333 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1334 if (!rt_is_raid1(rt)) {
1335 rs->ti->error = "write_mostly option is only valid for RAID1";
1336 return -EINVAL;
1337 }
1338
1339 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1340 rs->ti->error = "Invalid write_mostly index given";
1341 return -EINVAL;
1342 }
1343
1344 write_mostly++;
1345 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1346 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1347 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1348 if (!rt_is_raid1(rt)) {
1349 rs->ti->error = "max_write_behind option is only valid for RAID1";
1350 return -EINVAL;
1351 }
1352
1353 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1354 rs->ti->error = "Only one max_write_behind argument pair allowed";
1355 return -EINVAL;
1356 }
1357
1358 if (value < 0) {
1359 rs->ti->error = "Max write-behind limit out of range";
1360 return -EINVAL;
1361 }
1362
1363 rs->md.bitmap_info.max_write_behind = value / 2;
1364 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1365 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1366 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1367 return -EINVAL;
1368 }
1369 if (value < 0) {
1370 rs->ti->error = "daemon sleep period out of range";
1371 return -EINVAL;
1372 }
1373 rs->md.bitmap_info.daemon_sleep = value;
1374 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1375 /* Userspace passes new data_offset after having extended the data image LV */
1376 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1377 rs->ti->error = "Only one data_offset argument pair allowed";
1378 return -EINVAL;
1379 }
1380 /* Ensure sensible data offset */
1381 if (value < 0 ||
1382 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1383 rs->ti->error = "Bogus data_offset value";
1384 return -EINVAL;
1385 }
1386 rs->data_offset = value;
1387 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1388 /* Define the +/-# of disks to add to/remove from the given raid set */
1389 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1390 rs->ti->error = "Only one delta_disks argument pair allowed";
1391 return -EINVAL;
1392 }
1393 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1394 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1395 rs->ti->error = "Too many delta_disk requested";
1396 return -EINVAL;
1397 }
1398
1399 rs->delta_disks = value;
1400 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1401 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1402 rs->ti->error = "Only one stripe_cache argument pair allowed";
1403 return -EINVAL;
1404 }
1405
1406 if (!rt_is_raid456(rt)) {
1407 rs->ti->error = "Inappropriate argument: stripe_cache";
1408 return -EINVAL;
1409 }
1410
1411 if (value < 0) {
1412 rs->ti->error = "Bogus stripe cache entries value";
1413 return -EINVAL;
1414 }
1415 rs->stripe_cache_entries = value;
1416 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1417 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1418 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1419 return -EINVAL;
1420 }
1421
1422 if (value < 0) {
1423 rs->ti->error = "min_recovery_rate out of range";
1424 return -EINVAL;
1425 }
1426 rs->md.sync_speed_min = value;
1427 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1428 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1429 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1430 return -EINVAL;
1431 }
1432
1433 if (value < 0) {
1434 rs->ti->error = "max_recovery_rate out of range";
1435 return -EINVAL;
1436 }
1437 rs->md.sync_speed_max = value;
1438 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1439 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1440 rs->ti->error = "Only one region_size argument pair allowed";
1441 return -EINVAL;
1442 }
1443
1444 region_size = value;
1445 rs->requested_bitmap_chunk_sectors = value;
1446 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1447 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1448 rs->ti->error = "Only one raid10_copies argument pair allowed";
1449 return -EINVAL;
1450 }
1451
1452 if (!__within_range(value, 2, rs->md.raid_disks)) {
1453 rs->ti->error = "Bad value for 'raid10_copies'";
1454 return -EINVAL;
1455 }
1456
1457 raid10_copies = value;
1458 } else {
1459 DMERR("Unable to parse RAID parameter: %s", key);
1460 rs->ti->error = "Unable to parse RAID parameter";
1461 return -EINVAL;
1462 }
1463 }
1464
1465 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1466 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1467 rs->ti->error = "sync and nosync are mutually exclusive";
1468 return -EINVAL;
1469 }
1470
1471 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1472 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1473 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1474 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1475 return -EINVAL;
1476 }
1477
1478 if (write_mostly >= rs->md.raid_disks) {
1479 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1480 return -EINVAL;
1481 }
1482
1483 if (rs->md.sync_speed_max &&
1484 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1485 rs->ti->error = "Bogus recovery rates";
1486 return -EINVAL;
1487 }
1488
1489 if (validate_region_size(rs, region_size))
1490 return -EINVAL;
1491
1492 if (rs->md.chunk_sectors)
1493 max_io_len = rs->md.chunk_sectors;
1494 else
1495 max_io_len = region_size;
1496
1497 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1498 return -EINVAL;
1499
1500 if (rt_is_raid10(rt)) {
1501 if (raid10_copies > rs->md.raid_disks) {
1502 rs->ti->error = "Not enough devices to satisfy specification";
1503 return -EINVAL;
1504 }
1505
1506 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1507 if (rs->md.new_layout < 0) {
1508 rs->ti->error = "Error getting raid10 format";
1509 return rs->md.new_layout;
1510 }
1511
1512 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1513 if (!rt) {
1514 rs->ti->error = "Failed to recognize new raid10 layout";
1515 return -EINVAL;
1516 }
1517
1518 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1519 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1520 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1521 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1522 return -EINVAL;
1523 }
1524 }
1525
1526 rs->raid10_copies = raid10_copies;
1527
1528 /* Assume there are no metadata devices until the drives are parsed */
1529 rs->md.persistent = 0;
1530 rs->md.external = 1;
1531
1532 /* Check, if any invalid ctr arguments have been passed in for the raid level */
1533 return rs_check_for_valid_flags(rs);
1534}
1535
1536/* Set raid4/5/6 cache size */
1537static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1538{
1539 int r;
1540 struct r5conf *conf;
1541 struct mddev *mddev = &rs->md;
1542 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1543 uint32_t nr_stripes = rs->stripe_cache_entries;
1544
1545 if (!rt_is_raid456(rs->raid_type)) {
1546 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1547 return -EINVAL;
1548 }
1549
1550 if (nr_stripes < min_stripes) {
1551 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1552 nr_stripes, min_stripes);
1553 nr_stripes = min_stripes;
1554 }
1555
1556 conf = mddev->private;
1557 if (!conf) {
1558 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1559 return -EINVAL;
1560 }
1561
1562 /* Try setting number of stripes in raid456 stripe cache */
1563 if (conf->min_nr_stripes != nr_stripes) {
1564 r = raid5_set_cache_size(mddev, nr_stripes);
1565 if (r) {
1566 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1567 return r;
1568 }
1569
1570 DMINFO("%u stripe cache entries", nr_stripes);
1571 }
1572
1573 return 0;
1574}
1575
1576/* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1577static unsigned int mddev_data_stripes(struct raid_set *rs)
1578{
1579 return rs->md.raid_disks - rs->raid_type->parity_devs;
1580}
1581
1582/* Return # of data stripes of @rs (i.e. as of ctr) */
1583static unsigned int rs_data_stripes(struct raid_set *rs)
1584{
1585 return rs->raid_disks - rs->raid_type->parity_devs;
1586}
1587
1588/*
1589 * Retrieve rdev->sectors from any valid raid device of @rs
1590 * to allow userpace to pass in arbitray "- -" device tupples.
1591 */
1592static sector_t __rdev_sectors(struct raid_set *rs)
1593{
1594 int i;
1595
1596 for (i = 0; i < rs->raid_disks; i++) {
1597 struct md_rdev *rdev = &rs->dev[i].rdev;
1598
1599 if (!test_bit(Journal, &rdev->flags) &&
1600 rdev->bdev && rdev->sectors)
1601 return rdev->sectors;
1602 }
1603
1604 return 0;
1605}
1606
1607/* Check that calculated dev_sectors fits all component devices. */
1608static int _check_data_dev_sectors(struct raid_set *rs)
1609{
1610 sector_t ds = ~0;
1611 struct md_rdev *rdev;
1612
1613 rdev_for_each(rdev, &rs->md)
1614 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1615 ds = min(ds, bdev_nr_sectors(rdev->bdev));
1616 if (ds < rs->md.dev_sectors) {
1617 rs->ti->error = "Component device(s) too small";
1618 return -EINVAL;
1619 }
1620 }
1621
1622 return 0;
1623}
1624
1625/* Get reshape sectors from data_offsets or raid set */
1626static sector_t _get_reshape_sectors(struct raid_set *rs)
1627{
1628 struct md_rdev *rdev;
1629 sector_t reshape_sectors = 0;
1630
1631 rdev_for_each(rdev, &rs->md)
1632 if (!test_bit(Journal, &rdev->flags)) {
1633 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
1634 rdev->data_offset - rdev->new_data_offset :
1635 rdev->new_data_offset - rdev->data_offset;
1636 break;
1637 }
1638
1639 return max(reshape_sectors, (sector_t) rs->data_offset);
1640}
1641
1642/* Calculate the sectors per device and per array used for @rs */
1643static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
1644{
1645 int delta_disks;
1646 unsigned int data_stripes;
1647 sector_t array_sectors = sectors, dev_sectors = sectors;
1648 struct mddev *mddev = &rs->md;
1649
1650 if (use_mddev) {
1651 delta_disks = mddev->delta_disks;
1652 data_stripes = mddev_data_stripes(rs);
1653 } else {
1654 delta_disks = rs->delta_disks;
1655 data_stripes = rs_data_stripes(rs);
1656 }
1657
1658 /* Special raid1 case w/o delta_disks support (yet) */
1659 if (rt_is_raid1(rs->raid_type))
1660 ;
1661 else if (rt_is_raid10(rs->raid_type)) {
1662 if (rs->raid10_copies < 2 ||
1663 delta_disks < 0) {
1664 rs->ti->error = "Bogus raid10 data copies or delta disks";
1665 return -EINVAL;
1666 }
1667
1668 dev_sectors *= rs->raid10_copies;
1669 if (sector_div(dev_sectors, data_stripes))
1670 goto bad;
1671
1672 array_sectors = (data_stripes + delta_disks) * (dev_sectors - _get_reshape_sectors(rs));
1673 if (sector_div(array_sectors, rs->raid10_copies))
1674 goto bad;
1675
1676 } else if (sector_div(dev_sectors, data_stripes))
1677 goto bad;
1678
1679 else
1680 /* Striped layouts */
1681 array_sectors = (data_stripes + delta_disks) * (dev_sectors - _get_reshape_sectors(rs));
1682
1683 mddev->array_sectors = array_sectors;
1684 mddev->dev_sectors = dev_sectors;
1685 rs_set_rdev_sectors(rs);
1686
1687 return _check_data_dev_sectors(rs);
1688bad:
1689 rs->ti->error = "Target length not divisible by number of data devices";
1690 return -EINVAL;
1691}
1692
1693/* Setup recovery on @rs */
1694static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1695{
1696 /* raid0 does not recover */
1697 if (rs_is_raid0(rs))
1698 rs->md.resync_offset = MaxSector;
1699 /*
1700 * A raid6 set has to be recovered either
1701 * completely or for the grown part to
1702 * ensure proper parity and Q-Syndrome
1703 */
1704 else if (rs_is_raid6(rs))
1705 rs->md.resync_offset = dev_sectors;
1706 /*
1707 * Other raid set types may skip recovery
1708 * depending on the 'nosync' flag.
1709 */
1710 else
1711 rs->md.resync_offset = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1712 ? MaxSector : dev_sectors;
1713}
1714
1715static void do_table_event(struct work_struct *ws)
1716{
1717 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1718
1719 smp_rmb(); /* Make sure we access most actual mddev properties */
1720
1721 /* Only grow size resulting from added stripe(s) after reshape ended. */
1722 if (!rs_is_reshaping(rs) &&
1723 rs->array_sectors > rs->md.array_sectors &&
1724 !rs->md.delta_disks &&
1725 rs->md.raid_disks == rs->raid_disks) {
1726 /* The raid10 personality doesn't provide proper device sizes -> correct. */
1727 if (rs_is_raid10(rs))
1728 rs_set_rdev_sectors(rs);
1729
1730 rs->md.array_sectors = rs->array_sectors;
1731 rs_set_capacity(rs);
1732 }
1733
1734 dm_table_event(rs->ti->table);
1735}
1736
1737/*
1738 * Make sure a valid takover (level switch) is being requested on @rs
1739 *
1740 * Conversions of raid sets from one MD personality to another
1741 * have to conform to restrictions which are enforced here.
1742 */
1743static int rs_check_takeover(struct raid_set *rs)
1744{
1745 struct mddev *mddev = &rs->md;
1746 unsigned int near_copies;
1747
1748 if (rs->md.degraded) {
1749 rs->ti->error = "Can't takeover degraded raid set";
1750 return -EPERM;
1751 }
1752
1753 if (rs_is_reshaping(rs)) {
1754 rs->ti->error = "Can't takeover reshaping raid set";
1755 return -EPERM;
1756 }
1757
1758 switch (mddev->level) {
1759 case 0:
1760 /* raid0 -> raid1/5 with one disk */
1761 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1762 mddev->raid_disks == 1)
1763 return 0;
1764
1765 /* raid0 -> raid10 */
1766 if (mddev->new_level == 10 &&
1767 !(rs->raid_disks % mddev->raid_disks))
1768 return 0;
1769
1770 /* raid0 with multiple disks -> raid4/5/6 */
1771 if (__within_range(mddev->new_level, 4, 6) &&
1772 mddev->new_layout == ALGORITHM_PARITY_N &&
1773 mddev->raid_disks > 1)
1774 return 0;
1775
1776 break;
1777
1778 case 10:
1779 /* Can't takeover raid10_offset! */
1780 if (__is_raid10_offset(mddev->layout))
1781 break;
1782
1783 near_copies = __raid10_near_copies(mddev->layout);
1784
1785 /* raid10* -> raid0 */
1786 if (mddev->new_level == 0) {
1787 /* Can takeover raid10_near with raid disks divisable by data copies! */
1788 if (near_copies > 1 &&
1789 !(mddev->raid_disks % near_copies)) {
1790 mddev->raid_disks /= near_copies;
1791 mddev->delta_disks = mddev->raid_disks;
1792 return 0;
1793 }
1794
1795 /* Can takeover raid10_far */
1796 if (near_copies == 1 &&
1797 __raid10_far_copies(mddev->layout) > 1)
1798 return 0;
1799
1800 break;
1801 }
1802
1803 /* raid10_{near,far} -> raid1 */
1804 if (mddev->new_level == 1 &&
1805 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1806 return 0;
1807
1808 /* raid10_{near,far} with 2 disks -> raid4/5 */
1809 if (__within_range(mddev->new_level, 4, 5) &&
1810 mddev->raid_disks == 2)
1811 return 0;
1812 break;
1813
1814 case 1:
1815 /* raid1 with 2 disks -> raid4/5 */
1816 if (__within_range(mddev->new_level, 4, 5) &&
1817 mddev->raid_disks == 2) {
1818 mddev->degraded = 1;
1819 return 0;
1820 }
1821
1822 /* raid1 -> raid0 */
1823 if (mddev->new_level == 0 &&
1824 mddev->raid_disks == 1)
1825 return 0;
1826
1827 /* raid1 -> raid10 */
1828 if (mddev->new_level == 10)
1829 return 0;
1830 break;
1831
1832 case 4:
1833 /* raid4 -> raid0 */
1834 if (mddev->new_level == 0)
1835 return 0;
1836
1837 /* raid4 -> raid1/5 with 2 disks */
1838 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1839 mddev->raid_disks == 2)
1840 return 0;
1841
1842 /* raid4 -> raid5/6 with parity N */
1843 if (__within_range(mddev->new_level, 5, 6) &&
1844 mddev->layout == ALGORITHM_PARITY_N)
1845 return 0;
1846 break;
1847
1848 case 5:
1849 /* raid5 with parity N -> raid0 */
1850 if (mddev->new_level == 0 &&
1851 mddev->layout == ALGORITHM_PARITY_N)
1852 return 0;
1853
1854 /* raid5 with parity N -> raid4 */
1855 if (mddev->new_level == 4 &&
1856 mddev->layout == ALGORITHM_PARITY_N)
1857 return 0;
1858
1859 /* raid5 with 2 disks -> raid1/4/10 */
1860 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1861 mddev->raid_disks == 2)
1862 return 0;
1863
1864 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1865 if (mddev->new_level == 6 &&
1866 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1867 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1868 return 0;
1869 break;
1870
1871 case 6:
1872 /* raid6 with parity N -> raid0 */
1873 if (mddev->new_level == 0 &&
1874 mddev->layout == ALGORITHM_PARITY_N)
1875 return 0;
1876
1877 /* raid6 with parity N -> raid4 */
1878 if (mddev->new_level == 4 &&
1879 mddev->layout == ALGORITHM_PARITY_N)
1880 return 0;
1881
1882 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1883 if (mddev->new_level == 5 &&
1884 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1885 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1886 return 0;
1887 break;
1888
1889 default:
1890 break;
1891 }
1892
1893 rs->ti->error = "takeover not possible";
1894 return -EINVAL;
1895}
1896
1897/* True if @rs requested to be taken over */
1898static bool rs_takeover_requested(struct raid_set *rs)
1899{
1900 return rs->md.new_level != rs->md.level;
1901}
1902
1903/* True if layout is set to reshape. */
1904static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1905{
1906 return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1907 rs->md.new_layout != rs->md.layout ||
1908 rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1909}
1910
1911/* True if @rs is requested to reshape by ctr */
1912static bool rs_reshape_requested(struct raid_set *rs)
1913{
1914 bool change;
1915 struct mddev *mddev = &rs->md;
1916
1917 if (rs_takeover_requested(rs))
1918 return false;
1919
1920 if (rs_is_raid0(rs))
1921 return false;
1922
1923 change = rs_is_layout_change(rs, false);
1924
1925 /* Historical case to support raid1 reshape without delta disks */
1926 if (rs_is_raid1(rs)) {
1927 if (rs->delta_disks)
1928 return !!rs->delta_disks;
1929
1930 return !change &&
1931 mddev->raid_disks != rs->raid_disks;
1932 }
1933
1934 if (rs_is_raid10(rs))
1935 return change &&
1936 !__is_raid10_far(mddev->new_layout) &&
1937 rs->delta_disks >= 0;
1938
1939 return change;
1940}
1941
1942/* Features */
1943#define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
1944
1945/* State flags for sb->flags */
1946#define SB_FLAG_RESHAPE_ACTIVE 0x1
1947#define SB_FLAG_RESHAPE_BACKWARDS 0x2
1948
1949/*
1950 * This structure is never routinely used by userspace, unlike md superblocks.
1951 * Devices with this superblock should only ever be accessed via device-mapper.
1952 */
1953#define DM_RAID_MAGIC 0x64526D44
1954struct dm_raid_superblock {
1955 __le32 magic; /* "DmRd" */
1956 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1957
1958 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1959 __le32 array_position; /* The position of this drive in the raid set */
1960
1961 __le64 events; /* Incremented by md when superblock updated */
1962 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
1963 /* indicate failures (see extension below) */
1964
1965 /*
1966 * This offset tracks the progress of the repair or replacement of
1967 * an individual drive.
1968 */
1969 __le64 disk_recovery_offset;
1970
1971 /*
1972 * This offset tracks the progress of the initial raid set
1973 * synchronisation/parity calculation.
1974 */
1975 __le64 array_resync_offset;
1976
1977 /*
1978 * raid characteristics
1979 */
1980 __le32 level;
1981 __le32 layout;
1982 __le32 stripe_sectors;
1983
1984 /********************************************************************
1985 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1986 *
1987 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1988 */
1989
1990 __le32 flags; /* Flags defining array states for reshaping */
1991
1992 /*
1993 * This offset tracks the progress of a raid
1994 * set reshape in order to be able to restart it
1995 */
1996 __le64 reshape_position;
1997
1998 /*
1999 * These define the properties of the array in case of an interrupted reshape
2000 */
2001 __le32 new_level;
2002 __le32 new_layout;
2003 __le32 new_stripe_sectors;
2004 __le32 delta_disks;
2005
2006 __le64 array_sectors; /* Array size in sectors */
2007
2008 /*
2009 * Sector offsets to data on devices (reshaping).
2010 * Needed to support out of place reshaping, thus
2011 * not writing over any stripes whilst converting
2012 * them from old to new layout
2013 */
2014 __le64 data_offset;
2015 __le64 new_data_offset;
2016
2017 __le64 sectors; /* Used device size in sectors */
2018
2019 /*
2020 * Additional Bit field of devices indicating failures to support
2021 * up to 256 devices with the 1.9.0 on-disk metadata format
2022 */
2023 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
2024
2025 __le32 incompat_features; /* Used to indicate any incompatible features */
2026
2027 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
2028} __packed;
2029
2030/*
2031 * Check for reshape constraints on raid set @rs:
2032 *
2033 * - reshape function non-existent
2034 * - degraded set
2035 * - ongoing recovery
2036 * - ongoing reshape
2037 *
2038 * Returns 0 if none or -EPERM if given constraint
2039 * and error message reference in @errmsg
2040 */
2041static int rs_check_reshape(struct raid_set *rs)
2042{
2043 struct mddev *mddev = &rs->md;
2044
2045 if (!mddev->pers || !mddev->pers->check_reshape)
2046 rs->ti->error = "Reshape not supported";
2047 else if (mddev->degraded)
2048 rs->ti->error = "Can't reshape degraded raid set";
2049 else if (rs_is_recovering(rs))
2050 rs->ti->error = "Convert request on recovering raid set prohibited";
2051 else if (rs_is_reshaping(rs))
2052 rs->ti->error = "raid set already reshaping!";
2053 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2054 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2055 else
2056 return 0;
2057
2058 return -EPERM;
2059}
2060
2061static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2062{
2063 BUG_ON(!rdev->sb_page);
2064
2065 if (rdev->sb_loaded && !force_reload)
2066 return 0;
2067
2068 rdev->sb_loaded = 0;
2069
2070 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true)) {
2071 DMERR("Failed to read superblock of device at position %d",
2072 rdev->raid_disk);
2073 md_error(rdev->mddev, rdev);
2074 set_bit(Faulty, &rdev->flags);
2075 return -EIO;
2076 }
2077
2078 rdev->sb_loaded = 1;
2079
2080 return 0;
2081}
2082
2083static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2084{
2085 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2086 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2087
2088 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2089 int i = ARRAY_SIZE(sb->extended_failed_devices);
2090
2091 while (i--)
2092 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2093 }
2094}
2095
2096static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2097{
2098 int i = ARRAY_SIZE(sb->extended_failed_devices);
2099
2100 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2101 while (i--)
2102 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2103}
2104
2105/*
2106 * Synchronize the superblock members with the raid set properties
2107 *
2108 * All superblock data is little endian.
2109 */
2110static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2111{
2112 bool update_failed_devices = false;
2113 unsigned int i;
2114 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2115 struct dm_raid_superblock *sb;
2116 struct raid_set *rs = container_of(mddev, struct raid_set, md);
2117
2118 /* No metadata device, no superblock */
2119 if (!rdev->meta_bdev)
2120 return;
2121
2122 BUG_ON(!rdev->sb_page);
2123
2124 sb = page_address(rdev->sb_page);
2125
2126 sb_retrieve_failed_devices(sb, failed_devices);
2127
2128 for (i = 0; i < rs->raid_disks; i++)
2129 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2130 update_failed_devices = true;
2131 set_bit(i, (void *) failed_devices);
2132 }
2133
2134 if (update_failed_devices)
2135 sb_update_failed_devices(sb, failed_devices);
2136
2137 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2138 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2139
2140 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2141 sb->array_position = cpu_to_le32(rdev->raid_disk);
2142
2143 sb->events = cpu_to_le64(mddev->events);
2144
2145 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2146 sb->array_resync_offset = cpu_to_le64(mddev->resync_offset);
2147
2148 sb->level = cpu_to_le32(mddev->level);
2149 sb->layout = cpu_to_le32(mddev->layout);
2150 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2151
2152 /********************************************************************
2153 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2154 *
2155 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2156 */
2157 sb->new_level = cpu_to_le32(mddev->new_level);
2158 sb->new_layout = cpu_to_le32(mddev->new_layout);
2159 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2160
2161 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2162
2163 smp_rmb(); /* Make sure we access most recent reshape position */
2164 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2165 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2166 /* Flag ongoing reshape */
2167 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2168
2169 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2170 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2171 } else {
2172 /* Clear reshape flags */
2173 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2174 }
2175
2176 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2177 sb->data_offset = cpu_to_le64(rdev->data_offset);
2178 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2179 sb->sectors = cpu_to_le64(rdev->sectors);
2180 sb->incompat_features = cpu_to_le32(0);
2181
2182 /* Zero out the rest of the payload after the size of the superblock */
2183 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2184}
2185
2186/*
2187 * super_load
2188 *
2189 * This function creates a superblock if one is not found on the device
2190 * and will decide which superblock to use if there's a choice.
2191 *
2192 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2193 */
2194static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2195{
2196 int r;
2197 struct dm_raid_superblock *sb;
2198 struct dm_raid_superblock *refsb;
2199 uint64_t events_sb, events_refsb;
2200
2201 r = read_disk_sb(rdev, rdev->sb_size, false);
2202 if (r)
2203 return r;
2204
2205 sb = page_address(rdev->sb_page);
2206
2207 /*
2208 * Two cases that we want to write new superblocks and rebuild:
2209 * 1) New device (no matching magic number)
2210 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2211 */
2212 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2213 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2214 super_sync(rdev->mddev, rdev);
2215
2216 set_bit(FirstUse, &rdev->flags);
2217 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2218
2219 /* Force writing of superblocks to disk */
2220 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2221
2222 /* Any superblock is better than none, choose that if given */
2223 return refdev ? 0 : 1;
2224 }
2225
2226 if (!refdev)
2227 return 1;
2228
2229 events_sb = le64_to_cpu(sb->events);
2230
2231 refsb = page_address(refdev->sb_page);
2232 events_refsb = le64_to_cpu(refsb->events);
2233
2234 return (events_sb > events_refsb) ? 1 : 0;
2235}
2236
2237static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2238{
2239 int role;
2240 struct mddev *mddev = &rs->md;
2241 uint64_t events_sb;
2242 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2243 struct dm_raid_superblock *sb;
2244 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2245 struct md_rdev *r;
2246 struct dm_raid_superblock *sb2;
2247
2248 sb = page_address(rdev->sb_page);
2249 events_sb = le64_to_cpu(sb->events);
2250
2251 /*
2252 * Initialise to 1 if this is a new superblock.
2253 */
2254 mddev->events = events_sb ? : 1;
2255
2256 mddev->reshape_position = MaxSector;
2257
2258 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2259 mddev->level = le32_to_cpu(sb->level);
2260 mddev->layout = le32_to_cpu(sb->layout);
2261 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2262
2263 /*
2264 * Reshaping is supported, e.g. reshape_position is valid
2265 * in superblock and superblock content is authoritative.
2266 */
2267 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2268 /* Superblock is authoritative wrt given raid set layout! */
2269 mddev->new_level = le32_to_cpu(sb->new_level);
2270 mddev->new_layout = le32_to_cpu(sb->new_layout);
2271 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2272 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2273 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2274
2275 /* raid was reshaping and got interrupted */
2276 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2277 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2278 DMERR("Reshape requested but raid set is still reshaping");
2279 return -EINVAL;
2280 }
2281
2282 if (mddev->delta_disks < 0 ||
2283 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2284 mddev->reshape_backwards = 1;
2285 else
2286 mddev->reshape_backwards = 0;
2287
2288 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2289 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2290 if (!rs->raid_type)
2291 return -EINVAL;
2292 }
2293
2294 } else {
2295 /*
2296 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2297 */
2298 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2299 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2300
2301 if (rs_takeover_requested(rs)) {
2302 if (rt_cur && rt_new)
2303 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2304 rt_cur->name, rt_new->name);
2305 else
2306 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2307 return -EINVAL;
2308 } else if (rs_reshape_requested(rs)) {
2309 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2310 if (mddev->layout != mddev->new_layout) {
2311 if (rt_cur && rt_new)
2312 DMERR(" current layout %s vs new layout %s",
2313 rt_cur->name, rt_new->name);
2314 else
2315 DMERR(" current layout 0x%X vs new layout 0x%X",
2316 le32_to_cpu(sb->layout), mddev->new_layout);
2317 }
2318 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2319 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2320 mddev->chunk_sectors, mddev->new_chunk_sectors);
2321 if (rs->delta_disks)
2322 DMERR(" current %u disks vs new %u disks",
2323 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2324 if (rs_is_raid10(rs)) {
2325 DMERR(" Old layout: %s w/ %u copies",
2326 raid10_md_layout_to_format(mddev->layout),
2327 raid10_md_layout_to_copies(mddev->layout));
2328 DMERR(" New layout: %s w/ %u copies",
2329 raid10_md_layout_to_format(mddev->new_layout),
2330 raid10_md_layout_to_copies(mddev->new_layout));
2331 }
2332 return -EINVAL;
2333 }
2334
2335 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2336 }
2337
2338 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2339 mddev->resync_offset = le64_to_cpu(sb->array_resync_offset);
2340
2341 /*
2342 * During load, we set FirstUse if a new superblock was written.
2343 * There are two reasons we might not have a superblock:
2344 * 1) The raid set is brand new - in which case, all of the
2345 * devices must have their In_sync bit set. Also,
2346 * resync_offset must be 0, unless forced.
2347 * 2) This is a new device being added to an old raid set
2348 * and the new device needs to be rebuilt - in which
2349 * case the In_sync bit will /not/ be set and
2350 * resync_offset must be MaxSector.
2351 * 3) This is/are a new device(s) being added to an old
2352 * raid set during takeover to a higher raid level
2353 * to provide capacity for redundancy or during reshape
2354 * to add capacity to grow the raid set.
2355 */
2356 rdev_for_each(r, mddev) {
2357 if (test_bit(Journal, &rdev->flags))
2358 continue;
2359
2360 if (test_bit(FirstUse, &r->flags))
2361 new_devs++;
2362
2363 if (!test_bit(In_sync, &r->flags)) {
2364 DMINFO("Device %d specified for rebuild; clearing superblock",
2365 r->raid_disk);
2366 rebuilds++;
2367
2368 if (test_bit(FirstUse, &r->flags))
2369 rebuild_and_new++;
2370 }
2371 }
2372
2373 if (new_devs == rs->raid_disks || !rebuilds) {
2374 /* Replace a broken device */
2375 if (new_devs == rs->raid_disks) {
2376 DMINFO("Superblocks created for new raid set");
2377 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2378 } else if (new_devs != rebuilds &&
2379 new_devs != rs->delta_disks) {
2380 DMERR("New device injected into existing raid set without "
2381 "'delta_disks' or 'rebuild' parameter specified");
2382 return -EINVAL;
2383 }
2384 } else if (new_devs && new_devs != rebuilds) {
2385 DMERR("%u 'rebuild' devices cannot be injected into"
2386 " a raid set with %u other first-time devices",
2387 rebuilds, new_devs);
2388 return -EINVAL;
2389 } else if (rebuilds) {
2390 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2391 DMERR("new device%s provided without 'rebuild'",
2392 new_devs > 1 ? "s" : "");
2393 return -EINVAL;
2394 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2395 DMERR("'rebuild' specified while raid set is not in-sync (resync_offset=%llu)",
2396 (unsigned long long) mddev->resync_offset);
2397 return -EINVAL;
2398 } else if (rs_is_reshaping(rs)) {
2399 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2400 (unsigned long long) mddev->reshape_position);
2401 return -EINVAL;
2402 }
2403 }
2404
2405 /*
2406 * Now we set the Faulty bit for those devices that are
2407 * recorded in the superblock as failed.
2408 */
2409 sb_retrieve_failed_devices(sb, failed_devices);
2410 rdev_for_each(r, mddev) {
2411 if (test_bit(Journal, &r->flags) ||
2412 !r->sb_page)
2413 continue;
2414 sb2 = page_address(r->sb_page);
2415 sb2->failed_devices = 0;
2416 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2417
2418 /*
2419 * Check for any device re-ordering.
2420 */
2421 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2422 role = le32_to_cpu(sb2->array_position);
2423 if (role < 0)
2424 continue;
2425
2426 if (role != r->raid_disk) {
2427 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2428 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2429 rs->raid_disks % rs->raid10_copies) {
2430 rs->ti->error =
2431 "Cannot change raid10 near set to odd # of devices!";
2432 return -EINVAL;
2433 }
2434
2435 sb2->array_position = cpu_to_le32(r->raid_disk);
2436
2437 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2438 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2439 !rt_is_raid1(rs->raid_type)) {
2440 rs->ti->error = "Cannot change device positions in raid set";
2441 return -EINVAL;
2442 }
2443
2444 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2445 }
2446
2447 /*
2448 * Partial recovery is performed on
2449 * returning failed devices.
2450 */
2451 if (test_bit(role, (void *) failed_devices))
2452 set_bit(Faulty, &r->flags);
2453 }
2454 }
2455
2456 return 0;
2457}
2458
2459static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2460{
2461 struct mddev *mddev = &rs->md;
2462 struct dm_raid_superblock *sb;
2463
2464 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2465 return 0;
2466
2467 sb = page_address(rdev->sb_page);
2468
2469 /*
2470 * If mddev->events is not set, we know we have not yet initialized
2471 * the array.
2472 */
2473 if (!mddev->events && super_init_validation(rs, rdev))
2474 return -EINVAL;
2475
2476 if (le32_to_cpu(sb->compat_features) &&
2477 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2478 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2479 return -EINVAL;
2480 }
2481
2482 if (sb->incompat_features) {
2483 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2484 return -EINVAL;
2485 }
2486
2487 /* Enable bitmap creation on @rs unless no metadevs or raid0 or journaled raid4/5/6 set. */
2488 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2489 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2490
2491 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2492 /*
2493 * Retrieve rdev size stored in superblock to be prepared for shrink.
2494 * Check extended superblock members are present otherwise the size
2495 * will not be set!
2496 */
2497 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2498 rdev->sectors = le64_to_cpu(sb->sectors);
2499
2500 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2501 if (rdev->recovery_offset == MaxSector)
2502 set_bit(In_sync, &rdev->flags);
2503 /*
2504 * If no reshape in progress -> we're recovering single
2505 * disk(s) and have to set the device(s) to out-of-sync
2506 */
2507 else if (!rs_is_reshaping(rs))
2508 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2509 }
2510
2511 /*
2512 * If a device comes back, set it as not In_sync and no longer faulty.
2513 */
2514 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2515 rdev->recovery_offset = 0;
2516 clear_bit(In_sync, &rdev->flags);
2517 rdev->saved_raid_disk = rdev->raid_disk;
2518 }
2519
2520 /* Reshape support -> restore respective data offsets */
2521 rdev->data_offset = le64_to_cpu(sb->data_offset);
2522 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2523
2524 return 0;
2525}
2526
2527/*
2528 * Analyse superblocks and select the freshest.
2529 */
2530static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2531{
2532 int r;
2533 struct md_rdev *rdev, *freshest;
2534 struct mddev *mddev = &rs->md;
2535
2536 /* Respect resynchronization requested with "sync" argument. */
2537 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2538 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2539
2540 freshest = NULL;
2541 rdev_for_each(rdev, mddev) {
2542 if (test_bit(Journal, &rdev->flags))
2543 continue;
2544
2545 if (!rdev->meta_bdev)
2546 continue;
2547
2548 /* Set superblock offset/size for metadata device. */
2549 rdev->sb_start = 0;
2550 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2551 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2552 DMERR("superblock size of a logical block is no longer valid");
2553 return -EINVAL;
2554 }
2555
2556 /*
2557 * Skipping super_load due to CTR_FLAG_SYNC will cause
2558 * the array to undergo initialization again as
2559 * though it were new. This is the intended effect
2560 * of the "sync" directive.
2561 *
2562 * With reshaping capability added, we must ensure that
2563 * the "sync" directive is disallowed during the reshape.
2564 */
2565 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2566 continue;
2567
2568 r = super_load(rdev, freshest);
2569
2570 switch (r) {
2571 case 1:
2572 freshest = rdev;
2573 break;
2574 case 0:
2575 break;
2576 default:
2577 /* This is a failure to read the superblock from the metadata device. */
2578 /*
2579 * We have to keep any raid0 data/metadata device pairs or
2580 * the MD raid0 personality will fail to start the array.
2581 */
2582 if (rs_is_raid0(rs))
2583 continue;
2584
2585 /*
2586 * We keep the dm_devs to be able to emit the device tuple
2587 * properly on the table line in raid_status() (rather than
2588 * mistakenly acting as if '- -' got passed into the constructor).
2589 *
2590 * The rdev has to stay on the same_set list to allow for
2591 * the attempt to restore faulty devices on second resume.
2592 */
2593 rdev->raid_disk = rdev->saved_raid_disk = -1;
2594 break;
2595 }
2596 }
2597
2598 if (!freshest)
2599 return 0;
2600
2601 /*
2602 * Validation of the freshest device provides the source of
2603 * validation for the remaining devices.
2604 */
2605 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2606 if (super_validate(rs, freshest))
2607 return -EINVAL;
2608
2609 if (validate_raid_redundancy(rs)) {
2610 rs->ti->error = "Insufficient redundancy to activate array";
2611 return -EINVAL;
2612 }
2613
2614 rdev_for_each(rdev, mddev)
2615 if (!test_bit(Journal, &rdev->flags) &&
2616 rdev != freshest &&
2617 super_validate(rs, rdev))
2618 return -EINVAL;
2619 return 0;
2620}
2621
2622/*
2623 * Adjust data_offset and new_data_offset on all disk members of @rs
2624 * for out of place reshaping if requested by constructor
2625 *
2626 * We need free space at the beginning of each raid disk for forward
2627 * and at the end for backward reshapes which userspace has to provide
2628 * via remapping/reordering of space.
2629 */
2630static int rs_adjust_data_offsets(struct raid_set *rs)
2631{
2632 sector_t data_offset = 0, new_data_offset = 0;
2633 struct md_rdev *rdev;
2634
2635 /* Constructor did not request data offset change */
2636 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2637 if (!rs_is_reshapable(rs))
2638 goto out;
2639
2640 return 0;
2641 }
2642
2643 /* HM FIXME: get In_Sync raid_dev? */
2644 rdev = &rs->dev[0].rdev;
2645
2646 if (rs->delta_disks < 0) {
2647 /*
2648 * Removing disks (reshaping backwards):
2649 *
2650 * - before reshape: data is at offset 0 and free space
2651 * is at end of each component LV
2652 *
2653 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2654 */
2655 data_offset = 0;
2656 new_data_offset = rs->data_offset;
2657
2658 } else if (rs->delta_disks > 0) {
2659 /*
2660 * Adding disks (reshaping forwards):
2661 *
2662 * - before reshape: data is at offset rs->data_offset != 0 and
2663 * free space is at begin of each component LV
2664 *
2665 * - after reshape: data is at offset 0 on each component LV
2666 */
2667 data_offset = rs->data_offset;
2668 new_data_offset = 0;
2669
2670 } else {
2671 /*
2672 * User space passes in 0 for data offset after having removed reshape space
2673 *
2674 * - or - (data offset != 0)
2675 *
2676 * Changing RAID layout or chunk size -> toggle offsets
2677 *
2678 * - before reshape: data is at offset rs->data_offset 0 and
2679 * free space is at end of each component LV
2680 * -or-
2681 * data is at offset rs->data_offset != 0 and
2682 * free space is at begin of each component LV
2683 *
2684 * - after reshape: data is at offset 0 if it was at offset != 0
2685 * or at offset != 0 if it was at offset 0
2686 * on each component LV
2687 *
2688 */
2689 data_offset = rs->data_offset ? rdev->data_offset : 0;
2690 new_data_offset = data_offset ? 0 : rs->data_offset;
2691 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2692 }
2693
2694 /*
2695 * Make sure we got a minimum amount of free sectors per device
2696 */
2697 if (rs->data_offset &&
2698 bdev_nr_sectors(rdev->bdev) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2699 rs->ti->error = data_offset ? "No space for forward reshape" :
2700 "No space for backward reshape";
2701 return -ENOSPC;
2702 }
2703out:
2704 /*
2705 * Raise resync_offset in case data_offset != 0 to
2706 * avoid false recovery positives in the constructor.
2707 */
2708 if (rs->md.resync_offset < rs->md.dev_sectors)
2709 rs->md.resync_offset += rs->dev[0].rdev.data_offset;
2710
2711 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2712 rdev_for_each(rdev, &rs->md) {
2713 if (!test_bit(Journal, &rdev->flags)) {
2714 rdev->data_offset = data_offset;
2715 rdev->new_data_offset = new_data_offset;
2716 }
2717 }
2718
2719 return 0;
2720}
2721
2722/* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2723static void __reorder_raid_disk_indexes(struct raid_set *rs)
2724{
2725 int i = 0;
2726 struct md_rdev *rdev;
2727
2728 rdev_for_each(rdev, &rs->md) {
2729 if (!test_bit(Journal, &rdev->flags)) {
2730 rdev->raid_disk = i++;
2731 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2732 }
2733 }
2734}
2735
2736/*
2737 * Setup @rs for takeover by a different raid level
2738 */
2739static int rs_setup_takeover(struct raid_set *rs)
2740{
2741 struct mddev *mddev = &rs->md;
2742 struct md_rdev *rdev;
2743 unsigned int d = mddev->raid_disks = rs->raid_disks;
2744 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2745
2746 if (rt_is_raid10(rs->raid_type)) {
2747 if (rs_is_raid0(rs)) {
2748 /* Userpace reordered disks -> adjust raid_disk indexes */
2749 __reorder_raid_disk_indexes(rs);
2750
2751 /* raid0 -> raid10_far layout */
2752 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2753 rs->raid10_copies);
2754 } else if (rs_is_raid1(rs))
2755 /* raid1 -> raid10_near layout */
2756 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2757 rs->raid_disks);
2758 else
2759 return -EINVAL;
2760
2761 }
2762
2763 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2764 mddev->resync_offset = MaxSector;
2765
2766 while (d--) {
2767 rdev = &rs->dev[d].rdev;
2768
2769 if (test_bit(d, (void *) rs->rebuild_disks)) {
2770 clear_bit(In_sync, &rdev->flags);
2771 clear_bit(Faulty, &rdev->flags);
2772 mddev->resync_offset = rdev->recovery_offset = 0;
2773 /* Bitmap has to be created when we do an "up" takeover */
2774 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2775 }
2776
2777 rdev->new_data_offset = new_data_offset;
2778 }
2779
2780 return 0;
2781}
2782
2783/* Prepare @rs for reshape */
2784static int rs_prepare_reshape(struct raid_set *rs)
2785{
2786 bool reshape;
2787 struct mddev *mddev = &rs->md;
2788
2789 if (rs_is_raid10(rs)) {
2790 if (rs->raid_disks != mddev->raid_disks &&
2791 __is_raid10_near(mddev->layout) &&
2792 rs->raid10_copies &&
2793 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2794 /*
2795 * raid disk have to be multiple of data copies to allow this conversion,
2796 *
2797 * This is actually not a reshape it is a
2798 * rebuild of any additional mirrors per group
2799 */
2800 if (rs->raid_disks % rs->raid10_copies) {
2801 rs->ti->error = "Can't reshape raid10 mirror groups";
2802 return -EINVAL;
2803 }
2804
2805 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2806 __reorder_raid_disk_indexes(rs);
2807 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2808 rs->raid10_copies);
2809 mddev->new_layout = mddev->layout;
2810 reshape = false;
2811 } else
2812 reshape = true;
2813
2814 } else if (rs_is_raid456(rs))
2815 reshape = true;
2816
2817 else if (rs_is_raid1(rs)) {
2818 if (rs->delta_disks) {
2819 /* Process raid1 via delta_disks */
2820 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2821 reshape = true;
2822 } else {
2823 /* Process raid1 without delta_disks */
2824 mddev->raid_disks = rs->raid_disks;
2825 reshape = false;
2826 }
2827 } else {
2828 rs->ti->error = "Called with bogus raid type";
2829 return -EINVAL;
2830 }
2831
2832 if (reshape) {
2833 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2834 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2835 } else if (mddev->raid_disks < rs->raid_disks)
2836 /* Create new superblocks and bitmaps, if any new disks */
2837 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2838
2839 return 0;
2840}
2841
2842/*
2843 * Reshape:
2844 * - change raid layout
2845 * - change chunk size
2846 * - add disks
2847 * - remove disks
2848 */
2849static int rs_setup_reshape(struct raid_set *rs)
2850{
2851 int r = 0;
2852 unsigned int cur_raid_devs, d;
2853 sector_t reshape_sectors = _get_reshape_sectors(rs);
2854 struct mddev *mddev = &rs->md;
2855 struct md_rdev *rdev;
2856
2857 mddev->delta_disks = rs->delta_disks;
2858 cur_raid_devs = mddev->raid_disks;
2859
2860 /* Ignore impossible layout change whilst adding/removing disks */
2861 if (mddev->delta_disks &&
2862 mddev->layout != mddev->new_layout) {
2863 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2864 mddev->new_layout = mddev->layout;
2865 }
2866
2867 /*
2868 * Adjust array size:
2869 *
2870 * - in case of adding disk(s), array size has
2871 * to grow after the disk adding reshape,
2872 * which'll happen in the event handler;
2873 * reshape will happen forward, so space has to
2874 * be available at the beginning of each disk
2875 *
2876 * - in case of removing disk(s), array size
2877 * has to shrink before starting the reshape,
2878 * which'll happen here;
2879 * reshape will happen backward, so space has to
2880 * be available at the end of each disk
2881 *
2882 * - data_offset and new_data_offset are
2883 * adjusted for aforementioned out of place
2884 * reshaping based on userspace passing in
2885 * the "data_offset <sectors>" key/value
2886 * pair via the constructor
2887 */
2888
2889 /* Add disk(s) */
2890 if (rs->delta_disks > 0) {
2891 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2892 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2893 rdev = &rs->dev[d].rdev;
2894 clear_bit(In_sync, &rdev->flags);
2895
2896 /*
2897 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2898 * by md, which'll store that erroneously in the superblock on reshape
2899 */
2900 rdev->saved_raid_disk = -1;
2901 rdev->raid_disk = d;
2902
2903 rdev->sectors = mddev->dev_sectors;
2904 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2905 }
2906
2907 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2908
2909 /* Remove disk(s) */
2910 } else if (rs->delta_disks < 0) {
2911 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
2912 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2913
2914 /* Change layout and/or chunk size */
2915 } else {
2916 /*
2917 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2918 *
2919 * keeping number of disks and do layout change ->
2920 *
2921 * toggle reshape_backward depending on data_offset:
2922 *
2923 * - free space upfront -> reshape forward
2924 *
2925 * - free space at the end -> reshape backward
2926 *
2927 *
2928 * This utilizes free reshape space avoiding the need
2929 * for userspace to move (parts of) LV segments in
2930 * case of layout/chunksize change (for disk
2931 * adding/removing reshape space has to be at
2932 * the proper address (see above with delta_disks):
2933 *
2934 * add disk(s) -> begin
2935 * remove disk(s)-> end
2936 */
2937 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2938 }
2939
2940 /*
2941 * Adjust device size for forward reshape
2942 * because md_finish_reshape() reduces it.
2943 */
2944 if (!mddev->reshape_backwards)
2945 rdev_for_each(rdev, &rs->md)
2946 if (!test_bit(Journal, &rdev->flags))
2947 rdev->sectors += reshape_sectors;
2948
2949 return r;
2950}
2951
2952/*
2953 * If the md resync thread has updated superblock with max reshape position
2954 * at the end of a reshape but not (yet) reset the layout configuration
2955 * changes -> reset the latter.
2956 */
2957static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2958{
2959 if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2960 rs_set_cur(rs);
2961 rs->md.delta_disks = 0;
2962 rs->md.reshape_backwards = 0;
2963 }
2964}
2965
2966/*
2967 * Enable/disable discard support on RAID set depending on
2968 * RAID level and discard properties of underlying RAID members.
2969 */
2970static void configure_discard_support(struct raid_set *rs)
2971{
2972 int i;
2973 bool raid456;
2974 struct dm_target *ti = rs->ti;
2975
2976 /*
2977 * XXX: RAID level 4,5,6 require zeroing for safety.
2978 */
2979 raid456 = rs_is_raid456(rs);
2980
2981 for (i = 0; i < rs->raid_disks; i++) {
2982 if (!rs->dev[i].rdev.bdev ||
2983 !bdev_max_discard_sectors(rs->dev[i].rdev.bdev))
2984 return;
2985
2986 if (raid456) {
2987 if (!devices_handle_discard_safely) {
2988 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2989 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2990 return;
2991 }
2992 }
2993 }
2994
2995 ti->num_discard_bios = 1;
2996}
2997
2998/*
2999 * Construct a RAID0/1/10/4/5/6 mapping:
3000 * Args:
3001 * <raid_type> <#raid_params> <raid_params>{0,} \
3002 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
3003 *
3004 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
3005 * details on possible <raid_params>.
3006 *
3007 * Userspace is free to initialize the metadata devices, hence the superblocks to
3008 * enforce recreation based on the passed in table parameters.
3009 *
3010 */
3011static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3012{
3013 int r;
3014 bool resize = false;
3015 struct raid_type *rt;
3016 unsigned int num_raid_params, num_raid_devs;
3017 sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
3018 struct raid_set *rs = NULL;
3019 const char *arg;
3020 struct rs_layout rs_layout;
3021 struct dm_arg_set as = { argc, argv }, as_nrd;
3022 struct dm_arg _args[] = {
3023 { 0, as.argc, "Cannot understand number of raid parameters" },
3024 { 1, 254, "Cannot understand number of raid devices parameters" }
3025 };
3026
3027 arg = dm_shift_arg(&as);
3028 if (!arg) {
3029 ti->error = "No arguments";
3030 return -EINVAL;
3031 }
3032
3033 rt = get_raid_type(arg);
3034 if (!rt) {
3035 ti->error = "Unrecognised raid_type";
3036 return -EINVAL;
3037 }
3038
3039 /* Must have <#raid_params> */
3040 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3041 return -EINVAL;
3042
3043 /* number of raid device tupples <meta_dev data_dev> */
3044 as_nrd = as;
3045 dm_consume_args(&as_nrd, num_raid_params);
3046 _args[1].max = (as_nrd.argc - 1) / 2;
3047 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3048 return -EINVAL;
3049
3050 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3051 ti->error = "Invalid number of supplied raid devices";
3052 return -EINVAL;
3053 }
3054
3055 rs = raid_set_alloc(ti, rt, num_raid_devs);
3056 if (IS_ERR(rs))
3057 return PTR_ERR(rs);
3058
3059 r = parse_raid_params(rs, &as, num_raid_params);
3060 if (r)
3061 goto bad;
3062
3063 r = parse_dev_params(rs, &as);
3064 if (r)
3065 goto bad;
3066
3067 rs->md.sync_super = super_sync;
3068
3069 /*
3070 * Calculate ctr requested array and device sizes to allow
3071 * for superblock analysis needing device sizes defined.
3072 *
3073 * Any existing superblock will overwrite the array and device sizes
3074 */
3075 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3076 if (r)
3077 goto bad;
3078
3079 /* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3080 rs->array_sectors = rs->md.array_sectors;
3081 rs->dev_sectors = rs->md.dev_sectors;
3082
3083 /*
3084 * Backup any new raid set level, layout, ...
3085 * requested to be able to compare to superblock
3086 * members for conversion decisions.
3087 */
3088 rs_config_backup(rs, &rs_layout);
3089
3090 r = analyse_superblocks(ti, rs);
3091 if (r)
3092 goto bad;
3093
3094 /* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3095 sb_array_sectors = rs->md.array_sectors;
3096 rdev_sectors = __rdev_sectors(rs);
3097 if (!rdev_sectors) {
3098 ti->error = "Invalid rdev size";
3099 r = -EINVAL;
3100 goto bad;
3101 }
3102
3103
3104 reshape_sectors = _get_reshape_sectors(rs);
3105 if (rs->dev_sectors != rdev_sectors) {
3106 resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3107 if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3108 set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3109 }
3110
3111 INIT_WORK(&rs->md.event_work, do_table_event);
3112 ti->private = rs;
3113 ti->num_flush_bios = 1;
3114 ti->needs_bio_set_dev = true;
3115
3116 /* Restore any requested new layout for conversion decision */
3117 rs_config_restore(rs, &rs_layout);
3118
3119 /*
3120 * Now that we have any superblock metadata available,
3121 * check for new, recovering, reshaping, to be taken over,
3122 * to be reshaped or an existing, unchanged raid set to
3123 * run in sequence.
3124 */
3125 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3126 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3127 if (rs_is_raid6(rs) &&
3128 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3129 ti->error = "'nosync' not allowed for new raid6 set";
3130 r = -EINVAL;
3131 goto bad;
3132 }
3133 rs_setup_recovery(rs, 0);
3134 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3135 rs_set_new(rs);
3136 } else if (rs_is_recovering(rs)) {
3137 /* A recovering raid set may be resized */
3138 goto size_check;
3139 } else if (rs_is_reshaping(rs)) {
3140 /* Have to reject size change request during reshape */
3141 if (resize) {
3142 ti->error = "Can't resize a reshaping raid set";
3143 r = -EPERM;
3144 goto bad;
3145 }
3146 /* skip setup rs */
3147 } else if (rs_takeover_requested(rs)) {
3148 if (rs_is_reshaping(rs)) {
3149 ti->error = "Can't takeover a reshaping raid set";
3150 r = -EPERM;
3151 goto bad;
3152 }
3153
3154 /* We can't takeover a journaled raid4/5/6 */
3155 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3156 ti->error = "Can't takeover a journaled raid4/5/6 set";
3157 r = -EPERM;
3158 goto bad;
3159 }
3160
3161 /*
3162 * If a takeover is needed, userspace sets any additional
3163 * devices to rebuild and we can check for a valid request here.
3164 *
3165 * If acceptable, set the level to the new requested
3166 * one, prohibit requesting recovery, allow the raid
3167 * set to run and store superblocks during resume.
3168 */
3169 r = rs_check_takeover(rs);
3170 if (r)
3171 goto bad;
3172
3173 r = rs_setup_takeover(rs);
3174 if (r)
3175 goto bad;
3176
3177 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3178 /* Takeover ain't recovery, so disable recovery */
3179 rs_setup_recovery(rs, MaxSector);
3180 rs_set_new(rs);
3181 } else if (rs_reshape_requested(rs)) {
3182 /* Only request grow on raid set size extensions, not on reshapes. */
3183 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3184
3185 /*
3186 * No need to check for 'ongoing' takeover here, because takeover
3187 * is an instant operation as oposed to an ongoing reshape.
3188 */
3189
3190 /* We can't reshape a journaled raid4/5/6 */
3191 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3192 ti->error = "Can't reshape a journaled raid4/5/6 set";
3193 r = -EPERM;
3194 goto bad;
3195 }
3196
3197 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3198 if (reshape_sectors || rs_is_raid1(rs)) {
3199 /*
3200 * We can only prepare for a reshape here, because the
3201 * raid set needs to run to provide the respective reshape
3202 * check functions via its MD personality instance.
3203 *
3204 * So do the reshape check after md_run() succeeded.
3205 */
3206 r = rs_prepare_reshape(rs);
3207 if (r)
3208 goto bad;
3209
3210 /* Reshaping ain't recovery, so disable recovery */
3211 rs_setup_recovery(rs, MaxSector);
3212 }
3213 rs_set_cur(rs);
3214 } else {
3215size_check:
3216 /* May not set recovery when a device rebuild is requested */
3217 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3218 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3219 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3220 rs_setup_recovery(rs, MaxSector);
3221 } else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3222 /*
3223 * Set raid set to current size, i.e. size as of
3224 * superblocks to grow to larger size in preresume.
3225 */
3226 r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3227 if (r)
3228 goto bad;
3229
3230 rs_setup_recovery(rs, rs->md.resync_offset < rs->md.dev_sectors ? rs->md.resync_offset : rs->md.dev_sectors);
3231 } else {
3232 /* This is no size change or it is shrinking, update size and record in superblocks */
3233 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3234 if (r)
3235 goto bad;
3236
3237 if (sb_array_sectors > rs->array_sectors)
3238 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3239 }
3240 rs_set_cur(rs);
3241 }
3242
3243 /* If constructor requested it, change data and new_data offsets */
3244 r = rs_adjust_data_offsets(rs);
3245 if (r)
3246 goto bad;
3247
3248 /* Catch any inconclusive reshape superblock content. */
3249 rs_reset_inconclusive_reshape(rs);
3250
3251 /* Start raid set read-only and assumed clean to change in raid_resume() */
3252 rs->md.ro = MD_RDONLY;
3253 rs->md.in_sync = 1;
3254
3255 /* Has to be held on running the array */
3256 mddev_suspend_and_lock_nointr(&rs->md);
3257
3258 /* Keep array frozen until resume. */
3259 md_frozen_sync_thread(&rs->md);
3260
3261 r = md_run(&rs->md);
3262 rs->md.in_sync = 0; /* Assume already marked dirty */
3263 if (r) {
3264 ti->error = "Failed to run raid array";
3265 mddev_unlock(&rs->md);
3266 goto bad;
3267 }
3268
3269 r = md_start(&rs->md);
3270 if (r) {
3271 ti->error = "Failed to start raid array";
3272 goto bad_unlock;
3273 }
3274
3275 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3276 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3277 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3278 if (r) {
3279 ti->error = "Failed to set raid4/5/6 journal mode";
3280 goto bad_unlock;
3281 }
3282 }
3283
3284 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3285
3286 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3287 if (rs_is_raid456(rs)) {
3288 r = rs_set_raid456_stripe_cache(rs);
3289 if (r)
3290 goto bad_unlock;
3291 }
3292
3293 /* Now do an early reshape check */
3294 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3295 r = rs_check_reshape(rs);
3296 if (r)
3297 goto bad_unlock;
3298
3299 /* Restore new, ctr requested layout to perform check */
3300 rs_config_restore(rs, &rs_layout);
3301
3302 if (rs->md.pers->start_reshape) {
3303 r = rs->md.pers->check_reshape(&rs->md);
3304 if (r) {
3305 ti->error = "Reshape check failed";
3306 goto bad_unlock;
3307 }
3308 }
3309 }
3310
3311 /* Disable/enable discard support on raid set. */
3312 configure_discard_support(rs);
3313 rs->md.dm_gendisk = dm_disk(dm_table_get_md(ti->table));
3314
3315 mddev_unlock(&rs->md);
3316 return 0;
3317
3318bad_unlock:
3319 md_stop(&rs->md);
3320 mddev_unlock(&rs->md);
3321bad:
3322 raid_set_free(rs);
3323
3324 return r;
3325}
3326
3327static void raid_dtr(struct dm_target *ti)
3328{
3329 struct raid_set *rs = ti->private;
3330
3331 mddev_lock_nointr(&rs->md);
3332 md_stop(&rs->md);
3333 rs->md.dm_gendisk = NULL;
3334 mddev_unlock(&rs->md);
3335
3336 if (work_pending(&rs->md.event_work))
3337 flush_work(&rs->md.event_work);
3338 raid_set_free(rs);
3339}
3340
3341static int raid_map(struct dm_target *ti, struct bio *bio)
3342{
3343 struct raid_set *rs = ti->private;
3344 struct mddev *mddev = &rs->md;
3345
3346 /*
3347 * If we're reshaping to add disk(s), ti->len and
3348 * mddev->array_sectors will differ during the process
3349 * (ti->len > mddev->array_sectors), so we have to requeue
3350 * bios with addresses > mddev->array_sectors here or
3351 * there will occur accesses past EOD of the component
3352 * data images thus erroring the raid set.
3353 */
3354 if (unlikely(bio_has_data(bio) && bio_end_sector(bio) > mddev->array_sectors))
3355 return DM_MAPIO_REQUEUE;
3356
3357 if (unlikely(!md_handle_request(mddev, bio)))
3358 return DM_MAPIO_REQUEUE;
3359
3360 return DM_MAPIO_SUBMITTED;
3361}
3362
3363/* Return sync state string for @state */
3364enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3365static const char *sync_str(enum sync_state state)
3366{
3367 /* Has to be in above sync_state order! */
3368 static const char *sync_strs[] = {
3369 "frozen",
3370 "reshape",
3371 "resync",
3372 "check",
3373 "repair",
3374 "recover",
3375 "idle"
3376 };
3377
3378 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3379};
3380
3381/* Return enum sync_state for @mddev derived from @recovery flags */
3382static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3383{
3384 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3385 return st_frozen;
3386
3387 /* The MD sync thread can be done with io or be interrupted but still be running */
3388 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3389 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3390 (md_is_rdwr(mddev) && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3391 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3392 return st_reshape;
3393
3394 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3395 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3396 return st_resync;
3397 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3398 return st_check;
3399 return st_repair;
3400 }
3401
3402 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3403 return st_recover;
3404
3405 if (mddev->reshape_position != MaxSector)
3406 return st_reshape;
3407 }
3408
3409 return st_idle;
3410}
3411
3412/*
3413 * Return status string for @rdev
3414 *
3415 * Status characters:
3416 *
3417 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
3418 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3419 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3420 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3421 */
3422static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3423{
3424 if (!rdev->bdev)
3425 return "-";
3426 else if (test_bit(Faulty, &rdev->flags))
3427 return "D";
3428 else if (test_bit(Journal, &rdev->flags))
3429 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3430 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3431 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3432 !test_bit(In_sync, &rdev->flags)))
3433 return "a";
3434 else
3435 return "A";
3436}
3437
3438/* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3439static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3440 enum sync_state state, sector_t resync_max_sectors)
3441{
3442 sector_t r;
3443 struct mddev *mddev = &rs->md;
3444
3445 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3446 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3447
3448 if (rs_is_raid0(rs)) {
3449 r = resync_max_sectors;
3450 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3451
3452 } else {
3453 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3454 r = mddev->resync_offset;
3455 else
3456 r = mddev->curr_resync_completed;
3457
3458 if (state == st_idle && r >= resync_max_sectors) {
3459 /*
3460 * Sync complete.
3461 */
3462 /* In case we have finished recovering, the array is in sync. */
3463 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3464 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3465
3466 } else if (state == st_recover)
3467 /*
3468 * In case we are recovering, the array is not in sync
3469 * and health chars should show the recovering legs.
3470 *
3471 * Already retrieved recovery offset from curr_resync_completed above.
3472 */
3473 ;
3474
3475 else if (state == st_resync || state == st_reshape)
3476 /*
3477 * If "resync/reshape" is occurring, the raid set
3478 * is or may be out of sync hence the health
3479 * characters shall be 'a'.
3480 */
3481 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3482
3483 else if (state == st_check || state == st_repair)
3484 /*
3485 * If "check" or "repair" is occurring, the raid set has
3486 * undergone an initial sync and the health characters
3487 * should not be 'a' anymore.
3488 */
3489 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3490
3491 else if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3492 /*
3493 * We are idle and recovery is needed, prevent 'A' chars race
3494 * caused by components still set to in-sync by constructor.
3495 */
3496 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3497
3498 else {
3499 /*
3500 * We are idle and the raid set may be doing an initial
3501 * sync, or it may be rebuilding individual components.
3502 * If all the devices are In_sync, then it is the raid set
3503 * that is being initialized.
3504 */
3505 struct md_rdev *rdev;
3506
3507 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3508 rdev_for_each(rdev, mddev)
3509 if (!test_bit(Journal, &rdev->flags) &&
3510 !test_bit(In_sync, &rdev->flags)) {
3511 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3512 break;
3513 }
3514 }
3515 }
3516
3517 return min(r, resync_max_sectors);
3518}
3519
3520/* Helper to return @dev name or "-" if !@dev */
3521static const char *__get_dev_name(struct dm_dev *dev)
3522{
3523 return dev ? dev->name : "-";
3524}
3525
3526static void raid_status(struct dm_target *ti, status_type_t type,
3527 unsigned int status_flags, char *result, unsigned int maxlen)
3528{
3529 struct raid_set *rs = ti->private;
3530 struct mddev *mddev = &rs->md;
3531 struct r5conf *conf = rs_is_raid456(rs) ? mddev->private : NULL;
3532 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3533 unsigned long recovery;
3534 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3535 unsigned int sz = 0;
3536 unsigned int rebuild_writemostly_count = 0;
3537 sector_t progress, resync_max_sectors, resync_mismatches;
3538 enum sync_state state;
3539 struct raid_type *rt;
3540
3541 switch (type) {
3542 case STATUSTYPE_INFO:
3543 /* *Should* always succeed */
3544 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3545 if (!rt)
3546 return;
3547
3548 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3549
3550 /* Access most recent mddev properties for status output */
3551 smp_rmb();
3552 /* Get sensible max sectors even if raid set not yet started */
3553 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3554 mddev->resync_max_sectors : mddev->dev_sectors;
3555 recovery = rs->md.recovery;
3556 state = decipher_sync_action(mddev, recovery);
3557 progress = rs_get_progress(rs, recovery, state, resync_max_sectors);
3558 resync_mismatches = mddev->last_sync_action == ACTION_CHECK ?
3559 atomic64_read(&mddev->resync_mismatches) : 0;
3560
3561 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3562 for (i = 0; i < rs->raid_disks; i++)
3563 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3564
3565 /*
3566 * In-sync/Reshape ratio:
3567 * The in-sync ratio shows the progress of:
3568 * - Initializing the raid set
3569 * - Rebuilding a subset of devices of the raid set
3570 * The user can distinguish between the two by referring
3571 * to the status characters.
3572 *
3573 * The reshape ratio shows the progress of
3574 * changing the raid layout or the number of
3575 * disks of a raid set
3576 */
3577 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3578 (unsigned long long) resync_max_sectors);
3579
3580 /*
3581 * v1.5.0+:
3582 *
3583 * Sync action:
3584 * See Documentation/admin-guide/device-mapper/dm-raid.rst for
3585 * information on each of these states.
3586 */
3587 DMEMIT(" %s", sync_str(state));
3588
3589 /*
3590 * v1.5.0+:
3591 *
3592 * resync_mismatches/mismatch_cnt
3593 * This field shows the number of discrepancies found when
3594 * performing a "check" of the raid set.
3595 */
3596 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3597
3598 /*
3599 * v1.9.0+:
3600 *
3601 * data_offset (needed for out of space reshaping)
3602 * This field shows the data offset into the data
3603 * image LV where the first stripes data starts.
3604 *
3605 * We keep data_offset equal on all raid disks of the set,
3606 * so retrieving it from the first raid disk is sufficient.
3607 */
3608 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3609
3610 /*
3611 * v1.10.0+:
3612 */
3613 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3614 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3615 break;
3616
3617 case STATUSTYPE_TABLE:
3618 /* Report the table line string you would use to construct this raid set */
3619
3620 /*
3621 * Count any rebuild or writemostly argument pairs and subtract the
3622 * hweight count being added below of any rebuild and writemostly ctr flags.
3623 */
3624 for (i = 0; i < rs->raid_disks; i++) {
3625 rebuild_writemostly_count += (test_bit(i, (void *) rs->rebuild_disks) ? 2 : 0) +
3626 (test_bit(WriteMostly, &rs->dev[i].rdev.flags) ? 2 : 0);
3627 }
3628 rebuild_writemostly_count -= (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) ? 2 : 0) +
3629 (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags) ? 2 : 0);
3630 /* Calculate raid parameter count based on ^ rebuild/writemostly argument counts and ctr flags set. */
3631 raid_param_cnt += rebuild_writemostly_count +
3632 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3633 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3634 /* Emit table line */
3635 /* This has to be in the documented order for userspace! */
3636 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3637 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3638 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3639 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3640 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3641 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags))
3642 for (i = 0; i < rs->raid_disks; i++)
3643 if (test_bit(i, (void *) rs->rebuild_disks))
3644 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD), i);
3645 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3646 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3647 mddev->bitmap_info.daemon_sleep);
3648 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3649 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3650 mddev->sync_speed_min);
3651 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3652 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3653 mddev->sync_speed_max);
3654 if (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags))
3655 for (i = 0; i < rs->raid_disks; i++)
3656 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3657 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3658 rs->dev[i].rdev.raid_disk);
3659 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3660 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3661 mddev->bitmap_info.max_write_behind);
3662 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3663 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3664 max_nr_stripes);
3665 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3666 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3667 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3668 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3669 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3670 raid10_md_layout_to_copies(mddev->layout));
3671 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3672 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3673 raid10_md_layout_to_format(mddev->layout));
3674 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3675 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3676 max(rs->delta_disks, mddev->delta_disks));
3677 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3678 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3679 (unsigned long long) rs->data_offset);
3680 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3681 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3682 __get_dev_name(rs->journal_dev.dev));
3683 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3684 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3685 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3686 DMEMIT(" %d", rs->raid_disks);
3687 for (i = 0; i < rs->raid_disks; i++)
3688 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3689 __get_dev_name(rs->dev[i].data_dev));
3690 break;
3691
3692 case STATUSTYPE_IMA:
3693 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3694 if (!rt)
3695 return;
3696
3697 DMEMIT_TARGET_NAME_VERSION(ti->type);
3698 DMEMIT(",raid_type=%s,raid_disks=%d", rt->name, mddev->raid_disks);
3699
3700 /* Access most recent mddev properties for status output */
3701 smp_rmb();
3702 recovery = rs->md.recovery;
3703 state = decipher_sync_action(mddev, recovery);
3704 DMEMIT(",raid_state=%s", sync_str(state));
3705
3706 for (i = 0; i < rs->raid_disks; i++) {
3707 DMEMIT(",raid_device_%d_status=", i);
3708 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3709 }
3710
3711 if (rt_is_raid456(rt)) {
3712 DMEMIT(",journal_dev_mode=");
3713 switch (rs->journal_dev.mode) {
3714 case R5C_JOURNAL_MODE_WRITE_THROUGH:
3715 DMEMIT("%s",
3716 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_THROUGH].param);
3717 break;
3718 case R5C_JOURNAL_MODE_WRITE_BACK:
3719 DMEMIT("%s",
3720 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_BACK].param);
3721 break;
3722 default:
3723 DMEMIT("invalid");
3724 break;
3725 }
3726 }
3727 DMEMIT(";");
3728 break;
3729 }
3730}
3731
3732static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3733 char *result, unsigned int maxlen)
3734{
3735 struct raid_set *rs = ti->private;
3736 struct mddev *mddev = &rs->md;
3737 int ret = 0;
3738
3739 if (!mddev->pers || !mddev->pers->sync_request)
3740 return -EINVAL;
3741
3742 if (test_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags) ||
3743 test_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags))
3744 return -EBUSY;
3745
3746 if (!strcasecmp(argv[0], "frozen")) {
3747 ret = mddev_lock(mddev);
3748 if (ret)
3749 return ret;
3750
3751 md_frozen_sync_thread(mddev);
3752 mddev_unlock(mddev);
3753 } else if (!strcasecmp(argv[0], "idle")) {
3754 ret = mddev_lock(mddev);
3755 if (ret)
3756 return ret;
3757
3758 md_idle_sync_thread(mddev);
3759 mddev_unlock(mddev);
3760 }
3761
3762 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3763 if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3764 return -EBUSY;
3765 else if (!strcasecmp(argv[0], "resync"))
3766 ; /* MD_RECOVERY_NEEDED set below */
3767 else if (!strcasecmp(argv[0], "recover"))
3768 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3769 else {
3770 if (!strcasecmp(argv[0], "check")) {
3771 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3772 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3773 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3774 } else if (!strcasecmp(argv[0], "repair")) {
3775 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3776 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3777 } else
3778 return -EINVAL;
3779 }
3780 if (mddev->ro == MD_AUTO_READ) {
3781 /* A write to sync_action is enough to justify
3782 * canceling read-auto mode
3783 */
3784 mddev->ro = MD_RDWR;
3785 if (!mddev->suspended)
3786 md_wakeup_thread(mddev->sync_thread);
3787 }
3788 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3789 if (!mddev->suspended)
3790 md_wakeup_thread(mddev->thread);
3791
3792 return 0;
3793}
3794
3795static int raid_iterate_devices(struct dm_target *ti,
3796 iterate_devices_callout_fn fn, void *data)
3797{
3798 struct raid_set *rs = ti->private;
3799 unsigned int i;
3800 int r = 0;
3801
3802 for (i = 0; !r && i < rs->raid_disks; i++) {
3803 if (rs->dev[i].data_dev) {
3804 r = fn(ti, rs->dev[i].data_dev,
3805 0, /* No offset on data devs */
3806 rs->md.dev_sectors, data);
3807 }
3808 }
3809
3810 return r;
3811}
3812
3813static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3814{
3815 struct raid_set *rs = ti->private;
3816 unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3817
3818 if (chunk_size_bytes) {
3819 limits->io_min = chunk_size_bytes;
3820 limits->io_opt = chunk_size_bytes * mddev_data_stripes(rs);
3821 }
3822}
3823
3824static void raid_presuspend(struct dm_target *ti)
3825{
3826 struct raid_set *rs = ti->private;
3827 struct mddev *mddev = &rs->md;
3828
3829 /*
3830 * From now on, disallow raid_message() to change sync_thread until
3831 * resume, raid_postsuspend() is too late.
3832 */
3833 set_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
3834 set_bit(MD_DM_SUSPENDING, &mddev->flags);
3835
3836 if (!reshape_interrupted(mddev))
3837 return;
3838
3839 /*
3840 * For raid456, if reshape is interrupted, IO across reshape position
3841 * will never make progress, while caller will wait for IO to be done.
3842 * Inform raid456 to handle those IO to prevent deadlock.
3843 */
3844 if (mddev->pers && mddev->pers->prepare_suspend)
3845 mddev->pers->prepare_suspend(mddev);
3846}
3847
3848static void raid_presuspend_undo(struct dm_target *ti)
3849{
3850 struct raid_set *rs = ti->private;
3851 struct mddev *mddev = &rs->md;
3852
3853 clear_bit(MD_DM_SUSPENDING, &mddev->flags);
3854 clear_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
3855}
3856
3857static void raid_postsuspend(struct dm_target *ti)
3858{
3859 struct raid_set *rs = ti->private;
3860 struct mddev *mddev = &rs->md;
3861
3862 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3863 /*
3864 * sync_thread must be stopped during suspend, and writes have
3865 * to be stopped before suspending to avoid deadlocks.
3866 */
3867 md_stop_writes(&rs->md);
3868 mddev_suspend(&rs->md, false);
3869 rs->md.ro = MD_RDONLY;
3870 }
3871 clear_bit(MD_DM_SUSPENDING, &mddev->flags);
3872
3873}
3874
3875static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3876{
3877 int i;
3878 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3879 unsigned long flags;
3880 bool cleared = false;
3881 struct dm_raid_superblock *sb;
3882 struct mddev *mddev = &rs->md;
3883 struct md_rdev *r;
3884
3885 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3886 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3887 return;
3888
3889 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3890
3891 for (i = 0; i < rs->raid_disks; i++) {
3892 r = &rs->dev[i].rdev;
3893 /* HM FIXME: enhance journal device recovery processing */
3894 if (test_bit(Journal, &r->flags))
3895 continue;
3896
3897 if (test_bit(Faulty, &r->flags) &&
3898 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3899 DMINFO("Faulty %s device #%d has readable super block."
3900 " Attempting to revive it.",
3901 rs->raid_type->name, i);
3902
3903 /*
3904 * Faulty bit may be set, but sometimes the array can
3905 * be suspended before the personalities can respond
3906 * by removing the device from the array (i.e. calling
3907 * 'hot_remove_disk'). If they haven't yet removed
3908 * the failed device, its 'raid_disk' number will be
3909 * '>= 0' - meaning we must call this function
3910 * ourselves.
3911 */
3912 flags = r->flags;
3913 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3914 if (r->raid_disk >= 0) {
3915 if (mddev->pers->hot_remove_disk(mddev, r)) {
3916 /* Failed to revive this device, try next */
3917 r->flags = flags;
3918 continue;
3919 }
3920 } else
3921 r->raid_disk = r->saved_raid_disk = i;
3922
3923 clear_bit(Faulty, &r->flags);
3924 clear_bit(WriteErrorSeen, &r->flags);
3925
3926 if (mddev->pers->hot_add_disk(mddev, r)) {
3927 /* Failed to revive this device, try next */
3928 r->raid_disk = r->saved_raid_disk = -1;
3929 r->flags = flags;
3930 } else {
3931 clear_bit(In_sync, &r->flags);
3932 r->recovery_offset = 0;
3933 set_bit(i, (void *) cleared_failed_devices);
3934 cleared = true;
3935 }
3936 }
3937 }
3938
3939 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3940 if (cleared) {
3941 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3942
3943 rdev_for_each(r, &rs->md) {
3944 if (test_bit(Journal, &r->flags))
3945 continue;
3946
3947 sb = page_address(r->sb_page);
3948 sb_retrieve_failed_devices(sb, failed_devices);
3949
3950 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3951 failed_devices[i] &= ~cleared_failed_devices[i];
3952
3953 sb_update_failed_devices(sb, failed_devices);
3954 }
3955 }
3956}
3957
3958static int __load_dirty_region_bitmap(struct raid_set *rs)
3959{
3960 int r = 0;
3961
3962 /* Try loading the bitmap unless "raid0", which does not have one */
3963 if (!rs_is_raid0(rs) &&
3964 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3965 struct mddev *mddev = &rs->md;
3966
3967 if (md_bitmap_enabled(mddev, false)) {
3968 r = mddev->bitmap_ops->load(mddev);
3969 if (r)
3970 DMERR("Failed to load bitmap");
3971 }
3972 }
3973
3974 return r;
3975}
3976
3977/* Enforce updating all superblocks */
3978static void rs_update_sbs(struct raid_set *rs)
3979{
3980 struct mddev *mddev = &rs->md;
3981 int ro = mddev->ro;
3982
3983 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3984 mddev->ro = MD_RDWR;
3985 md_update_sb(mddev, 1);
3986 mddev->ro = ro;
3987}
3988
3989/*
3990 * Reshape changes raid algorithm of @rs to new one within personality
3991 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3992 * disks from a raid set thus growing/shrinking it or resizes the set
3993 *
3994 * Call mddev_lock_nointr() before!
3995 */
3996static int rs_start_reshape(struct raid_set *rs)
3997{
3998 int r;
3999 struct mddev *mddev = &rs->md;
4000 struct md_personality *pers = mddev->pers;
4001
4002 /* Don't allow the sync thread to work until the table gets reloaded. */
4003 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
4004
4005 r = rs_setup_reshape(rs);
4006 if (r)
4007 return r;
4008
4009 /*
4010 * Check any reshape constraints enforced by the personalility
4011 *
4012 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
4013 */
4014 r = pers->check_reshape(mddev);
4015 if (r) {
4016 rs->ti->error = "pers->check_reshape() failed";
4017 return r;
4018 }
4019
4020 /*
4021 * Personality may not provide start reshape method in which
4022 * case check_reshape above has already covered everything
4023 */
4024 if (pers->start_reshape) {
4025 r = pers->start_reshape(mddev);
4026 if (r) {
4027 rs->ti->error = "pers->start_reshape() failed";
4028 return r;
4029 }
4030 }
4031
4032 /*
4033 * Now reshape got set up, update superblocks to
4034 * reflect the fact so that a table reload will
4035 * access proper superblock content in the ctr.
4036 */
4037 rs_update_sbs(rs);
4038
4039 return 0;
4040}
4041
4042static int raid_preresume(struct dm_target *ti)
4043{
4044 int r;
4045 struct raid_set *rs = ti->private;
4046 struct mddev *mddev = &rs->md;
4047
4048 /* This is a resume after a suspend of the set -> it's already started. */
4049 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
4050 return 0;
4051
4052 /* If different and no explicit grow request, expose MD array size as of superblock. */
4053 if (!test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) &&
4054 rs->array_sectors != mddev->array_sectors)
4055 rs_set_capacity(rs);
4056
4057 /*
4058 * The superblocks need to be updated on disk if the
4059 * array is new or new devices got added (thus zeroed
4060 * out by userspace) or __load_dirty_region_bitmap
4061 * will overwrite them in core with old data or fail.
4062 */
4063 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
4064 rs_update_sbs(rs);
4065
4066 /* Load the bitmap from disk unless raid0 */
4067 r = __load_dirty_region_bitmap(rs);
4068 if (r)
4069 return r;
4070
4071 /* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
4072 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
4073 mddev->array_sectors = rs->array_sectors;
4074 mddev->dev_sectors = rs->dev_sectors;
4075 rs_set_rdev_sectors(rs);
4076 rs_set_capacity(rs);
4077 }
4078
4079 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
4080 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
4081 (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
4082 (rs->requested_bitmap_chunk_sectors &&
4083 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
4084 int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
4085
4086 if (md_bitmap_enabled(mddev, false)) {
4087 r = mddev->bitmap_ops->resize(mddev, mddev->dev_sectors,
4088 chunksize);
4089 if (r)
4090 DMERR("Failed to resize bitmap");
4091 }
4092 }
4093
4094 /* Check for any resize/reshape on @rs and adjust/initiate */
4095 if (mddev->resync_offset && mddev->resync_offset < MaxSector) {
4096 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4097 mddev->resync_min = mddev->resync_offset;
4098 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
4099 mddev->resync_max_sectors = mddev->dev_sectors;
4100 }
4101
4102 /* Check for any reshape request unless new raid set */
4103 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
4104 /* Initiate a reshape. */
4105 rs_set_rdev_sectors(rs);
4106 mddev_lock_nointr(mddev);
4107 r = rs_start_reshape(rs);
4108 mddev_unlock(mddev);
4109 if (r)
4110 DMWARN("Failed to check/start reshape, continuing without change");
4111 r = 0;
4112 }
4113
4114 return r;
4115}
4116
4117static void raid_resume(struct dm_target *ti)
4118{
4119 struct raid_set *rs = ti->private;
4120 struct mddev *mddev = &rs->md;
4121
4122 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
4123 /*
4124 * A secondary resume while the device is active.
4125 * Take this opportunity to check whether any failed
4126 * devices are reachable again.
4127 */
4128 mddev_lock_nointr(mddev);
4129 attempt_restore_of_faulty_devices(rs);
4130 mddev_unlock(mddev);
4131 }
4132
4133 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4134 /* Only reduce raid set size before running a disk removing reshape. */
4135 if (mddev->delta_disks < 0)
4136 rs_set_capacity(rs);
4137
4138 mddev_lock_nointr(mddev);
4139 WARN_ON_ONCE(!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery));
4140 WARN_ON_ONCE(rcu_dereference_protected(mddev->sync_thread,
4141 lockdep_is_held(&mddev->reconfig_mutex)));
4142 clear_bit(RT_FLAG_RS_FROZEN, &rs->runtime_flags);
4143 mddev->ro = MD_RDWR;
4144 mddev->in_sync = 0;
4145 md_unfrozen_sync_thread(mddev);
4146 mddev_unlock_and_resume(mddev);
4147 }
4148}
4149
4150static struct target_type raid_target = {
4151 .name = "raid",
4152 .version = {1, 15, 1},
4153 .module = THIS_MODULE,
4154 .ctr = raid_ctr,
4155 .dtr = raid_dtr,
4156 .map = raid_map,
4157 .status = raid_status,
4158 .message = raid_message,
4159 .iterate_devices = raid_iterate_devices,
4160 .io_hints = raid_io_hints,
4161 .presuspend = raid_presuspend,
4162 .presuspend_undo = raid_presuspend_undo,
4163 .postsuspend = raid_postsuspend,
4164 .preresume = raid_preresume,
4165 .resume = raid_resume,
4166};
4167module_dm(raid);
4168
4169module_param(devices_handle_discard_safely, bool, 0644);
4170MODULE_PARM_DESC(devices_handle_discard_safely,
4171 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4172
4173MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4174MODULE_ALIAS("dm-raid0");
4175MODULE_ALIAS("dm-raid1");
4176MODULE_ALIAS("dm-raid10");
4177MODULE_ALIAS("dm-raid4");
4178MODULE_ALIAS("dm-raid5");
4179MODULE_ALIAS("dm-raid6");
4180MODULE_AUTHOR("Neil Brown <[email protected]>");
4181MODULE_AUTHOR("Heinz Mauelshagen <[email protected]>");
4182MODULE_LICENSE("GPL");