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