blob: ee75e3510be66f6f0689365791d44273574820e2 [file] [log] [blame]
NeilBrown9d09e662011-01-13 20:00:02 +00001/*
2 * Copyright (C) 2010-2011 Neil Brown
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02003 * Copyright (C) 2010-2016 Red Hat, Inc. All rights reserved.
NeilBrown9d09e662011-01-13 20:00:02 +00004 *
5 * This file is released under the GPL.
6 */
7
8#include <linux/slab.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -04009#include <linux/module.h>
NeilBrown9d09e662011-01-13 20:00:02 +000010
11#include "md.h"
Jonathan Brassow32737272011-08-02 12:32:07 +010012#include "raid1.h"
NeilBrown9d09e662011-01-13 20:00:02 +000013#include "raid5.h"
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -050014#include "raid10.h"
NeilBrown9d09e662011-01-13 20:00:02 +000015#include "bitmap.h"
16
Alasdair G Kergon3e8dbb72011-08-02 12:32:03 +010017#include <linux/device-mapper.h>
18
NeilBrown9d09e662011-01-13 20:00:02 +000019#define DM_MSG_PREFIX "raid"
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +020020#define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
NeilBrown9d09e662011-01-13 20:00:02 +000021
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +020022/*
23 * Minimum sectors of free reshape space per raid device
24 */
25#define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
26
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +020027static bool devices_handle_discard_safely = false;
28
NeilBrown9d09e662011-01-13 20:00:02 +000029/*
Jonathan Brassowb12d4372011-08-02 12:32:07 +010030 * The following flags are used by dm-raid.c to set up the array state.
31 * They must be cleared before md_run is called.
NeilBrown9d09e662011-01-13 20:00:02 +000032 */
Mike Snitzer43157842016-05-30 13:03:37 -040033#define FirstUse 10 /* rdev flag */
NeilBrown9d09e662011-01-13 20:00:02 +000034
35struct raid_dev {
36 /*
37 * Two DM devices, one to hold metadata and one to hold the
Mike Snitzer43157842016-05-30 13:03:37 -040038 * actual data/parity. The reason for this is to not confuse
NeilBrown9d09e662011-01-13 20:00:02 +000039 * ti->len and give more flexibility in altering size and
40 * characteristics.
41 *
42 * While it is possible for this device to be associated
43 * with a different physical device than the data_dev, it
44 * is intended for it to be the same.
45 * |--------- Physical Device ---------|
46 * |- meta_dev -|------ data_dev ------|
47 */
48 struct dm_dev *meta_dev;
49 struct dm_dev *data_dev;
NeilBrown3cb03002011-10-11 16:45:26 +110050 struct md_rdev rdev;
NeilBrown9d09e662011-01-13 20:00:02 +000051};
52
53/*
Mike Snitzer42863252016-06-02 12:27:46 -040054 * Bits for establishing rs->ctr_flags
Heinz Mauelshagen702108d2016-05-19 18:49:26 +020055 *
56 * 1 = no flag value
57 * 2 = flag with value
NeilBrown9d09e662011-01-13 20:00:02 +000058 */
Mike Snitzer42863252016-06-02 12:27:46 -040059#define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
60#define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
61#define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
62#define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
63#define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
64#define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
65#define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
66#define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
67#define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
68#define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
69#define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
70#define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
Mike Snitzer9b6e5422016-06-02 11:48:09 -040071/* New for v1.9.0 */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +020072#define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
Mike Snitzer42863252016-06-02 12:27:46 -040073#define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
74#define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
75
76/*
77 * Flags for rs->ctr_flags field.
78 */
79#define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
80#define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
81#define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
82#define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
83#define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
84#define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
85#define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
86#define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
87#define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
88#define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
89#define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
90#define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
91#define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
92#define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
93#define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -050094
Heinz Mauelshagenf0902792016-05-19 18:49:27 +020095/*
96 * Definitions of various constructor flags to
97 * be used in checks of valid / invalid flags
98 * per raid level.
99 */
100/* Define all any sync flags */
101#define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
102
103/* Define flags for options without argument (e.g. 'nosync') */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200104#define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
105 CTR_FLAG_RAID10_USE_NEAR_SETS)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200106
107/* Define flags for options with one argument (e.g. 'delta_disks +2') */
108#define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
109 CTR_FLAG_WRITE_MOSTLY | \
110 CTR_FLAG_DAEMON_SLEEP | \
111 CTR_FLAG_MIN_RECOVERY_RATE | \
112 CTR_FLAG_MAX_RECOVERY_RATE | \
113 CTR_FLAG_MAX_WRITE_BEHIND | \
114 CTR_FLAG_STRIPE_CACHE | \
115 CTR_FLAG_REGION_SIZE | \
116 CTR_FLAG_RAID10_COPIES | \
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200117 CTR_FLAG_RAID10_FORMAT | \
118 CTR_FLAG_DELTA_DISKS | \
119 CTR_FLAG_DATA_OFFSET)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200120
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200121/* Valid options definitions per raid level... */
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200122
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200123/* "raid0" does only accept data offset */
124#define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200125
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200126/* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
127#define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
128 CTR_FLAG_REBUILD | \
129 CTR_FLAG_WRITE_MOSTLY | \
130 CTR_FLAG_DAEMON_SLEEP | \
131 CTR_FLAG_MIN_RECOVERY_RATE | \
132 CTR_FLAG_MAX_RECOVERY_RATE | \
133 CTR_FLAG_MAX_WRITE_BEHIND | \
134 CTR_FLAG_REGION_SIZE | \
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +0200135 CTR_FLAG_DELTA_DISKS | \
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200136 CTR_FLAG_DATA_OFFSET)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200137
138/* "raid10" does not accept any raid1 or stripe cache options */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200139#define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
140 CTR_FLAG_REBUILD | \
141 CTR_FLAG_DAEMON_SLEEP | \
142 CTR_FLAG_MIN_RECOVERY_RATE | \
143 CTR_FLAG_MAX_RECOVERY_RATE | \
144 CTR_FLAG_REGION_SIZE | \
145 CTR_FLAG_RAID10_COPIES | \
146 CTR_FLAG_RAID10_FORMAT | \
147 CTR_FLAG_DELTA_DISKS | \
148 CTR_FLAG_DATA_OFFSET | \
149 CTR_FLAG_RAID10_USE_NEAR_SETS)
150
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200151/*
152 * "raid4/5/6" do not accept any raid1 or raid10 specific options
153 *
154 * "raid6" does not accept "nosync", because it is not guaranteed
155 * that both parity and q-syndrome are being written properly with
156 * any writes
157 */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200158#define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
159 CTR_FLAG_REBUILD | \
160 CTR_FLAG_DAEMON_SLEEP | \
161 CTR_FLAG_MIN_RECOVERY_RATE | \
162 CTR_FLAG_MAX_RECOVERY_RATE | \
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200163 CTR_FLAG_MAX_WRITE_BEHIND | \
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200164 CTR_FLAG_STRIPE_CACHE | \
165 CTR_FLAG_REGION_SIZE | \
166 CTR_FLAG_DELTA_DISKS | \
167 CTR_FLAG_DATA_OFFSET)
168
169#define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
170 CTR_FLAG_REBUILD | \
171 CTR_FLAG_DAEMON_SLEEP | \
172 CTR_FLAG_MIN_RECOVERY_RATE | \
173 CTR_FLAG_MAX_RECOVERY_RATE | \
174 CTR_FLAG_MAX_WRITE_BEHIND | \
175 CTR_FLAG_STRIPE_CACHE | \
176 CTR_FLAG_REGION_SIZE | \
177 CTR_FLAG_DELTA_DISKS | \
178 CTR_FLAG_DATA_OFFSET)
179/* ...valid options definitions per raid level */
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200180
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200181/*
182 * Flags for rs->runtime_flags field
183 * (RT_FLAG prefix meaning "runtime flag")
184 *
185 * These are all internal and used to define runtime state,
186 * e.g. to prevent another resume from preresume processing
187 * the raid set all over again.
188 */
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +0200189#define RT_FLAG_RS_PRERESUMED 0
190#define RT_FLAG_RS_RESUMED 1
191#define RT_FLAG_RS_BITMAP_LOADED 2
192#define RT_FLAG_UPDATE_SBS 3
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200193#define RT_FLAG_RESHAPE_RS 4
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200194
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200195/* Array elements of 64 bit needed for rebuild/failed disk bits */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200196#define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
197
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200198/*
199 * raid set level, layout and chunk sectors backup/restore
200 */
201struct rs_layout {
202 int new_level;
203 int new_layout;
204 int new_chunk_sectors;
205};
206
NeilBrown9d09e662011-01-13 20:00:02 +0000207struct raid_set {
208 struct dm_target *ti;
209
Jonathan Brassow34f8ac6d2012-01-27 14:53:53 -0600210 uint32_t bitmap_loaded;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200211 uint32_t stripe_cache_entries;
Mike Snitzer42863252016-06-02 12:27:46 -0400212 unsigned long ctr_flags;
213 unsigned long runtime_flags;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200214
215 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
NeilBrown9d09e662011-01-13 20:00:02 +0000216
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200217 int raid_disks;
218 int delta_disks;
Heinz Mauelshagen4763e542016-05-19 18:49:31 +0200219 int data_offset;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200220 int raid10_copies;
Heinz Mauelshagen4257e082016-06-14 01:46:03 +0200221 int requested_bitmap_chunk_sectors;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200222
NeilBrownfd01b882011-10-11 16:47:53 +1100223 struct mddev md;
NeilBrown9d09e662011-01-13 20:00:02 +0000224 struct raid_type *raid_type;
225 struct dm_target_callbacks callbacks;
226
227 struct raid_dev dev[0];
228};
229
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200230static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200231{
232 struct mddev *mddev = &rs->md;
233
234 l->new_level = mddev->new_level;
235 l->new_layout = mddev->new_layout;
236 l->new_chunk_sectors = mddev->new_chunk_sectors;
237}
238
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200239static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200240{
241 struct mddev *mddev = &rs->md;
242
243 mddev->new_level = l->new_level;
244 mddev->new_layout = l->new_layout;
245 mddev->new_chunk_sectors = l->new_chunk_sectors;
246}
247
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200248/* raid10 algorithms (i.e. formats) */
249#define ALGORITHM_RAID10_DEFAULT 0
250#define ALGORITHM_RAID10_NEAR 1
251#define ALGORITHM_RAID10_OFFSET 2
252#define ALGORITHM_RAID10_FAR 3
253
NeilBrown9d09e662011-01-13 20:00:02 +0000254/* Supported raid types and properties. */
255static struct raid_type {
256 const char *name; /* RAID algorithm. */
257 const char *descr; /* Descriptor text for logging. */
Heinz Mauelshagen094f3942016-07-19 14:03:51 +0200258 const unsigned int parity_devs; /* # of parity devices. */
259 const unsigned int minimal_devs;/* minimal # of devices in set. */
260 const unsigned int level; /* RAID level. */
261 const unsigned int algorithm; /* RAID algorithm. */
NeilBrown9d09e662011-01-13 20:00:02 +0000262} raid_types[] = {
Mike Snitzer43157842016-05-30 13:03:37 -0400263 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
264 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
265 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200266 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
Mike Snitzer43157842016-05-30 13:03:37 -0400267 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
268 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
Heinz Mauelshagenb052b072016-10-17 21:20:07 +0200269 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
Mike Snitzer43157842016-05-30 13:03:37 -0400270 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
271 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
272 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
273 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
274 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
275 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
276 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
277 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
278 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
279 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
280 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
281 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
282 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
NeilBrown9d09e662011-01-13 20:00:02 +0000283};
284
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200285/* True, if @v is in inclusive range [@min, @max] */
Mike Snitzerbb91a632016-06-02 12:06:54 -0400286static bool __within_range(long v, long min, long max)
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200287{
288 return v >= min && v <= max;
289}
290
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200291/* All table line arguments are defined here */
292static struct arg_name_flag {
Mike Snitzer42863252016-06-02 12:27:46 -0400293 const unsigned long flag;
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200294 const char *name;
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400295} __arg_name_flags[] = {
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200296 { CTR_FLAG_SYNC, "sync"},
297 { CTR_FLAG_NOSYNC, "nosync"},
298 { CTR_FLAG_REBUILD, "rebuild"},
299 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
300 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
301 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
302 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
Heinz Mauelshagen65359ee2016-06-24 21:32:25 +0200303 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200304 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
305 { CTR_FLAG_REGION_SIZE, "region_size"},
306 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
307 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
Heinz Mauelshagen4763e542016-05-19 18:49:31 +0200308 { CTR_FLAG_DATA_OFFSET, "data_offset"},
309 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
310 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200311};
312
313/* Return argument name string for given @flag */
Mike Snitzer3fa6cf32016-06-02 11:58:51 -0400314static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200315{
316 if (hweight32(flag) == 1) {
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400317 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200318
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400319 while (anf-- > __arg_name_flags)
Mike Snitzer42863252016-06-02 12:27:46 -0400320 if (flag & anf->flag)
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200321 return anf->name;
322
323 } else
324 DMERR("%s called with more than one flag!", __func__);
325
326 return NULL;
327}
328
329/*
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200330 * Bool helpers to test for various raid levels of a raid set.
331 * It's level as reported by the superblock rather than
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200332 * the requested raid_type passed to the constructor.
333 */
334/* Return true, if raid set in @rs is raid0 */
335static bool rs_is_raid0(struct raid_set *rs)
336{
337 return !rs->md.level;
338}
339
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200340/* Return true, if raid set in @rs is raid1 */
341static bool rs_is_raid1(struct raid_set *rs)
342{
343 return rs->md.level == 1;
344}
345
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200346/* Return true, if raid set in @rs is raid10 */
347static bool rs_is_raid10(struct raid_set *rs)
348{
349 return rs->md.level == 10;
350}
351
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +0200352/* Return true, if raid set in @rs is level 6 */
353static bool rs_is_raid6(struct raid_set *rs)
354{
355 return rs->md.level == 6;
356}
357
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +0200358/* Return true, if raid set in @rs is level 4, 5 or 6 */
359static bool rs_is_raid456(struct raid_set *rs)
360{
361 return __within_range(rs->md.level, 4, 6);
362}
363
364/* Return true, if raid set in @rs is reshapable */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200365static bool __is_raid10_far(int layout);
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +0200366static bool rs_is_reshapable(struct raid_set *rs)
367{
368 return rs_is_raid456(rs) ||
369 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
370}
371
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200372/* Return true, if raid set in @rs is recovering */
373static bool rs_is_recovering(struct raid_set *rs)
374{
Heinz Mauelshagen345a6cd2016-06-25 02:42:54 +0200375 return rs->md.recovery_cp < rs->dev[0].rdev.sectors;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200376}
377
378/* Return true, if raid set in @rs is reshaping */
379static bool rs_is_reshaping(struct raid_set *rs)
380{
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200381 return rs->md.reshape_position != MaxSector;
382}
383
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200384/*
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200385 * bool helpers to test for various raid levels of a raid type @rt
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200386 */
387
388/* Return true, if raid type in @rt is raid0 */
389static bool rt_is_raid0(struct raid_type *rt)
390{
391 return !rt->level;
392}
393
394/* Return true, if raid type in @rt is raid1 */
395static bool rt_is_raid1(struct raid_type *rt)
396{
397 return rt->level == 1;
398}
399
400/* Return true, if raid type in @rt is raid10 */
401static bool rt_is_raid10(struct raid_type *rt)
402{
403 return rt->level == 10;
404}
405
406/* Return true, if raid type in @rt is raid4/5 */
407static bool rt_is_raid45(struct raid_type *rt)
408{
Mike Snitzerbb91a632016-06-02 12:06:54 -0400409 return __within_range(rt->level, 4, 5);
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200410}
411
412/* Return true, if raid type in @rt is raid6 */
413static bool rt_is_raid6(struct raid_type *rt)
414{
415 return rt->level == 6;
416}
Heinz Mauelshagen676fa5a2016-05-19 18:49:29 +0200417
418/* Return true, if raid type in @rt is raid4/5/6 */
419static bool rt_is_raid456(struct raid_type *rt)
420{
Mike Snitzerbb91a632016-06-02 12:06:54 -0400421 return __within_range(rt->level, 4, 6);
Heinz Mauelshagen676fa5a2016-05-19 18:49:29 +0200422}
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200423/* END: raid level bools */
424
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200425/* Return valid ctr flags for the raid level of @rs */
426static unsigned long __valid_flags(struct raid_set *rs)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200427{
428 if (rt_is_raid0(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200429 return RAID0_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200430 else if (rt_is_raid1(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200431 return RAID1_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200432 else if (rt_is_raid10(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200433 return RAID10_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200434 else if (rt_is_raid45(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200435 return RAID45_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200436 else if (rt_is_raid6(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200437 return RAID6_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200438
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200439 return 0;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200440}
441
442/*
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200443 * Check for valid flags set on @rs
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200444 *
445 * Has to be called after parsing of the ctr flags!
446 */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200447static int rs_check_for_valid_flags(struct raid_set *rs)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200448{
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200449 if (rs->ctr_flags & ~__valid_flags(rs)) {
Mike Snitzer42863252016-06-02 12:27:46 -0400450 rs->ti->error = "Invalid flags combination";
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400451 return -EINVAL;
452 }
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200453
454 return 0;
455}
456
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200457/* MD raid10 bit definitions and helpers */
458#define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
459#define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
460#define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
461#define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
462
463/* Return md raid10 near copies for @layout */
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400464static unsigned int __raid10_near_copies(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200465{
466 return layout & 0xFF;
467}
468
469/* Return md raid10 far copies for @layout */
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400470static unsigned int __raid10_far_copies(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200471{
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400472 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200473}
474
475/* Return true if md raid10 offset for @layout */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200476static bool __is_raid10_offset(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200477{
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200478 return !!(layout & RAID10_OFFSET);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200479}
480
481/* Return true if md raid10 near for @layout */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200482static bool __is_raid10_near(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200483{
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400484 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200485}
486
487/* Return true if md raid10 far for @layout */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200488static bool __is_raid10_far(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200489{
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400490 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200491}
492
493/* Return md raid10 layout string for @layout */
494static const char *raid10_md_layout_to_format(int layout)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100495{
496 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200497 * Bit 16 stands for "offset"
498 * (i.e. adjacent stripes hold copies)
499 *
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100500 * Refer to MD's raid10.c for details
501 */
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400502 if (__is_raid10_offset(layout))
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100503 return "offset";
504
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400505 if (__raid10_near_copies(layout) > 1)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100506 return "near";
507
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400508 WARN_ON(__raid10_far_copies(layout) < 2);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200509
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100510 return "far";
511}
512
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200513/* Return md raid10 algorithm for @name */
Arnd Bergmann68c1c4d2016-06-16 11:03:25 +0200514static int raid10_name_to_format(const char *name)
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500515{
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200516 if (!strcasecmp(name, "near"))
517 return ALGORITHM_RAID10_NEAR;
518 else if (!strcasecmp(name, "offset"))
519 return ALGORITHM_RAID10_OFFSET;
520 else if (!strcasecmp(name, "far"))
521 return ALGORITHM_RAID10_FAR;
522
523 return -EINVAL;
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500524}
525
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200526/* Return md raid10 copies for @layout */
527static unsigned int raid10_md_layout_to_copies(int layout)
528{
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200529 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200530}
531
532/* Return md raid10 format id for @format string */
533static int raid10_format_to_md_layout(struct raid_set *rs,
534 unsigned int algorithm,
535 unsigned int copies)
536{
537 unsigned int n = 1, f = 1, r = 0;
538
539 /*
540 * MD resilienece flaw:
541 *
542 * enabling use_far_sets for far/offset formats causes copies
543 * to be colocated on the same devs together with their origins!
544 *
545 * -> disable it for now in the definition above
546 */
547 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
548 algorithm == ALGORITHM_RAID10_NEAR)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100549 n = copies;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200550
551 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100552 f = copies;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200553 r = RAID10_OFFSET;
Mike Snitzer42863252016-06-02 12:27:46 -0400554 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200555 r |= RAID10_USE_FAR_SETS;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100556
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200557 } else if (algorithm == ALGORITHM_RAID10_FAR) {
558 f = copies;
559 r = !RAID10_OFFSET;
Mike Snitzer42863252016-06-02 12:27:46 -0400560 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200561 r |= RAID10_USE_FAR_SETS;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100562
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200563 } else
564 return -EINVAL;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100565
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200566 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
567}
568/* END: MD raid10 bit definitions and helpers */
569
570/* Check for any of the raid10 algorithms */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200571static bool __got_raid10(struct raid_type *rtp, const int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200572{
573 if (rtp->level == 10) {
574 switch (rtp->algorithm) {
575 case ALGORITHM_RAID10_DEFAULT:
576 case ALGORITHM_RAID10_NEAR:
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400577 return __is_raid10_near(layout);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200578 case ALGORITHM_RAID10_OFFSET:
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400579 return __is_raid10_offset(layout);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200580 case ALGORITHM_RAID10_FAR:
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400581 return __is_raid10_far(layout);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200582 default:
583 break;
584 }
585 }
586
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200587 return false;
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500588}
589
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200590/* Return raid_type for @name */
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200591static struct raid_type *get_raid_type(const char *name)
NeilBrown9d09e662011-01-13 20:00:02 +0000592{
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200593 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
NeilBrown9d09e662011-01-13 20:00:02 +0000594
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200595 while (rtp-- > raid_types)
596 if (!strcasecmp(rtp->name, name))
597 return rtp;
NeilBrown9d09e662011-01-13 20:00:02 +0000598
599 return NULL;
600}
601
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200602/* Return raid_type for @name based derived from @level and @layout */
603static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
604{
605 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
606
607 while (rtp-- > raid_types) {
608 /* RAID10 special checks based on @layout flags/properties */
609 if (rtp->level == level &&
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400610 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200611 return rtp;
612 }
613
614 return NULL;
615}
616
617/*
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200618 * Conditionally change bdev capacity of @rs
619 * in case of a disk add/remove reshape
620 */
621static void rs_set_capacity(struct raid_set *rs)
622{
623 struct mddev *mddev = &rs->md;
Heinz Mauelshagenfbe63652016-06-24 00:36:08 +0200624 struct md_rdev *rdev;
Heinz Mauelshagen0095dbc2016-06-24 00:10:12 +0200625 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200626
Heinz Mauelshagenfbe63652016-06-24 00:36:08 +0200627 /*
628 * raid10 sets rdev->sector to the device size, which
629 * is unintended in case of out-of-place reshaping
630 */
631 rdev_for_each(rdev, mddev)
632 rdev->sectors = mddev->dev_sectors;
633
Heinz Mauelshagen0095dbc2016-06-24 00:10:12 +0200634 set_capacity(gendisk, mddev->array_sectors);
635 revalidate_disk(gendisk);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200636}
637
638/*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +0200639 * Set the mddev properties in @rs to the current
640 * ones retrieved from the freshest superblock
641 */
642static void rs_set_cur(struct raid_set *rs)
643{
644 struct mddev *mddev = &rs->md;
645
646 mddev->new_level = mddev->level;
647 mddev->new_layout = mddev->layout;
648 mddev->new_chunk_sectors = mddev->chunk_sectors;
649}
650
651/*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200652 * Set the mddev properties in @rs to the new
653 * ones requested by the ctr
654 */
655static void rs_set_new(struct raid_set *rs)
656{
657 struct mddev *mddev = &rs->md;
658
659 mddev->level = mddev->new_level;
660 mddev->layout = mddev->new_layout;
661 mddev->chunk_sectors = mddev->new_chunk_sectors;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +0200662 mddev->raid_disks = rs->raid_disks;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200663 mddev->delta_disks = 0;
664}
665
Mike Snitzerbfcee0e2016-06-02 15:08:09 -0400666static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
Heinz Mauelshagen094f3942016-07-19 14:03:51 +0200667 unsigned int raid_devs)
NeilBrown9d09e662011-01-13 20:00:02 +0000668{
Heinz Mauelshagen094f3942016-07-19 14:03:51 +0200669 unsigned int i;
NeilBrown9d09e662011-01-13 20:00:02 +0000670 struct raid_set *rs;
NeilBrown9d09e662011-01-13 20:00:02 +0000671
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400672 if (raid_devs <= raid_type->parity_devs) {
673 ti->error = "Insufficient number of devices";
674 return ERR_PTR(-EINVAL);
675 }
NeilBrown9d09e662011-01-13 20:00:02 +0000676
NeilBrown9d09e662011-01-13 20:00:02 +0000677 rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400678 if (!rs) {
679 ti->error = "Cannot allocate raid context";
680 return ERR_PTR(-ENOMEM);
681 }
NeilBrown9d09e662011-01-13 20:00:02 +0000682
683 mddev_init(&rs->md);
684
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200685 rs->raid_disks = raid_devs;
686 rs->delta_disks = 0;
687
NeilBrown9d09e662011-01-13 20:00:02 +0000688 rs->ti = ti;
689 rs->raid_type = raid_type;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200690 rs->stripe_cache_entries = 256;
NeilBrown9d09e662011-01-13 20:00:02 +0000691 rs->md.raid_disks = raid_devs;
692 rs->md.level = raid_type->level;
693 rs->md.new_level = rs->md.level;
NeilBrown9d09e662011-01-13 20:00:02 +0000694 rs->md.layout = raid_type->algorithm;
695 rs->md.new_layout = rs->md.layout;
696 rs->md.delta_disks = 0;
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +0200697 rs->md.recovery_cp = MaxSector;
NeilBrown9d09e662011-01-13 20:00:02 +0000698
699 for (i = 0; i < raid_devs; i++)
700 md_rdev_init(&rs->dev[i].rdev);
701
702 /*
703 * Remaining items to be initialized by further RAID params:
704 * rs->md.persistent
705 * rs->md.external
706 * rs->md.chunk_sectors
707 * rs->md.new_chunk_sectors
Jonathan E Brassowc039c332012-07-27 15:08:04 +0100708 * rs->md.dev_sectors
NeilBrown9d09e662011-01-13 20:00:02 +0000709 */
710
711 return rs;
712}
713
Mike Snitzerbfcee0e2016-06-02 15:08:09 -0400714static void raid_set_free(struct raid_set *rs)
NeilBrown9d09e662011-01-13 20:00:02 +0000715{
716 int i;
717
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +0200718 for (i = 0; i < rs->raid_disks; i++) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100719 if (rs->dev[i].meta_dev)
720 dm_put_device(rs->ti, rs->dev[i].meta_dev);
NeilBrown545c8792012-05-22 13:54:30 +1000721 md_rdev_clear(&rs->dev[i].rdev);
NeilBrown9d09e662011-01-13 20:00:02 +0000722 if (rs->dev[i].data_dev)
723 dm_put_device(rs->ti, rs->dev[i].data_dev);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100724 }
NeilBrown9d09e662011-01-13 20:00:02 +0000725
726 kfree(rs);
727}
728
729/*
730 * For every device we have two words
731 * <meta_dev>: meta device name or '-' if missing
732 * <data_dev>: data device name or '-' if missing
733 *
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100734 * The following are permitted:
735 * - -
736 * - <data_dev>
737 * <meta_dev> <data_dev>
738 *
739 * The following is not allowed:
740 * <meta_dev> -
741 *
742 * This code parses those words. If there is a failure,
Mike Snitzerbfcee0e2016-06-02 15:08:09 -0400743 * the caller must use raid_set_free() to unwind the operations.
NeilBrown9d09e662011-01-13 20:00:02 +0000744 */
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200745static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
NeilBrown9d09e662011-01-13 20:00:02 +0000746{
747 int i;
748 int rebuild = 0;
749 int metadata_available = 0;
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +0200750 int r = 0;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200751 const char *arg;
NeilBrown9d09e662011-01-13 20:00:02 +0000752
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200753 /* Put off the number of raid devices argument to get to dev pairs */
754 arg = dm_shift_arg(as);
755 if (!arg)
756 return -EINVAL;
757
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +0200758 for (i = 0; i < rs->raid_disks; i++) {
NeilBrown9d09e662011-01-13 20:00:02 +0000759 rs->dev[i].rdev.raid_disk = i;
760
761 rs->dev[i].meta_dev = NULL;
762 rs->dev[i].data_dev = NULL;
763
764 /*
765 * There are no offsets, since there is a separate device
766 * for data and metadata.
767 */
768 rs->dev[i].rdev.data_offset = 0;
769 rs->dev[i].rdev.mddev = &rs->md;
770
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200771 arg = dm_shift_arg(as);
772 if (!arg)
773 return -EINVAL;
774
775 if (strcmp(arg, "-")) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400776 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
777 &rs->dev[i].meta_dev);
778 if (r) {
779 rs->ti->error = "RAID metadata device lookup failure";
780 return r;
781 }
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100782
783 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400784 if (!rs->dev[i].rdev.sb_page) {
785 rs->ti->error = "Failed to allocate superblock page";
786 return -ENOMEM;
787 }
NeilBrown9d09e662011-01-13 20:00:02 +0000788 }
789
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200790 arg = dm_shift_arg(as);
791 if (!arg)
792 return -EINVAL;
793
794 if (!strcmp(arg, "-")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000795 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400796 (!rs->dev[i].rdev.recovery_offset)) {
797 rs->ti->error = "Drive designated for rebuild not specified";
798 return -EINVAL;
799 }
NeilBrown9d09e662011-01-13 20:00:02 +0000800
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400801 if (rs->dev[i].meta_dev) {
802 rs->ti->error = "No data device supplied with metadata device";
803 return -EINVAL;
804 }
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100805
NeilBrown9d09e662011-01-13 20:00:02 +0000806 continue;
807 }
808
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400809 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
810 &rs->dev[i].data_dev);
811 if (r) {
812 rs->ti->error = "RAID device lookup failure";
813 return r;
814 }
NeilBrown9d09e662011-01-13 20:00:02 +0000815
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100816 if (rs->dev[i].meta_dev) {
817 metadata_available = 1;
818 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
819 }
NeilBrown9d09e662011-01-13 20:00:02 +0000820 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +0200821 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
NeilBrown9d09e662011-01-13 20:00:02 +0000822 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
823 rebuild++;
824 }
825
826 if (metadata_available) {
827 rs->md.external = 0;
828 rs->md.persistent = 1;
829 rs->md.major_version = 2;
830 } else if (rebuild && !rs->md.recovery_cp) {
831 /*
832 * Without metadata, we will not be able to tell if the array
833 * is in-sync or not - we must assume it is not. Therefore,
834 * it is impossible to rebuild a drive.
835 *
836 * Even if there is metadata, the on-disk information may
837 * indicate that the array is not in-sync and it will then
838 * fail at that time.
839 *
840 * User could specify 'nosync' option if desperate.
841 */
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400842 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
843 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +0000844 }
845
846 return 0;
847}
848
849/*
Jonathan Brassowc1084562011-08-02 12:32:07 +0100850 * validate_region_size
851 * @rs
852 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
853 *
854 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
855 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
856 *
857 * Returns: 0 on success, -EINVAL on failure.
858 */
859static int validate_region_size(struct raid_set *rs, unsigned long region_size)
860{
861 unsigned long min_region_size = rs->ti->len / (1 << 21);
862
Heinz Mauelshagen9e7d9362016-08-17 15:36:44 +0200863 if (rs_is_raid0(rs))
864 return 0;
865
Jonathan Brassowc1084562011-08-02 12:32:07 +0100866 if (!region_size) {
867 /*
Mike Snitzer43157842016-05-30 13:03:37 -0400868 * Choose a reasonable default. All figures in sectors.
Jonathan Brassowc1084562011-08-02 12:32:07 +0100869 */
870 if (min_region_size > (1 << 13)) {
Jonathan Brassow3a0f9aa2012-12-21 20:23:33 +0000871 /* If not a power of 2, make it the next power of 2 */
Mikulas Patocka042745e2015-10-02 11:17:37 -0400872 region_size = roundup_pow_of_two(min_region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100873 DMINFO("Choosing default region size of %lu sectors",
874 region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100875 } else {
876 DMINFO("Choosing default region size of 4MiB");
877 region_size = 1 << 13; /* sectors */
878 }
879 } else {
880 /*
881 * Validate user-supplied value.
882 */
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400883 if (region_size > rs->ti->len) {
884 rs->ti->error = "Supplied region size is too large";
885 return -EINVAL;
886 }
Jonathan Brassowc1084562011-08-02 12:32:07 +0100887
888 if (region_size < min_region_size) {
889 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
890 region_size, min_region_size);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400891 rs->ti->error = "Supplied region size is too small";
892 return -EINVAL;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100893 }
894
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400895 if (!is_power_of_2(region_size)) {
896 rs->ti->error = "Region size is not a power of 2";
897 return -EINVAL;
898 }
Jonathan Brassowc1084562011-08-02 12:32:07 +0100899
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400900 if (region_size < rs->md.chunk_sectors) {
901 rs->ti->error = "Region size is smaller than the chunk size";
902 return -EINVAL;
903 }
Jonathan Brassowc1084562011-08-02 12:32:07 +0100904 }
905
906 /*
907 * Convert sectors to bytes.
908 */
Heinz Mauelshagen89d3d9a2016-07-19 13:16:24 +0200909 rs->md.bitmap_info.chunksize = to_bytes(region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100910
911 return 0;
912}
913
914/*
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600915 * validate_raid_redundancy
Jonathan Brassoweb649122012-10-11 13:40:09 +1100916 * @rs
917 *
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600918 * Determine if there are enough devices in the array that haven't
919 * failed (or are being rebuilt) to form a usable array.
Jonathan Brassoweb649122012-10-11 13:40:09 +1100920 *
921 * Returns: 0 on success, -EINVAL on failure.
922 */
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600923static int validate_raid_redundancy(struct raid_set *rs)
Jonathan Brassoweb649122012-10-11 13:40:09 +1100924{
Heinz Mauelshagen094f3942016-07-19 14:03:51 +0200925 unsigned int i, rebuild_cnt = 0;
926 unsigned int rebuilds_per_group = 0, copies;
927 unsigned int group_size, last_group_start;
Jonathan Brassoweb649122012-10-11 13:40:09 +1100928
Jonathan Brassoweb649122012-10-11 13:40:09 +1100929 for (i = 0; i < rs->md.raid_disks; i++)
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600930 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
931 !rs->dev[i].rdev.sb_page)
Jonathan Brassoweb649122012-10-11 13:40:09 +1100932 rebuild_cnt++;
933
934 switch (rs->raid_type->level) {
Heinz Mauelshagen9e7d9362016-08-17 15:36:44 +0200935 case 0:
936 break;
Jonathan Brassoweb649122012-10-11 13:40:09 +1100937 case 1:
938 if (rebuild_cnt >= rs->md.raid_disks)
939 goto too_many;
940 break;
941 case 4:
942 case 5:
943 case 6:
944 if (rebuild_cnt > rs->raid_type->parity_devs)
945 goto too_many;
946 break;
947 case 10:
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200948 copies = raid10_md_layout_to_copies(rs->md.new_layout);
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100949 if (rebuild_cnt < copies)
950 break;
951
952 /*
953 * It is possible to have a higher rebuild count for RAID10,
954 * as long as the failed devices occur in different mirror
955 * groups (i.e. different stripes).
956 *
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100957 * When checking "near" format, make sure no adjacent devices
958 * have failed beyond what can be handled. In addition to the
959 * simple case where the number of devices is a multiple of the
960 * number of copies, we must also handle cases where the number
961 * of devices is not a multiple of the number of copies.
Mike Snitzer43157842016-05-30 13:03:37 -0400962 * E.g. dev1 dev2 dev3 dev4 dev5
963 * A A B B C
964 * C D D E E
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100965 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200966 if (__is_raid10_near(rs->md.new_layout)) {
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +0200967 for (i = 0; i < rs->md.raid_disks; i++) {
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100968 if (!(i % copies))
969 rebuilds_per_group = 0;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200970 if ((!rs->dev[i].rdev.sb_page ||
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +0200971 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100972 (++rebuilds_per_group >= copies))
973 goto too_many;
974 }
975 break;
976 }
977
978 /*
979 * When checking "far" and "offset" formats, we need to ensure
980 * that the device that holds its copy is not also dead or
981 * being rebuilt. (Note that "far" and "offset" formats only
982 * support two copies right now. These formats also only ever
983 * use the 'use_far_sets' variant.)
984 *
985 * This check is somewhat complicated by the need to account
Mike Snitzer43157842016-05-30 13:03:37 -0400986 * for arrays that are not a multiple of (far) copies. This
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100987 * results in the need to treat the last (potentially larger)
988 * set differently.
989 */
990 group_size = (rs->md.raid_disks / copies);
991 last_group_start = (rs->md.raid_disks / group_size) - 1;
992 last_group_start *= group_size;
993 for (i = 0; i < rs->md.raid_disks; i++) {
994 if (!(i % copies) && !(i > last_group_start))
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600995 rebuilds_per_group = 0;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100996 if ((!rs->dev[i].rdev.sb_page ||
997 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
Jonathan Brassow4ec1e362012-10-11 13:40:24 +1100998 (++rebuilds_per_group >= copies))
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100999 goto too_many;
Jonathan Brassow4ec1e362012-10-11 13:40:24 +11001000 }
1001 break;
Jonathan Brassoweb649122012-10-11 13:40:09 +11001002 default:
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001003 if (rebuild_cnt)
1004 return -EINVAL;
Jonathan Brassoweb649122012-10-11 13:40:09 +11001005 }
1006
1007 return 0;
1008
1009too_many:
Jonathan Brassoweb649122012-10-11 13:40:09 +11001010 return -EINVAL;
1011}
1012
1013/*
NeilBrown9d09e662011-01-13 20:00:02 +00001014 * Possible arguments are...
NeilBrown9d09e662011-01-13 20:00:02 +00001015 * <chunk_size> [optional_args]
1016 *
Jonathan Brassow32737272011-08-02 12:32:07 +01001017 * Argument definitions
1018 * <chunk_size> The number of sectors per disk that
Mike Snitzer43157842016-05-30 13:03:37 -04001019 * will form the "stripe"
Jonathan Brassow32737272011-08-02 12:32:07 +01001020 * [[no]sync] Force or prevent recovery of the
Mike Snitzer43157842016-05-30 13:03:37 -04001021 * entire array
NeilBrown9d09e662011-01-13 20:00:02 +00001022 * [rebuild <idx>] Rebuild the drive indicated by the index
Jonathan Brassow32737272011-08-02 12:32:07 +01001023 * [daemon_sleep <ms>] Time between bitmap daemon work to
Mike Snitzer43157842016-05-30 13:03:37 -04001024 * clear bits
NeilBrown9d09e662011-01-13 20:00:02 +00001025 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1026 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001027 * [write_mostly <idx>] Indicate a write mostly drive via index
NeilBrown9d09e662011-01-13 20:00:02 +00001028 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1029 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
Mike Snitzer43157842016-05-30 13:03:37 -04001030 * [region_size <sectors>] Defines granularity of bitmap
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001031 *
1032 * RAID10-only options:
Mike Snitzer43157842016-05-30 13:03:37 -04001033 * [raid10_copies <# copies>] Number of copies. (Default: 2)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001034 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
NeilBrown9d09e662011-01-13 20:00:02 +00001035 */
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001036static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001037 unsigned int num_raid_params)
NeilBrown9d09e662011-01-13 20:00:02 +00001038{
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001039 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001040 unsigned int raid10_copies = 2;
1041 unsigned int i, write_mostly = 0;
1042 unsigned int region_size = 0;
Mike Snitzer542f9032012-07-27 15:08:00 +01001043 sector_t max_io_len;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001044 const char *arg, *key;
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001045 struct raid_dev *rd;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001046 struct raid_type *rt = rs->raid_type;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001047
1048 arg = dm_shift_arg(as);
1049 num_raid_params--; /* Account for chunk_size argument */
1050
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001051 if (kstrtoint(arg, 10, &value) < 0) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001052 rs->ti->error = "Bad numerical argument given for chunk_size";
1053 return -EINVAL;
1054 }
NeilBrown9d09e662011-01-13 20:00:02 +00001055
1056 /*
1057 * First, parse the in-order required arguments
Jonathan Brassow32737272011-08-02 12:32:07 +01001058 * "chunk_size" is the only argument of this type.
NeilBrown9d09e662011-01-13 20:00:02 +00001059 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001060 if (rt_is_raid1(rt)) {
Jonathan Brassow32737272011-08-02 12:32:07 +01001061 if (value)
1062 DMERR("Ignoring chunk size parameter for RAID 1");
1063 value = 0;
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001064 } else if (!is_power_of_2(value)) {
1065 rs->ti->error = "Chunk size must be a power of 2";
1066 return -EINVAL;
1067 } else if (value < 8) {
1068 rs->ti->error = "Chunk size value is too small";
1069 return -EINVAL;
1070 }
NeilBrown9d09e662011-01-13 20:00:02 +00001071
1072 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
NeilBrown9d09e662011-01-13 20:00:02 +00001073
1074 /*
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001075 * We set each individual device as In_sync with a completed
1076 * 'recovery_offset'. If there has been a device failure or
1077 * replacement then one of the following cases applies:
1078 *
1079 * 1) User specifies 'rebuild'.
Mike Snitzer43157842016-05-30 13:03:37 -04001080 * - Device is reset when param is read.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001081 * 2) A new device is supplied.
Mike Snitzer43157842016-05-30 13:03:37 -04001082 * - No matching superblock found, resets device.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001083 * 3) Device failure was transient and returns on reload.
Mike Snitzer43157842016-05-30 13:03:37 -04001084 * - Failure noticed, resets device for bitmap replay.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001085 * 4) Device hadn't completed recovery after previous failure.
Mike Snitzer43157842016-05-30 13:03:37 -04001086 * - Superblock is read and overrides recovery_offset.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001087 *
1088 * What is found in the superblocks of the devices is always
1089 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1090 */
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +02001091 for (i = 0; i < rs->raid_disks; i++) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001092 set_bit(In_sync, &rs->dev[i].rdev.flags);
1093 rs->dev[i].rdev.recovery_offset = MaxSector;
1094 }
1095
1096 /*
NeilBrown9d09e662011-01-13 20:00:02 +00001097 * Second, parse the unordered optional arguments
1098 */
NeilBrown9d09e662011-01-13 20:00:02 +00001099 for (i = 0; i < num_raid_params; i++) {
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001100 key = dm_shift_arg(as);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001101 if (!key) {
1102 rs->ti->error = "Not enough raid parameters given";
1103 return -EINVAL;
1104 }
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001105
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001106 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001107 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001108 rs->ti->error = "Only one 'nosync' argument allowed";
1109 return -EINVAL;
1110 }
NeilBrown9d09e662011-01-13 20:00:02 +00001111 continue;
1112 }
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001113 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001114 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001115 rs->ti->error = "Only one 'sync' argument allowed";
1116 return -EINVAL;
1117 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001118 continue;
1119 }
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001120 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001121 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001122 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1123 return -EINVAL;
1124 }
NeilBrown9d09e662011-01-13 20:00:02 +00001125 continue;
1126 }
1127
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001128 arg = dm_shift_arg(as);
1129 i++; /* Account for the argument pairs */
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001130 if (!arg) {
1131 rs->ti->error = "Wrong number of raid parameters given";
1132 return -EINVAL;
1133 }
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001134
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001135 /*
1136 * Parameters that take a string value are checked here.
1137 */
1138
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001139 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001140 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001141 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1142 return -EINVAL;
1143 }
1144 if (!rt_is_raid10(rt)) {
1145 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1146 return -EINVAL;
1147 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001148 raid10_format = raid10_name_to_format(arg);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001149 if (raid10_format < 0) {
1150 rs->ti->error = "Invalid 'raid10_format' value given";
1151 return raid10_format;
1152 }
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001153 continue;
1154 }
1155
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001156 if (kstrtoint(arg, 10, &value) < 0) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001157 rs->ti->error = "Bad numerical argument given in raid params";
1158 return -EINVAL;
1159 }
NeilBrown9d09e662011-01-13 20:00:02 +00001160
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001161 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001162 /*
1163 * "rebuild" is being passed in by userspace to provide
1164 * indexes of replaced devices and to set up additional
1165 * devices on raid level takeover.
Mike Snitzer43157842016-05-30 13:03:37 -04001166 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001167 if (!__within_range(value, 0, rs->raid_disks - 1)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001168 rs->ti->error = "Invalid rebuild index given";
1169 return -EINVAL;
1170 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001171
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001172 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1173 rs->ti->error = "rebuild for this index already given";
1174 return -EINVAL;
1175 }
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001176
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001177 rd = rs->dev + value;
1178 clear_bit(In_sync, &rd->rdev.flags);
1179 clear_bit(Faulty, &rd->rdev.flags);
1180 rd->rdev.recovery_offset = 0;
Mike Snitzer42863252016-06-02 12:27:46 -04001181 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001182 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001183 if (!rt_is_raid1(rt)) {
1184 rs->ti->error = "write_mostly option is only valid for RAID1";
1185 return -EINVAL;
1186 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001187
Mike Snitzerbb91a632016-06-02 12:06:54 -04001188 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001189 rs->ti->error = "Invalid write_mostly index given";
1190 return -EINVAL;
1191 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001192
Heinz Mauelshagen5fa146b2016-06-15 18:50:18 +02001193 write_mostly++;
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001194 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
Mike Snitzer42863252016-06-02 12:27:46 -04001195 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001196 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001197 if (!rt_is_raid1(rt)) {
1198 rs->ti->error = "max_write_behind option is only valid for RAID1";
1199 return -EINVAL;
1200 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001201
Mike Snitzer42863252016-06-02 12:27:46 -04001202 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001203 rs->ti->error = "Only one max_write_behind argument pair allowed";
1204 return -EINVAL;
1205 }
NeilBrown9d09e662011-01-13 20:00:02 +00001206
1207 /*
1208 * In device-mapper, we specify things in sectors, but
1209 * MD records this value in kB
1210 */
1211 value /= 2;
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001212 if (value > COUNTER_MAX) {
1213 rs->ti->error = "Max write-behind limit out of range";
1214 return -EINVAL;
1215 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001216
NeilBrown9d09e662011-01-13 20:00:02 +00001217 rs->md.bitmap_info.max_write_behind = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001218 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001219 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001220 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1221 return -EINVAL;
1222 }
1223 if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1224 rs->ti->error = "daemon sleep period out of range";
1225 return -EINVAL;
1226 }
NeilBrown9d09e662011-01-13 20:00:02 +00001227 rs->md.bitmap_info.daemon_sleep = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001228 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001229 /* Userspace passes new data_offset after having extended the the data image LV */
Mike Snitzer42863252016-06-02 12:27:46 -04001230 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001231 rs->ti->error = "Only one data_offset argument pair allowed";
1232 return -EINVAL;
1233 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001234 /* Ensure sensible data offset */
Heinz Mauelshagen75dd3b92016-06-15 22:27:08 +02001235 if (value < 0 ||
1236 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001237 rs->ti->error = "Bogus data_offset value";
1238 return -EINVAL;
1239 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001240 rs->data_offset = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001241 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001242 /* Define the +/-# of disks to add to/remove from the given raid set */
Mike Snitzer42863252016-06-02 12:27:46 -04001243 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001244 rs->ti->error = "Only one delta_disks argument pair allowed";
1245 return -EINVAL;
1246 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001247 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001248 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001249 rs->ti->error = "Too many delta_disk requested";
1250 return -EINVAL;
1251 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001252
1253 rs->delta_disks = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001254 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001255 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001256 rs->ti->error = "Only one stripe_cache argument pair allowed";
1257 return -EINVAL;
1258 }
1259
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001260 if (!rt_is_raid456(rt)) {
1261 rs->ti->error = "Inappropriate argument: stripe_cache";
1262 return -EINVAL;
1263 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001264
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001265 rs->stripe_cache_entries = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001266 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001267 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001268 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1269 return -EINVAL;
1270 }
1271 if (value > INT_MAX) {
1272 rs->ti->error = "min_recovery_rate out of range";
1273 return -EINVAL;
1274 }
NeilBrown9d09e662011-01-13 20:00:02 +00001275 rs->md.sync_speed_min = (int)value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001276 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
Heinz Mauelshagenf15f64d2016-07-27 23:34:01 +02001277 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001278 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1279 return -EINVAL;
1280 }
1281 if (value > INT_MAX) {
1282 rs->ti->error = "max_recovery_rate out of range";
1283 return -EINVAL;
1284 }
NeilBrown9d09e662011-01-13 20:00:02 +00001285 rs->md.sync_speed_max = (int)value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001286 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001287 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001288 rs->ti->error = "Only one region_size argument pair allowed";
1289 return -EINVAL;
1290 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001291
Jonathan Brassowc1084562011-08-02 12:32:07 +01001292 region_size = value;
Heinz Mauelshagen4257e082016-06-14 01:46:03 +02001293 rs->requested_bitmap_chunk_sectors = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001294 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001295 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001296 rs->ti->error = "Only one raid10_copies argument pair allowed";
1297 return -EINVAL;
1298 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001299
Mike Snitzerbb91a632016-06-02 12:06:54 -04001300 if (!__within_range(value, 2, rs->md.raid_disks)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001301 rs->ti->error = "Bad value for 'raid10_copies'";
1302 return -EINVAL;
1303 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001304
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001305 raid10_copies = value;
NeilBrown9d09e662011-01-13 20:00:02 +00001306 } else {
1307 DMERR("Unable to parse RAID parameter: %s", key);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001308 rs->ti->error = "Unable to parse RAID parameter";
1309 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +00001310 }
1311 }
1312
Heinz Mauelshagen0d851d12016-06-15 18:43:55 +02001313 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1314 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1315 rs->ti->error = "sync and nosync are mutually exclusive";
1316 return -EINVAL;
1317 }
1318
Heinz Mauelshagen37f10be2016-06-24 23:21:37 +02001319 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1320 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1321 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1322 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1323 return -EINVAL;
1324 }
1325
Heinz Mauelshagen5fa146b2016-06-15 18:50:18 +02001326 if (write_mostly >= rs->md.raid_disks) {
1327 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1328 return -EINVAL;
1329 }
1330
Jonathan Brassowc1084562011-08-02 12:32:07 +01001331 if (validate_region_size(rs, region_size))
1332 return -EINVAL;
1333
1334 if (rs->md.chunk_sectors)
Mike Snitzer542f9032012-07-27 15:08:00 +01001335 max_io_len = rs->md.chunk_sectors;
Jonathan Brassowc1084562011-08-02 12:32:07 +01001336 else
Mike Snitzer542f9032012-07-27 15:08:00 +01001337 max_io_len = region_size;
Jonathan Brassowc1084562011-08-02 12:32:07 +01001338
Mike Snitzer542f9032012-07-27 15:08:00 +01001339 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1340 return -EINVAL;
Jonathan Brassow32737272011-08-02 12:32:07 +01001341
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001342 if (rt_is_raid10(rt)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001343 if (raid10_copies > rs->md.raid_disks) {
1344 rs->ti->error = "Not enough devices to satisfy specification";
1345 return -EINVAL;
1346 }
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001347
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001348 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001349 if (rs->md.new_layout < 0) {
1350 rs->ti->error = "Error getting raid10 format";
1351 return rs->md.new_layout;
1352 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001353
1354 rt = get_raid_type_by_ll(10, rs->md.new_layout);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001355 if (!rt) {
1356 rs->ti->error = "Failed to recognize new raid10 layout";
1357 return -EINVAL;
1358 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001359
1360 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1361 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
Mike Snitzer42863252016-06-02 12:27:46 -04001362 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001363 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1364 return -EINVAL;
1365 }
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001366 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001367
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001368 rs->raid10_copies = raid10_copies;
Jonathan E Brassowc039c332012-07-27 15:08:04 +01001369
NeilBrown9d09e662011-01-13 20:00:02 +00001370 /* Assume there are no metadata devices until the drives are parsed */
1371 rs->md.persistent = 0;
1372 rs->md.external = 1;
1373
Heinz Mauelshagenf0902792016-05-19 18:49:27 +02001374 /* Check, if any invalid ctr arguments have been passed in for the raid level */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +02001375 return rs_check_for_valid_flags(rs);
NeilBrown9d09e662011-01-13 20:00:02 +00001376}
1377
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001378/* Set raid4/5/6 cache size */
1379static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1380{
1381 int r;
1382 struct r5conf *conf;
1383 struct mddev *mddev = &rs->md;
1384 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1385 uint32_t nr_stripes = rs->stripe_cache_entries;
1386
1387 if (!rt_is_raid456(rs->raid_type)) {
1388 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1389 return -EINVAL;
1390 }
1391
1392 if (nr_stripes < min_stripes) {
1393 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1394 nr_stripes, min_stripes);
1395 nr_stripes = min_stripes;
1396 }
1397
1398 conf = mddev->private;
1399 if (!conf) {
1400 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1401 return -EINVAL;
1402 }
1403
1404 /* Try setting number of stripes in raid456 stripe cache */
1405 if (conf->min_nr_stripes != nr_stripes) {
1406 r = raid5_set_cache_size(mddev, nr_stripes);
1407 if (r) {
1408 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1409 return r;
1410 }
1411
1412 DMINFO("%u stripe cache entries", nr_stripes);
1413 }
1414
1415 return 0;
1416}
1417
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02001418/* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1419static unsigned int mddev_data_stripes(struct raid_set *rs)
1420{
1421 return rs->md.raid_disks - rs->raid_type->parity_devs;
1422}
1423
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001424/* Return # of data stripes of @rs (i.e. as of ctr) */
1425static unsigned int rs_data_stripes(struct raid_set *rs)
1426{
1427 return rs->raid_disks - rs->raid_type->parity_devs;
1428}
1429
1430/* Calculate the sectors per device and per array used for @rs */
1431static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1432{
1433 int delta_disks;
1434 unsigned int data_stripes;
1435 struct mddev *mddev = &rs->md;
1436 struct md_rdev *rdev;
1437 sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1438
1439 if (use_mddev) {
1440 delta_disks = mddev->delta_disks;
1441 data_stripes = mddev_data_stripes(rs);
1442 } else {
1443 delta_disks = rs->delta_disks;
1444 data_stripes = rs_data_stripes(rs);
1445 }
1446
1447 /* Special raid1 case w/o delta_disks support (yet) */
1448 if (rt_is_raid1(rs->raid_type))
1449 ;
1450 else if (rt_is_raid10(rs->raid_type)) {
1451 if (rs->raid10_copies < 2 ||
1452 delta_disks < 0) {
1453 rs->ti->error = "Bogus raid10 data copies or delta disks";
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001454 return -EINVAL;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001455 }
1456
1457 dev_sectors *= rs->raid10_copies;
1458 if (sector_div(dev_sectors, data_stripes))
1459 goto bad;
1460
1461 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1462 if (sector_div(array_sectors, rs->raid10_copies))
1463 goto bad;
1464
1465 } else if (sector_div(dev_sectors, data_stripes))
1466 goto bad;
1467
1468 else
1469 /* Striped layouts */
1470 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1471
1472 rdev_for_each(rdev, mddev)
1473 rdev->sectors = dev_sectors;
1474
1475 mddev->array_sectors = array_sectors;
1476 mddev->dev_sectors = dev_sectors;
1477
1478 return 0;
1479bad:
1480 rs->ti->error = "Target length not divisible by number of data devices";
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001481 return -EINVAL;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001482}
1483
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02001484/* Setup recovery on @rs */
1485static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1486{
1487 /* raid0 does not recover */
1488 if (rs_is_raid0(rs))
1489 rs->md.recovery_cp = MaxSector;
1490 /*
1491 * A raid6 set has to be recovered either
1492 * completely or for the grown part to
1493 * ensure proper parity and Q-Syndrome
1494 */
1495 else if (rs_is_raid6(rs))
1496 rs->md.recovery_cp = dev_sectors;
1497 /*
1498 * Other raid set types may skip recovery
1499 * depending on the 'nosync' flag.
1500 */
1501 else
1502 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1503 ? MaxSector : dev_sectors;
1504}
1505
1506/* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1507static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1508{
1509 if (!dev_sectors)
1510 /* New raid set or 'sync' flag provided */
1511 __rs_setup_recovery(rs, 0);
1512 else if (dev_sectors == MaxSector)
1513 /* Prevent recovery */
1514 __rs_setup_recovery(rs, MaxSector);
1515 else if (rs->dev[0].rdev.sectors < dev_sectors)
1516 /* Grown raid set */
1517 __rs_setup_recovery(rs, rs->dev[0].rdev.sectors);
1518 else
1519 __rs_setup_recovery(rs, MaxSector);
1520}
1521
NeilBrown9d09e662011-01-13 20:00:02 +00001522static void do_table_event(struct work_struct *ws)
1523{
1524 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1525
Heinz Mauelshagen9d9d9392016-06-16 03:15:49 +02001526 smp_rmb(); /* Make sure we access most actual mddev properties */
1527 if (!rs_is_reshaping(rs))
1528 rs_set_capacity(rs);
NeilBrown9d09e662011-01-13 20:00:02 +00001529 dm_table_event(rs->ti->table);
1530}
1531
1532static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1533{
1534 struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1535
NeilBrown5c675f82014-12-15 12:56:56 +11001536 return mddev_congested(&rs->md, bits);
NeilBrown9d09e662011-01-13 20:00:02 +00001537}
1538
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001539/*
1540 * Make sure a valid takover (level switch) is being requested on @rs
1541 *
1542 * Conversions of raid sets from one MD personality to another
1543 * have to conform to restrictions which are enforced here.
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001544 */
1545static int rs_check_takeover(struct raid_set *rs)
1546{
1547 struct mddev *mddev = &rs->md;
1548 unsigned int near_copies;
1549
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001550 if (rs->md.degraded) {
1551 rs->ti->error = "Can't takeover degraded raid set";
1552 return -EPERM;
1553 }
1554
1555 if (rs_is_reshaping(rs)) {
1556 rs->ti->error = "Can't takeover reshaping raid set";
1557 return -EPERM;
1558 }
1559
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001560 switch (mddev->level) {
1561 case 0:
1562 /* raid0 -> raid1/5 with one disk */
1563 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1564 mddev->raid_disks == 1)
1565 return 0;
1566
1567 /* raid0 -> raid10 */
1568 if (mddev->new_level == 10 &&
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001569 !(rs->raid_disks % mddev->raid_disks))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001570 return 0;
1571
1572 /* raid0 with multiple disks -> raid4/5/6 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001573 if (__within_range(mddev->new_level, 4, 6) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001574 mddev->new_layout == ALGORITHM_PARITY_N &&
1575 mddev->raid_disks > 1)
1576 return 0;
1577
1578 break;
1579
1580 case 10:
1581 /* Can't takeover raid10_offset! */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001582 if (__is_raid10_offset(mddev->layout))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001583 break;
1584
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001585 near_copies = __raid10_near_copies(mddev->layout);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001586
1587 /* raid10* -> raid0 */
1588 if (mddev->new_level == 0) {
1589 /* Can takeover raid10_near with raid disks divisable by data copies! */
1590 if (near_copies > 1 &&
1591 !(mddev->raid_disks % near_copies)) {
1592 mddev->raid_disks /= near_copies;
1593 mddev->delta_disks = mddev->raid_disks;
1594 return 0;
1595 }
1596
1597 /* Can takeover raid10_far */
1598 if (near_copies == 1 &&
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001599 __raid10_far_copies(mddev->layout) > 1)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001600 return 0;
1601
1602 break;
1603 }
1604
1605 /* raid10_{near,far} -> raid1 */
1606 if (mddev->new_level == 1 &&
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001607 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001608 return 0;
1609
1610 /* raid10_{near,far} with 2 disks -> raid4/5 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001611 if (__within_range(mddev->new_level, 4, 5) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001612 mddev->raid_disks == 2)
1613 return 0;
1614 break;
1615
1616 case 1:
1617 /* raid1 with 2 disks -> raid4/5 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001618 if (__within_range(mddev->new_level, 4, 5) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001619 mddev->raid_disks == 2) {
1620 mddev->degraded = 1;
1621 return 0;
1622 }
1623
1624 /* raid1 -> raid0 */
1625 if (mddev->new_level == 0 &&
1626 mddev->raid_disks == 1)
1627 return 0;
1628
1629 /* raid1 -> raid10 */
1630 if (mddev->new_level == 10)
1631 return 0;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001632 break;
1633
1634 case 4:
1635 /* raid4 -> raid0 */
1636 if (mddev->new_level == 0)
1637 return 0;
1638
1639 /* raid4 -> raid1/5 with 2 disks */
1640 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1641 mddev->raid_disks == 2)
1642 return 0;
1643
1644 /* raid4 -> raid5/6 with parity N */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001645 if (__within_range(mddev->new_level, 5, 6) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001646 mddev->layout == ALGORITHM_PARITY_N)
1647 return 0;
1648 break;
1649
1650 case 5:
1651 /* raid5 with parity N -> raid0 */
1652 if (mddev->new_level == 0 &&
1653 mddev->layout == ALGORITHM_PARITY_N)
1654 return 0;
1655
1656 /* raid5 with parity N -> raid4 */
1657 if (mddev->new_level == 4 &&
1658 mddev->layout == ALGORITHM_PARITY_N)
1659 return 0;
1660
1661 /* raid5 with 2 disks -> raid1/4/10 */
1662 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1663 mddev->raid_disks == 2)
1664 return 0;
1665
Heinz Mauelshagen6ee0bae2016-06-15 22:29:09 +02001666 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001667 if (mddev->new_level == 6 &&
1668 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
Mike Snitzerbb91a632016-06-02 12:06:54 -04001669 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001670 return 0;
1671 break;
1672
1673 case 6:
1674 /* raid6 with parity N -> raid0 */
1675 if (mddev->new_level == 0 &&
1676 mddev->layout == ALGORITHM_PARITY_N)
1677 return 0;
1678
1679 /* raid6 with parity N -> raid4 */
1680 if (mddev->new_level == 4 &&
1681 mddev->layout == ALGORITHM_PARITY_N)
1682 return 0;
1683
Heinz Mauelshagen6ee0bae2016-06-15 22:29:09 +02001684 /* raid6_*_n with Q-Syndrome N -> raid5_* */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001685 if (mddev->new_level == 5 &&
1686 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
Mike Snitzerbb91a632016-06-02 12:06:54 -04001687 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001688 return 0;
1689
1690 default:
1691 break;
1692 }
1693
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001694 rs->ti->error = "takeover not possible";
1695 return -EINVAL;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001696}
1697
1698/* True if @rs requested to be taken over */
1699static bool rs_takeover_requested(struct raid_set *rs)
1700{
1701 return rs->md.new_level != rs->md.level;
1702}
1703
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001704/* True if @rs is requested to reshape by ctr */
1705static bool rs_reshape_requested(struct raid_set *rs)
1706{
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001707 bool change;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001708 struct mddev *mddev = &rs->md;
1709
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001710 if (rs_takeover_requested(rs))
1711 return false;
1712
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001713 if (!mddev->level)
1714 return false;
1715
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001716 change = mddev->new_layout != mddev->layout ||
1717 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1718 rs->delta_disks;
1719
1720 /* Historical case to support raid1 reshape without delta disks */
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02001721 if (mddev->level == 1) {
1722 if (rs->delta_disks)
1723 return !!rs->delta_disks;
1724
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001725 return !change &&
1726 mddev->raid_disks != rs->raid_disks;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02001727 }
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001728
1729 if (mddev->level == 10)
1730 return change &&
1731 !__is_raid10_far(mddev->new_layout) &&
1732 rs->delta_disks >= 0;
1733
1734 return change;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001735}
1736
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001737/* Features */
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001738#define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001739
1740/* State flags for sb->flags */
1741#define SB_FLAG_RESHAPE_ACTIVE 0x1
1742#define SB_FLAG_RESHAPE_BACKWARDS 0x2
1743
NeilBrown9d09e662011-01-13 20:00:02 +00001744/*
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001745 * This structure is never routinely used by userspace, unlike md superblocks.
1746 * Devices with this superblock should only ever be accessed via device-mapper.
1747 */
1748#define DM_RAID_MAGIC 0x64526D44
1749struct dm_raid_superblock {
1750 __le32 magic; /* "DmRd" */
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001751 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001752
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001753 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1754 __le32 array_position; /* The position of this drive in the raid set */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001755
1756 __le64 events; /* Incremented by md when superblock updated */
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001757 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001758 /* indicate failures (see extension below) */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001759
1760 /*
1761 * This offset tracks the progress of the repair or replacement of
1762 * an individual drive.
1763 */
1764 __le64 disk_recovery_offset;
1765
1766 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001767 * This offset tracks the progress of the initial raid set
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001768 * synchronisation/parity calculation.
1769 */
1770 __le64 array_resync_offset;
1771
1772 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001773 * raid characteristics
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001774 */
1775 __le32 level;
1776 __le32 layout;
1777 __le32 stripe_sectors;
1778
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001779 /********************************************************************
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001780 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001781 *
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001782 * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001783 */
1784
1785 __le32 flags; /* Flags defining array states for reshaping */
1786
1787 /*
1788 * This offset tracks the progress of a raid
1789 * set reshape in order to be able to restart it
1790 */
1791 __le64 reshape_position;
1792
1793 /*
1794 * These define the properties of the array in case of an interrupted reshape
1795 */
1796 __le32 new_level;
1797 __le32 new_layout;
1798 __le32 new_stripe_sectors;
1799 __le32 delta_disks;
1800
1801 __le64 array_sectors; /* Array size in sectors */
1802
1803 /*
1804 * Sector offsets to data on devices (reshaping).
1805 * Needed to support out of place reshaping, thus
1806 * not writing over any stripes whilst converting
1807 * them from old to new layout
1808 */
1809 __le64 data_offset;
1810 __le64 new_data_offset;
1811
1812 __le64 sectors; /* Used device size in sectors */
1813
1814 /*
1815 * Additonal Bit field of devices indicating failures to support
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001816 * up to 256 devices with the 1.9.0 on-disk metadata format
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001817 */
1818 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1819
1820 __le32 incompat_features; /* Used to indicate any incompatible features */
1821
1822 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001823} __packed;
1824
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001825/*
1826 * Check for reshape constraints on raid set @rs:
1827 *
1828 * - reshape function non-existent
1829 * - degraded set
1830 * - ongoing recovery
1831 * - ongoing reshape
1832 *
1833 * Returns 0 if none or -EPERM if given constraint
1834 * and error message reference in @errmsg
1835 */
1836static int rs_check_reshape(struct raid_set *rs)
1837{
1838 struct mddev *mddev = &rs->md;
1839
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001840 if (!mddev->pers || !mddev->pers->check_reshape)
1841 rs->ti->error = "Reshape not supported";
1842 else if (mddev->degraded)
1843 rs->ti->error = "Can't reshape degraded raid set";
1844 else if (rs_is_recovering(rs))
1845 rs->ti->error = "Convert request on recovering raid set prohibited";
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001846 else if (rs_is_reshaping(rs))
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001847 rs->ti->error = "raid set already reshaping!";
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02001848 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
1849 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001850 else
1851 return 0;
1852
1853 return -EPERM;
1854}
1855
NeilBrown3cb03002011-10-11 16:45:26 +11001856static int read_disk_sb(struct md_rdev *rdev, int size)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001857{
1858 BUG_ON(!rdev->sb_page);
1859
1860 if (rdev->sb_loaded)
1861 return 0;
1862
Heinz Mauelshagen0a7b8182016-06-15 18:45:56 +02001863 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
Jonathan E Brassow04475682012-03-28 18:41:26 +01001864 DMERR("Failed to read superblock of device at position %d",
1865 rdev->raid_disk);
Jonathan Brassowc32fb9e2012-05-22 13:55:31 +10001866 md_error(rdev->mddev, rdev);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001867 return -EINVAL;
1868 }
1869
1870 rdev->sb_loaded = 1;
1871
1872 return 0;
1873}
1874
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001875static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1876{
1877 failed_devices[0] = le64_to_cpu(sb->failed_devices);
1878 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
1879
Mike Snitzer42863252016-06-02 12:27:46 -04001880 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001881 int i = ARRAY_SIZE(sb->extended_failed_devices);
1882
1883 while (i--)
1884 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
1885 }
1886}
1887
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001888static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1889{
1890 int i = ARRAY_SIZE(sb->extended_failed_devices);
1891
1892 sb->failed_devices = cpu_to_le64(failed_devices[0]);
1893 while (i--)
1894 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
1895}
1896
1897/*
1898 * Synchronize the superblock members with the raid set properties
1899 *
1900 * All superblock data is little endian.
1901 */
NeilBrownfd01b882011-10-11 16:47:53 +11001902static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001903{
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001904 bool update_failed_devices = false;
1905 unsigned int i;
1906 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001907 struct dm_raid_superblock *sb;
Jonathan Brassow81f382f2012-05-22 13:55:30 +10001908 struct raid_set *rs = container_of(mddev, struct raid_set, md);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001909
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001910 /* No metadata device, no superblock */
1911 if (!rdev->meta_bdev)
1912 return;
1913
1914 BUG_ON(!rdev->sb_page);
1915
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001916 sb = page_address(rdev->sb_page);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001917
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001918 sb_retrieve_failed_devices(sb, failed_devices);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001919
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001920 for (i = 0; i < rs->raid_disks; i++)
1921 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
1922 update_failed_devices = true;
1923 set_bit(i, (void *) failed_devices);
1924 }
1925
1926 if (update_failed_devices)
1927 sb_update_failed_devices(sb, failed_devices);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001928
1929 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001930 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001931
1932 sb->num_devices = cpu_to_le32(mddev->raid_disks);
1933 sb->array_position = cpu_to_le32(rdev->raid_disk);
1934
1935 sb->events = cpu_to_le64(mddev->events);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001936
1937 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
1938 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
1939
1940 sb->level = cpu_to_le32(mddev->level);
1941 sb->layout = cpu_to_le32(mddev->layout);
1942 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001943
1944 sb->new_level = cpu_to_le32(mddev->new_level);
1945 sb->new_layout = cpu_to_le32(mddev->new_layout);
1946 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
1947
1948 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1949
1950 smp_rmb(); /* Make sure we access most recent reshape position */
1951 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1952 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
1953 /* Flag ongoing reshape */
1954 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
1955
1956 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
1957 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
Mike Snitzer42863252016-06-02 12:27:46 -04001958 } else {
1959 /* Clear reshape flags */
1960 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
1961 }
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001962
1963 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
1964 sb->data_offset = cpu_to_le64(rdev->data_offset);
1965 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
1966 sb->sectors = cpu_to_le64(rdev->sectors);
Heinz Mauelshagenb2a48722016-08-03 17:47:04 +02001967 sb->incompat_features = cpu_to_le32(0);
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02001968
1969 /* Zero out the rest of the payload after the size of the superblock */
1970 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001971}
1972
1973/*
1974 * super_load
1975 *
1976 * This function creates a superblock if one is not found on the device
1977 * and will decide which superblock to use if there's a choice.
1978 *
1979 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1980 */
NeilBrown3cb03002011-10-11 16:45:26 +11001981static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001982{
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02001983 int r;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001984 struct dm_raid_superblock *sb;
1985 struct dm_raid_superblock *refsb;
1986 uint64_t events_sb, events_refsb;
1987
1988 rdev->sb_start = 0;
Heinz Mauelshagen40d43c42014-10-17 13:38:50 +02001989 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
1990 if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
1991 DMERR("superblock size of a logical block is no longer valid");
1992 return -EINVAL;
1993 }
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001994
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02001995 r = read_disk_sb(rdev, rdev->sb_size);
1996 if (r)
1997 return r;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001998
1999 sb = page_address(rdev->sb_page);
Jonathan E Brassow3aa3b2b2012-03-07 19:09:47 +00002000
2001 /*
2002 * Two cases that we want to write new superblocks and rebuild:
2003 * 1) New device (no matching magic number)
2004 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2005 */
2006 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2007 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002008 super_sync(rdev->mddev, rdev);
2009
2010 set_bit(FirstUse, &rdev->flags);
Mike Snitzer9b6e5422016-06-02 11:48:09 -04002011 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002012
2013 /* Force writing of superblocks to disk */
2014 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
2015
2016 /* Any superblock is better than none, choose that if given */
2017 return refdev ? 0 : 1;
2018 }
2019
2020 if (!refdev)
2021 return 1;
2022
2023 events_sb = le64_to_cpu(sb->events);
2024
2025 refsb = page_address(refdev->sb_page);
2026 events_refsb = le64_to_cpu(refsb->events);
2027
2028 return (events_sb > events_refsb) ? 1 : 0;
2029}
2030
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002031static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002032{
2033 int role;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002034 unsigned int d;
2035 struct mddev *mddev = &rs->md;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002036 uint64_t events_sb;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002037 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002038 struct dm_raid_superblock *sb;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002039 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11002040 struct md_rdev *r;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002041 struct dm_raid_superblock *sb2;
2042
2043 sb = page_address(rdev->sb_page);
2044 events_sb = le64_to_cpu(sb->events);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002045
2046 /*
2047 * Initialise to 1 if this is a new superblock.
2048 */
2049 mddev->events = events_sb ? : 1;
2050
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002051 mddev->reshape_position = MaxSector;
2052
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002053 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002054 * Reshaping is supported, e.g. reshape_position is valid
2055 * in superblock and superblock content is authoritative.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002056 */
Mike Snitzer42863252016-06-02 12:27:46 -04002057 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002058 /* Superblock is authoritative wrt given raid set layout! */
2059 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2060 mddev->level = le32_to_cpu(sb->level);
2061 mddev->layout = le32_to_cpu(sb->layout);
2062 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2063 mddev->new_level = le32_to_cpu(sb->new_level);
2064 mddev->new_layout = le32_to_cpu(sb->new_layout);
2065 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2066 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2067 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2068
2069 /* raid was reshaping and got interrupted */
Mike Snitzer42863252016-06-02 12:27:46 -04002070 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2071 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002072 DMERR("Reshape requested but raid set is still reshaping");
2073 return -EINVAL;
2074 }
2075
2076 if (mddev->delta_disks < 0 ||
Mike Snitzer42863252016-06-02 12:27:46 -04002077 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002078 mddev->reshape_backwards = 1;
2079 else
2080 mddev->reshape_backwards = 0;
2081
2082 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2083 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2084 }
2085
2086 } else {
2087 /*
Mike Snitzer9b6e5422016-06-02 11:48:09 -04002088 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002089 */
Heinz Mauelshagenb052b072016-10-17 21:20:07 +02002090 if (le32_to_cpu(sb->level) != mddev->new_level) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002091 DMERR("Reshaping/takeover raid sets not yet supported. (raid level/stripes/size change)");
2092 return -EINVAL;
2093 }
Heinz Mauelshagenb052b072016-10-17 21:20:07 +02002094 if (le32_to_cpu(sb->layout) != mddev->new_layout) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002095 DMERR("Reshaping raid sets not yet supported. (raid layout change)");
Mike Snitzer43157842016-05-30 13:03:37 -04002096 DMERR(" 0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
2097 DMERR(" Old layout: %s w/ %d copies",
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002098 raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
2099 raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
Mike Snitzer43157842016-05-30 13:03:37 -04002100 DMERR(" New layout: %s w/ %d copies",
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002101 raid10_md_layout_to_format(mddev->layout),
2102 raid10_md_layout_to_copies(mddev->layout));
2103 return -EINVAL;
2104 }
Heinz Mauelshagenb052b072016-10-17 21:20:07 +02002105 if (le32_to_cpu(sb->stripe_sectors) != mddev->new_chunk_sectors) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002106 DMERR("Reshaping raid sets not yet supported. (stripe sectors change)");
2107 return -EINVAL;
2108 }
2109
2110 /* We can only change the number of devices in raid1 with old (i.e. pre 1.0.7) metadata */
2111 if (!rt_is_raid1(rs->raid_type) &&
2112 (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
2113 DMERR("Reshaping raid sets not yet supported. (device count change from %u to %u)",
2114 sb->num_devices, mddev->raid_disks);
2115 return -EINVAL;
2116 }
2117
Heinz Mauelshagenb052b072016-10-17 21:20:07 +02002118 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2119
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002120 /* Table line is checked vs. authoritative superblock */
2121 rs_set_new(rs);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002122 }
2123
Mike Snitzer42863252016-06-02 12:27:46 -04002124 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002125 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2126
2127 /*
2128 * During load, we set FirstUse if a new superblock was written.
2129 * There are two reasons we might not have a superblock:
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002130 * 1) The raid set is brand new - in which case, all of the
Mike Snitzer43157842016-05-30 13:03:37 -04002131 * devices must have their In_sync bit set. Also,
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002132 * recovery_cp must be 0, unless forced.
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002133 * 2) This is a new device being added to an old raid set
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002134 * and the new device needs to be rebuilt - in which
2135 * case the In_sync bit will /not/ be set and
2136 * recovery_cp must be MaxSector.
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002137 * 3) This is/are a new device(s) being added to an old
2138 * raid set during takeover to a higher raid level
2139 * to provide capacity for redundancy or during reshape
2140 * to add capacity to grow the raid set.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002141 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002142 d = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11002143 rdev_for_each(r, mddev) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002144 if (test_bit(FirstUse, &r->flags))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002145 new_devs++;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002146
2147 if (!test_bit(In_sync, &r->flags)) {
2148 DMINFO("Device %d specified for rebuild; clearing superblock",
2149 r->raid_disk);
2150 rebuilds++;
2151
2152 if (test_bit(FirstUse, &r->flags))
2153 rebuild_and_new++;
2154 }
2155
2156 d++;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002157 }
2158
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002159 if (new_devs == rs->raid_disks || !rebuilds) {
2160 /* Replace a broken device */
2161 if (new_devs == 1 && !rs->delta_disks)
2162 ;
2163 if (new_devs == rs->raid_disks) {
2164 DMINFO("Superblocks created for new raid set");
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002165 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002166 } else if (new_devs != rebuilds &&
2167 new_devs != rs->delta_disks) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002168 DMERR("New device injected into existing raid set without "
2169 "'delta_disks' or 'rebuild' parameter specified");
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002170 return -EINVAL;
2171 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002172 } else if (new_devs && new_devs != rebuilds) {
2173 DMERR("%u 'rebuild' devices cannot be injected into"
2174 " a raid set with %u other first-time devices",
2175 rebuilds, new_devs);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002176 return -EINVAL;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002177 } else if (rebuilds) {
2178 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2179 DMERR("new device%s provided without 'rebuild'",
2180 new_devs > 1 ? "s" : "");
2181 return -EINVAL;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002182 } else if (rs_is_recovering(rs)) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002183 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2184 (unsigned long long) mddev->recovery_cp);
2185 return -EINVAL;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002186 } else if (rs_is_reshaping(rs)) {
2187 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2188 (unsigned long long) mddev->reshape_position);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002189 return -EINVAL;
2190 }
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002191 }
2192
2193 /*
2194 * Now we set the Faulty bit for those devices that are
2195 * recorded in the superblock as failed.
2196 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002197 sb_retrieve_failed_devices(sb, failed_devices);
NeilBrowndafb20f2012-03-19 12:46:39 +11002198 rdev_for_each(r, mddev) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002199 if (!r->sb_page)
2200 continue;
2201 sb2 = page_address(r->sb_page);
2202 sb2->failed_devices = 0;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002203 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002204
2205 /*
2206 * Check for any device re-ordering.
2207 */
2208 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2209 role = le32_to_cpu(sb2->array_position);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002210 if (role < 0)
2211 continue;
2212
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002213 if (role != r->raid_disk) {
Mike Snitzere6ca5e12016-06-02 15:27:22 -04002214 if (__is_raid10_near(mddev->layout)) {
2215 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002216 rs->raid_disks % rs->raid10_copies) {
2217 rs->ti->error =
2218 "Cannot change raid10 near set to odd # of devices!";
2219 return -EINVAL;
2220 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002221
2222 sb2->array_position = cpu_to_le32(r->raid_disk);
2223
2224 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002225 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2226 !rt_is_raid1(rs->raid_type)) {
2227 rs->ti->error = "Cannot change device positions in raid set";
2228 return -EINVAL;
2229 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002230
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002231 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002232 }
2233
2234 /*
2235 * Partial recovery is performed on
2236 * returning failed devices.
2237 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002238 if (test_bit(role, (void *) failed_devices))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002239 set_bit(Faulty, &r->flags);
2240 }
2241 }
2242
2243 return 0;
2244}
2245
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002246static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002247{
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002248 struct mddev *mddev = &rs->md;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002249 struct dm_raid_superblock *sb;
2250
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02002251 if (rs_is_raid0(rs) || !rdev->sb_page)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002252 return 0;
2253
2254 sb = page_address(rdev->sb_page);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002255
2256 /*
2257 * If mddev->events is not set, we know we have not yet initialized
2258 * the array.
2259 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002260 if (!mddev->events && super_init_validation(rs, rdev))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002261 return -EINVAL;
2262
Andy Whitcroft5c336772016-10-11 15:16:57 +01002263 if (le32_to_cpu(sb->compat_features) &&
2264 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
Mike Snitzer9b6e5422016-06-02 11:48:09 -04002265 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2266 return -EINVAL;
2267 }
2268
2269 if (sb->incompat_features) {
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002270 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
Heinz Mauelshagen4c9971c2016-04-29 18:59:56 +02002271 return -EINVAL;
2272 }
2273
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002274 /* Enable bitmap creation for RAID levels != 0 */
Heinz Mauelshagen676fa5a2016-05-19 18:49:29 +02002275 mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002276 rdev->mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2277
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002278 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2279 /* Retrieve device size stored in superblock to be prepared for shrink */
2280 rdev->sectors = le64_to_cpu(sb->sectors);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002281 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002282 if (rdev->recovery_offset == MaxSector)
2283 set_bit(In_sync, &rdev->flags);
2284 /*
2285 * If no reshape in progress -> we're recovering single
2286 * disk(s) and have to set the device(s) to out-of-sync
2287 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002288 else if (!rs_is_reshaping(rs))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002289 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002290 }
2291
2292 /*
2293 * If a device comes back, set it as not In_sync and no longer faulty.
2294 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002295 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2296 rdev->recovery_offset = 0;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002297 clear_bit(In_sync, &rdev->flags);
2298 rdev->saved_raid_disk = rdev->raid_disk;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002299 }
2300
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002301 /* Reshape support -> restore repective data offsets */
2302 rdev->data_offset = le64_to_cpu(sb->data_offset);
2303 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002304
2305 return 0;
2306}
2307
2308/*
2309 * Analyse superblocks and select the freshest.
2310 */
2311static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2312{
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002313 int r;
Jonathan E Brassow04475682012-03-28 18:41:26 +01002314 struct raid_dev *dev;
Jonathan Brassowa9ad8522012-04-24 10:23:13 +10002315 struct md_rdev *rdev, *tmp, *freshest;
NeilBrownfd01b882011-10-11 16:47:53 +11002316 struct mddev *mddev = &rs->md;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002317
2318 freshest = NULL;
Jonathan Brassowa9ad8522012-04-24 10:23:13 +10002319 rdev_for_each_safe(rdev, tmp, mddev) {
Jonathan Brassow761becf2012-10-11 13:42:19 +11002320 /*
Heinz Mauelshagenc76d53f2015-04-29 14:03:00 +02002321 * Skipping super_load due to CTR_FLAG_SYNC will cause
Jonathan Brassow761becf2012-10-11 13:42:19 +11002322 * the array to undergo initialization again as
Mike Snitzer43157842016-05-30 13:03:37 -04002323 * though it were new. This is the intended effect
Jonathan Brassow761becf2012-10-11 13:42:19 +11002324 * of the "sync" directive.
2325 *
2326 * When reshaping capability is added, we must ensure
2327 * that the "sync" directive is disallowed during the
2328 * reshape.
2329 */
Mike Snitzer42863252016-06-02 12:27:46 -04002330 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
Jonathan Brassow761becf2012-10-11 13:42:19 +11002331 continue;
2332
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002333 if (!rdev->meta_bdev)
2334 continue;
2335
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002336 r = super_load(rdev, freshest);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002337
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002338 switch (r) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002339 case 1:
2340 freshest = rdev;
2341 break;
2342 case 0:
2343 break;
2344 default:
Heinz Mauelshagen9e7d9362016-08-17 15:36:44 +02002345 /*
2346 * We have to keep any raid0 data/metadata device pairs or
2347 * the MD raid0 personality will fail to start the array.
2348 */
2349 if (rs_is_raid0(rs))
2350 continue;
2351
Jonathan E Brassow04475682012-03-28 18:41:26 +01002352 dev = container_of(rdev, struct raid_dev, rdev);
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002353 if (dev->meta_dev)
2354 dm_put_device(ti, dev->meta_dev);
Jonathan E Brassow04475682012-03-28 18:41:26 +01002355
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002356 dev->meta_dev = NULL;
2357 rdev->meta_bdev = NULL;
Jonathan E Brassow04475682012-03-28 18:41:26 +01002358
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002359 if (rdev->sb_page)
2360 put_page(rdev->sb_page);
Jonathan E Brassow04475682012-03-28 18:41:26 +01002361
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002362 rdev->sb_page = NULL;
Jonathan E Brassow04475682012-03-28 18:41:26 +01002363
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002364 rdev->sb_loaded = 0;
Jonathan E Brassow04475682012-03-28 18:41:26 +01002365
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002366 /*
2367 * We might be able to salvage the data device
2368 * even though the meta device has failed. For
2369 * now, we behave as though '- -' had been
2370 * set for this device in the table.
2371 */
2372 if (dev->data_dev)
2373 dm_put_device(ti, dev->data_dev);
Jonathan E Brassow04475682012-03-28 18:41:26 +01002374
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002375 dev->data_dev = NULL;
2376 rdev->bdev = NULL;
Jonathan E Brassow04475682012-03-28 18:41:26 +01002377
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002378 list_del(&rdev->same_set);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002379 }
2380 }
2381
2382 if (!freshest)
2383 return 0;
2384
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002385 if (validate_raid_redundancy(rs)) {
2386 rs->ti->error = "Insufficient redundancy to activate array";
2387 return -EINVAL;
2388 }
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002389
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002390 /*
2391 * Validation of the freshest device provides the source of
2392 * validation for the remaining devices.
2393 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002394 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2395 if (super_validate(rs, freshest))
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002396 return -EINVAL;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002397
NeilBrowndafb20f2012-03-19 12:46:39 +11002398 rdev_for_each(rdev, mddev)
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002399 if ((rdev != freshest) && super_validate(rs, rdev))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002400 return -EINVAL;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002401 return 0;
2402}
2403
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002404/*
2405 * Adjust data_offset and new_data_offset on all disk members of @rs
2406 * for out of place reshaping if requested by contructor
2407 *
2408 * We need free space at the beginning of each raid disk for forward
2409 * and at the end for backward reshapes which userspace has to provide
2410 * via remapping/reordering of space.
2411 */
2412static int rs_adjust_data_offsets(struct raid_set *rs)
2413{
2414 sector_t data_offset = 0, new_data_offset = 0;
2415 struct md_rdev *rdev;
2416
2417 /* Constructor did not request data offset change */
2418 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2419 if (!rs_is_reshapable(rs))
2420 goto out;
2421
2422 return 0;
2423 }
2424
2425 /* HM FIXME: get InSync raid_dev? */
2426 rdev = &rs->dev[0].rdev;
2427
2428 if (rs->delta_disks < 0) {
2429 /*
2430 * Removing disks (reshaping backwards):
2431 *
2432 * - before reshape: data is at offset 0 and free space
2433 * is at end of each component LV
2434 *
2435 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2436 */
2437 data_offset = 0;
2438 new_data_offset = rs->data_offset;
2439
2440 } else if (rs->delta_disks > 0) {
2441 /*
2442 * Adding disks (reshaping forwards):
2443 *
2444 * - before reshape: data is at offset rs->data_offset != 0 and
2445 * free space is at begin of each component LV
2446 *
2447 * - after reshape: data is at offset 0 on each component LV
2448 */
2449 data_offset = rs->data_offset;
2450 new_data_offset = 0;
2451
2452 } else {
2453 /*
2454 * User space passes in 0 for data offset after having removed reshape space
2455 *
2456 * - or - (data offset != 0)
2457 *
2458 * Changing RAID layout or chunk size -> toggle offsets
2459 *
2460 * - before reshape: data is at offset rs->data_offset 0 and
2461 * free space is at end of each component LV
2462 * -or-
2463 * data is at offset rs->data_offset != 0 and
2464 * free space is at begin of each component LV
2465 *
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02002466 * - after reshape: data is at offset 0 if it was at offset != 0
2467 * or at offset != 0 if it was at offset 0
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002468 * on each component LV
2469 *
2470 */
2471 data_offset = rs->data_offset ? rdev->data_offset : 0;
2472 new_data_offset = data_offset ? 0 : rs->data_offset;
2473 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2474 }
2475
2476 /*
2477 * Make sure we got a minimum amount of free sectors per device
2478 */
2479 if (rs->data_offset &&
2480 to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2481 rs->ti->error = data_offset ? "No space for forward reshape" :
2482 "No space for backward reshape";
2483 return -ENOSPC;
2484 }
2485out:
2486 /* Adjust data offsets on all rdevs */
2487 rdev_for_each(rdev, &rs->md) {
2488 rdev->data_offset = data_offset;
2489 rdev->new_data_offset = new_data_offset;
2490 }
2491
2492 return 0;
2493}
2494
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002495/* Userpace reordered disks -> adjust raid_disk indexes in @rs */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04002496static void __reorder_raid_disk_indexes(struct raid_set *rs)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002497{
2498 int i = 0;
2499 struct md_rdev *rdev;
2500
2501 rdev_for_each(rdev, &rs->md) {
2502 rdev->raid_disk = i++;
2503 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2504 }
2505}
2506
2507/*
2508 * Setup @rs for takeover by a different raid level
2509 */
2510static int rs_setup_takeover(struct raid_set *rs)
2511{
2512 struct mddev *mddev = &rs->md;
2513 struct md_rdev *rdev;
2514 unsigned int d = mddev->raid_disks = rs->raid_disks;
2515 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2516
2517 if (rt_is_raid10(rs->raid_type)) {
2518 if (mddev->level == 0) {
2519 /* Userpace reordered disks -> adjust raid_disk indexes */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04002520 __reorder_raid_disk_indexes(rs);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002521
2522 /* raid0 -> raid10_far layout */
2523 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2524 rs->raid10_copies);
2525 } else if (mddev->level == 1)
2526 /* raid1 -> raid10_near layout */
2527 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2528 rs->raid_disks);
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02002529 else
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002530 return -EINVAL;
2531
2532 }
2533
2534 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2535 mddev->recovery_cp = MaxSector;
2536
2537 while (d--) {
2538 rdev = &rs->dev[d].rdev;
2539
2540 if (test_bit(d, (void *) rs->rebuild_disks)) {
2541 clear_bit(In_sync, &rdev->flags);
2542 clear_bit(Faulty, &rdev->flags);
2543 mddev->recovery_cp = rdev->recovery_offset = 0;
2544 /* Bitmap has to be created when we do an "up" takeover */
2545 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2546 }
2547
2548 rdev->new_data_offset = new_data_offset;
2549 }
2550
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002551 return 0;
2552}
2553
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002554/* Prepare @rs for reshape */
2555static int rs_prepare_reshape(struct raid_set *rs)
2556{
2557 bool reshape;
2558 struct mddev *mddev = &rs->md;
2559
2560 if (rs_is_raid10(rs)) {
2561 if (rs->raid_disks != mddev->raid_disks &&
2562 __is_raid10_near(mddev->layout) &&
2563 rs->raid10_copies &&
2564 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2565 /*
2566 * raid disk have to be multiple of data copies to allow this conversion,
2567 *
2568 * This is actually not a reshape it is a
2569 * rebuild of any additional mirrors per group
2570 */
2571 if (rs->raid_disks % rs->raid10_copies) {
2572 rs->ti->error = "Can't reshape raid10 mirror groups";
2573 return -EINVAL;
2574 }
2575
2576 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2577 __reorder_raid_disk_indexes(rs);
2578 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2579 rs->raid10_copies);
2580 mddev->new_layout = mddev->layout;
2581 reshape = false;
2582 } else
2583 reshape = true;
2584
2585 } else if (rs_is_raid456(rs))
2586 reshape = true;
2587
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002588 else if (rs_is_raid1(rs)) {
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002589 if (rs->delta_disks) {
2590 /* Process raid1 via delta_disks */
2591 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2592 reshape = true;
2593 } else {
2594 /* Process raid1 without delta_disks */
2595 mddev->raid_disks = rs->raid_disks;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002596 reshape = false;
2597 }
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002598 } else {
2599 rs->ti->error = "Called with bogus raid type";
2600 return -EINVAL;
2601 }
2602
2603 if (reshape) {
2604 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2605 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002606 } else if (mddev->raid_disks < rs->raid_disks)
2607 /* Create new superblocks and bitmaps, if any new disks */
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002608 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002609
2610 return 0;
2611}
2612
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002613/*
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002614 *
2615 * - change raid layout
2616 * - change chunk size
2617 * - add disks
2618 * - remove disks
2619 */
2620static int rs_setup_reshape(struct raid_set *rs)
2621{
2622 int r = 0;
2623 unsigned int cur_raid_devs, d;
2624 struct mddev *mddev = &rs->md;
2625 struct md_rdev *rdev;
2626
2627 mddev->delta_disks = rs->delta_disks;
2628 cur_raid_devs = mddev->raid_disks;
2629
2630 /* Ignore impossible layout change whilst adding/removing disks */
2631 if (mddev->delta_disks &&
2632 mddev->layout != mddev->new_layout) {
2633 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2634 mddev->new_layout = mddev->layout;
2635 }
2636
2637 /*
2638 * Adjust array size:
2639 *
2640 * - in case of adding disks, array size has
2641 * to grow after the disk adding reshape,
2642 * which'll hapen in the event handler;
2643 * reshape will happen forward, so space has to
2644 * be available at the beginning of each disk
2645 *
2646 * - in case of removing disks, array size
2647 * has to shrink before starting the reshape,
2648 * which'll happen here;
2649 * reshape will happen backward, so space has to
2650 * be available at the end of each disk
2651 *
2652 * - data_offset and new_data_offset are
Heinz Mauelshagenae3c6cf2016-06-15 22:27:40 +02002653 * adjusted for aforementioned out of place
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002654 * reshaping based on userspace passing in
2655 * the "data_offset <sectors>" key/value
Heinz Mauelshagenae3c6cf2016-06-15 22:27:40 +02002656 * pair via the constructor
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002657 */
2658
2659 /* Add disk(s) */
2660 if (rs->delta_disks > 0) {
2661 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2662 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2663 rdev = &rs->dev[d].rdev;
2664 clear_bit(In_sync, &rdev->flags);
2665
2666 /*
2667 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2668 * by md, which'll store that erroneously in the superblock on reshape
2669 */
2670 rdev->saved_raid_disk = -1;
2671 rdev->raid_disk = d;
2672
2673 rdev->sectors = mddev->dev_sectors;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002674 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002675 }
2676
2677 mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2678
2679 /* Remove disk(s) */
2680 } else if (rs->delta_disks < 0) {
2681 r = rs_set_dev_and_array_sectors(rs, true);
2682 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2683
2684 /* Change layout and/or chunk size */
2685 } else {
2686 /*
2687 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2688 *
2689 * keeping number of disks and do layout change ->
2690 *
2691 * toggle reshape_backward depending on data_offset:
2692 *
2693 * - free space upfront -> reshape forward
2694 *
2695 * - free space at the end -> reshape backward
2696 *
2697 *
2698 * This utilizes free reshape space avoiding the need
2699 * for userspace to move (parts of) LV segments in
2700 * case of layout/chunksize change (for disk
2701 * adding/removing reshape space has to be at
2702 * the proper address (see above with delta_disks):
2703 *
2704 * add disk(s) -> begin
2705 * remove disk(s)-> end
2706 */
2707 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2708 }
2709
2710 return r;
2711}
2712
2713/*
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002714 * Enable/disable discard support on RAID set depending on
2715 * RAID level and discard properties of underlying RAID members.
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002716 */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002717static void configure_discard_support(struct raid_set *rs)
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002718{
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002719 int i;
2720 bool raid456;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002721 struct dm_target *ti = rs->ti;
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002722
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002723 /* Assume discards not supported until after checks below. */
2724 ti->discards_supported = false;
2725
2726 /* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002727 raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002728
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +02002729 for (i = 0; i < rs->raid_disks; i++) {
Heinz Mauelshagend20c4b02014-10-29 19:02:27 +01002730 struct request_queue *q;
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002731
Heinz Mauelshagend20c4b02014-10-29 19:02:27 +01002732 if (!rs->dev[i].rdev.bdev)
2733 continue;
2734
2735 q = bdev_get_queue(rs->dev[i].rdev.bdev);
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002736 if (!q || !blk_queue_discard(q))
2737 return;
2738
2739 if (raid456) {
2740 if (!q->limits.discard_zeroes_data)
2741 return;
2742 if (!devices_handle_discard_safely) {
2743 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2744 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2745 return;
2746 }
2747 }
2748 }
2749
2750 /* All RAID members properly support discards */
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002751 ti->discards_supported = true;
2752
2753 /*
2754 * RAID1 and RAID10 personalities require bio splitting,
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002755 * RAID0/4/5/6 don't and process large discard bios properly.
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002756 */
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002757 ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002758 ti->num_discard_bios = 1;
2759}
2760
2761/*
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002762 * Construct a RAID0/1/10/4/5/6 mapping:
NeilBrown9d09e662011-01-13 20:00:02 +00002763 * Args:
Mike Snitzer43157842016-05-30 13:03:37 -04002764 * <raid_type> <#raid_params> <raid_params>{0,} \
2765 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
NeilBrown9d09e662011-01-13 20:00:02 +00002766 *
Mike Snitzer43157842016-05-30 13:03:37 -04002767 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
NeilBrown9d09e662011-01-13 20:00:02 +00002768 * details on possible <raid_params>.
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002769 *
2770 * Userspace is free to initialize the metadata devices, hence the superblocks to
2771 * enforce recreation based on the passed in table parameters.
2772 *
NeilBrown9d09e662011-01-13 20:00:02 +00002773 */
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02002774static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
NeilBrown9d09e662011-01-13 20:00:02 +00002775{
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002776 int r;
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002777 bool resize;
NeilBrown9d09e662011-01-13 20:00:02 +00002778 struct raid_type *rt;
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02002779 unsigned int num_raid_params, num_raid_devs;
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02002780 sector_t calculated_dev_sectors;
NeilBrown9d09e662011-01-13 20:00:02 +00002781 struct raid_set *rs = NULL;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002782 const char *arg;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002783 struct rs_layout rs_layout;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002784 struct dm_arg_set as = { argc, argv }, as_nrd;
2785 struct dm_arg _args[] = {
2786 { 0, as.argc, "Cannot understand number of raid parameters" },
2787 { 1, 254, "Cannot understand number of raid devices parameters" }
2788 };
NeilBrown9d09e662011-01-13 20:00:02 +00002789
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002790 /* Must have <raid_type> */
2791 arg = dm_shift_arg(&as);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002792 if (!arg) {
2793 ti->error = "No arguments";
2794 return -EINVAL;
2795 }
NeilBrown9d09e662011-01-13 20:00:02 +00002796
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002797 rt = get_raid_type(arg);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002798 if (!rt) {
2799 ti->error = "Unrecognised raid_type";
2800 return -EINVAL;
2801 }
NeilBrown9d09e662011-01-13 20:00:02 +00002802
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002803 /* Must have <#raid_params> */
2804 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
Mike Snitzer43157842016-05-30 13:03:37 -04002805 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +00002806
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002807 /* number of raid device tupples <meta_dev data_dev> */
2808 as_nrd = as;
2809 dm_consume_args(&as_nrd, num_raid_params);
2810 _args[1].max = (as_nrd.argc - 1) / 2;
2811 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
Mike Snitzer43157842016-05-30 13:03:37 -04002812 return -EINVAL;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002813
Mike Snitzerbb91a632016-06-02 12:06:54 -04002814 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002815 ti->error = "Invalid number of supplied raid devices";
2816 return -EINVAL;
2817 }
NeilBrown9d09e662011-01-13 20:00:02 +00002818
Mike Snitzerbfcee0e2016-06-02 15:08:09 -04002819 rs = raid_set_alloc(ti, rt, num_raid_devs);
NeilBrown9d09e662011-01-13 20:00:02 +00002820 if (IS_ERR(rs))
2821 return PTR_ERR(rs);
2822
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02002823 r = parse_raid_params(rs, &as, num_raid_params);
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002824 if (r)
NeilBrown9d09e662011-01-13 20:00:02 +00002825 goto bad;
2826
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02002827 r = parse_dev_params(rs, &as);
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002828 if (r)
NeilBrown9d09e662011-01-13 20:00:02 +00002829 goto bad;
2830
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002831 rs->md.sync_super = super_sync;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002832
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02002833 /*
2834 * Calculate ctr requested array and device sizes to allow
2835 * for superblock analysis needing device sizes defined.
2836 *
2837 * Any existing superblock will overwrite the array and device sizes
2838 */
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002839 r = rs_set_dev_and_array_sectors(rs, false);
2840 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002841 goto bad;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002842
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02002843 calculated_dev_sectors = rs->dev[0].rdev.sectors;
2844
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002845 /*
2846 * Backup any new raid set level, layout, ...
2847 * requested to be able to compare to superblock
2848 * members for conversion decisions.
2849 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002850 rs_config_backup(rs, &rs_layout);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002851
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002852 r = analyse_superblocks(ti, rs);
2853 if (r)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002854 goto bad;
2855
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02002856 resize = calculated_dev_sectors != rs->dev[0].rdev.sectors;
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02002857
NeilBrown9d09e662011-01-13 20:00:02 +00002858 INIT_WORK(&rs->md.event_work, do_table_event);
NeilBrown9d09e662011-01-13 20:00:02 +00002859 ti->private = rs;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00002860 ti->num_flush_bios = 1;
NeilBrown9d09e662011-01-13 20:00:02 +00002861
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002862 /* Restore any requested new layout for conversion decision */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002863 rs_config_restore(rs, &rs_layout);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002864
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002865 /*
2866 * Now that we have any superblock metadata available,
2867 * check for new, recovering, reshaping, to be taken over,
2868 * to be reshaped or an existing, unchanged raid set to
2869 * run in sequence.
2870 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002871 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02002872 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
2873 if (rs_is_raid6(rs) &&
2874 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
2875 ti->error = "'nosync' not allowed for new raid6 set";
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002876 r = -EINVAL;
2877 goto bad;
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02002878 }
2879 rs_setup_recovery(rs, 0);
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02002880 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2881 rs_set_new(rs);
2882 } else if (rs_is_recovering(rs)) {
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002883 /* A recovering raid set may be resized */
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02002884 ; /* skip setup rs */
2885 } else if (rs_is_reshaping(rs)) {
2886 /* Have to reject size change request during reshape */
2887 if (resize) {
2888 ti->error = "Can't resize a reshaping raid set";
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002889 r = -EPERM;
2890 goto bad;
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02002891 }
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002892 /* skip setup rs */
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02002893 } else if (rs_takeover_requested(rs)) {
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002894 if (rs_is_reshaping(rs)) {
2895 ti->error = "Can't takeover a reshaping raid set";
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002896 r = -EPERM;
2897 goto bad;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002898 }
2899
2900 /*
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02002901 * If a takeover is needed, userspace sets any additional
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002902 * devices to rebuild and we can check for a valid request here.
2903 *
2904 * If acceptible, set the level to the new requested
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02002905 * one, prohibit requesting recovery, allow the raid
2906 * set to run and store superblocks during resume.
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002907 */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002908 r = rs_check_takeover(rs);
2909 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002910 goto bad;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002911
2912 r = rs_setup_takeover(rs);
2913 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002914 goto bad;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002915
Mike Snitzer42863252016-06-02 12:27:46 -04002916 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002917 /* Takeover ain't recovery, so disable recovery */
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02002918 rs_setup_recovery(rs, MaxSector);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02002919 rs_set_new(rs);
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002920 } else if (rs_reshape_requested(rs)) {
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002921 /*
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002922 * We can only prepare for a reshape here, because the
2923 * raid set needs to run to provide the repective reshape
2924 * check functions via its MD personality instance.
2925 *
2926 * So do the reshape check after md_run() succeeded.
2927 */
2928 r = rs_prepare_reshape(rs);
2929 if (r)
2930 return r;
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04002931
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002932 /* Reshaping ain't recovery, so disable recovery */
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02002933 rs_setup_recovery(rs, MaxSector);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02002934 rs_set_cur(rs);
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02002935 } else {
2936 /* May not set recovery when a device rebuild is requested */
Heinz Mauelshagen37f10be2016-06-24 23:21:37 +02002937 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
2938 rs_setup_recovery(rs, MaxSector);
2939 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2940 } else
2941 rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02002942 0 : (resize ? calculated_dev_sectors : MaxSector));
2943 rs_set_cur(rs);
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02002944 }
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002945
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002946 /* If constructor requested it, change data and new_data offsets */
2947 r = rs_adjust_data_offsets(rs);
2948 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002949 goto bad;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002950
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002951 /* Start raid set read-only and assumed clean to change in raid_resume() */
2952 rs->md.ro = 1;
2953 rs->md.in_sync = 1;
2954 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002955
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002956 /* Has to be held on running the array */
2957 mddev_lock_nointr(&rs->md);
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002958 r = md_run(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00002959 rs->md.in_sync = 0; /* Assume already marked dirty */
NeilBrown9d09e662011-01-13 20:00:02 +00002960
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002961 if (r) {
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002962 ti->error = "Failed to run raid array";
2963 mddev_unlock(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00002964 goto bad;
2965 }
2966
2967 rs->callbacks.congested_fn = raid_is_congested;
NeilBrown9d09e662011-01-13 20:00:02 +00002968 dm_table_add_target_callbacks(ti->table, &rs->callbacks);
2969
Jonathan Brassow32737272011-08-02 12:32:07 +01002970 mddev_suspend(&rs->md);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002971
2972 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
2973 if (rs_is_raid456(rs)) {
2974 r = rs_set_raid456_stripe_cache(rs);
2975 if (r)
2976 goto bad_stripe_cache;
2977 }
2978
2979 /* Now do an early reshape check */
2980 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
2981 r = rs_check_reshape(rs);
2982 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02002983 goto bad_check_reshape;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002984
2985 /* Restore new, ctr requested layout to perform check */
2986 rs_config_restore(rs, &rs_layout);
2987
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002988 if (rs->md.pers->start_reshape) {
2989 r = rs->md.pers->check_reshape(&rs->md);
2990 if (r) {
2991 ti->error = "Reshape check failed";
2992 goto bad_check_reshape;
2993 }
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002994 }
2995 }
2996
Heinz Mauelshagen461f2722016-11-29 22:37:30 +01002997 /* Disable/enable discard support on raid set. */
2998 configure_discard_support(rs);
2999
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003000 mddev_unlock(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00003001 return 0;
3002
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003003bad_stripe_cache:
3004bad_check_reshape:
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05003005 md_stop(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00003006bad:
Mike Snitzerbfcee0e2016-06-02 15:08:09 -04003007 raid_set_free(rs);
NeilBrown9d09e662011-01-13 20:00:02 +00003008
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003009 return r;
NeilBrown9d09e662011-01-13 20:00:02 +00003010}
3011
3012static void raid_dtr(struct dm_target *ti)
3013{
3014 struct raid_set *rs = ti->private;
3015
3016 list_del_init(&rs->callbacks.list);
3017 md_stop(&rs->md);
Mike Snitzerbfcee0e2016-06-02 15:08:09 -04003018 raid_set_free(rs);
NeilBrown9d09e662011-01-13 20:00:02 +00003019}
3020
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00003021static int raid_map(struct dm_target *ti, struct bio *bio)
NeilBrown9d09e662011-01-13 20:00:02 +00003022{
3023 struct raid_set *rs = ti->private;
NeilBrownfd01b882011-10-11 16:47:53 +11003024 struct mddev *mddev = &rs->md;
NeilBrown9d09e662011-01-13 20:00:02 +00003025
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003026 /*
3027 * If we're reshaping to add disk(s)), ti->len and
3028 * mddev->array_sectors will differ during the process
3029 * (ti->len > mddev->array_sectors), so we have to requeue
3030 * bios with addresses > mddev->array_sectors here or
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02003031 * there will occur accesses past EOD of the component
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003032 * data images thus erroring the raid set.
3033 */
3034 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3035 return DM_MAPIO_REQUEUE;
3036
NeilBrown9d09e662011-01-13 20:00:02 +00003037 mddev->pers->make_request(mddev, bio);
3038
3039 return DM_MAPIO_SUBMITTED;
3040}
3041
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003042/* Return string describing the current sync action of @mddev */
Jonathan Brassowbe836512013-04-24 11:42:43 +10003043static const char *decipher_sync_action(struct mddev *mddev)
3044{
3045 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3046 return "frozen";
3047
3048 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3049 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3050 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3051 return "reshape";
3052
3053 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3054 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3055 return "resync";
3056 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3057 return "check";
3058 return "repair";
3059 }
3060
3061 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3062 return "recover";
3063 }
3064
3065 return "idle";
3066}
3067
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003068/*
3069 * Return status string @rdev
3070 *
3071 * Status characters:
3072 *
3073 * 'D' = Dead/Failed device
3074 * 'a' = Alive but not in-sync
3075 * 'A' = Alive and in-sync
3076 */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04003077static const char *__raid_dev_status(struct md_rdev *rdev, bool array_in_sync)
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003078{
3079 if (test_bit(Faulty, &rdev->flags))
3080 return "D";
3081 else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
3082 return "a";
3083 else
3084 return "A";
3085}
3086
3087/* Helper to return resync/reshape progress for @rs and @array_in_sync */
3088static sector_t rs_get_progress(struct raid_set *rs,
3089 sector_t resync_max_sectors, bool *array_in_sync)
3090{
3091 sector_t r, recovery_cp, curr_resync_completed;
3092 struct mddev *mddev = &rs->md;
3093
3094 curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
3095 recovery_cp = mddev->recovery_cp;
3096 *array_in_sync = false;
3097
3098 if (rs_is_raid0(rs)) {
3099 r = resync_max_sectors;
3100 *array_in_sync = true;
3101
3102 } else {
3103 r = mddev->reshape_position;
3104
3105 /* Reshape is relative to the array size */
3106 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
3107 r != MaxSector) {
3108 if (r == MaxSector) {
3109 *array_in_sync = true;
3110 r = resync_max_sectors;
3111 } else {
3112 /* Got to reverse on backward reshape */
3113 if (mddev->reshape_backwards)
3114 r = mddev->array_sectors - r;
3115
3116 /* Devide by # of data stripes */
3117 sector_div(r, mddev_data_stripes(rs));
3118 }
3119
3120 /* Sync is relative to the component device size */
3121 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3122 r = curr_resync_completed;
3123 else
3124 r = recovery_cp;
3125
3126 if (r == MaxSector) {
3127 /*
3128 * Sync complete.
3129 */
3130 *array_in_sync = true;
3131 r = resync_max_sectors;
3132 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3133 /*
3134 * If "check" or "repair" is occurring, the raid set has
3135 * undergone an initial sync and the health characters
3136 * should not be 'a' anymore.
3137 */
3138 *array_in_sync = true;
3139 } else {
3140 struct md_rdev *rdev;
3141
3142 /*
3143 * The raid set may be doing an initial sync, or it may
Mike Snitzer43157842016-05-30 13:03:37 -04003144 * be rebuilding individual components. If all the
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003145 * devices are In_sync, then it is the raid set that is
3146 * being initialized.
3147 */
3148 rdev_for_each(rdev, mddev)
3149 if (!test_bit(In_sync, &rdev->flags))
3150 *array_in_sync = true;
3151#if 0
3152 r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3153#endif
3154 }
3155 }
3156
3157 return r;
3158}
3159
3160/* Helper to return @dev name or "-" if !@dev */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04003161static const char *__get_dev_name(struct dm_dev *dev)
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003162{
3163 return dev ? dev->name : "-";
3164}
3165
Mikulas Patockafd7c0922013-03-01 22:45:44 +00003166static void raid_status(struct dm_target *ti, status_type_t type,
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003167 unsigned int status_flags, char *result, unsigned int maxlen)
NeilBrown9d09e662011-01-13 20:00:02 +00003168{
3169 struct raid_set *rs = ti->private;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003170 struct mddev *mddev = &rs->md;
3171 struct r5conf *conf = mddev->private;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003172 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003173 bool array_in_sync;
3174 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3175 unsigned int sz = 0;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003176 unsigned int rebuild_disks;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003177 unsigned int write_mostly_params = 0;
3178 sector_t progress, resync_max_sectors, resync_mismatches;
3179 const char *sync_action;
3180 struct raid_type *rt;
3181 struct md_rdev *rdev;
NeilBrown9d09e662011-01-13 20:00:02 +00003182
3183 switch (type) {
3184 case STATUSTYPE_INFO:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003185 /* *Should* always succeed */
3186 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3187 if (!rt)
3188 return;
NeilBrown9d09e662011-01-13 20:00:02 +00003189
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003190 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
NeilBrown9d09e662011-01-13 20:00:02 +00003191
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003192 /* Access most recent mddev properties for status output */
3193 smp_rmb();
3194 /* Get sensible max sectors even if raid set not yet started */
Mike Snitzer42863252016-06-02 12:27:46 -04003195 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003196 mddev->resync_max_sectors : mddev->dev_sectors;
3197 progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3198 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003199 atomic64_read(&mddev->resync_mismatches) : 0;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003200 sync_action = decipher_sync_action(&rs->md);
3201
3202 /* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
Heinz Mauelshagen32682402016-06-30 21:32:20 +02003203 rdev_for_each(rdev, mddev)
3204 DMEMIT(__raid_dev_status(rdev, array_in_sync));
Jonathan Brassowbe836512013-04-24 11:42:43 +10003205
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003206 /*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003207 * In-sync/Reshape ratio:
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003208 * The in-sync ratio shows the progress of:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003209 * - Initializing the raid set
3210 * - Rebuilding a subset of devices of the raid set
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003211 * The user can distinguish between the two by referring
3212 * to the status characters.
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003213 *
3214 * The reshape ratio shows the progress of
3215 * changing the raid layout or the number of
3216 * disks of a raid set
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003217 */
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003218 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3219 (unsigned long long) resync_max_sectors);
NeilBrown9d09e662011-01-13 20:00:02 +00003220
Jonathan Brassowbe836512013-04-24 11:42:43 +10003221 /*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003222 * v1.5.0+:
3223 *
Jonathan Brassowbe836512013-04-24 11:42:43 +10003224 * Sync action:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003225 * See Documentation/device-mapper/dm-raid.txt for
Jonathan Brassowbe836512013-04-24 11:42:43 +10003226 * information on each of these states.
3227 */
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003228 DMEMIT(" %s", sync_action);
Jonathan Brassowbe836512013-04-24 11:42:43 +10003229
3230 /*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003231 * v1.5.0+:
3232 *
Jonathan Brassowbe836512013-04-24 11:42:43 +10003233 * resync_mismatches/mismatch_cnt
3234 * This field shows the number of discrepancies found when
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003235 * performing a "check" of the raid set.
Jonathan Brassowbe836512013-04-24 11:42:43 +10003236 */
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003237 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3238
3239 /*
Mike Snitzer9b6e5422016-06-02 11:48:09 -04003240 * v1.9.0+:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003241 *
3242 * data_offset (needed for out of space reshaping)
3243 * This field shows the data offset into the data
3244 * image LV where the first stripes data starts.
3245 *
3246 * We keep data_offset equal on all raid disks of the set,
3247 * so retrieving it from the first raid disk is sufficient.
3248 */
3249 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
NeilBrown9d09e662011-01-13 20:00:02 +00003250 break;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003251
NeilBrown9d09e662011-01-13 20:00:02 +00003252 case STATUSTYPE_TABLE:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003253 /* Report the table line string you would use to construct this raid set */
NeilBrown9d09e662011-01-13 20:00:02 +00003254
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003255 /* Calculate raid parameter count */
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003256 for (i = 0; i < rs->raid_disks; i++)
3257 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003258 write_mostly_params += 2;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003259 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3260 raid_param_cnt += rebuild_disks * 2 +
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003261 write_mostly_params +
3262 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3263 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3264 /* Emit table line */
3265 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
Mike Snitzer42863252016-06-02 12:27:46 -04003266 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003267 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003268 raid10_md_layout_to_format(mddev->layout));
Mike Snitzer42863252016-06-02 12:27:46 -04003269 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003270 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003271 raid10_md_layout_to_copies(mddev->layout));
Mike Snitzer42863252016-06-02 12:27:46 -04003272 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003273 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
Mike Snitzer42863252016-06-02 12:27:46 -04003274 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003275 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
Mike Snitzer42863252016-06-02 12:27:46 -04003276 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003277 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003278 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
Mike Snitzer42863252016-06-02 12:27:46 -04003279 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003280 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003281 (unsigned long long) rs->data_offset);
Mike Snitzer42863252016-06-02 12:27:46 -04003282 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003283 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003284 mddev->bitmap_info.daemon_sleep);
Mike Snitzer42863252016-06-02 12:27:46 -04003285 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003286 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003287 max(rs->delta_disks, mddev->delta_disks));
Mike Snitzer42863252016-06-02 12:27:46 -04003288 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003289 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003290 max_nr_stripes);
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003291 if (rebuild_disks)
3292 for (i = 0; i < rs->raid_disks; i++)
3293 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3294 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3295 rs->dev[i].rdev.raid_disk);
3296 if (write_mostly_params)
3297 for (i = 0; i < rs->raid_disks; i++)
3298 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3299 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3300 rs->dev[i].rdev.raid_disk);
Mike Snitzer42863252016-06-02 12:27:46 -04003301 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003302 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003303 mddev->bitmap_info.max_write_behind);
Mike Snitzer42863252016-06-02 12:27:46 -04003304 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003305 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003306 mddev->sync_speed_max);
Mike Snitzer42863252016-06-02 12:27:46 -04003307 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003308 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003309 mddev->sync_speed_min);
3310 DMEMIT(" %d", rs->raid_disks);
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003311 for (i = 0; i < rs->raid_disks; i++)
3312 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3313 __get_dev_name(rs->dev[i].data_dev));
NeilBrown9d09e662011-01-13 20:00:02 +00003314 }
NeilBrown9d09e662011-01-13 20:00:02 +00003315}
3316
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02003317static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003318{
3319 struct raid_set *rs = ti->private;
3320 struct mddev *mddev = &rs->md;
3321
Jonathan Brassowbe836512013-04-24 11:42:43 +10003322 if (!mddev->pers || !mddev->pers->sync_request)
3323 return -EINVAL;
3324
3325 if (!strcasecmp(argv[0], "frozen"))
3326 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3327 else
3328 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3329
3330 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3331 if (mddev->sync_thread) {
3332 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3333 md_reap_sync_thread(mddev);
3334 }
3335 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3336 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3337 return -EBUSY;
3338 else if (!strcasecmp(argv[0], "resync"))
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003339 ; /* MD_RECOVERY_NEEDED set below */
3340 else if (!strcasecmp(argv[0], "recover"))
Jonathan Brassowbe836512013-04-24 11:42:43 +10003341 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003342 else {
Jonathan Brassowbe836512013-04-24 11:42:43 +10003343 if (!strcasecmp(argv[0], "check"))
3344 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3345 else if (!!strcasecmp(argv[0], "repair"))
3346 return -EINVAL;
3347 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3348 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3349 }
3350 if (mddev->ro == 2) {
3351 /* A write to sync_action is enough to justify
3352 * canceling read-auto mode
3353 */
3354 mddev->ro = 0;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003355 if (!mddev->suspended && mddev->sync_thread)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003356 md_wakeup_thread(mddev->sync_thread);
3357 }
3358 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003359 if (!mddev->suspended && mddev->thread)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003360 md_wakeup_thread(mddev->thread);
3361
3362 return 0;
3363}
3364
3365static int raid_iterate_devices(struct dm_target *ti,
3366 iterate_devices_callout_fn fn, void *data)
NeilBrown9d09e662011-01-13 20:00:02 +00003367{
3368 struct raid_set *rs = ti->private;
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02003369 unsigned int i;
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003370 int r = 0;
NeilBrown9d09e662011-01-13 20:00:02 +00003371
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003372 for (i = 0; !r && i < rs->md.raid_disks; i++)
NeilBrown9d09e662011-01-13 20:00:02 +00003373 if (rs->dev[i].data_dev)
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003374 r = fn(ti,
NeilBrown9d09e662011-01-13 20:00:02 +00003375 rs->dev[i].data_dev,
3376 0, /* No offset on data devs */
3377 rs->md.dev_sectors,
3378 data);
3379
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003380 return r;
NeilBrown9d09e662011-01-13 20:00:02 +00003381}
3382
3383static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3384{
3385 struct raid_set *rs = ti->private;
Heinz Mauelshagen89d3d9a2016-07-19 13:16:24 +02003386 unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
NeilBrown9d09e662011-01-13 20:00:02 +00003387
3388 blk_limits_io_min(limits, chunk_size);
Heinz Mauelshagen89d3d9a2016-07-19 13:16:24 +02003389 blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
NeilBrown9d09e662011-01-13 20:00:02 +00003390}
3391
3392static void raid_presuspend(struct dm_target *ti)
3393{
3394 struct raid_set *rs = ti->private;
3395
3396 md_stop_writes(&rs->md);
3397}
3398
3399static void raid_postsuspend(struct dm_target *ti)
3400{
3401 struct raid_set *rs = ti->private;
3402
Heinz Mauelshagen31e10a42016-08-10 02:45:59 +02003403 if (!rs->md.suspended)
3404 mddev_suspend(&rs->md);
3405
3406 rs->md.ro = 1;
NeilBrown9d09e662011-01-13 20:00:02 +00003407}
3408
Jonathan Brassowf381e712013-05-08 17:57:13 -05003409static void attempt_restore_of_faulty_devices(struct raid_set *rs)
NeilBrown9d09e662011-01-13 20:00:02 +00003410{
Jonathan Brassow9092c022013-05-02 14:19:24 -05003411 int i;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003412 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
Jonathan Brassow9092c022013-05-02 14:19:24 -05003413 unsigned long flags;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003414 bool cleared = false;
Jonathan Brassow9092c022013-05-02 14:19:24 -05003415 struct dm_raid_superblock *sb;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003416 struct mddev *mddev = &rs->md;
Jonathan Brassow9092c022013-05-02 14:19:24 -05003417 struct md_rdev *r;
NeilBrown9d09e662011-01-13 20:00:02 +00003418
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003419 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3420 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3421 return;
3422
3423 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3424
Jonathan Brassowf381e712013-05-08 17:57:13 -05003425 for (i = 0; i < rs->md.raid_disks; i++) {
3426 r = &rs->dev[i].rdev;
3427 if (test_bit(Faulty, &r->flags) && r->sb_page &&
Heinz Mauelshagen0a7b8182016-06-15 18:45:56 +02003428 sync_page_io(r, 0, r->sb_size, r->sb_page,
3429 REQ_OP_READ, 0, true)) {
Jonathan Brassowf381e712013-05-08 17:57:13 -05003430 DMINFO("Faulty %s device #%d has readable super block."
3431 " Attempting to revive it.",
3432 rs->raid_type->name, i);
Jonathan Brassowa4dc1632013-05-08 18:00:54 -05003433
3434 /*
3435 * Faulty bit may be set, but sometimes the array can
3436 * be suspended before the personalities can respond
3437 * by removing the device from the array (i.e. calling
Mike Snitzer43157842016-05-30 13:03:37 -04003438 * 'hot_remove_disk'). If they haven't yet removed
Jonathan Brassowa4dc1632013-05-08 18:00:54 -05003439 * the failed device, its 'raid_disk' number will be
3440 * '>= 0' - meaning we must call this function
3441 * ourselves.
3442 */
3443 if ((r->raid_disk >= 0) &&
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003444 (mddev->pers->hot_remove_disk(mddev, r) != 0))
Jonathan Brassowa4dc1632013-05-08 18:00:54 -05003445 /* Failed to revive this device, try next */
3446 continue;
3447
Jonathan Brassowf381e712013-05-08 17:57:13 -05003448 r->raid_disk = i;
3449 r->saved_raid_disk = i;
3450 flags = r->flags;
3451 clear_bit(Faulty, &r->flags);
3452 clear_bit(WriteErrorSeen, &r->flags);
3453 clear_bit(In_sync, &r->flags);
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003454 if (mddev->pers->hot_add_disk(mddev, r)) {
Jonathan Brassowf381e712013-05-08 17:57:13 -05003455 r->raid_disk = -1;
3456 r->saved_raid_disk = -1;
3457 r->flags = flags;
3458 } else {
3459 r->recovery_offset = 0;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003460 set_bit(i, (void *) cleared_failed_devices);
3461 cleared = true;
Jonathan Brassowf381e712013-05-08 17:57:13 -05003462 }
3463 }
3464 }
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003465
3466 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3467 if (cleared) {
3468 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3469
Jonathan Brassowf381e712013-05-08 17:57:13 -05003470 rdev_for_each(r, &rs->md) {
3471 sb = page_address(r->sb_page);
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003472 sb_retrieve_failed_devices(sb, failed_devices);
3473
3474 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3475 failed_devices[i] &= ~cleared_failed_devices[i];
3476
3477 sb_update_failed_devices(sb, failed_devices);
Jonathan Brassowf381e712013-05-08 17:57:13 -05003478 }
3479 }
3480}
3481
Mike Snitzere6ca5e12016-06-02 15:27:22 -04003482static int __load_dirty_region_bitmap(struct raid_set *rs)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003483{
3484 int r = 0;
3485
3486 /* Try loading the bitmap unless "raid0", which does not have one */
3487 if (!rs_is_raid0(rs) &&
Mike Snitzer42863252016-06-02 12:27:46 -04003488 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003489 r = bitmap_load(&rs->md);
3490 if (r)
3491 DMERR("Failed to load bitmap");
3492 }
3493
3494 return r;
3495}
3496
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003497/* Enforce updating all superblocks */
3498static void rs_update_sbs(struct raid_set *rs)
3499{
3500 struct mddev *mddev = &rs->md;
3501 int ro = mddev->ro;
3502
3503 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3504 mddev->ro = 0;
3505 md_update_sb(mddev, 1);
3506 mddev->ro = ro;
3507}
3508
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003509/*
3510 * Reshape changes raid algorithm of @rs to new one within personality
3511 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3512 * disks from a raid set thus growing/shrinking it or resizes the set
3513 *
3514 * Call mddev_lock_nointr() before!
3515 */
3516static int rs_start_reshape(struct raid_set *rs)
3517{
3518 int r;
3519 struct mddev *mddev = &rs->md;
3520 struct md_personality *pers = mddev->pers;
3521
3522 r = rs_setup_reshape(rs);
3523 if (r)
3524 return r;
3525
3526 /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3527 if (mddev->suspended)
3528 mddev_resume(mddev);
3529
3530 /*
3531 * Check any reshape constraints enforced by the personalility
3532 *
3533 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3534 */
3535 r = pers->check_reshape(mddev);
3536 if (r) {
3537 rs->ti->error = "pers->check_reshape() failed";
3538 return r;
3539 }
3540
3541 /*
3542 * Personality may not provide start reshape method in which
3543 * case check_reshape above has already covered everything
3544 */
3545 if (pers->start_reshape) {
3546 r = pers->start_reshape(mddev);
3547 if (r) {
3548 rs->ti->error = "pers->start_reshape() failed";
3549 return r;
3550 }
3551 }
3552
3553 /* Suspend because a resume will happen in raid_resume() */
3554 if (!mddev->suspended)
3555 mddev_suspend(mddev);
3556
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003557 /*
3558 * Now reshape got set up, update superblocks to
3559 * reflect the fact so that a table reload will
3560 * access proper superblock content in the ctr.
3561 */
3562 rs_update_sbs(rs);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003563
3564 return 0;
3565}
3566
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003567static int raid_preresume(struct dm_target *ti)
3568{
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003569 int r;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003570 struct raid_set *rs = ti->private;
3571 struct mddev *mddev = &rs->md;
3572
3573 /* This is a resume after a suspend of the set -> it's already started */
Mike Snitzer42863252016-06-02 12:27:46 -04003574 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003575 return 0;
3576
3577 /*
3578 * The superblocks need to be updated on disk if the
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003579 * array is new or new devices got added (thus zeroed
3580 * out by userspace) or __load_dirty_region_bitmap
3581 * will overwrite them in core with old data or fail.
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003582 */
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003583 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3584 rs_update_sbs(rs);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003585
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003586 /* Load the bitmap from disk unless raid0 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003587 r = __load_dirty_region_bitmap(rs);
3588 if (r)
3589 return r;
3590
Heinz Mauelshagen4257e082016-06-14 01:46:03 +02003591 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
Dmitry Bilunov27d382f2017-03-30 18:14:26 +03003592 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
Heinz Mauelshagen4257e082016-06-14 01:46:03 +02003593 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3594 r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3595 to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3596 if (r)
3597 DMERR("Failed to resize bitmap");
3598 }
3599
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003600 /* Check for any resize/reshape on @rs and adjust/initiate */
3601 /* Be prepared for mddev_resume() in raid_resume() */
3602 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3603 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003604 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3605 mddev->resync_min = mddev->recovery_cp;
3606 }
3607
3608 rs_set_capacity(rs);
3609
Heinz Mauelshagen345a6cd2016-06-25 02:42:54 +02003610 /* Check for any reshape request unless new raid set */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003611 if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3612 /* Initiate a reshape. */
3613 mddev_lock_nointr(mddev);
3614 r = rs_start_reshape(rs);
3615 mddev_unlock(mddev);
3616 if (r)
3617 DMWARN("Failed to check/start reshape, continuing without change");
3618 r = 0;
3619 }
3620
3621 return r;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003622}
3623
Heinz Mauelshagen2937e222017-02-28 19:17:49 +01003624#define RESUME_STAY_FROZEN_FLAGS (CTR_FLAG_DELTA_DISKS | CTR_FLAG_DATA_OFFSET)
3625
Jonathan Brassowf381e712013-05-08 17:57:13 -05003626static void raid_resume(struct dm_target *ti)
3627{
3628 struct raid_set *rs = ti->private;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003629 struct mddev *mddev = &rs->md;
Jonathan Brassowf381e712013-05-08 17:57:13 -05003630
Mike Snitzer42863252016-06-02 12:27:46 -04003631 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003632 /*
3633 * A secondary resume while the device is active.
3634 * Take this opportunity to check whether any failed
3635 * devices are reachable again.
3636 */
3637 attempt_restore_of_faulty_devices(rs);
Jonathan Brassow47525e52012-05-22 13:55:29 +10003638 }
Heinz Mauelshagen31e10a42016-08-10 02:45:59 +02003639
3640 mddev->ro = 0;
3641 mddev->in_sync = 0;
3642
Heinz Mauelshagen2937e222017-02-28 19:17:49 +01003643 /*
3644 * Keep the RAID set frozen if reshape/rebuild flags are set.
3645 * The RAID set is unfrozen once the next table load/resume,
3646 * which clears the reshape/rebuild flags, occurs.
3647 * This ensures that the constructor for the inactive table
3648 * retrieves an up-to-date reshape_position.
3649 */
3650 if (!(rs->ctr_flags & RESUME_STAY_FROZEN_FLAGS))
3651 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
Heinz Mauelshagen31e10a42016-08-10 02:45:59 +02003652
3653 if (mddev->suspended)
3654 mddev_resume(mddev);
NeilBrown9d09e662011-01-13 20:00:02 +00003655}
3656
3657static struct target_type raid_target = {
3658 .name = "raid",
Heinz Mauelshagenb052b072016-10-17 21:20:07 +02003659 .version = {1, 9, 1},
NeilBrown9d09e662011-01-13 20:00:02 +00003660 .module = THIS_MODULE,
3661 .ctr = raid_ctr,
3662 .dtr = raid_dtr,
3663 .map = raid_map,
3664 .status = raid_status,
Jonathan Brassowbe836512013-04-24 11:42:43 +10003665 .message = raid_message,
NeilBrown9d09e662011-01-13 20:00:02 +00003666 .iterate_devices = raid_iterate_devices,
3667 .io_hints = raid_io_hints,
3668 .presuspend = raid_presuspend,
3669 .postsuspend = raid_postsuspend,
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003670 .preresume = raid_preresume,
NeilBrown9d09e662011-01-13 20:00:02 +00003671 .resume = raid_resume,
3672};
3673
3674static int __init dm_raid_init(void)
3675{
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11003676 DMINFO("Loading target version %u.%u.%u",
3677 raid_target.version[0],
3678 raid_target.version[1],
3679 raid_target.version[2]);
NeilBrown9d09e662011-01-13 20:00:02 +00003680 return dm_register_target(&raid_target);
3681}
3682
3683static void __exit dm_raid_exit(void)
3684{
3685 dm_unregister_target(&raid_target);
3686}
3687
3688module_init(dm_raid_init);
3689module_exit(dm_raid_exit);
3690
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02003691module_param(devices_handle_discard_safely, bool, 0644);
3692MODULE_PARM_DESC(devices_handle_discard_safely,
3693 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3694
Mike Snitzeref9b85a2016-06-02 12:02:19 -04003695MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3696MODULE_ALIAS("dm-raid0");
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05003697MODULE_ALIAS("dm-raid1");
3698MODULE_ALIAS("dm-raid10");
NeilBrown9d09e662011-01-13 20:00:02 +00003699MODULE_ALIAS("dm-raid4");
3700MODULE_ALIAS("dm-raid5");
3701MODULE_ALIAS("dm-raid6");
3702MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003703MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
NeilBrown9d09e662011-01-13 20:00:02 +00003704MODULE_LICENSE("GPL");