blob: 60632b409b80fee2da65f891d0f883caff279ad0 [file] [log] [blame]
Heinz Mauelshagen3bd94002023-01-25 21:00:44 +01001// SPDX-License-Identifier: GPL-2.0-only
NeilBrown9d09e662011-01-13 20:00:02 +00002/*
3 * Copyright (C) 2010-2011 Neil Brown
Heinz Mauelshagen5380c052018-09-06 14:02:54 -04004 * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved.
NeilBrown9d09e662011-01-13 20:00:02 +00005 *
6 * This file is released under the GPL.
7 */
8
9#include <linux/slab.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040010#include <linux/module.h>
NeilBrown9d09e662011-01-13 20:00:02 +000011
12#include "md.h"
Jonathan Brassow32737272011-08-02 12:32:07 +010013#include "raid1.h"
NeilBrown9d09e662011-01-13 20:00:02 +000014#include "raid5.h"
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -050015#include "raid10.h"
Mike Snitzer935fe092017-10-10 17:02:41 -040016#include "md-bitmap.h"
NeilBrown9d09e662011-01-13 20:00:02 +000017
Alasdair G Kergon3e8dbb72011-08-02 12:32:03 +010018#include <linux/device-mapper.h>
19
NeilBrown9d09e662011-01-13 20:00:02 +000020#define DM_MSG_PREFIX "raid"
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +020021#define MAX_RAID_DEVICES 253 /* md-raid kernel limit */
NeilBrown9d09e662011-01-13 20:00:02 +000022
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +020023/*
24 * Minimum sectors of free reshape space per raid device
25 */
26#define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
27
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +010028/*
29 * Minimum journal space 4 MiB in sectors.
30 */
31#define MIN_RAID456_JOURNAL_SPACE (4*2048)
32
Heinz Mauelshagen2f06cd12023-01-30 21:28:24 +010033static bool devices_handle_discard_safely;
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +020034
NeilBrown9d09e662011-01-13 20:00:02 +000035/*
Heinz Mauelshagenb30f1602023-02-07 20:48:51 +010036 * The following flags are used by dm-raid to set up the array state.
Jonathan Brassowb12d4372011-08-02 12:32:07 +010037 * They must be cleared before md_run is called.
NeilBrown9d09e662011-01-13 20:00:02 +000038 */
Mike Snitzer43157842016-05-30 13:03:37 -040039#define FirstUse 10 /* rdev flag */
NeilBrown9d09e662011-01-13 20:00:02 +000040
41struct raid_dev {
42 /*
43 * Two DM devices, one to hold metadata and one to hold the
Mike Snitzer43157842016-05-30 13:03:37 -040044 * actual data/parity. The reason for this is to not confuse
NeilBrown9d09e662011-01-13 20:00:02 +000045 * ti->len and give more flexibility in altering size and
46 * characteristics.
47 *
48 * While it is possible for this device to be associated
49 * with a different physical device than the data_dev, it
50 * is intended for it to be the same.
51 * |--------- Physical Device ---------|
52 * |- meta_dev -|------ data_dev ------|
53 */
54 struct dm_dev *meta_dev;
55 struct dm_dev *data_dev;
NeilBrown3cb03002011-10-11 16:45:26 +110056 struct md_rdev rdev;
NeilBrown9d09e662011-01-13 20:00:02 +000057};
58
59/*
Mike Snitzer42863252016-06-02 12:27:46 -040060 * Bits for establishing rs->ctr_flags
Heinz Mauelshagen702108d2016-05-19 18:49:26 +020061 *
62 * 1 = no flag value
63 * 2 = flag with value
NeilBrown9d09e662011-01-13 20:00:02 +000064 */
Mike Snitzer42863252016-06-02 12:27:46 -040065#define __CTR_FLAG_SYNC 0 /* 1 */ /* Not with raid0! */
66#define __CTR_FLAG_NOSYNC 1 /* 1 */ /* Not with raid0! */
67#define __CTR_FLAG_REBUILD 2 /* 2 */ /* Not with raid0! */
68#define __CTR_FLAG_DAEMON_SLEEP 3 /* 2 */ /* Not with raid0! */
69#define __CTR_FLAG_MIN_RECOVERY_RATE 4 /* 2 */ /* Not with raid0! */
70#define __CTR_FLAG_MAX_RECOVERY_RATE 5 /* 2 */ /* Not with raid0! */
71#define __CTR_FLAG_MAX_WRITE_BEHIND 6 /* 2 */ /* Only with raid1! */
72#define __CTR_FLAG_WRITE_MOSTLY 7 /* 2 */ /* Only with raid1! */
73#define __CTR_FLAG_STRIPE_CACHE 8 /* 2 */ /* Only with raid4/5/6! */
74#define __CTR_FLAG_REGION_SIZE 9 /* 2 */ /* Not with raid0! */
75#define __CTR_FLAG_RAID10_COPIES 10 /* 2 */ /* Only with raid10 */
76#define __CTR_FLAG_RAID10_FORMAT 11 /* 2 */ /* Only with raid10 */
Mike Snitzer9b6e5422016-06-02 11:48:09 -040077/* New for v1.9.0 */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +020078#define __CTR_FLAG_DELTA_DISKS 12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
Mike Snitzer42863252016-06-02 12:27:46 -040079#define __CTR_FLAG_DATA_OFFSET 13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
80#define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
81
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +010082/* New for v1.10.0 */
Heinz Mauelshagen6e536362017-03-22 17:44:38 +010083#define __CTR_FLAG_JOURNAL_DEV 15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
84
85/* New for v1.11.1 */
86#define __CTR_FLAG_JOURNAL_MODE 16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +010087
Mike Snitzer42863252016-06-02 12:27:46 -040088/*
89 * Flags for rs->ctr_flags field.
90 */
91#define CTR_FLAG_SYNC (1 << __CTR_FLAG_SYNC)
92#define CTR_FLAG_NOSYNC (1 << __CTR_FLAG_NOSYNC)
93#define CTR_FLAG_REBUILD (1 << __CTR_FLAG_REBUILD)
94#define CTR_FLAG_DAEMON_SLEEP (1 << __CTR_FLAG_DAEMON_SLEEP)
95#define CTR_FLAG_MIN_RECOVERY_RATE (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
96#define CTR_FLAG_MAX_RECOVERY_RATE (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
97#define CTR_FLAG_MAX_WRITE_BEHIND (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
98#define CTR_FLAG_WRITE_MOSTLY (1 << __CTR_FLAG_WRITE_MOSTLY)
99#define CTR_FLAG_STRIPE_CACHE (1 << __CTR_FLAG_STRIPE_CACHE)
100#define CTR_FLAG_REGION_SIZE (1 << __CTR_FLAG_REGION_SIZE)
101#define CTR_FLAG_RAID10_COPIES (1 << __CTR_FLAG_RAID10_COPIES)
102#define CTR_FLAG_RAID10_FORMAT (1 << __CTR_FLAG_RAID10_FORMAT)
103#define CTR_FLAG_DELTA_DISKS (1 << __CTR_FLAG_DELTA_DISKS)
104#define CTR_FLAG_DATA_OFFSET (1 << __CTR_FLAG_DATA_OFFSET)
105#define CTR_FLAG_RAID10_USE_NEAR_SETS (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100106#define CTR_FLAG_JOURNAL_DEV (1 << __CTR_FLAG_JOURNAL_DEV)
Heinz Mauelshagen6e536362017-03-22 17:44:38 +0100107#define CTR_FLAG_JOURNAL_MODE (1 << __CTR_FLAG_JOURNAL_MODE)
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500108
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200109/*
110 * Definitions of various constructor flags to
111 * be used in checks of valid / invalid flags
112 * per raid level.
113 */
114/* Define all any sync flags */
115#define CTR_FLAGS_ANY_SYNC (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
116
117/* Define flags for options without argument (e.g. 'nosync') */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200118#define CTR_FLAG_OPTIONS_NO_ARGS (CTR_FLAGS_ANY_SYNC | \
119 CTR_FLAG_RAID10_USE_NEAR_SETS)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200120
121/* Define flags for options with one argument (e.g. 'delta_disks +2') */
122#define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
123 CTR_FLAG_WRITE_MOSTLY | \
124 CTR_FLAG_DAEMON_SLEEP | \
125 CTR_FLAG_MIN_RECOVERY_RATE | \
126 CTR_FLAG_MAX_RECOVERY_RATE | \
127 CTR_FLAG_MAX_WRITE_BEHIND | \
128 CTR_FLAG_STRIPE_CACHE | \
129 CTR_FLAG_REGION_SIZE | \
130 CTR_FLAG_RAID10_COPIES | \
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200131 CTR_FLAG_RAID10_FORMAT | \
132 CTR_FLAG_DELTA_DISKS | \
Heinz Mauelshagen43f39522019-12-19 17:58:46 +0100133 CTR_FLAG_DATA_OFFSET | \
134 CTR_FLAG_JOURNAL_DEV | \
135 CTR_FLAG_JOURNAL_MODE)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200136
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200137/* Valid options definitions per raid level... */
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200138
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200139/* "raid0" does only accept data offset */
140#define RAID0_VALID_FLAGS (CTR_FLAG_DATA_OFFSET)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200141
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200142/* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
143#define RAID1_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
144 CTR_FLAG_REBUILD | \
145 CTR_FLAG_WRITE_MOSTLY | \
146 CTR_FLAG_DAEMON_SLEEP | \
147 CTR_FLAG_MIN_RECOVERY_RATE | \
148 CTR_FLAG_MAX_RECOVERY_RATE | \
149 CTR_FLAG_MAX_WRITE_BEHIND | \
150 CTR_FLAG_REGION_SIZE | \
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +0200151 CTR_FLAG_DELTA_DISKS | \
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200152 CTR_FLAG_DATA_OFFSET)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200153
154/* "raid10" does not accept any raid1 or stripe cache options */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200155#define RAID10_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
156 CTR_FLAG_REBUILD | \
157 CTR_FLAG_DAEMON_SLEEP | \
158 CTR_FLAG_MIN_RECOVERY_RATE | \
159 CTR_FLAG_MAX_RECOVERY_RATE | \
160 CTR_FLAG_REGION_SIZE | \
161 CTR_FLAG_RAID10_COPIES | \
162 CTR_FLAG_RAID10_FORMAT | \
163 CTR_FLAG_DELTA_DISKS | \
164 CTR_FLAG_DATA_OFFSET | \
165 CTR_FLAG_RAID10_USE_NEAR_SETS)
166
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200167/*
168 * "raid4/5/6" do not accept any raid1 or raid10 specific options
169 *
170 * "raid6" does not accept "nosync", because it is not guaranteed
171 * that both parity and q-syndrome are being written properly with
172 * any writes
173 */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200174#define RAID45_VALID_FLAGS (CTR_FLAGS_ANY_SYNC | \
175 CTR_FLAG_REBUILD | \
176 CTR_FLAG_DAEMON_SLEEP | \
177 CTR_FLAG_MIN_RECOVERY_RATE | \
178 CTR_FLAG_MAX_RECOVERY_RATE | \
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200179 CTR_FLAG_STRIPE_CACHE | \
180 CTR_FLAG_REGION_SIZE | \
181 CTR_FLAG_DELTA_DISKS | \
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100182 CTR_FLAG_DATA_OFFSET | \
Heinz Mauelshagen6e536362017-03-22 17:44:38 +0100183 CTR_FLAG_JOURNAL_DEV | \
184 CTR_FLAG_JOURNAL_MODE)
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200185
186#define RAID6_VALID_FLAGS (CTR_FLAG_SYNC | \
187 CTR_FLAG_REBUILD | \
188 CTR_FLAG_DAEMON_SLEEP | \
189 CTR_FLAG_MIN_RECOVERY_RATE | \
190 CTR_FLAG_MAX_RECOVERY_RATE | \
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200191 CTR_FLAG_STRIPE_CACHE | \
192 CTR_FLAG_REGION_SIZE | \
193 CTR_FLAG_DELTA_DISKS | \
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100194 CTR_FLAG_DATA_OFFSET | \
Heinz Mauelshagen6e536362017-03-22 17:44:38 +0100195 CTR_FLAG_JOURNAL_DEV | \
196 CTR_FLAG_JOURNAL_MODE)
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200197/* ...valid options definitions per raid level */
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200198
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200199/*
200 * Flags for rs->runtime_flags field
201 * (RT_FLAG prefix meaning "runtime flag")
202 *
203 * These are all internal and used to define runtime state,
204 * e.g. to prevent another resume from preresume processing
205 * the raid set all over again.
206 */
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +0200207#define RT_FLAG_RS_PRERESUMED 0
208#define RT_FLAG_RS_RESUMED 1
209#define RT_FLAG_RS_BITMAP_LOADED 2
210#define RT_FLAG_UPDATE_SBS 3
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200211#define RT_FLAG_RESHAPE_RS 4
Heinz Mauelshagen0cf352e2017-07-13 17:34:24 +0200212#define RT_FLAG_RS_SUSPENDED 5
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +0100213#define RT_FLAG_RS_IN_SYNC 6
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +0100214#define RT_FLAG_RS_RESYNCING 7
Heinz Mauelshagen99273d92019-10-01 17:47:53 +0200215#define RT_FLAG_RS_GROW 8
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200216
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200217/* Array elements of 64 bit needed for rebuild/failed disk bits */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200218#define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
219
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200220/*
221 * raid set level, layout and chunk sectors backup/restore
222 */
223struct rs_layout {
224 int new_level;
225 int new_layout;
226 int new_chunk_sectors;
227};
228
NeilBrown9d09e662011-01-13 20:00:02 +0000229struct raid_set {
230 struct dm_target *ti;
231
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200232 uint32_t stripe_cache_entries;
Mike Snitzer42863252016-06-02 12:27:46 -0400233 unsigned long ctr_flags;
234 unsigned long runtime_flags;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200235
236 uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
NeilBrown9d09e662011-01-13 20:00:02 +0000237
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200238 int raid_disks;
239 int delta_disks;
Heinz Mauelshagen4763e542016-05-19 18:49:31 +0200240 int data_offset;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200241 int raid10_copies;
Heinz Mauelshagen4257e082016-06-14 01:46:03 +0200242 int requested_bitmap_chunk_sectors;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200243
NeilBrownfd01b882011-10-11 16:47:53 +1100244 struct mddev md;
NeilBrown9d09e662011-01-13 20:00:02 +0000245 struct raid_type *raid_type;
NeilBrown9d09e662011-01-13 20:00:02 +0000246
Heinz Mauelshagen99273d92019-10-01 17:47:53 +0200247 sector_t array_sectors;
248 sector_t dev_sectors;
249
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100250 /* Optional raid4/5/6 journal device */
251 struct journal_dev {
252 struct dm_dev *dev;
253 struct md_rdev rdev;
Heinz Mauelshagen6e536362017-03-22 17:44:38 +0100254 int mode;
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100255 } journal_dev;
256
Gustavo A. R. Silvab18ae8d2020-05-07 13:51:58 -0500257 struct raid_dev dev[];
NeilBrown9d09e662011-01-13 20:00:02 +0000258};
259
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200260static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200261{
262 struct mddev *mddev = &rs->md;
263
264 l->new_level = mddev->new_level;
265 l->new_layout = mddev->new_layout;
266 l->new_chunk_sectors = mddev->new_chunk_sectors;
267}
268
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200269static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +0200270{
271 struct mddev *mddev = &rs->md;
272
273 mddev->new_level = l->new_level;
274 mddev->new_layout = l->new_layout;
275 mddev->new_chunk_sectors = l->new_chunk_sectors;
276}
277
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200278/* raid10 algorithms (i.e. formats) */
279#define ALGORITHM_RAID10_DEFAULT 0
280#define ALGORITHM_RAID10_NEAR 1
281#define ALGORITHM_RAID10_OFFSET 2
282#define ALGORITHM_RAID10_FAR 3
283
NeilBrown9d09e662011-01-13 20:00:02 +0000284/* Supported raid types and properties. */
285static struct raid_type {
286 const char *name; /* RAID algorithm. */
287 const char *descr; /* Descriptor text for logging. */
Heinz Mauelshagen094f3942016-07-19 14:03:51 +0200288 const unsigned int parity_devs; /* # of parity devices. */
289 const unsigned int minimal_devs;/* minimal # of devices in set. */
290 const unsigned int level; /* RAID level. */
291 const unsigned int algorithm; /* RAID algorithm. */
NeilBrown9d09e662011-01-13 20:00:02 +0000292} raid_types[] = {
Mike Snitzer43157842016-05-30 13:03:37 -0400293 {"raid0", "raid0 (striping)", 0, 2, 0, 0 /* NONE */},
294 {"raid1", "raid1 (mirroring)", 0, 2, 1, 0 /* NONE */},
295 {"raid10_far", "raid10 far (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_FAR},
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200296 {"raid10_offset", "raid10 offset (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_OFFSET},
Mike Snitzer43157842016-05-30 13:03:37 -0400297 {"raid10_near", "raid10 near (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_NEAR},
298 {"raid10", "raid10 (striped mirrors)", 0, 2, 10, ALGORITHM_RAID10_DEFAULT},
Heinz Mauelshagenb052b072016-10-17 21:20:07 +0200299 {"raid4", "raid4 (dedicated first parity disk)", 1, 2, 5, ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
Mike Snitzer43157842016-05-30 13:03:37 -0400300 {"raid5_n", "raid5 (dedicated last parity disk)", 1, 2, 5, ALGORITHM_PARITY_N},
301 {"raid5_ls", "raid5 (left symmetric)", 1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
302 {"raid5_rs", "raid5 (right symmetric)", 1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
303 {"raid5_la", "raid5 (left asymmetric)", 1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
304 {"raid5_ra", "raid5 (right asymmetric)", 1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
305 {"raid6_zr", "raid6 (zero restart)", 2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
306 {"raid6_nr", "raid6 (N restart)", 2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
307 {"raid6_nc", "raid6 (N continue)", 2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE},
308 {"raid6_n_6", "raid6 (dedicated parity/Q n/6)", 2, 4, 6, ALGORITHM_PARITY_N_6},
309 {"raid6_ls_6", "raid6 (left symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_SYMMETRIC_6},
310 {"raid6_rs_6", "raid6 (right symmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_SYMMETRIC_6},
311 {"raid6_la_6", "raid6 (left asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_LEFT_ASYMMETRIC_6},
312 {"raid6_ra_6", "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6, ALGORITHM_RIGHT_ASYMMETRIC_6}
NeilBrown9d09e662011-01-13 20:00:02 +0000313};
314
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200315/* True, if @v is in inclusive range [@min, @max] */
Mike Snitzerbb91a632016-06-02 12:06:54 -0400316static bool __within_range(long v, long min, long max)
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200317{
318 return v >= min && v <= max;
319}
320
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200321/* All table line arguments are defined here */
322static struct arg_name_flag {
Mike Snitzer42863252016-06-02 12:27:46 -0400323 const unsigned long flag;
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200324 const char *name;
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400325} __arg_name_flags[] = {
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200326 { CTR_FLAG_SYNC, "sync"},
327 { CTR_FLAG_NOSYNC, "nosync"},
328 { CTR_FLAG_REBUILD, "rebuild"},
329 { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
330 { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
331 { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
332 { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
Heinz Mauelshagen65359ee2016-06-24 21:32:25 +0200333 { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200334 { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
335 { CTR_FLAG_REGION_SIZE, "region_size"},
336 { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
337 { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
Heinz Mauelshagen4763e542016-05-19 18:49:31 +0200338 { CTR_FLAG_DATA_OFFSET, "data_offset"},
339 { CTR_FLAG_DELTA_DISKS, "delta_disks"},
340 { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100341 { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
Heinz Mauelshagen6e536362017-03-22 17:44:38 +0100342 { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200343};
344
345/* Return argument name string for given @flag */
Mike Snitzer3fa6cf32016-06-02 11:58:51 -0400346static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200347{
348 if (hweight32(flag) == 1) {
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400349 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200350
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400351 while (anf-- > __arg_name_flags)
Mike Snitzer42863252016-06-02 12:27:46 -0400352 if (flag & anf->flag)
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200353 return anf->name;
354
355 } else
356 DMERR("%s called with more than one flag!", __func__);
357
358 return NULL;
359}
360
Heinz Mauelshagen6e536362017-03-22 17:44:38 +0100361/* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
362static struct {
363 const int mode;
364 const char *param;
365} _raid456_journal_mode[] = {
Heinz Mauelshagen255e2642023-01-25 23:31:55 +0100366 { R5C_JOURNAL_MODE_WRITE_THROUGH, "writethrough" },
367 { R5C_JOURNAL_MODE_WRITE_BACK, "writeback" }
Heinz Mauelshagen6e536362017-03-22 17:44:38 +0100368};
369
370/* Return MD raid4/5/6 journal mode for dm @journal_mode one */
371static int dm_raid_journal_mode_to_md(const char *mode)
372{
373 int m = ARRAY_SIZE(_raid456_journal_mode);
374
375 while (m--)
376 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
377 return _raid456_journal_mode[m].mode;
378
379 return -EINVAL;
380}
381
382/* Return dm-raid raid4/5/6 journal mode string for @mode */
383static const char *md_journal_mode_to_dm_raid(const int mode)
384{
385 int m = ARRAY_SIZE(_raid456_journal_mode);
386
387 while (m--)
388 if (mode == _raid456_journal_mode[m].mode)
389 return _raid456_journal_mode[m].param;
390
391 return "unknown";
392}
393
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200394/*
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200395 * Bool helpers to test for various raid levels of a raid set.
396 * It's level as reported by the superblock rather than
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200397 * the requested raid_type passed to the constructor.
398 */
399/* Return true, if raid set in @rs is raid0 */
400static bool rs_is_raid0(struct raid_set *rs)
401{
402 return !rs->md.level;
403}
404
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200405/* Return true, if raid set in @rs is raid1 */
406static bool rs_is_raid1(struct raid_set *rs)
407{
408 return rs->md.level == 1;
409}
410
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200411/* Return true, if raid set in @rs is raid10 */
412static bool rs_is_raid10(struct raid_set *rs)
413{
414 return rs->md.level == 10;
415}
416
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +0200417/* Return true, if raid set in @rs is level 6 */
418static bool rs_is_raid6(struct raid_set *rs)
419{
420 return rs->md.level == 6;
421}
422
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +0200423/* Return true, if raid set in @rs is level 4, 5 or 6 */
424static bool rs_is_raid456(struct raid_set *rs)
425{
426 return __within_range(rs->md.level, 4, 6);
427}
428
429/* Return true, if raid set in @rs is reshapable */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200430static bool __is_raid10_far(int layout);
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +0200431static bool rs_is_reshapable(struct raid_set *rs)
432{
433 return rs_is_raid456(rs) ||
434 (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
435}
436
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200437/* Return true, if raid set in @rs is recovering */
438static bool rs_is_recovering(struct raid_set *rs)
439{
Heinz Mauelshagen50c4feb2017-01-14 03:53:08 +0100440 return rs->md.recovery_cp < rs->md.dev_sectors;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200441}
442
443/* Return true, if raid set in @rs is reshaping */
444static bool rs_is_reshaping(struct raid_set *rs)
445{
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200446 return rs->md.reshape_position != MaxSector;
447}
448
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200449/*
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200450 * bool helpers to test for various raid levels of a raid type @rt
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200451 */
452
453/* Return true, if raid type in @rt is raid0 */
454static bool rt_is_raid0(struct raid_type *rt)
455{
456 return !rt->level;
457}
458
459/* Return true, if raid type in @rt is raid1 */
460static bool rt_is_raid1(struct raid_type *rt)
461{
462 return rt->level == 1;
463}
464
465/* Return true, if raid type in @rt is raid10 */
466static bool rt_is_raid10(struct raid_type *rt)
467{
468 return rt->level == 10;
469}
470
471/* Return true, if raid type in @rt is raid4/5 */
472static bool rt_is_raid45(struct raid_type *rt)
473{
Mike Snitzerbb91a632016-06-02 12:06:54 -0400474 return __within_range(rt->level, 4, 5);
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200475}
476
477/* Return true, if raid type in @rt is raid6 */
478static bool rt_is_raid6(struct raid_type *rt)
479{
480 return rt->level == 6;
481}
Heinz Mauelshagen676fa5a2016-05-19 18:49:29 +0200482
483/* Return true, if raid type in @rt is raid4/5/6 */
484static bool rt_is_raid456(struct raid_type *rt)
485{
Mike Snitzerbb91a632016-06-02 12:06:54 -0400486 return __within_range(rt->level, 4, 6);
Heinz Mauelshagen676fa5a2016-05-19 18:49:29 +0200487}
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200488/* END: raid level bools */
489
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200490/* Return valid ctr flags for the raid level of @rs */
491static unsigned long __valid_flags(struct raid_set *rs)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200492{
493 if (rt_is_raid0(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200494 return RAID0_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200495 else if (rt_is_raid1(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200496 return RAID1_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200497 else if (rt_is_raid10(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200498 return RAID10_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200499 else if (rt_is_raid45(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200500 return RAID45_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200501 else if (rt_is_raid6(rs->raid_type))
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200502 return RAID6_VALID_FLAGS;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200503
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200504 return 0;
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200505}
506
507/*
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200508 * Check for valid flags set on @rs
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200509 *
510 * Has to be called after parsing of the ctr flags!
511 */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200512static int rs_check_for_valid_flags(struct raid_set *rs)
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200513{
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +0200514 if (rs->ctr_flags & ~__valid_flags(rs)) {
Mike Snitzer42863252016-06-02 12:27:46 -0400515 rs->ti->error = "Invalid flags combination";
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400516 return -EINVAL;
517 }
Heinz Mauelshagenf0902792016-05-19 18:49:27 +0200518
519 return 0;
520}
521
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200522/* MD raid10 bit definitions and helpers */
523#define RAID10_OFFSET (1 << 16) /* stripes with data copies area adjacent on devices */
524#define RAID10_BROCKEN_USE_FAR_SETS (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
525#define RAID10_USE_FAR_SETS (1 << 18) /* Use sets instead of whole stripe rotation */
526#define RAID10_FAR_COPIES_SHIFT 8 /* raid10 # far copies shift (2nd byte of layout) */
527
528/* Return md raid10 near copies for @layout */
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400529static unsigned int __raid10_near_copies(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200530{
531 return layout & 0xFF;
532}
533
534/* Return md raid10 far copies for @layout */
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400535static unsigned int __raid10_far_copies(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200536{
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400537 return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200538}
539
540/* Return true if md raid10 offset for @layout */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200541static bool __is_raid10_offset(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200542{
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200543 return !!(layout & RAID10_OFFSET);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200544}
545
546/* Return true if md raid10 near for @layout */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200547static bool __is_raid10_near(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200548{
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400549 return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200550}
551
552/* Return true if md raid10 far for @layout */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200553static bool __is_raid10_far(int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200554{
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400555 return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200556}
557
558/* Return md raid10 layout string for @layout */
559static const char *raid10_md_layout_to_format(int layout)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100560{
561 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200562 * Bit 16 stands for "offset"
563 * (i.e. adjacent stripes hold copies)
564 *
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100565 * Refer to MD's raid10.c for details
566 */
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400567 if (__is_raid10_offset(layout))
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100568 return "offset";
569
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400570 if (__raid10_near_copies(layout) > 1)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100571 return "near";
572
Heinz Mauelshagenbbac1e02017-07-13 17:33:22 +0200573 if (__raid10_far_copies(layout) > 1)
574 return "far";
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200575
Heinz Mauelshagenbbac1e02017-07-13 17:33:22 +0200576 return "unknown";
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100577}
578
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200579/* Return md raid10 algorithm for @name */
Arnd Bergmannf2ccaa52018-06-22 10:01:19 +0200580static int raid10_name_to_format(const char *name)
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500581{
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200582 if (!strcasecmp(name, "near"))
583 return ALGORITHM_RAID10_NEAR;
584 else if (!strcasecmp(name, "offset"))
585 return ALGORITHM_RAID10_OFFSET;
586 else if (!strcasecmp(name, "far"))
587 return ALGORITHM_RAID10_FAR;
588
589 return -EINVAL;
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500590}
591
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200592/* Return md raid10 copies for @layout */
593static unsigned int raid10_md_layout_to_copies(int layout)
594{
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200595 return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200596}
597
598/* Return md raid10 format id for @format string */
599static int raid10_format_to_md_layout(struct raid_set *rs,
600 unsigned int algorithm,
601 unsigned int copies)
602{
603 unsigned int n = 1, f = 1, r = 0;
604
605 /*
606 * MD resilienece flaw:
607 *
608 * enabling use_far_sets for far/offset formats causes copies
609 * to be colocated on the same devs together with their origins!
610 *
611 * -> disable it for now in the definition above
612 */
613 if (algorithm == ALGORITHM_RAID10_DEFAULT ||
614 algorithm == ALGORITHM_RAID10_NEAR)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100615 n = copies;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200616
617 else if (algorithm == ALGORITHM_RAID10_OFFSET) {
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100618 f = copies;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200619 r = RAID10_OFFSET;
Mike Snitzer42863252016-06-02 12:27:46 -0400620 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200621 r |= RAID10_USE_FAR_SETS;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100622
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200623 } else if (algorithm == ALGORITHM_RAID10_FAR) {
624 f = copies;
Mike Snitzer42863252016-06-02 12:27:46 -0400625 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200626 r |= RAID10_USE_FAR_SETS;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100627
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200628 } else
629 return -EINVAL;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +1100630
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200631 return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
632}
633/* END: MD raid10 bit definitions and helpers */
634
635/* Check for any of the raid10 algorithms */
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200636static bool __got_raid10(struct raid_type *rtp, const int layout)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200637{
638 if (rtp->level == 10) {
639 switch (rtp->algorithm) {
640 case ALGORITHM_RAID10_DEFAULT:
641 case ALGORITHM_RAID10_NEAR:
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400642 return __is_raid10_near(layout);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200643 case ALGORITHM_RAID10_OFFSET:
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400644 return __is_raid10_offset(layout);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200645 case ALGORITHM_RAID10_FAR:
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400646 return __is_raid10_far(layout);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200647 default:
648 break;
649 }
650 }
651
Heinz Mauelshagend7ccc2e2016-07-06 18:29:22 +0200652 return false;
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -0500653}
654
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200655/* Return raid_type for @name */
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200656static struct raid_type *get_raid_type(const char *name)
NeilBrown9d09e662011-01-13 20:00:02 +0000657{
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200658 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
NeilBrown9d09e662011-01-13 20:00:02 +0000659
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200660 while (rtp-- > raid_types)
661 if (!strcasecmp(rtp->name, name))
662 return rtp;
NeilBrown9d09e662011-01-13 20:00:02 +0000663
664 return NULL;
665}
666
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200667/* Return raid_type for @name based derived from @level and @layout */
668static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
669{
670 struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
671
672 while (rtp-- > raid_types) {
673 /* RAID10 special checks based on @layout flags/properties */
674 if (rtp->level == level &&
Mike Snitzere6ca5e12016-06-02 15:27:22 -0400675 (__got_raid10(rtp, layout) || rtp->algorithm == layout))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200676 return rtp;
677 }
678
679 return NULL;
680}
681
Heinz Mauelshagen61e06e22017-12-02 01:03:51 +0100682/* Adjust rdev sectors */
683static void rs_set_rdev_sectors(struct raid_set *rs)
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200684{
685 struct mddev *mddev = &rs->md;
Heinz Mauelshagenfbe63652016-06-24 00:36:08 +0200686 struct md_rdev *rdev;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200687
Heinz Mauelshagenfbe63652016-06-24 00:36:08 +0200688 /*
689 * raid10 sets rdev->sector to the device size, which
690 * is unintended in case of out-of-place reshaping
691 */
692 rdev_for_each(rdev, mddev)
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100693 if (!test_bit(Journal, &rdev->flags))
694 rdev->sectors = mddev->dev_sectors;
Heinz Mauelshagen61e06e22017-12-02 01:03:51 +0100695}
Heinz Mauelshagenfbe63652016-06-24 00:36:08 +0200696
Heinz Mauelshagen61e06e22017-12-02 01:03:51 +0100697/*
698 * Change bdev capacity of @rs in case of a disk add/remove reshape
699 */
700static void rs_set_capacity(struct raid_set *rs)
701{
702 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
703
Christoph Hellwigdc2985a2020-11-16 15:57:10 +0100704 set_capacity_and_notify(gendisk, rs->md.array_sectors);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200705}
706
707/*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +0200708 * Set the mddev properties in @rs to the current
709 * ones retrieved from the freshest superblock
710 */
711static void rs_set_cur(struct raid_set *rs)
712{
713 struct mddev *mddev = &rs->md;
714
715 mddev->new_level = mddev->level;
716 mddev->new_layout = mddev->layout;
717 mddev->new_chunk_sectors = mddev->chunk_sectors;
718}
719
720/*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200721 * Set the mddev properties in @rs to the new
722 * ones requested by the ctr
723 */
724static void rs_set_new(struct raid_set *rs)
725{
726 struct mddev *mddev = &rs->md;
727
728 mddev->level = mddev->new_level;
729 mddev->layout = mddev->new_layout;
730 mddev->chunk_sectors = mddev->new_chunk_sectors;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +0200731 mddev->raid_disks = rs->raid_disks;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200732 mddev->delta_disks = 0;
733}
734
Mike Snitzerbfcee0e2016-06-02 15:08:09 -0400735static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
Heinz Mauelshagen094f3942016-07-19 14:03:51 +0200736 unsigned int raid_devs)
NeilBrown9d09e662011-01-13 20:00:02 +0000737{
Heinz Mauelshagen094f3942016-07-19 14:03:51 +0200738 unsigned int i;
NeilBrown9d09e662011-01-13 20:00:02 +0000739 struct raid_set *rs;
NeilBrown9d09e662011-01-13 20:00:02 +0000740
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400741 if (raid_devs <= raid_type->parity_devs) {
742 ti->error = "Insufficient number of devices";
743 return ERR_PTR(-EINVAL);
744 }
NeilBrown9d09e662011-01-13 20:00:02 +0000745
Kees Cookacafe7e2018-05-08 13:45:50 -0700746 rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400747 if (!rs) {
748 ti->error = "Cannot allocate raid context";
749 return ERR_PTR(-ENOMEM);
750 }
NeilBrown9d09e662011-01-13 20:00:02 +0000751
752 mddev_init(&rs->md);
753
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +0200754 rs->raid_disks = raid_devs;
755 rs->delta_disks = 0;
756
NeilBrown9d09e662011-01-13 20:00:02 +0000757 rs->ti = ti;
758 rs->raid_type = raid_type;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +0200759 rs->stripe_cache_entries = 256;
NeilBrown9d09e662011-01-13 20:00:02 +0000760 rs->md.raid_disks = raid_devs;
761 rs->md.level = raid_type->level;
762 rs->md.new_level = rs->md.level;
NeilBrown9d09e662011-01-13 20:00:02 +0000763 rs->md.layout = raid_type->algorithm;
764 rs->md.new_layout = rs->md.layout;
765 rs->md.delta_disks = 0;
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +0200766 rs->md.recovery_cp = MaxSector;
NeilBrown9d09e662011-01-13 20:00:02 +0000767
768 for (i = 0; i < raid_devs; i++)
769 md_rdev_init(&rs->dev[i].rdev);
770
771 /*
772 * Remaining items to be initialized by further RAID params:
773 * rs->md.persistent
774 * rs->md.external
775 * rs->md.chunk_sectors
776 * rs->md.new_chunk_sectors
Jonathan E Brassowc039c332012-07-27 15:08:04 +0100777 * rs->md.dev_sectors
NeilBrown9d09e662011-01-13 20:00:02 +0000778 */
779
780 return rs;
781}
782
Heinz Mauelshagen38b0bd02018-09-06 18:33:38 +0200783/* Free all @rs allocations */
Mike Snitzerbfcee0e2016-06-02 15:08:09 -0400784static void raid_set_free(struct raid_set *rs)
NeilBrown9d09e662011-01-13 20:00:02 +0000785{
786 int i;
787
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100788 if (rs->journal_dev.dev) {
789 md_rdev_clear(&rs->journal_dev.rdev);
790 dm_put_device(rs->ti, rs->journal_dev.dev);
791 }
792
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +0200793 for (i = 0; i < rs->raid_disks; i++) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100794 if (rs->dev[i].meta_dev)
795 dm_put_device(rs->ti, rs->dev[i].meta_dev);
NeilBrown545c8792012-05-22 13:54:30 +1000796 md_rdev_clear(&rs->dev[i].rdev);
NeilBrown9d09e662011-01-13 20:00:02 +0000797 if (rs->dev[i].data_dev)
798 dm_put_device(rs->ti, rs->dev[i].data_dev);
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100799 }
NeilBrown9d09e662011-01-13 20:00:02 +0000800
801 kfree(rs);
802}
803
804/*
805 * For every device we have two words
806 * <meta_dev>: meta device name or '-' if missing
807 * <data_dev>: data device name or '-' if missing
808 *
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100809 * The following are permitted:
810 * - -
811 * - <data_dev>
812 * <meta_dev> <data_dev>
813 *
814 * The following is not allowed:
815 * <meta_dev> -
816 *
817 * This code parses those words. If there is a failure,
Mike Snitzerbfcee0e2016-06-02 15:08:09 -0400818 * the caller must use raid_set_free() to unwind the operations.
NeilBrown9d09e662011-01-13 20:00:02 +0000819 */
Heinz Mauelshagen702108d2016-05-19 18:49:26 +0200820static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
NeilBrown9d09e662011-01-13 20:00:02 +0000821{
822 int i;
823 int rebuild = 0;
824 int metadata_available = 0;
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +0200825 int r = 0;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200826 const char *arg;
NeilBrown9d09e662011-01-13 20:00:02 +0000827
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200828 /* Put off the number of raid devices argument to get to dev pairs */
829 arg = dm_shift_arg(as);
830 if (!arg)
831 return -EINVAL;
832
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +0200833 for (i = 0; i < rs->raid_disks; i++) {
NeilBrown9d09e662011-01-13 20:00:02 +0000834 rs->dev[i].rdev.raid_disk = i;
835
836 rs->dev[i].meta_dev = NULL;
837 rs->dev[i].data_dev = NULL;
838
839 /*
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100840 * There are no offsets initially.
841 * Out of place reshape will set them accordingly.
NeilBrown9d09e662011-01-13 20:00:02 +0000842 */
843 rs->dev[i].rdev.data_offset = 0;
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100844 rs->dev[i].rdev.new_data_offset = 0;
NeilBrown9d09e662011-01-13 20:00:02 +0000845 rs->dev[i].rdev.mddev = &rs->md;
846
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200847 arg = dm_shift_arg(as);
848 if (!arg)
849 return -EINVAL;
850
851 if (strcmp(arg, "-")) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400852 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
853 &rs->dev[i].meta_dev);
854 if (r) {
855 rs->ti->error = "RAID metadata device lookup failure";
856 return r;
857 }
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100858
859 rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400860 if (!rs->dev[i].rdev.sb_page) {
861 rs->ti->error = "Failed to allocate superblock page";
862 return -ENOMEM;
863 }
NeilBrown9d09e662011-01-13 20:00:02 +0000864 }
865
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +0200866 arg = dm_shift_arg(as);
867 if (!arg)
868 return -EINVAL;
869
870 if (!strcmp(arg, "-")) {
NeilBrown9d09e662011-01-13 20:00:02 +0000871 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400872 (!rs->dev[i].rdev.recovery_offset)) {
873 rs->ti->error = "Drive designated for rebuild not specified";
874 return -EINVAL;
875 }
NeilBrown9d09e662011-01-13 20:00:02 +0000876
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400877 if (rs->dev[i].meta_dev) {
878 rs->ti->error = "No data device supplied with metadata device";
879 return -EINVAL;
880 }
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100881
NeilBrown9d09e662011-01-13 20:00:02 +0000882 continue;
883 }
884
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400885 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
886 &rs->dev[i].data_dev);
887 if (r) {
888 rs->ti->error = "RAID device lookup failure";
889 return r;
890 }
NeilBrown9d09e662011-01-13 20:00:02 +0000891
Jonathan Brassowb12d4372011-08-02 12:32:07 +0100892 if (rs->dev[i].meta_dev) {
893 metadata_available = 1;
894 rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
895 }
NeilBrown9d09e662011-01-13 20:00:02 +0000896 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +0200897 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
NeilBrown9d09e662011-01-13 20:00:02 +0000898 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
899 rebuild++;
900 }
901
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +0100902 if (rs->journal_dev.dev)
903 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
904
NeilBrown9d09e662011-01-13 20:00:02 +0000905 if (metadata_available) {
906 rs->md.external = 0;
907 rs->md.persistent = 1;
908 rs->md.major_version = 2;
909 } else if (rebuild && !rs->md.recovery_cp) {
910 /*
911 * Without metadata, we will not be able to tell if the array
912 * is in-sync or not - we must assume it is not. Therefore,
913 * it is impossible to rebuild a drive.
914 *
915 * Even if there is metadata, the on-disk information may
916 * indicate that the array is not in-sync and it will then
917 * fail at that time.
918 *
919 * User could specify 'nosync' option if desperate.
920 */
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400921 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
922 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +0000923 }
924
925 return 0;
926}
927
928/*
Jonathan Brassowc1084562011-08-02 12:32:07 +0100929 * validate_region_size
930 * @rs
931 * @region_size: region size in sectors. If 0, pick a size (4MiB default).
932 *
933 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
934 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
935 *
936 * Returns: 0 on success, -EINVAL on failure.
937 */
938static int validate_region_size(struct raid_set *rs, unsigned long region_size)
939{
940 unsigned long min_region_size = rs->ti->len / (1 << 21);
941
Heinz Mauelshagen9e7d9362016-08-17 15:36:44 +0200942 if (rs_is_raid0(rs))
943 return 0;
944
Jonathan Brassowc1084562011-08-02 12:32:07 +0100945 if (!region_size) {
946 /*
Mike Snitzer43157842016-05-30 13:03:37 -0400947 * Choose a reasonable default. All figures in sectors.
Jonathan Brassowc1084562011-08-02 12:32:07 +0100948 */
949 if (min_region_size > (1 << 13)) {
Jonathan Brassow3a0f9aa2012-12-21 20:23:33 +0000950 /* If not a power of 2, make it the next power of 2 */
Mikulas Patocka042745e2015-10-02 11:17:37 -0400951 region_size = roundup_pow_of_two(min_region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100952 DMINFO("Choosing default region size of %lu sectors",
953 region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100954 } else {
955 DMINFO("Choosing default region size of 4MiB");
956 region_size = 1 << 13; /* sectors */
957 }
958 } else {
959 /*
960 * Validate user-supplied value.
961 */
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400962 if (region_size > rs->ti->len) {
963 rs->ti->error = "Supplied region size is too large";
964 return -EINVAL;
965 }
Jonathan Brassowc1084562011-08-02 12:32:07 +0100966
967 if (region_size < min_region_size) {
968 DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
969 region_size, min_region_size);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400970 rs->ti->error = "Supplied region size is too small";
971 return -EINVAL;
Jonathan Brassowc1084562011-08-02 12:32:07 +0100972 }
973
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400974 if (!is_power_of_2(region_size)) {
975 rs->ti->error = "Region size is not a power of 2";
976 return -EINVAL;
977 }
Jonathan Brassowc1084562011-08-02 12:32:07 +0100978
Mike Snitzerbd83a4c2016-05-31 14:26:52 -0400979 if (region_size < rs->md.chunk_sectors) {
980 rs->ti->error = "Region size is smaller than the chunk size";
981 return -EINVAL;
982 }
Jonathan Brassowc1084562011-08-02 12:32:07 +0100983 }
984
985 /*
986 * Convert sectors to bytes.
987 */
Heinz Mauelshagen89d3d9a2016-07-19 13:16:24 +0200988 rs->md.bitmap_info.chunksize = to_bytes(region_size);
Jonathan Brassowc1084562011-08-02 12:32:07 +0100989
990 return 0;
991}
992
993/*
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600994 * validate_raid_redundancy
Jonathan Brassoweb649122012-10-11 13:40:09 +1100995 * @rs
996 *
Jonathan Brassow55ebbb52013-01-22 21:42:18 -0600997 * Determine if there are enough devices in the array that haven't
998 * failed (or are being rebuilt) to form a usable array.
Jonathan Brassoweb649122012-10-11 13:40:09 +1100999 *
1000 * Returns: 0 on success, -EINVAL on failure.
1001 */
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001002static int validate_raid_redundancy(struct raid_set *rs)
Jonathan Brassoweb649122012-10-11 13:40:09 +11001003{
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001004 unsigned int i, rebuild_cnt = 0;
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02001005 unsigned int rebuilds_per_group = 0, copies, raid_disks;
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001006 unsigned int group_size, last_group_start;
Jonathan Brassoweb649122012-10-11 13:40:09 +11001007
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02001008 for (i = 0; i < rs->raid_disks; i++)
1009 if (!test_bit(FirstUse, &rs->dev[i].rdev.flags) &&
1010 ((!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1011 !rs->dev[i].rdev.sb_page)))
Jonathan Brassoweb649122012-10-11 13:40:09 +11001012 rebuild_cnt++;
1013
Heinz Mauelshagen53bf5382017-12-13 17:13:17 +01001014 switch (rs->md.level) {
Heinz Mauelshagen9e7d9362016-08-17 15:36:44 +02001015 case 0:
1016 break;
Jonathan Brassoweb649122012-10-11 13:40:09 +11001017 case 1:
1018 if (rebuild_cnt >= rs->md.raid_disks)
1019 goto too_many;
1020 break;
1021 case 4:
1022 case 5:
1023 case 6:
1024 if (rebuild_cnt > rs->raid_type->parity_devs)
1025 goto too_many;
1026 break;
1027 case 10:
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001028 copies = raid10_md_layout_to_copies(rs->md.new_layout);
Heinz Mauelshagen53bf5382017-12-13 17:13:17 +01001029 if (copies < 2) {
1030 DMERR("Bogus raid10 data copies < 2!");
1031 return -EINVAL;
1032 }
1033
Jonathan Brassow4ec1e362012-10-11 13:40:24 +11001034 if (rebuild_cnt < copies)
1035 break;
1036
1037 /*
1038 * It is possible to have a higher rebuild count for RAID10,
1039 * as long as the failed devices occur in different mirror
1040 * groups (i.e. different stripes).
1041 *
Jonathan Brassow4ec1e362012-10-11 13:40:24 +11001042 * When checking "near" format, make sure no adjacent devices
1043 * have failed beyond what can be handled. In addition to the
1044 * simple case where the number of devices is a multiple of the
1045 * number of copies, we must also handle cases where the number
1046 * of devices is not a multiple of the number of copies.
Mike Snitzer43157842016-05-30 13:03:37 -04001047 * E.g. dev1 dev2 dev3 dev4 dev5
1048 * A A B B C
1049 * C D D E E
Jonathan Brassow4ec1e362012-10-11 13:40:24 +11001050 */
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02001051 raid_disks = min(rs->raid_disks, rs->md.raid_disks);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001052 if (__is_raid10_near(rs->md.new_layout)) {
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02001053 for (i = 0; i < raid_disks; i++) {
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001054 if (!(i % copies))
1055 rebuilds_per_group = 0;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001056 if ((!rs->dev[i].rdev.sb_page ||
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001057 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001058 (++rebuilds_per_group >= copies))
1059 goto too_many;
1060 }
1061 break;
1062 }
1063
1064 /*
1065 * When checking "far" and "offset" formats, we need to ensure
1066 * that the device that holds its copy is not also dead or
1067 * being rebuilt. (Note that "far" and "offset" formats only
1068 * support two copies right now. These formats also only ever
1069 * use the 'use_far_sets' variant.)
1070 *
1071 * This check is somewhat complicated by the need to account
Mike Snitzer43157842016-05-30 13:03:37 -04001072 * for arrays that are not a multiple of (far) copies. This
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001073 * results in the need to treat the last (potentially larger)
1074 * set differently.
1075 */
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02001076 group_size = (raid_disks / copies);
1077 last_group_start = (raid_disks / group_size) - 1;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001078 last_group_start *= group_size;
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02001079 for (i = 0; i < raid_disks; i++) {
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001080 if (!(i % copies) && !(i > last_group_start))
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001081 rebuilds_per_group = 0;
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001082 if ((!rs->dev[i].rdev.sb_page ||
1083 !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
Jonathan Brassow4ec1e362012-10-11 13:40:24 +11001084 (++rebuilds_per_group >= copies))
Heinz Mauelshagen23fda2e2023-02-07 23:06:47 +01001085 goto too_many;
Jonathan Brassow4ec1e362012-10-11 13:40:24 +11001086 }
1087 break;
Jonathan Brassoweb649122012-10-11 13:40:09 +11001088 default:
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06001089 if (rebuild_cnt)
1090 return -EINVAL;
Jonathan Brassoweb649122012-10-11 13:40:09 +11001091 }
1092
1093 return 0;
1094
1095too_many:
Jonathan Brassoweb649122012-10-11 13:40:09 +11001096 return -EINVAL;
1097}
1098
1099/*
NeilBrown9d09e662011-01-13 20:00:02 +00001100 * Possible arguments are...
NeilBrown9d09e662011-01-13 20:00:02 +00001101 * <chunk_size> [optional_args]
1102 *
Jonathan Brassow32737272011-08-02 12:32:07 +01001103 * Argument definitions
1104 * <chunk_size> The number of sectors per disk that
Mike Snitzer43157842016-05-30 13:03:37 -04001105 * will form the "stripe"
Jonathan Brassow32737272011-08-02 12:32:07 +01001106 * [[no]sync] Force or prevent recovery of the
Mike Snitzer43157842016-05-30 13:03:37 -04001107 * entire array
NeilBrown9d09e662011-01-13 20:00:02 +00001108 * [rebuild <idx>] Rebuild the drive indicated by the index
Jonathan Brassow32737272011-08-02 12:32:07 +01001109 * [daemon_sleep <ms>] Time between bitmap daemon work to
Mike Snitzer43157842016-05-30 13:03:37 -04001110 * clear bits
NeilBrown9d09e662011-01-13 20:00:02 +00001111 * [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1112 * [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001113 * [write_mostly <idx>] Indicate a write mostly drive via index
NeilBrown9d09e662011-01-13 20:00:02 +00001114 * [max_write_behind <sectors>] See '-write-behind=' (man mdadm)
1115 * [stripe_cache <sectors>] Stripe cache size for higher RAIDs
Mike Snitzer43157842016-05-30 13:03:37 -04001116 * [region_size <sectors>] Defines granularity of bitmap
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01001117 * [journal_dev <dev>] raid4/5/6 journaling deviice
Heinz Mauelshagen255e2642023-01-25 23:31:55 +01001118 * (i.e. write hole closing log)
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001119 *
1120 * RAID10-only options:
Mike Snitzer43157842016-05-30 13:03:37 -04001121 * [raid10_copies <# copies>] Number of copies. (Default: 2)
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11001122 * [raid10_format <near|far|offset>] Layout algorithm. (Default: near)
NeilBrown9d09e662011-01-13 20:00:02 +00001123 */
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001124static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001125 unsigned int num_raid_params)
NeilBrown9d09e662011-01-13 20:00:02 +00001126{
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001127 int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001128 unsigned int raid10_copies = 2;
1129 unsigned int i, write_mostly = 0;
1130 unsigned int region_size = 0;
Mike Snitzer542f9032012-07-27 15:08:00 +01001131 sector_t max_io_len;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001132 const char *arg, *key;
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001133 struct raid_dev *rd;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001134 struct raid_type *rt = rs->raid_type;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001135
1136 arg = dm_shift_arg(as);
1137 num_raid_params--; /* Account for chunk_size argument */
1138
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001139 if (kstrtoint(arg, 10, &value) < 0) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001140 rs->ti->error = "Bad numerical argument given for chunk_size";
1141 return -EINVAL;
1142 }
NeilBrown9d09e662011-01-13 20:00:02 +00001143
1144 /*
1145 * First, parse the in-order required arguments
Jonathan Brassow32737272011-08-02 12:32:07 +01001146 * "chunk_size" is the only argument of this type.
NeilBrown9d09e662011-01-13 20:00:02 +00001147 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001148 if (rt_is_raid1(rt)) {
Jonathan Brassow32737272011-08-02 12:32:07 +01001149 if (value)
1150 DMERR("Ignoring chunk size parameter for RAID 1");
1151 value = 0;
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001152 } else if (!is_power_of_2(value)) {
1153 rs->ti->error = "Chunk size must be a power of 2";
1154 return -EINVAL;
1155 } else if (value < 8) {
1156 rs->ti->error = "Chunk size value is too small";
1157 return -EINVAL;
1158 }
NeilBrown9d09e662011-01-13 20:00:02 +00001159
1160 rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
NeilBrown9d09e662011-01-13 20:00:02 +00001161
1162 /*
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001163 * We set each individual device as In_sync with a completed
1164 * 'recovery_offset'. If there has been a device failure or
1165 * replacement then one of the following cases applies:
1166 *
1167 * 1) User specifies 'rebuild'.
Mike Snitzer43157842016-05-30 13:03:37 -04001168 * - Device is reset when param is read.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001169 * 2) A new device is supplied.
Mike Snitzer43157842016-05-30 13:03:37 -04001170 * - No matching superblock found, resets device.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001171 * 3) Device failure was transient and returns on reload.
Mike Snitzer43157842016-05-30 13:03:37 -04001172 * - Failure noticed, resets device for bitmap replay.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001173 * 4) Device hadn't completed recovery after previous failure.
Mike Snitzer43157842016-05-30 13:03:37 -04001174 * - Superblock is read and overrides recovery_offset.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001175 *
1176 * What is found in the superblocks of the devices is always
1177 * authoritative, unless 'rebuild' or '[no]sync' was specified.
1178 */
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +02001179 for (i = 0; i < rs->raid_disks; i++) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001180 set_bit(In_sync, &rs->dev[i].rdev.flags);
1181 rs->dev[i].rdev.recovery_offset = MaxSector;
1182 }
1183
1184 /*
NeilBrown9d09e662011-01-13 20:00:02 +00001185 * Second, parse the unordered optional arguments
1186 */
NeilBrown9d09e662011-01-13 20:00:02 +00001187 for (i = 0; i < num_raid_params; i++) {
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001188 key = dm_shift_arg(as);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001189 if (!key) {
1190 rs->ti->error = "Not enough raid parameters given";
1191 return -EINVAL;
1192 }
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001193
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001194 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001195 if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001196 rs->ti->error = "Only one 'nosync' argument allowed";
1197 return -EINVAL;
1198 }
NeilBrown9d09e662011-01-13 20:00:02 +00001199 continue;
1200 }
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001201 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001202 if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001203 rs->ti->error = "Only one 'sync' argument allowed";
1204 return -EINVAL;
1205 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001206 continue;
1207 }
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001208 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001209 if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001210 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1211 return -EINVAL;
1212 }
NeilBrown9d09e662011-01-13 20:00:02 +00001213 continue;
1214 }
1215
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02001216 arg = dm_shift_arg(as);
1217 i++; /* Account for the argument pairs */
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001218 if (!arg) {
1219 rs->ti->error = "Wrong number of raid parameters given";
1220 return -EINVAL;
1221 }
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001222
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001223 /*
1224 * Parameters that take a string value are checked here.
1225 */
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01001226 /* "raid10_format {near|offset|far} */
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001227 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001228 if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001229 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1230 return -EINVAL;
1231 }
1232 if (!rt_is_raid10(rt)) {
1233 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1234 return -EINVAL;
1235 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001236 raid10_format = raid10_name_to_format(arg);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001237 if (raid10_format < 0) {
1238 rs->ti->error = "Invalid 'raid10_format' value given";
1239 return raid10_format;
1240 }
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001241 continue;
1242 }
1243
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01001244 /* "journal_dev <dev>" */
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01001245 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1246 int r;
1247 struct md_rdev *jdev;
1248
1249 if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1250 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1251 return -EINVAL;
1252 }
1253 if (!rt_is_raid456(rt)) {
1254 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1255 return -EINVAL;
1256 }
1257 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1258 &rs->journal_dev.dev);
1259 if (r) {
1260 rs->ti->error = "raid4/5/6 journal device lookup failure";
1261 return r;
1262 }
1263 jdev = &rs->journal_dev.rdev;
1264 md_rdev_init(jdev);
1265 jdev->mddev = &rs->md;
1266 jdev->bdev = rs->journal_dev.dev->bdev;
Christoph Hellwig6dcbb522021-10-18 12:11:05 +02001267 jdev->sectors = bdev_nr_sectors(jdev->bdev);
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01001268 if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1269 rs->ti->error = "No space for raid4/5/6 journal";
1270 return -ENOSPC;
1271 }
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01001272 rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01001273 set_bit(Journal, &jdev->flags);
1274 continue;
1275 }
1276
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01001277 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1278 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1279 int r;
1280
1281 if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1282 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1283 return -EINVAL;
1284 }
1285 if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1286 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1287 return -EINVAL;
1288 }
1289 r = dm_raid_journal_mode_to_md(arg);
1290 if (r < 0) {
1291 rs->ti->error = "Invalid 'journal_mode' argument";
1292 return r;
1293 }
1294 rs->journal_dev.mode = r;
1295 continue;
1296 }
1297
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01001298 /*
1299 * Parameters with number values from here on.
1300 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001301 if (kstrtoint(arg, 10, &value) < 0) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001302 rs->ti->error = "Bad numerical argument given in raid params";
1303 return -EINVAL;
1304 }
NeilBrown9d09e662011-01-13 20:00:02 +00001305
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001306 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001307 /*
1308 * "rebuild" is being passed in by userspace to provide
1309 * indexes of replaced devices and to set up additional
1310 * devices on raid level takeover.
Mike Snitzer43157842016-05-30 13:03:37 -04001311 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001312 if (!__within_range(value, 0, rs->raid_disks - 1)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001313 rs->ti->error = "Invalid rebuild index given";
1314 return -EINVAL;
1315 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001316
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001317 if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1318 rs->ti->error = "rebuild for this index already given";
1319 return -EINVAL;
1320 }
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001321
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001322 rd = rs->dev + value;
1323 clear_bit(In_sync, &rd->rdev.flags);
1324 clear_bit(Faulty, &rd->rdev.flags);
1325 rd->rdev.recovery_offset = 0;
Mike Snitzer42863252016-06-02 12:27:46 -04001326 set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001327 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001328 if (!rt_is_raid1(rt)) {
1329 rs->ti->error = "write_mostly option is only valid for RAID1";
1330 return -EINVAL;
1331 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001332
Mike Snitzerbb91a632016-06-02 12:06:54 -04001333 if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001334 rs->ti->error = "Invalid write_mostly index given";
1335 return -EINVAL;
1336 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001337
Heinz Mauelshagen5fa146b2016-06-15 18:50:18 +02001338 write_mostly++;
Jonathan Brassow46bed2b2011-08-02 12:32:07 +01001339 set_bit(WriteMostly, &rs->dev[value].rdev.flags);
Mike Snitzer42863252016-06-02 12:27:46 -04001340 set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001341 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001342 if (!rt_is_raid1(rt)) {
1343 rs->ti->error = "max_write_behind option is only valid for RAID1";
1344 return -EINVAL;
1345 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001346
Mike Snitzer42863252016-06-02 12:27:46 -04001347 if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001348 rs->ti->error = "Only one max_write_behind argument pair allowed";
1349 return -EINVAL;
1350 }
NeilBrown9d09e662011-01-13 20:00:02 +00001351
1352 /*
1353 * In device-mapper, we specify things in sectors, but
1354 * MD records this value in kB
1355 */
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001356 if (value < 0 || value / 2 > COUNTER_MAX) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001357 rs->ti->error = "Max write-behind limit out of range";
1358 return -EINVAL;
1359 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001360
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001361 rs->md.bitmap_info.max_write_behind = value / 2;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001362 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001363 if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001364 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1365 return -EINVAL;
1366 }
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001367 if (value < 0) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001368 rs->ti->error = "daemon sleep period out of range";
1369 return -EINVAL;
1370 }
NeilBrown9d09e662011-01-13 20:00:02 +00001371 rs->md.bitmap_info.daemon_sleep = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001372 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
Jiang Jiance92fc42022-06-21 19:32:34 +08001373 /* Userspace passes new data_offset after having extended the data image LV */
Mike Snitzer42863252016-06-02 12:27:46 -04001374 if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001375 rs->ti->error = "Only one data_offset argument pair allowed";
1376 return -EINVAL;
1377 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001378 /* Ensure sensible data offset */
Heinz Mauelshagen75dd3b92016-06-15 22:27:08 +02001379 if (value < 0 ||
1380 (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001381 rs->ti->error = "Bogus data_offset value";
1382 return -EINVAL;
1383 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001384 rs->data_offset = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001385 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001386 /* Define the +/-# of disks to add to/remove from the given raid set */
Mike Snitzer42863252016-06-02 12:27:46 -04001387 if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001388 rs->ti->error = "Only one delta_disks argument pair allowed";
1389 return -EINVAL;
1390 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001391 /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001392 if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001393 rs->ti->error = "Too many delta_disk requested";
1394 return -EINVAL;
1395 }
Heinz Mauelshagen4763e542016-05-19 18:49:31 +02001396
1397 rs->delta_disks = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001398 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001399 if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001400 rs->ti->error = "Only one stripe_cache argument pair allowed";
1401 return -EINVAL;
1402 }
1403
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001404 if (!rt_is_raid456(rt)) {
1405 rs->ti->error = "Inappropriate argument: stripe_cache";
1406 return -EINVAL;
1407 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001408
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001409 if (value < 0) {
1410 rs->ti->error = "Bogus stripe cache entries value";
1411 return -EINVAL;
1412 }
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001413 rs->stripe_cache_entries = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001414 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001415 if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001416 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1417 return -EINVAL;
1418 }
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001419
1420 if (value < 0) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001421 rs->ti->error = "min_recovery_rate out of range";
1422 return -EINVAL;
1423 }
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001424 rs->md.sync_speed_min = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001425 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
Heinz Mauelshagenf15f64d2016-07-27 23:34:01 +02001426 if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001427 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1428 return -EINVAL;
1429 }
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001430
1431 if (value < 0) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001432 rs->ti->error = "max_recovery_rate out of range";
1433 return -EINVAL;
1434 }
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001435 rs->md.sync_speed_max = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001436 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001437 if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001438 rs->ti->error = "Only one region_size argument pair allowed";
1439 return -EINVAL;
1440 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001441
Jonathan Brassowc1084562011-08-02 12:32:07 +01001442 region_size = value;
Heinz Mauelshagen4257e082016-06-14 01:46:03 +02001443 rs->requested_bitmap_chunk_sectors = value;
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04001444 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
Mike Snitzer42863252016-06-02 12:27:46 -04001445 if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001446 rs->ti->error = "Only one raid10_copies argument pair allowed";
1447 return -EINVAL;
1448 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001449
Mike Snitzerbb91a632016-06-02 12:06:54 -04001450 if (!__within_range(value, 2, rs->md.raid_disks)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001451 rs->ti->error = "Bad value for 'raid10_copies'";
1452 return -EINVAL;
1453 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001454
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001455 raid10_copies = value;
NeilBrown9d09e662011-01-13 20:00:02 +00001456 } else {
1457 DMERR("Unable to parse RAID parameter: %s", key);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001458 rs->ti->error = "Unable to parse RAID parameter";
1459 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +00001460 }
1461 }
1462
Heinz Mauelshagen0d851d12016-06-15 18:43:55 +02001463 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1464 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1465 rs->ti->error = "sync and nosync are mutually exclusive";
1466 return -EINVAL;
1467 }
1468
Heinz Mauelshagen37f10be2016-06-24 23:21:37 +02001469 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1470 (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1471 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1472 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1473 return -EINVAL;
1474 }
1475
Heinz Mauelshagen5fa146b2016-06-15 18:50:18 +02001476 if (write_mostly >= rs->md.raid_disks) {
1477 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1478 return -EINVAL;
1479 }
1480
Heinz Mauelshagen13bc62d2018-03-28 17:07:14 +02001481 if (rs->md.sync_speed_max &&
1482 rs->md.sync_speed_min > rs->md.sync_speed_max) {
1483 rs->ti->error = "Bogus recovery rates";
1484 return -EINVAL;
1485 }
1486
Jonathan Brassowc1084562011-08-02 12:32:07 +01001487 if (validate_region_size(rs, region_size))
1488 return -EINVAL;
1489
1490 if (rs->md.chunk_sectors)
Mike Snitzer542f9032012-07-27 15:08:00 +01001491 max_io_len = rs->md.chunk_sectors;
Jonathan Brassowc1084562011-08-02 12:32:07 +01001492 else
Mike Snitzer542f9032012-07-27 15:08:00 +01001493 max_io_len = region_size;
Jonathan Brassowc1084562011-08-02 12:32:07 +01001494
Mike Snitzer542f9032012-07-27 15:08:00 +01001495 if (dm_set_target_max_io_len(rs->ti, max_io_len))
1496 return -EINVAL;
Jonathan Brassow32737272011-08-02 12:32:07 +01001497
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001498 if (rt_is_raid10(rt)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001499 if (raid10_copies > rs->md.raid_disks) {
1500 rs->ti->error = "Not enough devices to satisfy specification";
1501 return -EINVAL;
1502 }
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05001503
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001504 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001505 if (rs->md.new_layout < 0) {
1506 rs->ti->error = "Error getting raid10 format";
1507 return rs->md.new_layout;
1508 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001509
1510 rt = get_raid_type_by_ll(10, rs->md.new_layout);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001511 if (!rt) {
1512 rs->ti->error = "Failed to recognize new raid10 layout";
1513 return -EINVAL;
1514 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001515
1516 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1517 rt->algorithm == ALGORITHM_RAID10_NEAR) &&
Mike Snitzer42863252016-06-02 12:27:46 -04001518 test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001519 rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1520 return -EINVAL;
1521 }
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001522 }
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02001523
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001524 rs->raid10_copies = raid10_copies;
Jonathan E Brassowc039c332012-07-27 15:08:04 +01001525
NeilBrown9d09e662011-01-13 20:00:02 +00001526 /* Assume there are no metadata devices until the drives are parsed */
1527 rs->md.persistent = 0;
1528 rs->md.external = 1;
1529
Heinz Mauelshagenf0902792016-05-19 18:49:27 +02001530 /* Check, if any invalid ctr arguments have been passed in for the raid level */
Heinz Mauelshagena30cbc02016-06-09 16:42:16 +02001531 return rs_check_for_valid_flags(rs);
NeilBrown9d09e662011-01-13 20:00:02 +00001532}
1533
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001534/* Set raid4/5/6 cache size */
1535static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1536{
1537 int r;
1538 struct r5conf *conf;
1539 struct mddev *mddev = &rs->md;
1540 uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1541 uint32_t nr_stripes = rs->stripe_cache_entries;
1542
1543 if (!rt_is_raid456(rs->raid_type)) {
1544 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1545 return -EINVAL;
1546 }
1547
1548 if (nr_stripes < min_stripes) {
1549 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1550 nr_stripes, min_stripes);
1551 nr_stripes = min_stripes;
1552 }
1553
1554 conf = mddev->private;
1555 if (!conf) {
1556 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1557 return -EINVAL;
1558 }
1559
1560 /* Try setting number of stripes in raid456 stripe cache */
1561 if (conf->min_nr_stripes != nr_stripes) {
1562 r = raid5_set_cache_size(mddev, nr_stripes);
1563 if (r) {
1564 rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1565 return r;
1566 }
1567
1568 DMINFO("%u stripe cache entries", nr_stripes);
1569 }
1570
1571 return 0;
1572}
1573
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02001574/* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1575static unsigned int mddev_data_stripes(struct raid_set *rs)
1576{
1577 return rs->md.raid_disks - rs->raid_type->parity_devs;
1578}
1579
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001580/* Return # of data stripes of @rs (i.e. as of ctr) */
1581static unsigned int rs_data_stripes(struct raid_set *rs)
1582{
1583 return rs->raid_disks - rs->raid_type->parity_devs;
1584}
1585
Heinz Mauelshagen50c4feb2017-01-14 03:53:08 +01001586/*
1587 * Retrieve rdev->sectors from any valid raid device of @rs
1588 * to allow userpace to pass in arbitray "- -" device tupples.
1589 */
1590static sector_t __rdev_sectors(struct raid_set *rs)
1591{
1592 int i;
1593
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02001594 for (i = 0; i < rs->raid_disks; i++) {
Heinz Mauelshagen50c4feb2017-01-14 03:53:08 +01001595 struct md_rdev *rdev = &rs->dev[i].rdev;
1596
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01001597 if (!test_bit(Journal, &rdev->flags) &&
1598 rdev->bdev && rdev->sectors)
Heinz Mauelshagen50c4feb2017-01-14 03:53:08 +01001599 return rdev->sectors;
1600 }
1601
Heinz Mauelshagen4d49f1b2017-06-30 15:45:58 +02001602 return 0;
Heinz Mauelshagen50c4feb2017-01-14 03:53:08 +01001603}
1604
Heinz Mauelshagen188a212d2017-12-02 01:03:59 +01001605/* Check that calculated dev_sectors fits all component devices. */
1606static int _check_data_dev_sectors(struct raid_set *rs)
1607{
1608 sector_t ds = ~0;
1609 struct md_rdev *rdev;
1610
1611 rdev_for_each(rdev, &rs->md)
1612 if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
Christoph Hellwig6dcbb522021-10-18 12:11:05 +02001613 ds = min(ds, bdev_nr_sectors(rdev->bdev));
Heinz Mauelshagen188a212d2017-12-02 01:03:59 +01001614 if (ds < rs->md.dev_sectors) {
1615 rs->ti->error = "Component device(s) too small";
1616 return -EINVAL;
1617 }
1618 }
1619
1620 return 0;
1621}
1622
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001623/* Calculate the sectors per device and per array used for @rs */
Heinz Mauelshagen22c992e2019-10-01 17:47:52 +02001624static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001625{
1626 int delta_disks;
1627 unsigned int data_stripes;
Heinz Mauelshagen22c992e2019-10-01 17:47:52 +02001628 sector_t array_sectors = sectors, dev_sectors = sectors;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001629 struct mddev *mddev = &rs->md;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001630
1631 if (use_mddev) {
1632 delta_disks = mddev->delta_disks;
1633 data_stripes = mddev_data_stripes(rs);
1634 } else {
1635 delta_disks = rs->delta_disks;
1636 data_stripes = rs_data_stripes(rs);
1637 }
1638
1639 /* Special raid1 case w/o delta_disks support (yet) */
1640 if (rt_is_raid1(rs->raid_type))
1641 ;
1642 else if (rt_is_raid10(rs->raid_type)) {
1643 if (rs->raid10_copies < 2 ||
1644 delta_disks < 0) {
1645 rs->ti->error = "Bogus raid10 data copies or delta disks";
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001646 return -EINVAL;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001647 }
1648
1649 dev_sectors *= rs->raid10_copies;
1650 if (sector_div(dev_sectors, data_stripes))
1651 goto bad;
1652
1653 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1654 if (sector_div(array_sectors, rs->raid10_copies))
1655 goto bad;
1656
1657 } else if (sector_div(dev_sectors, data_stripes))
1658 goto bad;
1659
1660 else
1661 /* Striped layouts */
1662 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1663
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001664 mddev->array_sectors = array_sectors;
1665 mddev->dev_sectors = dev_sectors;
Heinz Mauelshagen22c992e2019-10-01 17:47:52 +02001666 rs_set_rdev_sectors(rs);
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001667
Heinz Mauelshagen188a212d2017-12-02 01:03:59 +01001668 return _check_data_dev_sectors(rs);
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001669bad:
1670 rs->ti->error = "Target length not divisible by number of data devices";
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02001671 return -EINVAL;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001672}
1673
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02001674/* Setup recovery on @rs */
Heinz Mauelshagenf9f3ee92019-10-01 17:47:54 +02001675static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02001676{
1677 /* raid0 does not recover */
1678 if (rs_is_raid0(rs))
1679 rs->md.recovery_cp = MaxSector;
1680 /*
1681 * A raid6 set has to be recovered either
1682 * completely or for the grown part to
1683 * ensure proper parity and Q-Syndrome
1684 */
1685 else if (rs_is_raid6(rs))
1686 rs->md.recovery_cp = dev_sectors;
1687 /*
1688 * Other raid set types may skip recovery
1689 * depending on the 'nosync' flag.
1690 */
1691 else
1692 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1693 ? MaxSector : dev_sectors;
1694}
1695
NeilBrown9d09e662011-01-13 20:00:02 +00001696static void do_table_event(struct work_struct *ws)
1697{
1698 struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1699
Heinz Mauelshagen9d9d9392016-06-16 03:15:49 +02001700 smp_rmb(); /* Make sure we access most actual mddev properties */
Heinz Mauelshagen61e06e22017-12-02 01:03:51 +01001701 if (!rs_is_reshaping(rs)) {
1702 if (rs_is_raid10(rs))
1703 rs_set_rdev_sectors(rs);
Heinz Mauelshagen9d9d9392016-06-16 03:15:49 +02001704 rs_set_capacity(rs);
Heinz Mauelshagen61e06e22017-12-02 01:03:51 +01001705 }
NeilBrown9d09e662011-01-13 20:00:02 +00001706 dm_table_event(rs->ti->table);
1707}
1708
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001709/*
1710 * Make sure a valid takover (level switch) is being requested on @rs
1711 *
1712 * Conversions of raid sets from one MD personality to another
1713 * have to conform to restrictions which are enforced here.
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001714 */
1715static int rs_check_takeover(struct raid_set *rs)
1716{
1717 struct mddev *mddev = &rs->md;
1718 unsigned int near_copies;
1719
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001720 if (rs->md.degraded) {
1721 rs->ti->error = "Can't takeover degraded raid set";
1722 return -EPERM;
1723 }
1724
1725 if (rs_is_reshaping(rs)) {
1726 rs->ti->error = "Can't takeover reshaping raid set";
1727 return -EPERM;
1728 }
1729
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001730 switch (mddev->level) {
1731 case 0:
1732 /* raid0 -> raid1/5 with one disk */
1733 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1734 mddev->raid_disks == 1)
1735 return 0;
1736
1737 /* raid0 -> raid10 */
1738 if (mddev->new_level == 10 &&
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02001739 !(rs->raid_disks % mddev->raid_disks))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001740 return 0;
1741
1742 /* raid0 with multiple disks -> raid4/5/6 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001743 if (__within_range(mddev->new_level, 4, 6) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001744 mddev->new_layout == ALGORITHM_PARITY_N &&
1745 mddev->raid_disks > 1)
1746 return 0;
1747
1748 break;
1749
1750 case 10:
1751 /* Can't takeover raid10_offset! */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001752 if (__is_raid10_offset(mddev->layout))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001753 break;
1754
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001755 near_copies = __raid10_near_copies(mddev->layout);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001756
1757 /* raid10* -> raid0 */
1758 if (mddev->new_level == 0) {
1759 /* Can takeover raid10_near with raid disks divisable by data copies! */
1760 if (near_copies > 1 &&
1761 !(mddev->raid_disks % near_copies)) {
1762 mddev->raid_disks /= near_copies;
1763 mddev->delta_disks = mddev->raid_disks;
1764 return 0;
1765 }
1766
1767 /* Can takeover raid10_far */
1768 if (near_copies == 1 &&
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001769 __raid10_far_copies(mddev->layout) > 1)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001770 return 0;
1771
1772 break;
1773 }
1774
1775 /* raid10_{near,far} -> raid1 */
1776 if (mddev->new_level == 1 &&
Mike Snitzere6ca5e12016-06-02 15:27:22 -04001777 max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001778 return 0;
1779
1780 /* raid10_{near,far} with 2 disks -> raid4/5 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001781 if (__within_range(mddev->new_level, 4, 5) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001782 mddev->raid_disks == 2)
1783 return 0;
1784 break;
1785
1786 case 1:
1787 /* raid1 with 2 disks -> raid4/5 */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001788 if (__within_range(mddev->new_level, 4, 5) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001789 mddev->raid_disks == 2) {
1790 mddev->degraded = 1;
1791 return 0;
1792 }
1793
1794 /* raid1 -> raid0 */
1795 if (mddev->new_level == 0 &&
1796 mddev->raid_disks == 1)
1797 return 0;
1798
1799 /* raid1 -> raid10 */
1800 if (mddev->new_level == 10)
1801 return 0;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001802 break;
1803
1804 case 4:
1805 /* raid4 -> raid0 */
1806 if (mddev->new_level == 0)
1807 return 0;
1808
1809 /* raid4 -> raid1/5 with 2 disks */
1810 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1811 mddev->raid_disks == 2)
1812 return 0;
1813
1814 /* raid4 -> raid5/6 with parity N */
Mike Snitzerbb91a632016-06-02 12:06:54 -04001815 if (__within_range(mddev->new_level, 5, 6) &&
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001816 mddev->layout == ALGORITHM_PARITY_N)
1817 return 0;
1818 break;
1819
1820 case 5:
1821 /* raid5 with parity N -> raid0 */
1822 if (mddev->new_level == 0 &&
1823 mddev->layout == ALGORITHM_PARITY_N)
1824 return 0;
1825
1826 /* raid5 with parity N -> raid4 */
1827 if (mddev->new_level == 4 &&
1828 mddev->layout == ALGORITHM_PARITY_N)
1829 return 0;
1830
1831 /* raid5 with 2 disks -> raid1/4/10 */
1832 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1833 mddev->raid_disks == 2)
1834 return 0;
1835
Heinz Mauelshagen6ee0bae2016-06-15 22:29:09 +02001836 /* raid5_* -> raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001837 if (mddev->new_level == 6 &&
1838 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
Mike Snitzerbb91a632016-06-02 12:06:54 -04001839 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001840 return 0;
1841 break;
1842
1843 case 6:
1844 /* raid6 with parity N -> raid0 */
1845 if (mddev->new_level == 0 &&
1846 mddev->layout == ALGORITHM_PARITY_N)
1847 return 0;
1848
1849 /* raid6 with parity N -> raid4 */
1850 if (mddev->new_level == 4 &&
1851 mddev->layout == ALGORITHM_PARITY_N)
1852 return 0;
1853
Heinz Mauelshagen6ee0bae2016-06-15 22:29:09 +02001854 /* raid6_*_n with Q-Syndrome N -> raid5_* */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001855 if (mddev->new_level == 5 &&
1856 ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
Mike Snitzerbb91a632016-06-02 12:06:54 -04001857 __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001858 return 0;
Gustavo A. R. Silvabe962b22021-04-20 18:09:02 -04001859 break;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001860
1861 default:
1862 break;
1863 }
1864
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04001865 rs->ti->error = "takeover not possible";
1866 return -EINVAL;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02001867}
1868
1869/* True if @rs requested to be taken over */
1870static bool rs_takeover_requested(struct raid_set *rs)
1871{
1872 return rs->md.new_level != rs->md.level;
1873}
1874
Heinz Mauelshagenf99a8e42021-04-21 23:32:36 +02001875/* True if layout is set to reshape. */
1876static bool rs_is_layout_change(struct raid_set *rs, bool use_mddev)
1877{
1878 return (use_mddev ? rs->md.delta_disks : rs->delta_disks) ||
1879 rs->md.new_layout != rs->md.layout ||
1880 rs->md.new_chunk_sectors != rs->md.chunk_sectors;
1881}
1882
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001883/* True if @rs is requested to reshape by ctr */
1884static bool rs_reshape_requested(struct raid_set *rs)
1885{
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001886 bool change;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001887 struct mddev *mddev = &rs->md;
1888
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001889 if (rs_takeover_requested(rs))
1890 return false;
1891
Heinz Mauelshagen552aa672017-12-13 17:13:21 +01001892 if (rs_is_raid0(rs))
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001893 return false;
1894
Heinz Mauelshagenf99a8e42021-04-21 23:32:36 +02001895 change = rs_is_layout_change(rs, false);
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001896
1897 /* Historical case to support raid1 reshape without delta disks */
Heinz Mauelshagen552aa672017-12-13 17:13:21 +01001898 if (rs_is_raid1(rs)) {
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02001899 if (rs->delta_disks)
1900 return !!rs->delta_disks;
1901
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001902 return !change &&
1903 mddev->raid_disks != rs->raid_disks;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02001904 }
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001905
Heinz Mauelshagen552aa672017-12-13 17:13:21 +01001906 if (rs_is_raid10(rs))
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02001907 return change &&
1908 !__is_raid10_far(mddev->new_layout) &&
1909 rs->delta_disks >= 0;
1910
1911 return change;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02001912}
1913
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001914/* Features */
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001915#define FEATURE_FLAG_SUPPORTS_V190 0x1 /* Supports extended superblock */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001916
1917/* State flags for sb->flags */
1918#define SB_FLAG_RESHAPE_ACTIVE 0x1
1919#define SB_FLAG_RESHAPE_BACKWARDS 0x2
1920
NeilBrown9d09e662011-01-13 20:00:02 +00001921/*
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001922 * This structure is never routinely used by userspace, unlike md superblocks.
1923 * Devices with this superblock should only ever be accessed via device-mapper.
1924 */
1925#define DM_RAID_MAGIC 0x64526D44
1926struct dm_raid_superblock {
1927 __le32 magic; /* "DmRd" */
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001928 __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001929
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001930 __le32 num_devices; /* Number of devices in this raid set. (Max 64) */
1931 __le32 array_position; /* The position of this drive in the raid set */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001932
1933 __le64 events; /* Incremented by md when superblock updated */
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001934 __le64 failed_devices; /* Pre 1.9.0 part of bit field of devices to */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001935 /* indicate failures (see extension below) */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001936
1937 /*
1938 * This offset tracks the progress of the repair or replacement of
1939 * an individual drive.
1940 */
1941 __le64 disk_recovery_offset;
1942
1943 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001944 * This offset tracks the progress of the initial raid set
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001945 * synchronisation/parity calculation.
1946 */
1947 __le64 array_resync_offset;
1948
1949 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001950 * raid characteristics
Jonathan Brassowb12d4372011-08-02 12:32:07 +01001951 */
1952 __le32 level;
1953 __le32 layout;
1954 __le32 stripe_sectors;
1955
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001956 /********************************************************************
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001957 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001958 *
Heinz Mauelshagenc4d097d2017-06-23 17:27:01 +02001959 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001960 */
1961
1962 __le32 flags; /* Flags defining array states for reshaping */
1963
1964 /*
1965 * This offset tracks the progress of a raid
1966 * set reshape in order to be able to restart it
1967 */
1968 __le64 reshape_position;
1969
1970 /*
1971 * These define the properties of the array in case of an interrupted reshape
1972 */
1973 __le32 new_level;
1974 __le32 new_layout;
1975 __le32 new_stripe_sectors;
1976 __le32 delta_disks;
1977
1978 __le64 array_sectors; /* Array size in sectors */
1979
1980 /*
1981 * Sector offsets to data on devices (reshaping).
1982 * Needed to support out of place reshaping, thus
1983 * not writing over any stripes whilst converting
1984 * them from old to new layout
1985 */
1986 __le64 data_offset;
1987 __le64 new_data_offset;
1988
1989 __le64 sectors; /* Used device size in sectors */
1990
1991 /*
Yu Zheefdd3c32023-02-06 11:27:39 +08001992 * Additional Bit field of devices indicating failures to support
Mike Snitzer9b6e5422016-06-02 11:48:09 -04001993 * up to 256 devices with the 1.9.0 on-disk metadata format
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02001994 */
1995 __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1996
1997 __le32 incompat_features; /* Used to indicate any incompatible features */
1998
1999 /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002000} __packed;
2001
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002002/*
2003 * Check for reshape constraints on raid set @rs:
2004 *
2005 * - reshape function non-existent
2006 * - degraded set
2007 * - ongoing recovery
2008 * - ongoing reshape
2009 *
2010 * Returns 0 if none or -EPERM if given constraint
2011 * and error message reference in @errmsg
2012 */
2013static int rs_check_reshape(struct raid_set *rs)
2014{
2015 struct mddev *mddev = &rs->md;
2016
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002017 if (!mddev->pers || !mddev->pers->check_reshape)
2018 rs->ti->error = "Reshape not supported";
2019 else if (mddev->degraded)
2020 rs->ti->error = "Can't reshape degraded raid set";
2021 else if (rs_is_recovering(rs))
2022 rs->ti->error = "Convert request on recovering raid set prohibited";
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002023 else if (rs_is_reshaping(rs))
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002024 rs->ti->error = "raid set already reshaping!";
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002025 else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2026 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002027 else
2028 return 0;
2029
2030 return -EPERM;
2031}
2032
Heinz Mauelshagene2568462017-01-14 03:53:10 +01002033static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002034{
2035 BUG_ON(!rdev->sb_page);
2036
Heinz Mauelshagene2568462017-01-14 03:53:10 +01002037 if (rdev->sb_loaded && !force_reload)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002038 return 0;
2039
Heinz Mauelshagene2568462017-01-14 03:53:10 +01002040 rdev->sb_loaded = 0;
2041
Bart Van Assche4ce4c732022-07-14 11:06:57 -07002042 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true)) {
Jonathan E Brassow04475682012-03-28 18:41:26 +01002043 DMERR("Failed to read superblock of device at position %d",
2044 rdev->raid_disk);
Jonathan Brassowc32fb9e2012-05-22 13:55:31 +10002045 md_error(rdev->mddev, rdev);
Heinz Mauelshagene2568462017-01-14 03:53:10 +01002046 set_bit(Faulty, &rdev->flags);
2047 return -EIO;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002048 }
2049
2050 rdev->sb_loaded = 1;
2051
2052 return 0;
2053}
2054
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002055static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2056{
2057 failed_devices[0] = le64_to_cpu(sb->failed_devices);
2058 memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2059
Mike Snitzer42863252016-06-02 12:27:46 -04002060 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002061 int i = ARRAY_SIZE(sb->extended_failed_devices);
2062
2063 while (i--)
2064 failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2065 }
2066}
2067
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002068static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2069{
2070 int i = ARRAY_SIZE(sb->extended_failed_devices);
2071
2072 sb->failed_devices = cpu_to_le64(failed_devices[0]);
2073 while (i--)
2074 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2075}
2076
2077/*
2078 * Synchronize the superblock members with the raid set properties
2079 *
2080 * All superblock data is little endian.
2081 */
NeilBrownfd01b882011-10-11 16:47:53 +11002082static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002083{
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002084 bool update_failed_devices = false;
2085 unsigned int i;
2086 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002087 struct dm_raid_superblock *sb;
Jonathan Brassow81f382f2012-05-22 13:55:30 +10002088 struct raid_set *rs = container_of(mddev, struct raid_set, md);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002089
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002090 /* No metadata device, no superblock */
2091 if (!rdev->meta_bdev)
2092 return;
2093
2094 BUG_ON(!rdev->sb_page);
2095
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002096 sb = page_address(rdev->sb_page);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002097
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002098 sb_retrieve_failed_devices(sb, failed_devices);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002099
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002100 for (i = 0; i < rs->raid_disks; i++)
2101 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2102 update_failed_devices = true;
2103 set_bit(i, (void *) failed_devices);
2104 }
2105
2106 if (update_failed_devices)
2107 sb_update_failed_devices(sb, failed_devices);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002108
2109 sb->magic = cpu_to_le32(DM_RAID_MAGIC);
Mike Snitzer9b6e5422016-06-02 11:48:09 -04002110 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002111
2112 sb->num_devices = cpu_to_le32(mddev->raid_disks);
2113 sb->array_position = cpu_to_le32(rdev->raid_disk);
2114
2115 sb->events = cpu_to_le64(mddev->events);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002116
2117 sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2118 sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2119
2120 sb->level = cpu_to_le32(mddev->level);
2121 sb->layout = cpu_to_le32(mddev->layout);
2122 sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002123
Heinz Mauelshagenc4d097d2017-06-23 17:27:01 +02002124 /********************************************************************
2125 * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2126 *
2127 * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2128 */
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002129 sb->new_level = cpu_to_le32(mddev->new_level);
2130 sb->new_layout = cpu_to_le32(mddev->new_layout);
2131 sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2132
2133 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2134
2135 smp_rmb(); /* Make sure we access most recent reshape position */
2136 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2137 if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2138 /* Flag ongoing reshape */
2139 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2140
2141 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2142 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
Mike Snitzer42863252016-06-02 12:27:46 -04002143 } else {
2144 /* Clear reshape flags */
2145 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2146 }
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002147
2148 sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2149 sb->data_offset = cpu_to_le64(rdev->data_offset);
2150 sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2151 sb->sectors = cpu_to_le64(rdev->sectors);
Heinz Mauelshagenb2a48722016-08-03 17:47:04 +02002152 sb->incompat_features = cpu_to_le32(0);
Heinz Mauelshagen7b34df72016-05-19 18:49:32 +02002153
2154 /* Zero out the rest of the payload after the size of the superblock */
2155 memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002156}
2157
2158/*
2159 * super_load
2160 *
2161 * This function creates a superblock if one is not found on the device
2162 * and will decide which superblock to use if there's a choice.
2163 *
2164 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2165 */
NeilBrown3cb03002011-10-11 16:45:26 +11002166static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002167{
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002168 int r;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002169 struct dm_raid_superblock *sb;
2170 struct dm_raid_superblock *refsb;
2171 uint64_t events_sb, events_refsb;
2172
Heinz Mauelshagene2568462017-01-14 03:53:10 +01002173 r = read_disk_sb(rdev, rdev->sb_size, false);
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002174 if (r)
2175 return r;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002176
2177 sb = page_address(rdev->sb_page);
Jonathan E Brassow3aa3b2b2012-03-07 19:09:47 +00002178
2179 /*
2180 * Two cases that we want to write new superblocks and rebuild:
2181 * 1) New device (no matching magic number)
2182 * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2183 */
2184 if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2185 (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002186 super_sync(rdev->mddev, rdev);
2187
2188 set_bit(FirstUse, &rdev->flags);
Mike Snitzer9b6e5422016-06-02 11:48:09 -04002189 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002190
2191 /* Force writing of superblocks to disk */
Shaohua Li29530792016-12-08 15:48:19 -08002192 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002193
2194 /* Any superblock is better than none, choose that if given */
2195 return refdev ? 0 : 1;
2196 }
2197
2198 if (!refdev)
2199 return 1;
2200
2201 events_sb = le64_to_cpu(sb->events);
2202
2203 refsb = page_address(refdev->sb_page);
2204 events_refsb = le64_to_cpu(refsb->events);
2205
2206 return (events_sb > events_refsb) ? 1 : 0;
2207}
2208
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002209static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002210{
2211 int role;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002212 unsigned int d;
2213 struct mddev *mddev = &rs->md;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002214 uint64_t events_sb;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002215 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002216 struct dm_raid_superblock *sb;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002217 uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11002218 struct md_rdev *r;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002219 struct dm_raid_superblock *sb2;
2220
2221 sb = page_address(rdev->sb_page);
2222 events_sb = le64_to_cpu(sb->events);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002223
2224 /*
2225 * Initialise to 1 if this is a new superblock.
2226 */
2227 mddev->events = events_sb ? : 1;
2228
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002229 mddev->reshape_position = MaxSector;
2230
Heinz Mauelshagen453c2a82016-10-18 17:46:45 +02002231 mddev->raid_disks = le32_to_cpu(sb->num_devices);
2232 mddev->level = le32_to_cpu(sb->level);
2233 mddev->layout = le32_to_cpu(sb->layout);
2234 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2235
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002236 /*
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002237 * Reshaping is supported, e.g. reshape_position is valid
2238 * in superblock and superblock content is authoritative.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002239 */
Mike Snitzer42863252016-06-02 12:27:46 -04002240 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002241 /* Superblock is authoritative wrt given raid set layout! */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002242 mddev->new_level = le32_to_cpu(sb->new_level);
2243 mddev->new_layout = le32_to_cpu(sb->new_layout);
2244 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2245 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2246 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2247
2248 /* raid was reshaping and got interrupted */
Mike Snitzer42863252016-06-02 12:27:46 -04002249 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2250 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002251 DMERR("Reshape requested but raid set is still reshaping");
2252 return -EINVAL;
2253 }
2254
2255 if (mddev->delta_disks < 0 ||
Mike Snitzer42863252016-06-02 12:27:46 -04002256 (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002257 mddev->reshape_backwards = 1;
2258 else
2259 mddev->reshape_backwards = 0;
2260
2261 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2262 rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2263 }
2264
2265 } else {
2266 /*
Mike Snitzer9b6e5422016-06-02 11:48:09 -04002267 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002268 */
Heinz Mauelshagen453c2a82016-10-18 17:46:45 +02002269 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2270 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002271
Heinz Mauelshagen453c2a82016-10-18 17:46:45 +02002272 if (rs_takeover_requested(rs)) {
2273 if (rt_cur && rt_new)
2274 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2275 rt_cur->name, rt_new->name);
2276 else
2277 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2278 return -EINVAL;
2279 } else if (rs_reshape_requested(rs)) {
2280 DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2281 if (mddev->layout != mddev->new_layout) {
2282 if (rt_cur && rt_new)
2283 DMERR(" current layout %s vs new layout %s",
2284 rt_cur->name, rt_new->name);
2285 else
2286 DMERR(" current layout 0x%X vs new layout 0x%X",
2287 le32_to_cpu(sb->layout), mddev->new_layout);
2288 }
2289 if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2290 DMERR(" current stripe sectors %u vs new stripe sectors %u",
2291 mddev->chunk_sectors, mddev->new_chunk_sectors);
2292 if (rs->delta_disks)
2293 DMERR(" current %u disks vs new %u disks",
2294 mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2295 if (rs_is_raid10(rs)) {
2296 DMERR(" Old layout: %s w/ %u copies",
2297 raid10_md_layout_to_format(mddev->layout),
2298 raid10_md_layout_to_copies(mddev->layout));
2299 DMERR(" New layout: %s w/ %u copies",
2300 raid10_md_layout_to_format(mddev->new_layout),
2301 raid10_md_layout_to_copies(mddev->new_layout));
2302 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002303 return -EINVAL;
2304 }
2305
Heinz Mauelshagenb052b072016-10-17 21:20:07 +02002306 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002307 }
2308
Mike Snitzer42863252016-06-02 12:27:46 -04002309 if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002310 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2311
2312 /*
2313 * During load, we set FirstUse if a new superblock was written.
2314 * There are two reasons we might not have a superblock:
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002315 * 1) The raid set is brand new - in which case, all of the
Mike Snitzer43157842016-05-30 13:03:37 -04002316 * devices must have their In_sync bit set. Also,
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002317 * recovery_cp must be 0, unless forced.
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002318 * 2) This is a new device being added to an old raid set
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002319 * and the new device needs to be rebuilt - in which
2320 * case the In_sync bit will /not/ be set and
2321 * recovery_cp must be MaxSector.
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002322 * 3) This is/are a new device(s) being added to an old
2323 * raid set during takeover to a higher raid level
2324 * to provide capacity for redundancy or during reshape
2325 * to add capacity to grow the raid set.
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002326 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002327 d = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11002328 rdev_for_each(r, mddev) {
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01002329 if (test_bit(Journal, &rdev->flags))
2330 continue;
2331
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002332 if (test_bit(FirstUse, &r->flags))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002333 new_devs++;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002334
2335 if (!test_bit(In_sync, &r->flags)) {
2336 DMINFO("Device %d specified for rebuild; clearing superblock",
2337 r->raid_disk);
2338 rebuilds++;
2339
2340 if (test_bit(FirstUse, &r->flags))
2341 rebuild_and_new++;
2342 }
2343
2344 d++;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002345 }
2346
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002347 if (new_devs == rs->raid_disks || !rebuilds) {
2348 /* Replace a broken device */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002349 if (new_devs == rs->raid_disks) {
2350 DMINFO("Superblocks created for new raid set");
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002351 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002352 } else if (new_devs != rebuilds &&
2353 new_devs != rs->delta_disks) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002354 DMERR("New device injected into existing raid set without "
2355 "'delta_disks' or 'rebuild' parameter specified");
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002356 return -EINVAL;
2357 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002358 } else if (new_devs && new_devs != rebuilds) {
2359 DMERR("%u 'rebuild' devices cannot be injected into"
2360 " a raid set with %u other first-time devices",
2361 rebuilds, new_devs);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002362 return -EINVAL;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002363 } else if (rebuilds) {
2364 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2365 DMERR("new device%s provided without 'rebuild'",
2366 new_devs > 1 ? "s" : "");
2367 return -EINVAL;
Heinz Mauelshagen11e47232017-12-13 17:13:18 +01002368 } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002369 DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2370 (unsigned long long) mddev->recovery_cp);
2371 return -EINVAL;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002372 } else if (rs_is_reshaping(rs)) {
2373 DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2374 (unsigned long long) mddev->reshape_position);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002375 return -EINVAL;
2376 }
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002377 }
2378
2379 /*
2380 * Now we set the Faulty bit for those devices that are
2381 * recorded in the superblock as failed.
2382 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002383 sb_retrieve_failed_devices(sb, failed_devices);
NeilBrowndafb20f2012-03-19 12:46:39 +11002384 rdev_for_each(r, mddev) {
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01002385 if (test_bit(Journal, &rdev->flags) ||
2386 !r->sb_page)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002387 continue;
2388 sb2 = page_address(r->sb_page);
2389 sb2->failed_devices = 0;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002390 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002391
2392 /*
2393 * Check for any device re-ordering.
2394 */
2395 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2396 role = le32_to_cpu(sb2->array_position);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002397 if (role < 0)
2398 continue;
2399
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002400 if (role != r->raid_disk) {
Heinz Mauelshagen453c2a82016-10-18 17:46:45 +02002401 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
Mike Snitzere6ca5e12016-06-02 15:27:22 -04002402 if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002403 rs->raid_disks % rs->raid10_copies) {
2404 rs->ti->error =
2405 "Cannot change raid10 near set to odd # of devices!";
2406 return -EINVAL;
2407 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002408
2409 sb2->array_position = cpu_to_le32(r->raid_disk);
2410
2411 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002412 !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2413 !rt_is_raid1(rs->raid_type)) {
2414 rs->ti->error = "Cannot change device positions in raid set";
2415 return -EINVAL;
2416 }
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002417
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002418 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002419 }
2420
2421 /*
2422 * Partial recovery is performed on
2423 * returning failed devices.
2424 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002425 if (test_bit(role, (void *) failed_devices))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002426 set_bit(Faulty, &r->flags);
2427 }
2428 }
2429
2430 return 0;
2431}
2432
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002433static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002434{
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002435 struct mddev *mddev = &rs->md;
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002436 struct dm_raid_superblock *sb;
2437
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01002438 if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002439 return 0;
2440
2441 sb = page_address(rdev->sb_page);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002442
2443 /*
2444 * If mddev->events is not set, we know we have not yet initialized
2445 * the array.
2446 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002447 if (!mddev->events && super_init_validation(rs, rdev))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002448 return -EINVAL;
2449
Andy Whitcroft5c336772016-10-11 15:16:57 +01002450 if (le32_to_cpu(sb->compat_features) &&
2451 le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
Mike Snitzer9b6e5422016-06-02 11:48:09 -04002452 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2453 return -EINVAL;
2454 }
2455
2456 if (sb->incompat_features) {
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002457 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
Heinz Mauelshagen4c9971c2016-04-29 18:59:56 +02002458 return -EINVAL;
2459 }
2460
Heinz Mauelshagenf9f3ee92019-10-01 17:47:54 +02002461 /* Enable bitmap creation on @rs unless no metadevs or raid0 or journaled raid4/5/6 set. */
Heinz Mauelshagend857ad72018-10-12 20:24:25 +02002462 mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
Heinz Mauelshagen977f1a02017-01-14 03:53:11 +01002463 mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02002464
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002465 if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
Heinz Mauelshagenc4d097d2017-06-23 17:27:01 +02002466 /*
2467 * Retrieve rdev size stored in superblock to be prepared for shrink.
2468 * Check extended superblock members are present otherwise the size
2469 * will not be set!
2470 */
2471 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2472 rdev->sectors = le64_to_cpu(sb->sectors);
2473
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002474 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002475 if (rdev->recovery_offset == MaxSector)
2476 set_bit(In_sync, &rdev->flags);
2477 /*
2478 * If no reshape in progress -> we're recovering single
2479 * disk(s) and have to set the device(s) to out-of-sync
2480 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002481 else if (!rs_is_reshaping(rs))
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002482 clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002483 }
2484
2485 /*
2486 * If a device comes back, set it as not In_sync and no longer faulty.
2487 */
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002488 if (test_and_clear_bit(Faulty, &rdev->flags)) {
2489 rdev->recovery_offset = 0;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002490 clear_bit(In_sync, &rdev->flags);
2491 rdev->saved_raid_disk = rdev->raid_disk;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002492 }
2493
Heinz Mauelshagen33e53f02016-05-19 18:49:30 +02002494 /* Reshape support -> restore repective data offsets */
2495 rdev->data_offset = le64_to_cpu(sb->data_offset);
2496 rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002497
2498 return 0;
2499}
2500
2501/*
2502 * Analyse superblocks and select the freshest.
2503 */
2504static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2505{
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002506 int r;
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01002507 struct md_rdev *rdev, *freshest;
NeilBrownfd01b882011-10-11 16:47:53 +11002508 struct mddev *mddev = &rs->md;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002509
2510 freshest = NULL;
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01002511 rdev_for_each(rdev, mddev) {
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01002512 if (test_bit(Journal, &rdev->flags))
2513 continue;
2514
Heinz Mauelshagen23397842017-11-02 19:58:28 +01002515 if (!rdev->meta_bdev)
2516 continue;
2517
2518 /* Set superblock offset/size for metadata device. */
2519 rdev->sb_start = 0;
2520 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2521 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2522 DMERR("superblock size of a logical block is no longer valid");
2523 return -EINVAL;
2524 }
2525
Jonathan Brassow761becf2012-10-11 13:42:19 +11002526 /*
Heinz Mauelshagenc76d53f2015-04-29 14:03:00 +02002527 * Skipping super_load due to CTR_FLAG_SYNC will cause
Jonathan Brassow761becf2012-10-11 13:42:19 +11002528 * the array to undergo initialization again as
Mike Snitzer43157842016-05-30 13:03:37 -04002529 * though it were new. This is the intended effect
Jonathan Brassow761becf2012-10-11 13:42:19 +11002530 * of the "sync" directive.
2531 *
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01002532 * With reshaping capability added, we must ensure that
Jilin Yuancea44662022-08-30 23:33:45 +08002533 * the "sync" directive is disallowed during the reshape.
Jonathan Brassow761becf2012-10-11 13:42:19 +11002534 */
Mike Snitzer42863252016-06-02 12:27:46 -04002535 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
Jonathan Brassow761becf2012-10-11 13:42:19 +11002536 continue;
2537
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002538 r = super_load(rdev, freshest);
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002539
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002540 switch (r) {
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002541 case 1:
2542 freshest = rdev;
2543 break;
2544 case 0:
2545 break;
2546 default:
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01002547 /* This is a failure to read the superblock from the metadata device. */
Heinz Mauelshagen9e7d9362016-08-17 15:36:44 +02002548 /*
2549 * We have to keep any raid0 data/metadata device pairs or
2550 * the MD raid0 personality will fail to start the array.
2551 */
2552 if (rs_is_raid0(rs))
2553 continue;
2554
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002555 /*
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01002556 * We keep the dm_devs to be able to emit the device tuple
2557 * properly on the table line in raid_status() (rather than
2558 * mistakenly acting as if '- -' got passed into the constructor).
2559 *
2560 * The rdev has to stay on the same_set list to allow for
2561 * the attempt to restore faulty devices on second resume.
Jonathan Brassow55ebbb52013-01-22 21:42:18 -06002562 */
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01002563 rdev->raid_disk = rdev->saved_raid_disk = -1;
2564 break;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002565 }
2566 }
2567
2568 if (!freshest)
2569 return 0;
2570
2571 /*
2572 * Validation of the freshest device provides the source of
2573 * validation for the remaining devices.
2574 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002575 rs->ti->error = "Unable to assemble array: Invalid superblocks";
2576 if (super_validate(rs, freshest))
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04002577 return -EINVAL;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002578
Heinz Mauelshagenf4af3f82017-07-13 17:36:12 +02002579 if (validate_raid_redundancy(rs)) {
2580 rs->ti->error = "Insufficient redundancy to activate array";
2581 return -EINVAL;
2582 }
2583
NeilBrowndafb20f2012-03-19 12:46:39 +11002584 rdev_for_each(rdev, mddev)
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01002585 if (!test_bit(Journal, &rdev->flags) &&
2586 rdev != freshest &&
2587 super_validate(rs, rdev))
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002588 return -EINVAL;
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002589 return 0;
2590}
2591
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002592/*
2593 * Adjust data_offset and new_data_offset on all disk members of @rs
Jiangshan Yi96fccdc2022-09-05 10:45:52 +08002594 * for out of place reshaping if requested by constructor
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002595 *
2596 * We need free space at the beginning of each raid disk for forward
2597 * and at the end for backward reshapes which userspace has to provide
2598 * via remapping/reordering of space.
2599 */
2600static int rs_adjust_data_offsets(struct raid_set *rs)
2601{
2602 sector_t data_offset = 0, new_data_offset = 0;
2603 struct md_rdev *rdev;
2604
2605 /* Constructor did not request data offset change */
2606 if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2607 if (!rs_is_reshapable(rs))
2608 goto out;
2609
2610 return 0;
2611 }
2612
Heinz Mauelshagen5380c052018-09-06 14:02:54 -04002613 /* HM FIXME: get In_Sync raid_dev? */
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002614 rdev = &rs->dev[0].rdev;
2615
2616 if (rs->delta_disks < 0) {
2617 /*
2618 * Removing disks (reshaping backwards):
2619 *
2620 * - before reshape: data is at offset 0 and free space
2621 * is at end of each component LV
2622 *
2623 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2624 */
2625 data_offset = 0;
2626 new_data_offset = rs->data_offset;
2627
2628 } else if (rs->delta_disks > 0) {
2629 /*
2630 * Adding disks (reshaping forwards):
2631 *
2632 * - before reshape: data is at offset rs->data_offset != 0 and
2633 * free space is at begin of each component LV
2634 *
2635 * - after reshape: data is at offset 0 on each component LV
2636 */
2637 data_offset = rs->data_offset;
2638 new_data_offset = 0;
2639
2640 } else {
2641 /*
2642 * User space passes in 0 for data offset after having removed reshape space
2643 *
2644 * - or - (data offset != 0)
2645 *
2646 * Changing RAID layout or chunk size -> toggle offsets
2647 *
2648 * - before reshape: data is at offset rs->data_offset 0 and
2649 * free space is at end of each component LV
2650 * -or-
2651 * data is at offset rs->data_offset != 0 and
2652 * free space is at begin of each component LV
2653 *
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02002654 * - after reshape: data is at offset 0 if it was at offset != 0
2655 * or at offset != 0 if it was at offset 0
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002656 * on each component LV
2657 *
2658 */
2659 data_offset = rs->data_offset ? rdev->data_offset : 0;
2660 new_data_offset = data_offset ? 0 : rs->data_offset;
2661 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2662 }
2663
2664 /*
2665 * Make sure we got a minimum amount of free sectors per device
2666 */
2667 if (rs->data_offset &&
Christoph Hellwig6dcbb522021-10-18 12:11:05 +02002668 bdev_nr_sectors(rdev->bdev) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002669 rs->ti->error = data_offset ? "No space for forward reshape" :
2670 "No space for backward reshape";
2671 return -ENOSPC;
2672 }
2673out:
Heinz Mauelshagen052b2b12017-12-02 01:03:49 +01002674 /*
2675 * Raise recovery_cp in case data_offset != 0 to
2676 * avoid false recovery positives in the constructor.
2677 */
2678 if (rs->md.recovery_cp < rs->md.dev_sectors)
2679 rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2680
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01002681 /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002682 rdev_for_each(rdev, &rs->md) {
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01002683 if (!test_bit(Journal, &rdev->flags)) {
2684 rdev->data_offset = data_offset;
2685 rdev->new_data_offset = new_data_offset;
2686 }
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02002687 }
2688
2689 return 0;
2690}
2691
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002692/* Userpace reordered disks -> adjust raid_disk indexes in @rs */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04002693static void __reorder_raid_disk_indexes(struct raid_set *rs)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002694{
2695 int i = 0;
2696 struct md_rdev *rdev;
2697
2698 rdev_for_each(rdev, &rs->md) {
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01002699 if (!test_bit(Journal, &rdev->flags)) {
2700 rdev->raid_disk = i++;
2701 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2702 }
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002703 }
2704}
2705
2706/*
2707 * Setup @rs for takeover by a different raid level
2708 */
2709static int rs_setup_takeover(struct raid_set *rs)
2710{
2711 struct mddev *mddev = &rs->md;
2712 struct md_rdev *rdev;
2713 unsigned int d = mddev->raid_disks = rs->raid_disks;
2714 sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2715
2716 if (rt_is_raid10(rs->raid_type)) {
Heinz Mauelshagen552aa672017-12-13 17:13:21 +01002717 if (rs_is_raid0(rs)) {
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002718 /* Userpace reordered disks -> adjust raid_disk indexes */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04002719 __reorder_raid_disk_indexes(rs);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002720
2721 /* raid0 -> raid10_far layout */
2722 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2723 rs->raid10_copies);
Heinz Mauelshagen552aa672017-12-13 17:13:21 +01002724 } else if (rs_is_raid1(rs))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002725 /* raid1 -> raid10_near layout */
2726 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2727 rs->raid_disks);
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02002728 else
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002729 return -EINVAL;
2730
2731 }
2732
2733 clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2734 mddev->recovery_cp = MaxSector;
2735
2736 while (d--) {
2737 rdev = &rs->dev[d].rdev;
2738
2739 if (test_bit(d, (void *) rs->rebuild_disks)) {
2740 clear_bit(In_sync, &rdev->flags);
2741 clear_bit(Faulty, &rdev->flags);
2742 mddev->recovery_cp = rdev->recovery_offset = 0;
2743 /* Bitmap has to be created when we do an "up" takeover */
2744 set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2745 }
2746
2747 rdev->new_data_offset = new_data_offset;
2748 }
2749
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002750 return 0;
2751}
2752
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002753/* Prepare @rs for reshape */
2754static int rs_prepare_reshape(struct raid_set *rs)
2755{
2756 bool reshape;
2757 struct mddev *mddev = &rs->md;
2758
2759 if (rs_is_raid10(rs)) {
2760 if (rs->raid_disks != mddev->raid_disks &&
2761 __is_raid10_near(mddev->layout) &&
2762 rs->raid10_copies &&
2763 rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2764 /*
2765 * raid disk have to be multiple of data copies to allow this conversion,
2766 *
2767 * This is actually not a reshape it is a
2768 * rebuild of any additional mirrors per group
2769 */
2770 if (rs->raid_disks % rs->raid10_copies) {
2771 rs->ti->error = "Can't reshape raid10 mirror groups";
2772 return -EINVAL;
2773 }
2774
2775 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2776 __reorder_raid_disk_indexes(rs);
2777 mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2778 rs->raid10_copies);
2779 mddev->new_layout = mddev->layout;
2780 reshape = false;
2781 } else
2782 reshape = true;
2783
2784 } else if (rs_is_raid456(rs))
2785 reshape = true;
2786
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002787 else if (rs_is_raid1(rs)) {
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002788 if (rs->delta_disks) {
2789 /* Process raid1 via delta_disks */
2790 mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2791 reshape = true;
2792 } else {
2793 /* Process raid1 without delta_disks */
2794 mddev->raid_disks = rs->raid_disks;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002795 reshape = false;
2796 }
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002797 } else {
2798 rs->ti->error = "Called with bogus raid type";
2799 return -EINVAL;
2800 }
2801
2802 if (reshape) {
2803 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2804 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002805 } else if (mddev->raid_disks < rs->raid_disks)
2806 /* Create new superblocks and bitmaps, if any new disks */
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002807 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02002808
2809 return 0;
2810}
2811
Heinz Mauelshagen052b2b12017-12-02 01:03:49 +01002812/* Get reshape sectors from data_offsets or raid set */
2813static sector_t _get_reshape_sectors(struct raid_set *rs)
2814{
2815 struct md_rdev *rdev;
2816 sector_t reshape_sectors = 0;
2817
2818 rdev_for_each(rdev, &rs->md)
2819 if (!test_bit(Journal, &rdev->flags)) {
2820 reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2821 rdev->data_offset - rdev->new_data_offset :
2822 rdev->new_data_offset - rdev->data_offset;
2823 break;
2824 }
2825
2826 return max(reshape_sectors, (sector_t) rs->data_offset);
2827}
2828
Jonathan Brassowb12d4372011-08-02 12:32:07 +01002829/*
Heinz Mauelshagenf99a8e42021-04-21 23:32:36 +02002830 * Reshape:
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002831 * - change raid layout
2832 * - change chunk size
2833 * - add disks
2834 * - remove disks
2835 */
2836static int rs_setup_reshape(struct raid_set *rs)
2837{
2838 int r = 0;
2839 unsigned int cur_raid_devs, d;
Heinz Mauelshagen052b2b12017-12-02 01:03:49 +01002840 sector_t reshape_sectors = _get_reshape_sectors(rs);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002841 struct mddev *mddev = &rs->md;
2842 struct md_rdev *rdev;
2843
2844 mddev->delta_disks = rs->delta_disks;
2845 cur_raid_devs = mddev->raid_disks;
2846
2847 /* Ignore impossible layout change whilst adding/removing disks */
2848 if (mddev->delta_disks &&
2849 mddev->layout != mddev->new_layout) {
2850 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2851 mddev->new_layout = mddev->layout;
2852 }
2853
2854 /*
2855 * Adjust array size:
2856 *
Heinz Mauelshagen052b2b12017-12-02 01:03:49 +01002857 * - in case of adding disk(s), array size has
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002858 * to grow after the disk adding reshape,
Yu Zheefdd3c32023-02-06 11:27:39 +08002859 * which'll happen in the event handler;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002860 * reshape will happen forward, so space has to
2861 * be available at the beginning of each disk
2862 *
Heinz Mauelshagen052b2b12017-12-02 01:03:49 +01002863 * - in case of removing disk(s), array size
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002864 * has to shrink before starting the reshape,
2865 * which'll happen here;
2866 * reshape will happen backward, so space has to
2867 * be available at the end of each disk
2868 *
2869 * - data_offset and new_data_offset are
Heinz Mauelshagenae3c6cf2016-06-15 22:27:40 +02002870 * adjusted for aforementioned out of place
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002871 * reshaping based on userspace passing in
2872 * the "data_offset <sectors>" key/value
Heinz Mauelshagenae3c6cf2016-06-15 22:27:40 +02002873 * pair via the constructor
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002874 */
2875
2876 /* Add disk(s) */
2877 if (rs->delta_disks > 0) {
2878 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2879 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2880 rdev = &rs->dev[d].rdev;
2881 clear_bit(In_sync, &rdev->flags);
2882
2883 /*
2884 * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2885 * by md, which'll store that erroneously in the superblock on reshape
2886 */
2887 rdev->saved_raid_disk = -1;
2888 rdev->raid_disk = d;
2889
2890 rdev->sectors = mddev->dev_sectors;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02002891 rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002892 }
2893
Heinz Mauelshagen052b2b12017-12-02 01:03:49 +01002894 mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002895
2896 /* Remove disk(s) */
2897 } else if (rs->delta_disks < 0) {
Heinz Mauelshagen22c992e2019-10-01 17:47:52 +02002898 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002899 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2900
2901 /* Change layout and/or chunk size */
2902 } else {
2903 /*
2904 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2905 *
2906 * keeping number of disks and do layout change ->
2907 *
2908 * toggle reshape_backward depending on data_offset:
2909 *
2910 * - free space upfront -> reshape forward
2911 *
2912 * - free space at the end -> reshape backward
2913 *
2914 *
2915 * This utilizes free reshape space avoiding the need
2916 * for userspace to move (parts of) LV segments in
2917 * case of layout/chunksize change (for disk
2918 * adding/removing reshape space has to be at
2919 * the proper address (see above with delta_disks):
2920 *
2921 * add disk(s) -> begin
2922 * remove disk(s)-> end
2923 */
2924 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2925 }
2926
Heinz Mauelshagen052b2b12017-12-02 01:03:49 +01002927 /*
2928 * Adjust device size for forward reshape
2929 * because md_finish_reshape() reduces it.
2930 */
2931 if (!mddev->reshape_backwards)
2932 rdev_for_each(rdev, &rs->md)
2933 if (!test_bit(Journal, &rdev->flags))
2934 rdev->sectors += reshape_sectors;
2935
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02002936 return r;
2937}
2938
2939/*
Heinz Mauelshagenf99a8e42021-04-21 23:32:36 +02002940 * If the md resync thread has updated superblock with max reshape position
2941 * at the end of a reshape but not (yet) reset the layout configuration
2942 * changes -> reset the latter.
2943 */
2944static void rs_reset_inconclusive_reshape(struct raid_set *rs)
2945{
2946 if (!rs_is_reshaping(rs) && rs_is_layout_change(rs, true)) {
2947 rs_set_cur(rs);
2948 rs->md.delta_disks = 0;
2949 rs->md.reshape_backwards = 0;
2950 }
2951}
2952
2953/*
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002954 * Enable/disable discard support on RAID set depending on
2955 * RAID level and discard properties of underlying RAID members.
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002956 */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002957static void configure_discard_support(struct raid_set *rs)
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002958{
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002959 int i;
2960 bool raid456;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02002961 struct dm_target *ti = rs->ti;
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002962
Christoph Hellwig48920ff2017-04-05 19:21:23 +02002963 /*
2964 * XXX: RAID level 4,5,6 require zeroing for safety.
2965 */
Heinz Mauelshagen552aa672017-12-13 17:13:21 +01002966 raid456 = rs_is_raid456(rs);
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002967
Heinz Mauelshagenffeeac72016-06-30 14:37:50 +02002968 for (i = 0; i < rs->raid_disks; i++) {
Christoph Hellwig70200572022-04-15 06:52:55 +02002969 if (!rs->dev[i].rdev.bdev ||
2970 !bdev_max_discard_sectors(rs->dev[i].rdev.bdev))
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002971 return;
2972
2973 if (raid456) {
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02002974 if (!devices_handle_discard_safely) {
2975 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2976 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2977 return;
2978 }
2979 }
2980 }
2981
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02002982 ti->num_discard_bios = 1;
2983}
2984
2985/*
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002986 * Construct a RAID0/1/10/4/5/6 mapping:
NeilBrown9d09e662011-01-13 20:00:02 +00002987 * Args:
Mike Snitzer43157842016-05-30 13:03:37 -04002988 * <raid_type> <#raid_params> <raid_params>{0,} \
2989 * <#raid_devs> [<meta_dev1> <dev1>]{1,}
NeilBrown9d09e662011-01-13 20:00:02 +00002990 *
Mike Snitzer43157842016-05-30 13:03:37 -04002991 * <raid_params> varies by <raid_type>. See 'parse_raid_params' for
NeilBrown9d09e662011-01-13 20:00:02 +00002992 * details on possible <raid_params>.
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02002993 *
2994 * Userspace is free to initialize the metadata devices, hence the superblocks to
2995 * enforce recreation based on the passed in table parameters.
2996 *
NeilBrown9d09e662011-01-13 20:00:02 +00002997 */
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02002998static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
NeilBrown9d09e662011-01-13 20:00:02 +00002999{
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003000 int r;
Heinz Mauelshagen75015372017-12-02 01:03:50 +01003001 bool resize = false;
NeilBrown9d09e662011-01-13 20:00:02 +00003002 struct raid_type *rt;
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02003003 unsigned int num_raid_params, num_raid_devs;
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003004 sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
NeilBrown9d09e662011-01-13 20:00:02 +00003005 struct raid_set *rs = NULL;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003006 const char *arg;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003007 struct rs_layout rs_layout;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003008 struct dm_arg_set as = { argc, argv }, as_nrd;
3009 struct dm_arg _args[] = {
3010 { 0, as.argc, "Cannot understand number of raid parameters" },
3011 { 1, 254, "Cannot understand number of raid devices parameters" }
3012 };
NeilBrown9d09e662011-01-13 20:00:02 +00003013
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003014 arg = dm_shift_arg(&as);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04003015 if (!arg) {
3016 ti->error = "No arguments";
3017 return -EINVAL;
3018 }
NeilBrown9d09e662011-01-13 20:00:02 +00003019
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003020 rt = get_raid_type(arg);
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04003021 if (!rt) {
3022 ti->error = "Unrecognised raid_type";
3023 return -EINVAL;
3024 }
NeilBrown9d09e662011-01-13 20:00:02 +00003025
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003026 /* Must have <#raid_params> */
3027 if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
Mike Snitzer43157842016-05-30 13:03:37 -04003028 return -EINVAL;
NeilBrown9d09e662011-01-13 20:00:02 +00003029
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003030 /* number of raid device tupples <meta_dev data_dev> */
3031 as_nrd = as;
3032 dm_consume_args(&as_nrd, num_raid_params);
3033 _args[1].max = (as_nrd.argc - 1) / 2;
3034 if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
Mike Snitzer43157842016-05-30 13:03:37 -04003035 return -EINVAL;
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003036
Mike Snitzerbb91a632016-06-02 12:06:54 -04003037 if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
Mike Snitzerbd83a4c2016-05-31 14:26:52 -04003038 ti->error = "Invalid number of supplied raid devices";
3039 return -EINVAL;
3040 }
NeilBrown9d09e662011-01-13 20:00:02 +00003041
Mike Snitzerbfcee0e2016-06-02 15:08:09 -04003042 rs = raid_set_alloc(ti, rt, num_raid_devs);
NeilBrown9d09e662011-01-13 20:00:02 +00003043 if (IS_ERR(rs))
3044 return PTR_ERR(rs);
3045
Heinz Mauelshagen92c83d72016-05-19 18:49:25 +02003046 r = parse_raid_params(rs, &as, num_raid_params);
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003047 if (r)
NeilBrown9d09e662011-01-13 20:00:02 +00003048 goto bad;
3049
Heinz Mauelshagen702108d2016-05-19 18:49:26 +02003050 r = parse_dev_params(rs, &as);
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003051 if (r)
NeilBrown9d09e662011-01-13 20:00:02 +00003052 goto bad;
3053
Jonathan Brassowb12d4372011-08-02 12:32:07 +01003054 rs->md.sync_super = super_sync;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003055
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02003056 /*
3057 * Calculate ctr requested array and device sizes to allow
3058 * for superblock analysis needing device sizes defined.
3059 *
3060 * Any existing superblock will overwrite the array and device sizes
3061 */
Heinz Mauelshagen22c992e2019-10-01 17:47:52 +02003062 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02003063 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003064 goto bad;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02003065
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003066 /* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3067 rs->array_sectors = rs->md.array_sectors;
3068 rs->dev_sectors = rs->md.dev_sectors;
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02003069
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003070 /*
3071 * Backup any new raid set level, layout, ...
3072 * requested to be able to compare to superblock
3073 * members for conversion decisions.
3074 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003075 rs_config_backup(rs, &rs_layout);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003076
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003077 r = analyse_superblocks(ti, rs);
3078 if (r)
Jonathan Brassowb12d4372011-08-02 12:32:07 +01003079 goto bad;
3080
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003081 /* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3082 sb_array_sectors = rs->md.array_sectors;
Heinz Mauelshagen4d49f1b2017-06-30 15:45:58 +02003083 rdev_sectors = __rdev_sectors(rs);
3084 if (!rdev_sectors) {
3085 ti->error = "Invalid rdev size";
3086 r = -EINVAL;
3087 goto bad;
3088 }
3089
Heinz Mauelshagen75015372017-12-02 01:03:50 +01003090
3091 reshape_sectors = _get_reshape_sectors(rs);
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003092 if (rs->dev_sectors != rdev_sectors) {
3093 resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3094 if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3095 set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3096 }
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02003097
NeilBrown9d09e662011-01-13 20:00:02 +00003098 INIT_WORK(&rs->md.event_work, do_table_event);
NeilBrown9d09e662011-01-13 20:00:02 +00003099 ti->private = rs;
Alasdair G Kergon55a62ee2013-03-01 22:45:47 +00003100 ti->num_flush_bios = 1;
Mike Snitzer9dd1cd32022-07-20 13:58:04 -04003101 ti->needs_bio_set_dev = true;
NeilBrown9d09e662011-01-13 20:00:02 +00003102
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003103 /* Restore any requested new layout for conversion decision */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003104 rs_config_restore(rs, &rs_layout);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003105
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02003106 /*
3107 * Now that we have any superblock metadata available,
3108 * check for new, recovering, reshaping, to be taken over,
3109 * to be reshaped or an existing, unchanged raid set to
3110 * run in sequence.
3111 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003112 if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02003113 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3114 if (rs_is_raid6(rs) &&
3115 test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3116 ti->error = "'nosync' not allowed for new raid6 set";
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003117 r = -EINVAL;
3118 goto bad;
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02003119 }
3120 rs_setup_recovery(rs, 0);
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02003121 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3122 rs_set_new(rs);
3123 } else if (rs_is_recovering(rs)) {
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02003124 /* A recovering raid set may be resized */
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003125 goto size_check;
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02003126 } else if (rs_is_reshaping(rs)) {
3127 /* Have to reject size change request during reshape */
3128 if (resize) {
3129 ti->error = "Can't resize a reshaping raid set";
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003130 r = -EPERM;
3131 goto bad;
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02003132 }
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02003133 /* skip setup rs */
Heinz Mauelshagen2d92a3c2016-06-24 00:32:58 +02003134 } else if (rs_takeover_requested(rs)) {
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003135 if (rs_is_reshaping(rs)) {
3136 ti->error = "Can't takeover a reshaping raid set";
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003137 r = -EPERM;
3138 goto bad;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003139 }
3140
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003141 /* We can't takeover a journaled raid4/5/6 */
3142 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3143 ti->error = "Can't takeover a journaled raid4/5/6 set";
3144 r = -EPERM;
3145 goto bad;
3146 }
3147
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003148 /*
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02003149 * If a takeover is needed, userspace sets any additional
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02003150 * devices to rebuild and we can check for a valid request here.
3151 *
Yu Zheefdd3c32023-02-06 11:27:39 +08003152 * If acceptable, set the level to the new requested
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02003153 * one, prohibit requesting recovery, allow the raid
3154 * set to run and store superblocks during resume.
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003155 */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003156 r = rs_check_takeover(rs);
3157 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003158 goto bad;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003159
3160 r = rs_setup_takeover(rs);
3161 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003162 goto bad;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003163
Mike Snitzer42863252016-06-02 12:27:46 -04003164 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
Heinz Mauelshagen469b3042016-06-29 18:13:58 +02003165 /* Takeover ain't recovery, so disable recovery */
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02003166 rs_setup_recovery(rs, MaxSector);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003167 rs_set_new(rs);
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02003168 } else if (rs_reshape_requested(rs)) {
Heinz Mauelshagenf9f3ee92019-10-01 17:47:54 +02003169 /* Only request grow on raid set size extensions, not on reshapes. */
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003170 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3171
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003172 /*
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003173 * No need to check for 'ongoing' takeover here, because takeover
3174 * is an instant operation as oposed to an ongoing reshape.
3175 */
3176
3177 /* We can't reshape a journaled raid4/5/6 */
3178 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3179 ti->error = "Can't reshape a journaled raid4/5/6 set";
3180 r = -EPERM;
3181 goto bad;
3182 }
3183
Heinz Mauelshagen11e47232017-12-13 17:13:18 +01003184 /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3185 if (reshape_sectors || rs_is_raid1(rs)) {
3186 /*
Heinz Mauelshagena4a82ce2023-01-26 15:48:30 +01003187 * We can only prepare for a reshape here, because the
3188 * raid set needs to run to provide the repective reshape
3189 * check functions via its MD personality instance.
3190 *
3191 * So do the reshape check after md_run() succeeded.
3192 */
Heinz Mauelshagen11e47232017-12-13 17:13:18 +01003193 r = rs_prepare_reshape(rs);
3194 if (r)
Wenwen Wangdc1a3e82019-08-18 19:18:34 -05003195 goto bad;
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003196
Heinz Mauelshagen11e47232017-12-13 17:13:18 +01003197 /* Reshaping ain't recovery, so disable recovery */
3198 rs_setup_recovery(rs, MaxSector);
3199 }
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003200 rs_set_cur(rs);
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02003201 } else {
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003202size_check:
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02003203 /* May not set recovery when a device rebuild is requested */
Heinz Mauelshagen37f10be2016-06-24 23:21:37 +02003204 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003205 clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
Heinz Mauelshagen37f10be2016-06-24 23:21:37 +02003206 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
Heinz Mauelshagenf9f3ee92019-10-01 17:47:54 +02003207 rs_setup_recovery(rs, MaxSector);
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003208 } else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3209 /*
Heinz Mauelshagenf9f3ee92019-10-01 17:47:54 +02003210 * Set raid set to current size, i.e. size as of
3211 * superblocks to grow to larger size in preresume.
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003212 */
3213 r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3214 if (r)
3215 goto bad;
3216
3217 rs_setup_recovery(rs, rs->md.recovery_cp < rs->md.dev_sectors ? rs->md.recovery_cp : rs->md.dev_sectors);
3218 } else {
3219 /* This is no size change or it is shrinking, update size and record in superblocks */
3220 r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3221 if (r)
3222 goto bad;
3223
3224 if (sb_array_sectors > rs->array_sectors)
3225 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3226 }
Heinz Mauelshagen2a5556c2016-06-27 14:44:09 +02003227 rs_set_cur(rs);
Heinz Mauelshagen4dff2f12016-06-24 00:21:09 +02003228 }
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003229
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02003230 /* If constructor requested it, change data and new_data offsets */
3231 r = rs_adjust_data_offsets(rs);
3232 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003233 goto bad;
Heinz Mauelshagen40ba37e2016-06-13 17:55:13 +02003234
Heinz Mauelshagenf99a8e42021-04-21 23:32:36 +02003235 /* Catch any inconclusive reshape superblock content. */
3236 rs_reset_inconclusive_reshape(rs);
3237
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003238 /* Start raid set read-only and assumed clean to change in raid_resume() */
3239 rs->md.ro = 1;
3240 rs->md.in_sync = 1;
Heinz Mauelshagen5380c052018-09-06 14:02:54 -04003241
Heinz Mauelshagenf99a8e42021-04-21 23:32:36 +02003242 /* Keep array frozen until resume. */
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003243 set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
Heinz Mauelshagen75b8e042014-09-24 17:47:18 +02003244
Heinz Mauelshagen0cf45032015-04-29 14:03:04 +02003245 /* Has to be held on running the array */
3246 mddev_lock_nointr(&rs->md);
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003247 r = md_run(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00003248 rs->md.in_sync = 0; /* Assume already marked dirty */
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003249 if (r) {
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003250 ti->error = "Failed to run raid array";
3251 mddev_unlock(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00003252 goto bad;
3253 }
3254
Song Liud5d885f2017-11-19 22:17:01 -08003255 r = md_start(&rs->md);
Song Liud5d885f2017-11-19 22:17:01 -08003256 if (r) {
3257 ti->error = "Failed to start raid array";
3258 mddev_unlock(&rs->md);
3259 goto bad_md_start;
3260 }
3261
Heinz Mauelshagen5380c052018-09-06 14:02:54 -04003262 /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01003263 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3264 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3265 if (r) {
3266 ti->error = "Failed to set raid4/5/6 journal mode";
3267 mddev_unlock(&rs->md);
3268 goto bad_journal_mode_set;
3269 }
3270 }
3271
Jonathan Brassow32737272011-08-02 12:32:07 +01003272 mddev_suspend(&rs->md);
Heinz Mauelshagen0cf352e2017-07-13 17:34:24 +02003273 set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003274
3275 /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3276 if (rs_is_raid456(rs)) {
3277 r = rs_set_raid456_stripe_cache(rs);
3278 if (r)
3279 goto bad_stripe_cache;
3280 }
3281
3282 /* Now do an early reshape check */
3283 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3284 r = rs_check_reshape(rs);
3285 if (r)
Heinz Mauelshagenb1956dc2016-06-24 21:49:26 +02003286 goto bad_check_reshape;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003287
3288 /* Restore new, ctr requested layout to perform check */
3289 rs_config_restore(rs, &rs_layout);
3290
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003291 if (rs->md.pers->start_reshape) {
3292 r = rs->md.pers->check_reshape(&rs->md);
3293 if (r) {
3294 ti->error = "Reshape check failed";
3295 goto bad_check_reshape;
3296 }
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003297 }
3298 }
3299
Heinz Mauelshagen11e29682016-11-29 22:37:30 +01003300 /* Disable/enable discard support on raid set. */
3301 configure_discard_support(rs);
3302
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003303 mddev_unlock(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00003304 return 0;
3305
Song Liud5d885f2017-11-19 22:17:01 -08003306bad_md_start:
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01003307bad_journal_mode_set:
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003308bad_stripe_cache:
3309bad_check_reshape:
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05003310 md_stop(&rs->md);
NeilBrown9d09e662011-01-13 20:00:02 +00003311bad:
Mike Snitzerbfcee0e2016-06-02 15:08:09 -04003312 raid_set_free(rs);
NeilBrown9d09e662011-01-13 20:00:02 +00003313
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003314 return r;
NeilBrown9d09e662011-01-13 20:00:02 +00003315}
3316
3317static void raid_dtr(struct dm_target *ti)
3318{
3319 struct raid_set *rs = ti->private;
3320
NeilBrown9d09e662011-01-13 20:00:02 +00003321 md_stop(&rs->md);
Mike Snitzerbfcee0e2016-06-02 15:08:09 -04003322 raid_set_free(rs);
NeilBrown9d09e662011-01-13 20:00:02 +00003323}
3324
Mikulas Patocka7de3ee52012-12-21 20:23:41 +00003325static int raid_map(struct dm_target *ti, struct bio *bio)
NeilBrown9d09e662011-01-13 20:00:02 +00003326{
3327 struct raid_set *rs = ti->private;
NeilBrownfd01b882011-10-11 16:47:53 +11003328 struct mddev *mddev = &rs->md;
NeilBrown9d09e662011-01-13 20:00:02 +00003329
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003330 /*
3331 * If we're reshaping to add disk(s)), ti->len and
3332 * mddev->array_sectors will differ during the process
3333 * (ti->len > mddev->array_sectors), so we have to requeue
3334 * bios with addresses > mddev->array_sectors here or
Heinz Mauelshagen2527b562016-06-24 01:03:19 +02003335 * there will occur accesses past EOD of the component
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003336 * data images thus erroring the raid set.
3337 */
3338 if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3339 return DM_MAPIO_REQUEUE;
3340
Shaohua Lic4d6a1b2017-09-21 10:29:22 -07003341 md_handle_request(mddev, bio);
NeilBrown9d09e662011-01-13 20:00:02 +00003342
3343 return DM_MAPIO_SUBMITTED;
3344}
3345
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003346/* Return sync state string for @state */
3347enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3348static const char *sync_str(enum sync_state state)
3349{
3350 /* Has to be in above sync_state order! */
3351 static const char *sync_strs[] = {
3352 "frozen",
3353 "reshape",
3354 "resync",
3355 "check",
3356 "repair",
3357 "recover",
3358 "idle"
3359 };
3360
3361 return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3362};
3363
3364/* Return enum sync_state for @mddev derived from @recovery flags */
Geert Uytterhoeven0328ba92018-09-17 10:02:34 +02003365static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003366{
Heinz Mauelshagen67143512017-12-02 01:03:53 +01003367 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003368 return st_frozen;
Jonathan Brassowbe836512013-04-24 11:42:43 +10003369
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003370 /* The MD sync thread can be done with io or be interrupted but still be running */
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003371 if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3372 (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3373 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
Heinz Mauelshagen67143512017-12-02 01:03:53 +01003374 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003375 return st_reshape;
Jonathan Brassowbe836512013-04-24 11:42:43 +10003376
Heinz Mauelshagen67143512017-12-02 01:03:53 +01003377 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3378 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003379 return st_resync;
3380 if (test_bit(MD_RECOVERY_CHECK, &recovery))
3381 return st_check;
3382 return st_repair;
Jonathan Brassowbe836512013-04-24 11:42:43 +10003383 }
3384
Heinz Mauelshagen67143512017-12-02 01:03:53 +01003385 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003386 return st_recover;
3387
3388 if (mddev->reshape_position != MaxSector)
3389 return st_reshape;
Jonathan Brassowbe836512013-04-24 11:42:43 +10003390 }
3391
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003392 return st_idle;
Jonathan Brassowbe836512013-04-24 11:42:43 +10003393}
3394
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003395/*
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003396 * Return status string for @rdev
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003397 *
3398 * Status characters:
3399 *
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003400 * 'D' = Dead/Failed raid set component or raid4/5/6 journal device
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01003401 * 'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3402 * 'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01003403 * '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003404 */
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +01003405static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003406{
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01003407 if (!rdev->bdev)
3408 return "-";
3409 else if (test_bit(Faulty, &rdev->flags))
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003410 return "D";
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003411 else if (test_bit(Journal, &rdev->flags))
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01003412 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003413 else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3414 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3415 !test_bit(In_sync, &rdev->flags)))
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003416 return "a";
3417 else
3418 return "A";
3419}
3420
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +01003421/* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
Heinz Mauelshagen67143512017-12-02 01:03:53 +01003422static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003423 enum sync_state state, sector_t resync_max_sectors)
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003424{
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003425 sector_t r;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003426 struct mddev *mddev = &rs->md;
3427
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +01003428 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003429 clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003430
3431 if (rs_is_raid0(rs)) {
3432 r = resync_max_sectors;
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +01003433 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003434
3435 } else {
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003436 if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3437 r = mddev->recovery_cp;
3438 else
3439 r = mddev->curr_resync_completed;
3440
3441 if (state == st_idle && r >= resync_max_sectors) {
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003442 /*
3443 * Sync complete.
3444 */
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003445 /* In case we have finished recovering, the array is in sync. */
3446 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3447 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3448
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003449 } else if (state == st_recover)
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003450 /*
3451 * In case we are recovering, the array is not in sync
3452 * and health chars should show the recovering legs.
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003453 *
3454 * Already retrieved recovery offset from curr_resync_completed above.
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003455 */
3456 ;
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003457
3458 else if (state == st_resync || state == st_reshape)
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003459 /*
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003460 * If "resync/reshape" is occurring, the raid set
Heinz Mauelshagendc15b9432017-12-13 17:13:19 +01003461 * is or may be out of sync hence the health
3462 * characters shall be 'a'.
3463 */
3464 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3465
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003466 else if (state == st_check || state == st_repair)
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003467 /*
3468 * If "check" or "repair" is occurring, the raid set has
3469 * undergone an initial sync and the health characters
3470 * should not be 'a' anymore.
3471 */
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +01003472 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003473
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003474 else if (test_bit(MD_RECOVERY_NEEDED, &recovery))
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003475 /*
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003476 * We are idle and recovery is needed, prevent 'A' chars race
Heinz Mauelshagen36a240a2018-09-06 22:54:29 +02003477 * caused by components still set to in-sync by constructor.
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003478 */
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003479 set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003480
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003481 else {
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003482 /*
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003483 * We are idle and the raid set may be doing an initial
3484 * sync, or it may be rebuilding individual components.
3485 * If all the devices are In_sync, then it is the raid set
3486 * that is being initialized.
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003487 */
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003488 struct md_rdev *rdev;
3489
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003490 set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003491 rdev_for_each(rdev, mddev)
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003492 if (!test_bit(Journal, &rdev->flags) &&
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003493 !test_bit(In_sync, &rdev->flags)) {
3494 clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3495 break;
3496 }
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003497 }
3498 }
3499
Heinz Mauelshagen4102d9d2017-12-02 01:03:55 +01003500 return min(r, resync_max_sectors);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003501}
3502
3503/* Helper to return @dev name or "-" if !@dev */
Mike Snitzere6ca5e12016-06-02 15:27:22 -04003504static const char *__get_dev_name(struct dm_dev *dev)
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003505{
3506 return dev ? dev->name : "-";
3507}
3508
Mikulas Patockafd7c092e2013-03-01 22:45:44 +00003509static void raid_status(struct dm_target *ti, status_type_t type,
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003510 unsigned int status_flags, char *result, unsigned int maxlen)
NeilBrown9d09e662011-01-13 20:00:02 +00003511{
3512 struct raid_set *rs = ti->private;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003513 struct mddev *mddev = &rs->md;
Mikulas Patocka1fbeea212022-07-24 14:31:35 -04003514 struct r5conf *conf = rs_is_raid456(rs) ? mddev->private : NULL;
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003515 int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
Heinz Mauelshagen67143512017-12-02 01:03:53 +01003516 unsigned long recovery;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003517 unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3518 unsigned int sz = 0;
Heinz Mauelshagen43f39522019-12-19 17:58:46 +01003519 unsigned int rebuild_writemostly_count = 0;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003520 sector_t progress, resync_max_sectors, resync_mismatches;
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003521 enum sync_state state;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003522 struct raid_type *rt;
NeilBrown9d09e662011-01-13 20:00:02 +00003523
3524 switch (type) {
3525 case STATUSTYPE_INFO:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003526 /* *Should* always succeed */
3527 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3528 if (!rt)
3529 return;
NeilBrown9d09e662011-01-13 20:00:02 +00003530
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003531 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
NeilBrown9d09e662011-01-13 20:00:02 +00003532
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003533 /* Access most recent mddev properties for status output */
3534 smp_rmb();
3535 /* Get sensible max sectors even if raid set not yet started */
Mike Snitzer42863252016-06-02 12:27:46 -04003536 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003537 mddev->resync_max_sectors : mddev->dev_sectors;
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003538 recovery = rs->md.recovery;
3539 state = decipher_sync_action(mddev, recovery);
3540 progress = rs_get_progress(rs, recovery, state, resync_max_sectors);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003541 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003542 atomic64_read(&mddev->resync_mismatches) : 0;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003543
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01003544 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3545 for (i = 0; i < rs->raid_disks; i++)
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +01003546 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
Jonathan Brassowbe836512013-04-24 11:42:43 +10003547
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003548 /*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003549 * In-sync/Reshape ratio:
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003550 * The in-sync ratio shows the progress of:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003551 * - Initializing the raid set
3552 * - Rebuilding a subset of devices of the raid set
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003553 * The user can distinguish between the two by referring
3554 * to the status characters.
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003555 *
3556 * The reshape ratio shows the progress of
3557 * changing the raid layout or the number of
3558 * disks of a raid set
Jonathan E Brassow2e727c32011-10-31 20:21:26 +00003559 */
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003560 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3561 (unsigned long long) resync_max_sectors);
NeilBrown9d09e662011-01-13 20:00:02 +00003562
Jonathan Brassowbe836512013-04-24 11:42:43 +10003563 /*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003564 * v1.5.0+:
3565 *
Jonathan Brassowbe836512013-04-24 11:42:43 +10003566 * Sync action:
Mauro Carvalho Chehab6cf2a732019-06-18 12:40:23 -03003567 * See Documentation/admin-guide/device-mapper/dm-raid.rst for
Jonathan Brassowbe836512013-04-24 11:42:43 +10003568 * information on each of these states.
3569 */
Heinz Mauelshagen53be73a2019-10-01 17:47:55 +02003570 DMEMIT(" %s", sync_str(state));
Jonathan Brassowbe836512013-04-24 11:42:43 +10003571
3572 /*
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003573 * v1.5.0+:
3574 *
Jonathan Brassowbe836512013-04-24 11:42:43 +10003575 * resync_mismatches/mismatch_cnt
3576 * This field shows the number of discrepancies found when
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003577 * performing a "check" of the raid set.
Jonathan Brassowbe836512013-04-24 11:42:43 +10003578 */
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003579 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3580
3581 /*
Mike Snitzer9b6e5422016-06-02 11:48:09 -04003582 * v1.9.0+:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003583 *
3584 * data_offset (needed for out of space reshaping)
3585 * This field shows the data offset into the data
3586 * image LV where the first stripes data starts.
3587 *
3588 * We keep data_offset equal on all raid disks of the set,
3589 * so retrieving it from the first raid disk is sufficient.
3590 */
3591 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003592
3593 /*
3594 * v1.10.0+:
3595 */
3596 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
Heinz Mauelshagen242ea5a2017-12-02 01:03:54 +01003597 __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
NeilBrown9d09e662011-01-13 20:00:02 +00003598 break;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003599
NeilBrown9d09e662011-01-13 20:00:02 +00003600 case STATUSTYPE_TABLE:
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003601 /* Report the table line string you would use to construct this raid set */
NeilBrown9d09e662011-01-13 20:00:02 +00003602
Heinz Mauelshagen43f39522019-12-19 17:58:46 +01003603 /*
3604 * Count any rebuild or writemostly argument pairs and subtract the
3605 * hweight count being added below of any rebuild and writemostly ctr flags.
3606 */
3607 for (i = 0; i < rs->raid_disks; i++) {
3608 rebuild_writemostly_count += (test_bit(i, (void *) rs->rebuild_disks) ? 2 : 0) +
3609 (test_bit(WriteMostly, &rs->dev[i].rdev.flags) ? 2 : 0);
3610 }
3611 rebuild_writemostly_count -= (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) ? 2 : 0) +
3612 (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags) ? 2 : 0);
3613 /* Calculate raid parameter count based on ^ rebuild/writemostly argument counts and ctr flags set. */
3614 raid_param_cnt += rebuild_writemostly_count +
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003615 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
Heinz Mauelshagen43f39522019-12-19 17:58:46 +01003616 hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003617 /* Emit table line */
Heinz Mauelshagen4464e362017-03-18 01:39:12 +01003618 /* This has to be in the documented order for userspace! */
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003619 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
Mike Snitzer42863252016-06-02 12:27:46 -04003620 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003621 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
Heinz Mauelshagen4464e362017-03-18 01:39:12 +01003622 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3623 DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
Heinz Mauelshagen43f39522019-12-19 17:58:46 +01003624 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags))
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003625 for (i = 0; i < rs->raid_disks; i++)
Heinz Mauelshagen43f39522019-12-19 17:58:46 +01003626 if (test_bit(i, (void *) rs->rebuild_disks))
3627 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD), i);
Heinz Mauelshagen4464e362017-03-18 01:39:12 +01003628 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3629 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3630 mddev->bitmap_info.daemon_sleep);
3631 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3632 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3633 mddev->sync_speed_min);
3634 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3635 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3636 mddev->sync_speed_max);
Heinz Mauelshagen43f39522019-12-19 17:58:46 +01003637 if (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags))
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003638 for (i = 0; i < rs->raid_disks; i++)
3639 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3640 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3641 rs->dev[i].rdev.raid_disk);
Mike Snitzer42863252016-06-02 12:27:46 -04003642 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
Mike Snitzer3fa6cf32016-06-02 11:58:51 -04003643 DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003644 mddev->bitmap_info.max_write_behind);
Heinz Mauelshagen4464e362017-03-18 01:39:12 +01003645 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3646 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3647 max_nr_stripes);
3648 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3649 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3650 (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3651 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3652 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3653 raid10_md_layout_to_copies(mddev->layout));
3654 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3655 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3656 raid10_md_layout_to_format(mddev->layout));
3657 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3658 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3659 max(rs->delta_disks, mddev->delta_disks));
3660 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3661 DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3662 (unsigned long long) rs->data_offset);
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003663 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3664 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3665 __get_dev_name(rs->journal_dev.dev));
Heinz Mauelshagen6e536362017-03-22 17:44:38 +01003666 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3667 DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3668 md_journal_mode_to_dm_raid(rs->journal_dev.mode));
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003669 DMEMIT(" %d", rs->raid_disks);
Heinz Mauelshagen7a7c3302016-06-30 13:57:08 +02003670 for (i = 0; i < rs->raid_disks; i++)
3671 DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3672 __get_dev_name(rs->dev[i].data_dev));
Tushar Sugandhi8ec45662021-07-12 17:49:03 -07003673 break;
3674
3675 case STATUSTYPE_IMA:
3676 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3677 if (!rt)
3678 return;
3679
3680 DMEMIT_TARGET_NAME_VERSION(ti->type);
3681 DMEMIT(",raid_type=%s,raid_disks=%d", rt->name, mddev->raid_disks);
3682
3683 /* Access most recent mddev properties for status output */
3684 smp_rmb();
3685 recovery = rs->md.recovery;
3686 state = decipher_sync_action(mddev, recovery);
3687 DMEMIT(",raid_state=%s", sync_str(state));
3688
3689 for (i = 0; i < rs->raid_disks; i++) {
3690 DMEMIT(",raid_device_%d_status=", i);
3691 DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3692 }
3693
3694 if (rt_is_raid456(rt)) {
3695 DMEMIT(",journal_dev_mode=");
3696 switch (rs->journal_dev.mode) {
3697 case R5C_JOURNAL_MODE_WRITE_THROUGH:
3698 DMEMIT("%s",
3699 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_THROUGH].param);
3700 break;
3701 case R5C_JOURNAL_MODE_WRITE_BACK:
3702 DMEMIT("%s",
3703 _raid456_journal_mode[R5C_JOURNAL_MODE_WRITE_BACK].param);
3704 break;
3705 default:
3706 DMEMIT("invalid");
3707 break;
3708 }
3709 }
3710 DMEMIT(";");
3711 break;
NeilBrown9d09e662011-01-13 20:00:02 +00003712 }
NeilBrown9d09e662011-01-13 20:00:02 +00003713}
3714
Mike Snitzer1eb5fa82018-02-28 15:59:59 -05003715static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
Heinz Mauelshagen86a32382023-01-25 21:14:58 +01003716 char *result, unsigned int maxlen)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003717{
3718 struct raid_set *rs = ti->private;
3719 struct mddev *mddev = &rs->md;
3720
Jonathan Brassowbe836512013-04-24 11:42:43 +10003721 if (!mddev->pers || !mddev->pers->sync_request)
3722 return -EINVAL;
3723
3724 if (!strcasecmp(argv[0], "frozen"))
3725 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3726 else
3727 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3728
3729 if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3730 if (mddev->sync_thread) {
3731 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Guoqing Jiang9dfbdaf2022-06-21 11:11:29 +08003732 md_unregister_thread(&mddev->sync_thread);
Guoqing Jiangd0a18032022-06-07 10:03:56 +08003733 md_reap_sync_thread(mddev);
Jonathan Brassowbe836512013-04-24 11:42:43 +10003734 }
Heinz Mauelshagen74694bc2018-12-18 17:35:41 +01003735 } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003736 return -EBUSY;
3737 else if (!strcasecmp(argv[0], "resync"))
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003738 ; /* MD_RECOVERY_NEEDED set below */
3739 else if (!strcasecmp(argv[0], "recover"))
Jonathan Brassowbe836512013-04-24 11:42:43 +10003740 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003741 else {
Mike Snitzerad470472017-02-28 11:55:16 -05003742 if (!strcasecmp(argv[0], "check")) {
Jonathan Brassowbe836512013-04-24 11:42:43 +10003743 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
Mike Snitzerad470472017-02-28 11:55:16 -05003744 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3745 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3746 } else if (!strcasecmp(argv[0], "repair")) {
Mike Snitzer105db592017-01-06 15:38:08 -05003747 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3748 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3749 } else
Jonathan Brassowbe836512013-04-24 11:42:43 +10003750 return -EINVAL;
Jonathan Brassowbe836512013-04-24 11:42:43 +10003751 }
3752 if (mddev->ro == 2) {
3753 /* A write to sync_action is enough to justify
3754 * canceling read-auto mode
3755 */
3756 mddev->ro = 0;
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003757 if (!mddev->suspended && mddev->sync_thread)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003758 md_wakeup_thread(mddev->sync_thread);
3759 }
3760 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02003761 if (!mddev->suspended && mddev->thread)
Jonathan Brassowbe836512013-04-24 11:42:43 +10003762 md_wakeup_thread(mddev->thread);
3763
3764 return 0;
3765}
3766
3767static int raid_iterate_devices(struct dm_target *ti,
3768 iterate_devices_callout_fn fn, void *data)
NeilBrown9d09e662011-01-13 20:00:02 +00003769{
3770 struct raid_set *rs = ti->private;
Heinz Mauelshagen094f3942016-07-19 14:03:51 +02003771 unsigned int i;
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003772 int r = 0;
NeilBrown9d09e662011-01-13 20:00:02 +00003773
Heinz Mauelshagen332bd072022-06-28 00:37:22 +02003774 for (i = 0; !r && i < rs->raid_disks; i++) {
3775 if (rs->dev[i].data_dev) {
3776 r = fn(ti, rs->dev[i].data_dev,
3777 0, /* No offset on data devs */
3778 rs->md.dev_sectors, data);
3779 }
3780 }
NeilBrown9d09e662011-01-13 20:00:02 +00003781
Heinz Mauelshagen73c6f232016-05-19 18:49:24 +02003782 return r;
NeilBrown9d09e662011-01-13 20:00:02 +00003783}
3784
3785static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3786{
3787 struct raid_set *rs = ti->private;
Ming Leic8156fc2019-09-11 19:31:33 +08003788 unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
NeilBrown9d09e662011-01-13 20:00:02 +00003789
Ming Leic8156fc2019-09-11 19:31:33 +08003790 blk_limits_io_min(limits, chunk_size_bytes);
3791 blk_limits_io_opt(limits, chunk_size_bytes * mddev_data_stripes(rs));
NeilBrown9d09e662011-01-13 20:00:02 +00003792}
3793
NeilBrown9d09e662011-01-13 20:00:02 +00003794static void raid_postsuspend(struct dm_target *ti)
3795{
3796 struct raid_set *rs = ti->private;
3797
NeilBrown4d5324f2017-10-19 12:17:16 +11003798 if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
Heinz Mauelshagen1af20482017-12-02 01:03:48 +01003799 /* Writes have to be stopped before suspending to avoid deadlocks. */
3800 if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3801 md_stop_writes(&rs->md);
3802
NeilBrown4d5324f2017-10-19 12:17:16 +11003803 mddev_lock_nointr(&rs->md);
Heinz Mauelshagen31e10a42016-08-10 02:45:59 +02003804 mddev_suspend(&rs->md);
NeilBrown4d5324f2017-10-19 12:17:16 +11003805 mddev_unlock(&rs->md);
3806 }
NeilBrown9d09e662011-01-13 20:00:02 +00003807}
3808
Jonathan Brassowf381e712013-05-08 17:57:13 -05003809static void attempt_restore_of_faulty_devices(struct raid_set *rs)
NeilBrown9d09e662011-01-13 20:00:02 +00003810{
Jonathan Brassow9092c022013-05-02 14:19:24 -05003811 int i;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003812 uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
Jonathan Brassow9092c022013-05-02 14:19:24 -05003813 unsigned long flags;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003814 bool cleared = false;
Jonathan Brassow9092c022013-05-02 14:19:24 -05003815 struct dm_raid_superblock *sb;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003816 struct mddev *mddev = &rs->md;
Jonathan Brassow9092c022013-05-02 14:19:24 -05003817 struct md_rdev *r;
NeilBrown9d09e662011-01-13 20:00:02 +00003818
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003819 /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3820 if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3821 return;
3822
3823 memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3824
Mikulas Patocka7dad24d2022-07-24 14:33:52 -04003825 for (i = 0; i < rs->raid_disks; i++) {
Jonathan Brassowf381e712013-05-08 17:57:13 -05003826 r = &rs->dev[i].rdev;
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003827 /* HM FIXME: enhance journal device recovery processing */
3828 if (test_bit(Journal, &r->flags))
3829 continue;
3830
Heinz Mauelshagene2568462017-01-14 03:53:10 +01003831 if (test_bit(Faulty, &r->flags) &&
3832 r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
Jonathan Brassowf381e712013-05-08 17:57:13 -05003833 DMINFO("Faulty %s device #%d has readable super block."
3834 " Attempting to revive it.",
3835 rs->raid_type->name, i);
Jonathan Brassowa4dc1632013-05-08 18:00:54 -05003836
3837 /*
3838 * Faulty bit may be set, but sometimes the array can
3839 * be suspended before the personalities can respond
3840 * by removing the device from the array (i.e. calling
Mike Snitzer43157842016-05-30 13:03:37 -04003841 * 'hot_remove_disk'). If they haven't yet removed
Jonathan Brassowa4dc1632013-05-08 18:00:54 -05003842 * the failed device, its 'raid_disk' number will be
3843 * '>= 0' - meaning we must call this function
3844 * ourselves.
3845 */
Jonathan Brassowf381e712013-05-08 17:57:13 -05003846 flags = r->flags;
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01003847 clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3848 if (r->raid_disk >= 0) {
3849 if (mddev->pers->hot_remove_disk(mddev, r)) {
3850 /* Failed to revive this device, try next */
3851 r->flags = flags;
3852 continue;
3853 }
3854 } else
3855 r->raid_disk = r->saved_raid_disk = i;
3856
Jonathan Brassowf381e712013-05-08 17:57:13 -05003857 clear_bit(Faulty, &r->flags);
3858 clear_bit(WriteErrorSeen, &r->flags);
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01003859
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003860 if (mddev->pers->hot_add_disk(mddev, r)) {
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01003861 /* Failed to revive this device, try next */
3862 r->raid_disk = r->saved_raid_disk = -1;
Jonathan Brassowf381e712013-05-08 17:57:13 -05003863 r->flags = flags;
3864 } else {
Heinz Mauelshagenc63ede32017-01-14 03:53:07 +01003865 clear_bit(In_sync, &r->flags);
Jonathan Brassowf381e712013-05-08 17:57:13 -05003866 r->recovery_offset = 0;
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003867 set_bit(i, (void *) cleared_failed_devices);
3868 cleared = true;
Jonathan Brassowf381e712013-05-08 17:57:13 -05003869 }
3870 }
3871 }
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003872
3873 /* If any failed devices could be cleared, update all sbs failed_devices bits */
3874 if (cleared) {
3875 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3876
Jonathan Brassowf381e712013-05-08 17:57:13 -05003877 rdev_for_each(r, &rs->md) {
Heinz Mauelshagen63c32ed2016-11-30 22:31:05 +01003878 if (test_bit(Journal, &r->flags))
3879 continue;
3880
Jonathan Brassowf381e712013-05-08 17:57:13 -05003881 sb = page_address(r->sb_page);
Heinz Mauelshagena3c06a32016-08-09 14:55:35 +02003882 sb_retrieve_failed_devices(sb, failed_devices);
3883
3884 for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3885 failed_devices[i] &= ~cleared_failed_devices[i];
3886
3887 sb_update_failed_devices(sb, failed_devices);
Jonathan Brassowf381e712013-05-08 17:57:13 -05003888 }
3889 }
3890}
3891
Mike Snitzere6ca5e12016-06-02 15:27:22 -04003892static int __load_dirty_region_bitmap(struct raid_set *rs)
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003893{
3894 int r = 0;
3895
3896 /* Try loading the bitmap unless "raid0", which does not have one */
3897 if (!rs_is_raid0(rs) &&
Mike Snitzer42863252016-06-02 12:27:46 -04003898 !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
Andy Shevchenkoe64e40182018-08-01 15:20:50 -07003899 r = md_bitmap_load(&rs->md);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003900 if (r)
3901 DMERR("Failed to load bitmap");
3902 }
3903
3904 return r;
3905}
3906
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003907/* Enforce updating all superblocks */
3908static void rs_update_sbs(struct raid_set *rs)
3909{
3910 struct mddev *mddev = &rs->md;
3911 int ro = mddev->ro;
3912
Shaohua Li29530792016-12-08 15:48:19 -08003913 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003914 mddev->ro = 0;
3915 md_update_sb(mddev, 1);
3916 mddev->ro = ro;
3917}
3918
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003919/*
3920 * Reshape changes raid algorithm of @rs to new one within personality
3921 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3922 * disks from a raid set thus growing/shrinking it or resizes the set
3923 *
3924 * Call mddev_lock_nointr() before!
3925 */
3926static int rs_start_reshape(struct raid_set *rs)
3927{
3928 int r;
3929 struct mddev *mddev = &rs->md;
3930 struct md_personality *pers = mddev->pers;
3931
Heinz Mauelshagen644e2532018-09-06 18:33:39 +02003932 /* Don't allow the sync thread to work until the table gets reloaded. */
3933 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3934
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003935 r = rs_setup_reshape(rs);
3936 if (r)
3937 return r;
3938
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003939 /*
3940 * Check any reshape constraints enforced by the personalility
3941 *
3942 * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3943 */
3944 r = pers->check_reshape(mddev);
3945 if (r) {
3946 rs->ti->error = "pers->check_reshape() failed";
3947 return r;
3948 }
3949
3950 /*
3951 * Personality may not provide start reshape method in which
3952 * case check_reshape above has already covered everything
3953 */
3954 if (pers->start_reshape) {
3955 r = pers->start_reshape(mddev);
3956 if (r) {
3957 rs->ti->error = "pers->start_reshape() failed";
3958 return r;
3959 }
3960 }
3961
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003962 /*
3963 * Now reshape got set up, update superblocks to
3964 * reflect the fact so that a table reload will
3965 * access proper superblock content in the ctr.
3966 */
3967 rs_update_sbs(rs);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003968
3969 return 0;
3970}
3971
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003972static int raid_preresume(struct dm_target *ti)
3973{
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003974 int r;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003975 struct raid_set *rs = ti->private;
3976 struct mddev *mddev = &rs->md;
3977
Heinz Mauelshagen11e47232017-12-13 17:13:18 +01003978 /* This is a resume after a suspend of the set -> it's already started. */
Mike Snitzer42863252016-06-02 12:27:46 -04003979 if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003980 return 0;
3981
3982 /*
3983 * The superblocks need to be updated on disk if the
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003984 * array is new or new devices got added (thus zeroed
3985 * out by userspace) or __load_dirty_region_bitmap
3986 * will overwrite them in core with old data or fail.
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003987 */
Heinz Mauelshagen6e209022016-06-14 15:23:13 -04003988 if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3989 rs_update_sbs(rs);
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003990
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02003991 /* Load the bitmap from disk unless raid0 */
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02003992 r = __load_dirty_region_bitmap(rs);
3993 if (r)
3994 return r;
3995
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02003996 /* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
3997 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3998 mddev->array_sectors = rs->array_sectors;
3999 mddev->dev_sectors = rs->dev_sectors;
4000 rs_set_rdev_sectors(rs);
4001 rs_set_capacity(rs);
4002 }
4003
4004 /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
Heinz Mauelshagen255e2642023-01-25 23:31:55 +01004005 if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02004006 (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
4007 (rs->requested_bitmap_chunk_sectors &&
4008 mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
4009 int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
4010
4011 r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors, chunksize, 0);
Heinz Mauelshagen4257e082016-06-14 01:46:03 +02004012 if (r)
4013 DMERR("Failed to resize bitmap");
4014 }
4015
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02004016 /* Check for any resize/reshape on @rs and adjust/initiate */
4017 /* Be prepared for mddev_resume() in raid_resume() */
4018 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4019 if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02004020 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02004021 mddev->resync_min = mddev->recovery_cp;
Heinz Mauelshagen99273d92019-10-01 17:47:53 +02004022 if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
4023 mddev->resync_max_sectors = mddev->dev_sectors;
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02004024 }
4025
Heinz Mauelshagen345a6cd2016-06-25 02:42:54 +02004026 /* Check for any reshape request unless new raid set */
Heinz Mauelshagend39f0012017-12-02 01:03:52 +01004027 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02004028 /* Initiate a reshape. */
Heinz Mauelshagen61e06e22017-12-02 01:03:51 +01004029 rs_set_rdev_sectors(rs);
Heinz Mauelshagen9dbd1aa2016-06-13 17:55:14 +02004030 mddev_lock_nointr(mddev);
4031 r = rs_start_reshape(rs);
4032 mddev_unlock(mddev);
4033 if (r)
4034 DMWARN("Failed to check/start reshape, continuing without change");
4035 r = 0;
4036 }
4037
4038 return r;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02004039}
4040
Jonathan Brassowf381e712013-05-08 17:57:13 -05004041static void raid_resume(struct dm_target *ti)
4042{
4043 struct raid_set *rs = ti->private;
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02004044 struct mddev *mddev = &rs->md;
Jonathan Brassowf381e712013-05-08 17:57:13 -05004045
Mike Snitzer42863252016-06-02 12:27:46 -04004046 if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02004047 /*
4048 * A secondary resume while the device is active.
4049 * Take this opportunity to check whether any failed
4050 * devices are reachable again.
4051 */
4052 attempt_restore_of_faulty_devices(rs);
Jonathan Brassow47525e52012-05-22 13:55:29 +10004053 }
Heinz Mauelshagen31e10a42016-08-10 02:45:59 +02004054
NeilBrown4d5324f2017-10-19 12:17:16 +11004055 if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
Heinz Mauelshagen11e47232017-12-13 17:13:18 +01004056 /* Only reduce raid set size before running a disk removing reshape. */
4057 if (mddev->delta_disks < 0)
4058 rs_set_capacity(rs);
4059
NeilBrown4d5324f2017-10-19 12:17:16 +11004060 mddev_lock_nointr(mddev);
Heinz Mauelshagen11e47232017-12-13 17:13:18 +01004061 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
Heinz Mauelshagen78a75d12017-12-02 01:03:56 +01004062 mddev->ro = 0;
4063 mddev->in_sync = 0;
Heinz Mauelshagen31e10a42016-08-10 02:45:59 +02004064 mddev_resume(mddev);
NeilBrown4d5324f2017-10-19 12:17:16 +11004065 mddev_unlock(mddev);
4066 }
NeilBrown9d09e662011-01-13 20:00:02 +00004067}
4068
4069static struct target_type raid_target = {
4070 .name = "raid",
Heinz Mauelshagen43f39522019-12-19 17:58:46 +01004071 .version = {1, 15, 1},
NeilBrown9d09e662011-01-13 20:00:02 +00004072 .module = THIS_MODULE,
4073 .ctr = raid_ctr,
4074 .dtr = raid_dtr,
4075 .map = raid_map,
4076 .status = raid_status,
Jonathan Brassowbe836512013-04-24 11:42:43 +10004077 .message = raid_message,
NeilBrown9d09e662011-01-13 20:00:02 +00004078 .iterate_devices = raid_iterate_devices,
4079 .io_hints = raid_io_hints,
NeilBrown9d09e662011-01-13 20:00:02 +00004080 .postsuspend = raid_postsuspend,
Heinz Mauelshagenecbfb9f2016-05-19 18:49:33 +02004081 .preresume = raid_preresume,
NeilBrown9d09e662011-01-13 20:00:02 +00004082 .resume = raid_resume,
4083};
4084
4085static int __init dm_raid_init(void)
4086{
Jonathan Brassowfe5d2f42013-02-21 13:28:10 +11004087 DMINFO("Loading target version %u.%u.%u",
4088 raid_target.version[0],
4089 raid_target.version[1],
4090 raid_target.version[2]);
NeilBrown9d09e662011-01-13 20:00:02 +00004091 return dm_register_target(&raid_target);
4092}
4093
4094static void __exit dm_raid_exit(void)
4095{
4096 dm_unregister_target(&raid_target);
4097}
4098
4099module_init(dm_raid_init);
4100module_exit(dm_raid_exit);
4101
Heinz Mauelshagen48cf06b2014-09-24 17:47:19 +02004102module_param(devices_handle_discard_safely, bool, 0644);
4103MODULE_PARM_DESC(devices_handle_discard_safely,
4104 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4105
Mike Snitzeref9b85a2016-06-02 12:02:19 -04004106MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4107MODULE_ALIAS("dm-raid0");
Jonathan Brassow63f33b8d2012-07-31 21:44:26 -05004108MODULE_ALIAS("dm-raid1");
4109MODULE_ALIAS("dm-raid10");
NeilBrown9d09e662011-01-13 20:00:02 +00004110MODULE_ALIAS("dm-raid4");
4111MODULE_ALIAS("dm-raid5");
4112MODULE_ALIAS("dm-raid6");
4113MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
Heinz Mauelshagen3a1c1ef2016-05-19 18:49:34 +02004114MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
NeilBrown9d09e662011-01-13 20:00:02 +00004115MODULE_LICENSE("GPL");