2 * raid6check - extended consistency check for RAID-6
4 * Copyright (C) 2011 Piergiorgio Sartor
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 * Author: Piergiorgio Sartor
22 * Based on "restripe.c" from "mdadm" codebase
30 #define CHECK_PAGE_BITS (12)
31 #define CHECK_PAGE_SIZE (1 << CHECK_PAGE_BITS)
33 char const Name[] = "raid6check";
41 int geo_map(int block, unsigned long long stripe, int raid_disks,
42 int level, int layout);
43 int is_ddf(int layout);
44 void qsyndrome(uint8_t *p, uint8_t *q, uint8_t **sources, int disks, int size);
45 void make_tables(void);
46 void ensure_zero_has_size(int chunk_size);
47 void raid6_datap_recov(int disks, size_t bytes, int faila, uint8_t **ptrs,
49 void raid6_2data_recov(int disks, size_t bytes, int faila, int failb,
50 uint8_t **ptrs, int neg_offset);
51 void xor_blocks(char *target, char **sources, int disks, int size);
53 /* Collect per stripe consistency information */
54 void raid6_collect(int chunk_size, uint8_t *p, uint8_t *q,
55 char *chunkP, char *chunkQ, int *results)
60 extern uint8_t raid6_gflog[];
62 for(i = 0; i < chunk_size; i++) {
63 Px = (uint8_t)chunkP[i] ^ (uint8_t)p[i];
64 Qx = (uint8_t)chunkQ[i] ^ (uint8_t)q[i];
66 if((Px != 0) && (Qx == 0))
69 if((Px == 0) && (Qx != 0))
72 if((Px != 0) && (Qx != 0)) {
73 data_id = (raid6_gflog[Qx] - raid6_gflog[Px]);
74 if(data_id < 0) data_id += 255;
78 if((Px == 0) && (Qx == 0))
83 /* Try to find out if a specific disk has problems in a CHECK_PAGE_SIZE page size */
84 int raid6_stats_blk(int *results, int raid_disks)
87 int curr_broken_disk = -255;
88 int prev_broken_disk = -255;
89 int broken_status = 0;
91 for(i = 0; i < CHECK_PAGE_SIZE; i++) {
93 if(results[i] != -255)
94 curr_broken_disk = results[i];
96 if(curr_broken_disk >= raid_disks)
99 switch(broken_status) {
101 if(curr_broken_disk != -255) {
102 prev_broken_disk = curr_broken_disk;
108 if(curr_broken_disk != prev_broken_disk)
114 curr_broken_disk = prev_broken_disk = -65535;
119 return curr_broken_disk;
122 /* Collect disks status for a strip in CHECK_PAGE_SIZE page size blocks */
123 void raid6_stats(int *disk, int *results, int raid_disks, int chunk_size)
127 for(i = 0, j = 0; i < chunk_size; i += CHECK_PAGE_SIZE, j++) {
128 disk[j] = raid6_stats_blk(&results[i], raid_disks);
132 int lock_stripe(struct mdinfo *info, unsigned long long start,
133 int chunk_size, int data_disks, sighandler_t *sig) {
135 if(mlockall(MCL_CURRENT | MCL_FUTURE) != 0) {
139 sig[0] = signal(SIGTERM, SIG_IGN);
140 sig[1] = signal(SIGINT, SIG_IGN);
141 sig[2] = signal(SIGQUIT, SIG_IGN);
143 rv = sysfs_set_num(info, NULL, "suspend_lo", start * chunk_size * data_disks);
144 rv |= sysfs_set_num(info, NULL, "suspend_hi", (start + 1) * chunk_size * data_disks);
148 int unlock_all_stripes(struct mdinfo *info, sighandler_t *sig) {
150 rv = sysfs_set_num(info, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
151 rv |= sysfs_set_num(info, NULL, "suspend_hi", 0);
152 rv |= sysfs_set_num(info, NULL, "suspend_lo", 0);
154 signal(SIGQUIT, sig[2]);
155 signal(SIGINT, sig[1]);
156 signal(SIGTERM, sig[0]);
158 if(munlockall() != 0)
164 int autorepair(int *disk, unsigned long long start, int chunk_size,
165 char *name[], int raid_disks, int syndrome_disks, char **blocks_page,
166 char **blocks, uint8_t *p, int *block_index_for_slot,
167 int *source, unsigned long long *offsets)
170 int pages_to_write_count = 0;
171 int page_to_write[chunk_size >> CHECK_PAGE_BITS];
172 for(j = 0; j < (chunk_size >> CHECK_PAGE_BITS); j++) {
173 if (disk[j] >= -2 && block_index_for_slot[disk[j]] >= 0) {
174 int slot = block_index_for_slot[disk[j]];
175 printf("Auto-repairing slot %d (%s)\n", slot, name[slot]);
176 pages_to_write_count++;
177 page_to_write[j] = 1;
178 for(i = -2; i < syndrome_disks; i++) {
179 blocks_page[i] = blocks[i] + j * CHECK_PAGE_SIZE;
182 qsyndrome(p, (uint8_t*)blocks_page[-2],
183 (uint8_t**)blocks_page,
184 syndrome_disks, CHECK_PAGE_SIZE);
187 char *all_but_failed_blocks[syndrome_disks];
188 for(i = 0; i < syndrome_disks; i++) {
190 all_but_failed_blocks[i] = blocks_page[-1];
192 all_but_failed_blocks[i] = blocks_page[i];
194 xor_blocks(blocks_page[disk[j]],
195 all_but_failed_blocks, syndrome_disks,
200 page_to_write[j] = 0;
204 if(pages_to_write_count > 0) {
206 for(j = 0; j < (chunk_size >> CHECK_PAGE_BITS); j++) {
207 if(page_to_write[j] == 1) {
208 int slot = block_index_for_slot[disk[j]];
209 lseek64(source[slot], offsets[slot] + start * chunk_size + j * CHECK_PAGE_SIZE, SEEK_SET);
210 write_res += write(source[slot],
211 blocks[disk[j]] + j * CHECK_PAGE_SIZE,
216 if (write_res != (CHECK_PAGE_SIZE * pages_to_write_count)) {
217 fprintf(stderr, "Failed to write a full chunk.\n");
226 int manual_repair(int chunk_size, int syndrome_disks,
227 int failed_slot1, int failed_slot2,
228 unsigned long long start, int *block_index_for_slot,
229 char *name[], char **stripes, char **blocks, uint8_t *p,
230 int *source, unsigned long long *offsets)
233 int fd1 = block_index_for_slot[failed_slot1];
234 int fd2 = block_index_for_slot[failed_slot2];
235 printf("Repairing stripe %llu\n", start);
236 printf("Assuming slots %d (%s) and %d (%s) are incorrect\n",
240 if (failed_slot1 == -2 || failed_slot2 == -2) {
241 char *all_but_failed_blocks[syndrome_disks];
242 int failed_data_or_p;
244 if (failed_slot1 == -2)
245 failed_data_or_p = failed_slot2;
247 failed_data_or_p = failed_slot1;
249 printf("Repairing D/P(%d) and Q\n", failed_data_or_p);
251 for (i = 0; i < syndrome_disks; i++) {
252 if (i == failed_data_or_p)
253 all_but_failed_blocks[i] = blocks[-1];
255 all_but_failed_blocks[i] = blocks[i];
257 xor_blocks(blocks[failed_data_or_p],
258 all_but_failed_blocks, syndrome_disks, chunk_size);
259 qsyndrome(p, (uint8_t*)blocks[-2], (uint8_t**)blocks,
260 syndrome_disks, chunk_size);
262 ensure_zero_has_size(chunk_size);
263 if (failed_slot1 == -1 || failed_slot2 == -1) {
265 if (failed_slot1 == -1)
266 failed_data = failed_slot2;
268 failed_data = failed_slot1;
270 printf("Repairing D(%d) and P\n", failed_data);
271 raid6_datap_recov(syndrome_disks+2, chunk_size,
272 failed_data, (uint8_t**)blocks, 1);
274 printf("Repairing D and D\n");
275 raid6_2data_recov(syndrome_disks+2, chunk_size,
276 failed_slot1, failed_slot2,
277 (uint8_t**)blocks, 1);
281 int write_res1, write_res2;
284 seek_res = lseek64(source[fd1],
285 offsets[fd1] + start * chunk_size, SEEK_SET);
287 fprintf(stderr, "lseek failed for failed_disk1\n");
290 write_res1 = write(source[fd1], blocks[failed_slot1], chunk_size);
292 seek_res = lseek64(source[fd2],
293 offsets[fd2] + start * chunk_size, SEEK_SET);
295 fprintf(stderr, "lseek failed for failed_disk2\n");
298 write_res2 = write(source[fd2], blocks[failed_slot2], chunk_size);
300 if (write_res1 != chunk_size || write_res2 != chunk_size) {
301 fprintf(stderr, "Failed to write a complete chunk.\n");
308 int check_stripes(struct mdinfo *info, int *source, unsigned long long *offsets,
309 int raid_disks, int chunk_size, int level, int layout,
310 unsigned long long start, unsigned long long length, char *name[],
311 enum repair repair, int failed_disk1, int failed_disk2)
313 /* read the data and p and q blocks, and check we got them right */
314 int data_disks = raid_disks - 2;
315 int syndrome_disks = data_disks + is_ddf(layout) * 2;
318 /* stripes[] is indexed by raid_disk and holds chunks from each device */
319 char **stripes = xmalloc(raid_disks * sizeof(char*));
321 /* blocks[] is indexed by syndrome number and points to either one of the
322 * chunks from 'stripes[]', or to a chunk of zeros. -1 and -2 are
324 char **blocks = xmalloc((syndrome_disks + 2) * sizeof(char*));
326 /* blocks_page[] is a temporary index to just one page of the chunks
327 * that blocks[] points to. */
328 char **blocks_page = xmalloc((syndrome_disks + 2) * sizeof(char*));
330 /* block_index_for_slot[] provides the reverse mapping from blocks to stripes.
331 * The index is a syndrome position, the content is a raid_disk number.
332 * indicies -1 and -2 work, and are P and Q disks */
333 int *block_index_for_slot = xmalloc((syndrome_disks+2) * sizeof(int));
335 /* 'p' and 'q' contain calcualted P and Q, to be compared with
336 * blocks[-1] and blocks[-2];
338 uint8_t *p = xmalloc(chunk_size);
339 uint8_t *q = xmalloc(chunk_size);
340 char *zero = xmalloc(chunk_size);
341 int *results = xmalloc(chunk_size * sizeof(int));
342 sighandler_t *sig = xmalloc(3 * sizeof(sighandler_t));
345 int diskP, diskQ, diskD;
348 extern int tables_ready;
353 if (posix_memalign((void**)&stripe_buf, 4096, raid_disks * chunk_size) != 0)
355 block_index_for_slot += 2;
359 memset(zero, 0, chunk_size);
360 for ( i = 0 ; i < raid_disks ; i++)
361 stripes[i] = stripe_buf + i * chunk_size;
364 /* The syndrome number of the broken disk is recorded
365 * in 'disk[]' which allows a different broken disk for
368 int disk[chunk_size >> CHECK_PAGE_BITS];
370 err = lock_stripe(info, start, chunk_size, data_disks, sig);
373 unlock_all_stripes(info, sig);
376 for (i = 0 ; i < raid_disks ; i++) {
377 off64_t seek_res = lseek64(source[i], offsets[i] + start * chunk_size,
380 fprintf(stderr, "lseek to source %d failed\n", i);
381 unlock_all_stripes(info, sig);
385 int read_res = read(source[i], stripes[i], chunk_size);
386 if (read_res < chunk_size) {
387 fprintf(stderr, "Failed to read complete chunk disk %d, aborting\n", i);
388 unlock_all_stripes(info, sig);
394 diskP = geo_map(-1, start, raid_disks, level, layout);
395 block_index_for_slot[-1] = diskP;
396 blocks[-1] = stripes[diskP];
398 diskQ = geo_map(-2, start, raid_disks, level, layout);
399 block_index_for_slot[-2] = diskQ;
400 blocks[-2] = stripes[diskQ];
402 if (!is_ddf(layout)) {
403 /* The syndrome-order of disks starts immediately after 'Q',
406 for (i = 0 ; i < data_disks ; i++) {
408 if (diskD >= raid_disks)
412 if (diskD >= raid_disks)
414 blocks[i] = stripes[diskD];
415 block_index_for_slot[i] = diskD;
418 /* The syndrome-order exactly follows raid-disk
419 * numbers, with ZERO in place of P and Q
421 for (i = 0 ; i < raid_disks; i++) {
422 if (i == diskP || i == diskQ) {
424 block_index_for_slot[i] = -1;
426 blocks[i] = stripes[i];
427 block_index_for_slot[i] = i;
432 qsyndrome(p, q, (uint8_t**)blocks, syndrome_disks, chunk_size);
434 raid6_collect(chunk_size, p, q, stripes[diskP], stripes[diskQ], results);
435 raid6_stats(disk, results, raid_disks, chunk_size);
437 for(j = 0; j < (chunk_size >> CHECK_PAGE_BITS); j++) {
440 int slot = block_index_for_slot[role];
442 printf("Error detected at stripe %llu, page %d: possible failed disk slot %d: %d --> %s\n",
443 start, j, role, slot, name[slot]);
445 printf("Error detected at stripe %llu, page %d: failed slot %d should be zeros\n",
447 } else if(disk[j] == -65535) {
448 printf("Error detected at stripe %llu, page %d: disk slot unknown\n", start, j);
452 if(repair == AUTO_REPAIR) {
453 err = autorepair(disk, start, chunk_size,
454 name, raid_disks, syndrome_disks, blocks_page,
455 blocks, p, block_index_for_slot,
458 unlock_all_stripes(info, sig);
463 if(repair == MANUAL_REPAIR) {
464 int failed_slot1 = -1, failed_slot2 = -1;
465 for (i = -2; i < syndrome_disks; i++) {
466 if (block_index_for_slot[i] == failed_disk1)
468 if (block_index_for_slot[i] == failed_disk2)
471 err = manual_repair(chunk_size, syndrome_disks,
472 failed_slot1, failed_slot2,
473 start, block_index_for_slot,
474 name, stripes, blocks, p,
477 unlock_all_stripes(info, sig);
482 err = unlock_all_stripes(info, sig);
497 free(block_index_for_slot-2);
506 unsigned long long getnum(char *str, char **err)
509 unsigned long long rv = strtoull(str, &e, 10);
517 int main(int argc, char *argv[])
519 /* md_device start length */
522 char **disk_name = NULL;
523 unsigned long long *offsets = NULL;
529 enum repair repair = NO_REPAIR;
530 int failed_disk1 = -1;
531 int failed_disk2 = -1;
532 unsigned long long start, length;
535 struct mdinfo *info = NULL, *comp = NULL;
539 char *prg = strrchr(argv[0], '/');
547 fprintf(stderr, "Usage: %s md_device start_stripe length_stripes [autorepair]\n", prg);
548 fprintf(stderr, " or: %s md_device repair stripe failed_slot_1 failed_slot_2\n", prg);
553 mdfd = open(argv[1], O_RDONLY);
556 fprintf(stderr, "%s: cannot open %s\n", prg, argv[1]);
561 info = sysfs_read(mdfd, NULL,
573 fprintf(stderr, "%s: Error reading sysfs information of %s\n", prg, argv[1]);
578 if(info->array.level != level) {
579 fprintf(stderr, "%s: %s not a RAID-6\n", prg, argv[1]);
584 if(info->array.failed_disks > 0) {
585 fprintf(stderr, "%s: %s degraded array\n", prg, argv[1]);
590 printf("layout: %d\n", info->array.layout);
591 printf("disks: %d\n", info->array.raid_disks);
592 printf("component size: %llu\n", info->component_size * 512);
593 printf("total stripes: %llu\n", (info->component_size * 512) / info->array.chunk_size);
594 printf("chunk size: %d\n", info->array.chunk_size);
598 for(i = 0, active_disks = 0; active_disks < info->array.raid_disks; i++) {
599 printf("disk: %d - offset: %llu - size: %llu - name: %s - slot: %d\n",
600 i, comp->data_offset * 512, comp->component_size * 512,
601 map_dev(comp->disk.major, comp->disk.minor, 0),
602 comp->disk.raid_disk);
603 if(comp->disk.raid_disk >= 0)
611 raid_disks = info->array.raid_disks;
612 chunk_size = info->array.chunk_size;
613 layout = info->array.layout;
614 if (strcmp(argv[2], "repair")==0) {
616 fprintf(stderr, "For repair mode, call %s md_device repair stripe failed_slot_1 failed_slot_2\n", prg);
620 repair = MANUAL_REPAIR;
621 start = getnum(argv[3], &err);
623 failed_disk1 = getnum(argv[4], &err);
624 failed_disk2 = getnum(argv[5], &err);
626 if(failed_disk1 >= info->array.raid_disks) {
627 fprintf(stderr, "%s: failed_slot_1 index is higher than number of devices in raid\n", prg);
631 if(failed_disk2 >= info->array.raid_disks) {
632 fprintf(stderr, "%s: failed_slot_2 index is higher than number of devices in raid\n", prg);
636 if(failed_disk1 == failed_disk2) {
637 fprintf(stderr, "%s: failed_slot_1 and failed_slot_2 are the same\n", prg);
643 start = getnum(argv[2], &err);
644 length = getnum(argv[3], &err);
645 if (argc >= 5 && strcmp(argv[4], "autorepair")==0)
646 repair = AUTO_REPAIR;
650 fprintf(stderr, "%s: Bad number: %s\n", prg, err);
655 if(start > ((info->component_size * 512) / chunk_size)) {
656 start = (info->component_size * 512) / chunk_size;
657 fprintf(stderr, "%s: start beyond disks size\n", prg);
661 ((length + start) > ((info->component_size * 512) / chunk_size))) {
662 length = (info->component_size * 512) / chunk_size - start;
665 disk_name = xmalloc(raid_disks * sizeof(*disk_name));
666 fds = xmalloc(raid_disks * sizeof(*fds));
667 offsets = xcalloc(raid_disks, sizeof(*offsets));
668 buf = xmalloc(raid_disks * chunk_size);
670 for(i=0; i<raid_disks; i++) {
676 for (i=0, active_disks=0; active_disks<raid_disks; i++) {
677 int disk_slot = comp->disk.raid_disk;
679 disk_name[disk_slot] = map_dev(comp->disk.major, comp->disk.minor, 0);
680 offsets[disk_slot] = comp->data_offset * 512;
681 fds[disk_slot] = open(disk_name[disk_slot], O_RDWR | O_DIRECT);
682 if (fds[disk_slot] < 0) {
683 perror(disk_name[disk_slot]);
684 fprintf(stderr,"%s: cannot open %s\n", prg, disk_name[disk_slot]);
693 int rv = check_stripes(info, fds, offsets,
694 raid_disks, chunk_size, level, layout,
695 start, length, disk_name, repair, failed_disk1, failed_disk2);
697 fprintf(stderr, "%s: check_stripes returned %d\n", prg, rv);
705 for(i = 0; i < raid_disks; i++)