/*
* mdadm - manage Linux "md" devices aka RAID arrays.
*
- * Copyright (C) 2006 Neil Brown <neilb@suse.de>
+ * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
*
*
* This program is free software; you can redistribute it and/or modify
*/
#include "mdadm.h"
+#include <stdint.h>
/* To restripe, we read from old geometry to a buffer, and
* read from buffer to new geometry.
- * When reading we don't worry about parity. When writing we do.
+ * When reading, we might have missing devices and so could need
+ * to reconstruct.
+ * When writing, we need to create correct parity and Q.
*
*/
-static int geo_map(int block, unsigned long long stripe, int raid_disks, int level, int layout)
+static int geo_map(int block, unsigned long long stripe, int raid_disks,
+ int level, int layout)
{
- /* On the given stripe, find which disk in the array with have
+ /* On the given stripe, find which disk in the array will have
* block numbered 'block'.
+ * '-1' means the parity block.
+ * '-2' means the Q syndrome.
*/
int pd;
switch(level*100 + layout) {
case 000:
case 400:
+ case 500 + ALGORITHM_PARITY_N:
/* raid 4 isn't messed around by parity blocks */
if (block == -1)
return raid_disks-1; /* parity block */
if (block == -1) return pd;
return (pd + 1 + block) % raid_disks;
+ case 500 + ALGORITHM_PARITY_0:
+ return block + 1;
+
+
+ case 600 + ALGORITHM_PARITY_N_6:
+ if (block == -2)
+ return raid_disks - 1;
+ if (block == -1)
+ return raid_disks - 2; /* parity block */
+ return block;
+ case 600 + ALGORITHM_LEFT_ASYMMETRIC_6:
+ if (block == -2)
+ return raid_disks - 1;
+ raid_disks--;
+ pd = (raid_disks-1) - stripe % raid_disks;
+ if (block == -1) return pd;
+ if (block >= pd)
+ block++;
+ return block;
+
+ case 600 + ALGORITHM_RIGHT_ASYMMETRIC_6:
+ if (block == -2)
+ return raid_disks - 1;
+ raid_disks--;
+ pd = stripe % raid_disks;
+ if (block == -1) return pd;
+ if (block >= pd)
+ block++;
+ return block;
+
+ case 600 + ALGORITHM_LEFT_SYMMETRIC_6:
+ if (block == -2)
+ return raid_disks - 1;
+ raid_disks--;
+ pd = (raid_disks - 1) - stripe % raid_disks;
+ if (block == -1) return pd;
+ return (pd + 1 + block) % raid_disks;
+
+ case 600 + ALGORITHM_RIGHT_SYMMETRIC_6:
+ if (block == -2)
+ return raid_disks - 1;
+ raid_disks--;
+ pd = stripe % raid_disks;
+ if (block == -1) return pd;
+ return (pd + 1 + block) % raid_disks;
+
+ case 600 + ALGORITHM_PARITY_0_6:
+ if (block == -2)
+ return raid_disks - 1;
+ return block + 1;
+
+
+ case 600 + ALGORITHM_PARITY_0:
+ if (block == -1)
+ return 0;
+ if (block == -2)
+ return 1;
+ return block + 2;
+
case 600 + ALGORITHM_LEFT_ASYMMETRIC:
pd = raid_disks - 1 - (stripe % raid_disks);
if (block == -1) return pd;
+ if (block == -2) return (pd+1) % raid_disks;
if (pd == raid_disks - 1)
return block+1;
if (block >= pd)
return block+2;
return block;
+ case 600 + ALGORITHM_ROTATING_ZERO_RESTART:
+ /* Different order for calculating Q, otherwize same as ... */
case 600 + ALGORITHM_RIGHT_ASYMMETRIC:
pd = stripe % raid_disks;
if (block == -1) return pd;
+ if (block == -2) return (pd+1) % raid_disks;
if (pd == raid_disks - 1)
return block+1;
if (block >= pd)
case 600 + ALGORITHM_LEFT_SYMMETRIC:
pd = raid_disks - 1 - (stripe % raid_disks);
if (block == -1) return pd;
+ if (block == -2) return (pd+1) % raid_disks;
return (pd + 2 + block) % raid_disks;
case 600 + ALGORITHM_RIGHT_SYMMETRIC:
pd = stripe % raid_disks;
if (block == -1) return pd;
+ if (block == -2) return (pd+1) % raid_disks;
return (pd + 2 + block) % raid_disks;
+
+
+ case 600 + ALGORITHM_ROTATING_N_RESTART:
+ /* Same a left_asymmetric, by first stripe is
+ * D D D P Q rather than
+ * Q D D D P
+ */
+ pd = raid_disks - 1 - ((stripe + 1) % raid_disks);
+ if (block == -1) return pd;
+ if (block == -2) return (pd+1) % raid_disks;
+ if (pd == raid_disks - 1)
+ return block+1;
+ if (block >= pd)
+ return block+2;
+ return block;
+
+ case 600 + ALGORITHM_ROTATING_N_CONTINUE:
+ /* Same as left_symmetric but Q is before P */
+ pd = raid_disks - 1 - (stripe % raid_disks);
+ if (block == -1) return pd;
+ if (block == -2) return (pd+raid_disks-1) % raid_disks;
+ return (pd + 1 + block) % raid_disks;
}
return -1;
}
+static int is_ddf(int layout)
+{
+ switch (layout)
+ {
+ default:
+ return 0;
+ case ALGORITHM_ROTATING_N_CONTINUE:
+ case ALGORITHM_ROTATING_N_RESTART:
+ case ALGORITHM_ROTATING_ZERO_RESTART:
+ return 1;
+ }
+}
static void xor_blocks(char *target, char **sources, int disks, int size)
}
}
+static void qsyndrome(uint8_t *p, uint8_t *q, uint8_t **sources, int disks, int size)
+{
+ int d, z;
+ uint8_t wq0, wp0, wd0, w10, w20;
+ for ( d = 0; d < size; d++) {
+ wq0 = wp0 = sources[disks-1][d];
+ for ( z = disks-2 ; z >= 0 ; z-- ) {
+ wd0 = sources[z][d];
+ wp0 ^= wd0;
+ w20 = (wq0&0x80) ? 0xff : 0x00;
+ w10 = (wq0 << 1) & 0xff;
+ w20 &= 0x1d;
+ w10 ^= w20;
+ wq0 = w10 ^ wd0;
+ }
+ p[d] = wp0;
+ q[d] = wq0;
+ }
+}
+
+
+/*
+ * The following was taken from linux/drivers/md/mktables.c, and modified
+ * to create in-memory tables rather than C code
+ */
+static uint8_t gfmul(uint8_t a, uint8_t b)
+{
+ uint8_t v = 0;
+
+ while (b) {
+ if (b & 1)
+ v ^= a;
+ a = (a << 1) ^ (a & 0x80 ? 0x1d : 0);
+ b >>= 1;
+ }
+
+ return v;
+}
+
+static uint8_t gfpow(uint8_t a, int b)
+{
+ uint8_t v = 1;
+
+ b %= 255;
+ if (b < 0)
+ b += 255;
+
+ while (b) {
+ if (b & 1)
+ v = gfmul(v, a);
+ a = gfmul(a, a);
+ b >>= 1;
+ }
+
+ return v;
+}
+
+int tables_ready = 0;
+uint8_t raid6_gfmul[256][256];
+uint8_t raid6_gfexp[256];
+uint8_t raid6_gfinv[256];
+uint8_t raid6_gfexi[256];
+void make_tables(void)
+{
+ int i, j;
+ uint8_t v;
+
+ /* Compute multiplication table */
+ for (i = 0; i < 256; i++)
+ for (j = 0; j < 256; j++)
+ raid6_gfmul[i][j] = gfmul(i, j);
+
+ /* Compute power-of-2 table (exponent) */
+ v = 1;
+ for (i = 0; i < 256; i++) {
+ raid6_gfexp[i] = v;
+ v = gfmul(v, 2);
+ if (v == 1)
+ v = 0; /* For entry 255, not a real entry */
+ }
+
+ /* Compute inverse table x^-1 == x^254 */
+ for (i = 0; i < 256; i++)
+ raid6_gfinv[i] = gfpow(i, 254);
+
+ /* Compute inv(2^x + 1) (exponent-xor-inverse) table */
+ for (i = 0; i < 256; i ++)
+ raid6_gfexi[i] = raid6_gfinv[raid6_gfexp[i] ^ 1];
+
+ tables_ready = 1;
+}
+
+uint8_t *zero;
+/* Following was taken from linux/drivers/md/raid6recov.c */
+
+/* Recover two failed data blocks. */
+void raid6_2data_recov(int disks, size_t bytes, int faila, int failb,
+ uint8_t **ptrs)
+{
+ uint8_t *p, *q, *dp, *dq;
+ uint8_t px, qx, db;
+ const uint8_t *pbmul; /* P multiplier table for B data */
+ const uint8_t *qmul; /* Q multiplier table (for both) */
+
+ p = ptrs[disks-2];
+ q = ptrs[disks-1];
+
+ /* Compute syndrome with zero for the missing data pages
+ Use the dead data pages as temporary storage for
+ delta p and delta q */
+ dp = ptrs[faila];
+ ptrs[faila] = zero;
+ dq = ptrs[failb];
+ ptrs[failb] = zero;
+
+ qsyndrome(dp, dq, ptrs, disks-2, bytes);
+
+ /* Restore pointer table */
+ ptrs[faila] = dp;
+ ptrs[failb] = dq;
+
+ /* Now, pick the proper data tables */
+ pbmul = raid6_gfmul[raid6_gfexi[failb-faila]];
+ qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]]];
+
+ /* Now do it... */
+ while ( bytes-- ) {
+ px = *p ^ *dp;
+ qx = qmul[*q ^ *dq];
+ *dq++ = db = pbmul[px] ^ qx; /* Reconstructed B */
+ *dp++ = db ^ px; /* Reconstructed A */
+ p++; q++;
+ }
+}
+
+/* Recover failure of one data block plus the P block */
+void raid6_datap_recov(int disks, size_t bytes, int faila, uint8_t **ptrs)
+{
+ uint8_t *p, *q, *dq;
+ const uint8_t *qmul; /* Q multiplier table */
+
+ p = ptrs[disks-2];
+ q = ptrs[disks-1];
+
+ /* Compute syndrome with zero for the missing data page
+ Use the dead data page as temporary storage for delta q */
+ dq = ptrs[faila];
+ ptrs[faila] = zero;
+
+ qsyndrome(p, dq, ptrs, disks-2, bytes);
+
+ /* Restore pointer table */
+ ptrs[faila] = dq;
+
+ /* Now, pick the proper data tables */
+ qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]]];
+
+ /* Now do it... */
+ while ( bytes-- ) {
+ *p++ ^= *dq = qmul[*q ^ *dq];
+ q++; dq++;
+ }
+}
+
/* Save data:
* We are given:
- * A list of 'fds' of the active disks. For now we require all to be present.
- * A geomtry: raid_disks, chunk_size, level, layout
+ * A list of 'fds' of the active disks. Some may be absent.
+ * A geometry: raid_disks, chunk_size, level, layout
* A list of 'fds' for mirrored targets. They are already seeked to
* right (Write) location
- * A start and length
+ * A start and length which must be stripe-aligned
+ * 'buf' is large enough to hold one stripe, and is aligned
*/
int save_stripes(int *source, unsigned long long *offsets,
int raid_disks, int chunk_size, int level, int layout,
int nwrites, int *dest,
- unsigned long long start, unsigned long long length)
+ unsigned long long start, unsigned long long length,
+ char *buf)
{
- char buf[8192];
- int cpos = start % chunk_size; /* where in chunk we are up to */
int len;
int data_disks = raid_disks - (level == 0 ? 0 : level <=5 ? 1 : 2);
int disk;
+ int i;
+ if (!tables_ready)
+ make_tables();
+
+ if (zero == NULL) {
+ zero = malloc(chunk_size);
+ memset(zero, 0, chunk_size);
+ }
+
+ len = data_disks * chunk_size;
while (length > 0) {
- unsigned long long offset;
- int i;
- len = chunk_size - cpos;
- if (len > sizeof(buf)) len = sizeof(buf);
- if (len > length) len = length;
- /* len bytes to be moved from one device */
-
- offset = (start/chunk_size/data_disks)*chunk_size + cpos;
- disk = start/chunk_size % data_disks;
- disk = geo_map(disk, start/chunk_size/data_disks,
- raid_disks, level, layout);
- if (lseek64(source[disk], offsets[disk]+offset, 0) < 0)
- return -1;
- if (read(source[disk], buf, len) != len)
+ int failed = 0;
+ int fdisk[3], fblock[3];
+ for (disk = 0; disk < raid_disks ; disk++) {
+ unsigned long long offset;
+ int dnum;
+
+ offset = (start/chunk_size/data_disks)*chunk_size;
+ dnum = geo_map(disk < data_disks ? disk : data_disks - disk - 1,
+ start/chunk_size/data_disks,
+ raid_disks, level, layout);
+ if (dnum < 0) abort();
+ if (source[dnum] < 0 ||
+ lseek64(source[dnum], offsets[dnum]+offset, 0) < 0 ||
+ read(source[dnum], buf+disk * chunk_size, chunk_size)
+ != chunk_size)
+ if (failed <= 2) {
+ fdisk[failed] = dnum;
+ fblock[failed] = disk;
+ failed++;
+ }
+ }
+ if (failed == 0 || fblock[0] >= data_disks)
+ /* all data disks are good */
+ ;
+ else if (failed == 1 || fblock[1] >= data_disks+1) {
+ /* one failed data disk and good parity */
+ char *bufs[data_disks];
+ for (i=0; i < data_disks; i++)
+ if (fblock[0] == i)
+ bufs[i] = buf + data_disks*chunk_size;
+ else
+ bufs[i] = buf + i*chunk_size;
+
+ xor_blocks(buf + fblock[0]*chunk_size,
+ bufs, data_disks, chunk_size);
+ } else if (failed > 2 || level != 6)
+ /* too much failure */
return -1;
+ else {
+ /* RAID6 computations needed. */
+ uint8_t *bufs[data_disks+4];
+ int qdisk;
+ int syndrome_disks;
+ disk = geo_map(-1, start/chunk_size/data_disks,
+ raid_disks, level, layout);
+ qdisk = geo_map(-2, start/chunk_size/data_disks,
+ raid_disks, level, layout);
+ if (is_ddf(layout)) {
+ /* q over 'raid_disks' blocks, in device order.
+ * 'p' and 'q' get to be all zero
+ */
+ for (i = 0; i < raid_disks; i++)
+ bufs[i] = zero;
+ for (i = 0; i < data_disks; i++) {
+ int dnum = geo_map(i,
+ start/chunk_size/data_disks,
+ raid_disks, level, layout);
+ int snum;
+ /* i is the logical block number, so is index to 'buf'.
+ * dnum is physical disk number
+ * and thus the syndrome number.
+ */
+ snum = dnum;
+ bufs[snum] = (uint8_t*)buf + chunk_size * i;
+ }
+ syndrome_disks = raid_disks;
+ } else {
+ /* for md, q is over 'data_disks' blocks,
+ * starting immediately after 'q'
+ * Note that for the '_6' variety, the p block
+ * makes a hole that we need to be careful of.
+ */
+ int j;
+ int snum = 0;
+ for (j = 0; j < raid_disks; j++) {
+ int dnum = (qdisk + 1 + j) % raid_disks;
+ if (dnum == disk || dnum == qdisk)
+ continue;
+ for (i = 0; i < data_disks; i++)
+ if (geo_map(i,
+ start/chunk_size/data_disks,
+ raid_disks, level, layout) == dnum)
+ break;
+ /* i is the logical block number, so is index to 'buf'.
+ * dnum is physical disk number
+ * snum is syndrome disk for which 0 is immediately after Q
+ */
+ bufs[snum] = (uint8_t*)buf + chunk_size * i;
+
+ if (fblock[0] == i)
+ fdisk[0] = snum;
+ if (fblock[1] == i)
+ fdisk[1] = snum;
+ snum++;
+ }
+
+ syndrome_disks = data_disks;
+ }
+
+ /* Place P and Q blocks at end of bufs */
+ bufs[syndrome_disks] = (uint8_t*)buf + chunk_size * data_disks;
+ bufs[syndrome_disks+1] = (uint8_t*)buf + chunk_size * (data_disks+1);
+
+ if (fblock[1] == data_disks)
+ /* One data failed, and parity failed */
+ raid6_datap_recov(syndrome_disks+2, chunk_size,
+ fdisk[0], bufs);
+ else {
+ if (fdisk[0] > fdisk[1]) {
+ int t = fdisk[0];
+ fdisk[0] = fdisk[1];
+ fdisk[1] = t;
+ }
+ /* Two data blocks failed, P,Q OK */
+ raid6_2data_recov(syndrome_disks+2, chunk_size,
+ fdisk[0], fdisk[1], bufs);
+ }
+ }
+
for (i=0; i<nwrites; i++)
if (write(dest[i], buf, len) != len)
return -1;
+
length -= len;
start += len;
- cpos += len;
- while (cpos >= chunk_size) cpos -= chunk_size;
}
return 0;
}
int source, unsigned long long read_offset,
unsigned long long start, unsigned long long length)
{
- char *stripe_buf = malloc(raid_disks * chunk_size);
+ char *stripe_buf;
char **stripes = malloc(raid_disks * sizeof(char*));
char **blocks = malloc(raid_disks * sizeof(char*));
int i;
- int data_disks = raid_disks - (level == 0 ? 0 : level <=5 ? 1 : 2);
+ int data_disks = raid_disks - (level == 0 ? 0 : level <= 5 ? 1 : 2);
- if (stripe_buf == NULL || stripes == NULL || blocks == NULL) {
+ if (posix_memalign((void**)&stripe_buf, 4096, raid_disks * chunk_size))
+ stripe_buf = NULL;
+ if (zero == NULL) {
+ zero = malloc(chunk_size);
+ if (zero)
+ memset(zero, 0, chunk_size);
+ }
+ if (stripe_buf == NULL || stripes == NULL || blocks == NULL
+ || zero == NULL) {
free(stripe_buf);
free(stripes);
free(blocks);
+ free(zero);
return -2;
}
for (i=0; i<raid_disks; i++)
while (length > 0) {
int len = data_disks * chunk_size;
unsigned long long offset;
+ int disk, qdisk;
+ int syndrome_disks;
if (length < len)
return -3;
for (i=0; i < data_disks; i++) {
int disk = geo_map(i, start/chunk_size/data_disks,
raid_disks, level, layout);
- blocks[i] = stripes[disk];
if (lseek64(source, read_offset, 0) != read_offset)
return -1;
if (read(source, stripes[disk], chunk_size) != chunk_size)
}
/* We have the data, now do the parity */
offset = (start/chunk_size/data_disks) * chunk_size;
- if (level >= 4) {
- int disk = geo_map(-1, start/chunk_size/data_disks,
+ switch (level) {
+ case 4:
+ case 5:
+ disk = geo_map(-1, start/chunk_size/data_disks,
raid_disks, level, layout);
+ for (i = 0; i < data_disks; i++)
+ blocks[i] = stripes[(disk+1+i) % raid_disks];
xor_blocks(stripes[disk], blocks, data_disks, chunk_size);
- /* FIXME need to do raid6 Q as well */
+ break;
+ case 6:
+ disk = geo_map(-1, start/chunk_size/data_disks,
+ raid_disks, level, layout);
+ qdisk = geo_map(-2, start/chunk_size/data_disks,
+ raid_disks, level, layout);
+ if (is_ddf(layout)) {
+ /* q over 'raid_disks' blocks, in device order.
+ * 'p' and 'q' get to be all zero
+ */
+ for (i = 0; i < raid_disks; i++)
+ if (i == disk || i == qdisk)
+ blocks[i] = (char*)zero;
+ else
+ blocks[i] = stripes[i];
+ syndrome_disks = raid_disks;
+ } else {
+ /* for md, q is over 'data_disks' blocks,
+ * starting immediately after 'q'
+ */
+ for (i = 0; i < data_disks; i++)
+ blocks[i] = stripes[(qdisk+1+i) % raid_disks];
+
+ syndrome_disks = data_disks;
+ }
+ qsyndrome((uint8_t*)stripes[disk],
+ (uint8_t*)stripes[qdisk],
+ (uint8_t**)blocks,
+ syndrome_disks, chunk_size);
+ break;
}
for (i=0; i < raid_disks ; i++)
if (dest[i] >= 0) {
#ifdef MAIN
+int test_stripes(int *source, unsigned long long *offsets,
+ int raid_disks, int chunk_size, int level, int layout,
+ unsigned long long start, unsigned long long length)
+{
+ /* ready the data and p (and q) blocks, and check we got them right */
+ char *stripe_buf = malloc(raid_disks * chunk_size);
+ char **stripes = malloc(raid_disks * sizeof(char*));
+ char **blocks = malloc(raid_disks * sizeof(char*));
+ char *p = malloc(chunk_size);
+ char *q = malloc(chunk_size);
+
+ int i;
+ int data_disks = raid_disks - (level == 5 ? 1: 2);
+ for ( i = 0 ; i < raid_disks ; i++)
+ stripes[i] = stripe_buf + i * chunk_size;
+
+ while (length > 0) {
+ int disk;
+
+ for (i = 0 ; i < raid_disks ; i++) {
+ lseek64(source[i], offsets[i]+start, 0);
+ read(source[i], stripes[i], chunk_size);
+ }
+ for (i = 0 ; i < data_disks ; i++) {
+ int disk = geo_map(i, start/chunk_size, raid_disks,
+ level, layout);
+ blocks[i] = stripes[disk];
+ printf("%d->%d\n", i, disk);
+ }
+ switch(level) {
+ case 6:
+ qsyndrome(p, q, (uint8_t**)blocks, data_disks, chunk_size);
+ disk = geo_map(-1, start/chunk_size, raid_disks,
+ level, layout);
+ if (memcmp(p, stripes[disk], chunk_size) != 0) {
+ printf("P(%d) wrong at %llu\n", disk,
+ start / chunk_size);
+ }
+ disk = geo_map(-2, start/chunk_size, raid_disks,
+ level, layout);
+ if (memcmp(q, stripes[disk], chunk_size) != 0) {
+ printf("Q(%d) wrong at %llu\n", disk,
+ start / chunk_size);
+ }
+ break;
+ }
+ length -= chunk_size;
+ start += chunk_size;
+ }
+ return 0;
+}
+
unsigned long long getnum(char *str, char **err)
{
char *e;
int save;
int *fds;
char *file;
+ char *buf;
int storefd;
unsigned long long *offsets;
int raid_disks, chunk_size, level, layout;
save = 1;
else if (strcmp(argv[1], "restore") == 0)
save = 0;
+ else if (strcmp(argv[1], "test") == 0)
+ save = 2;
else {
fprintf(stderr, "test_stripe: must give 'save' or 'restore'.\n");
exit(2);
}
}
- if (save) {
+ buf = malloc(raid_disks * chunk_size);
+
+ if (save == 1) {
int rv = save_stripes(fds, offsets,
raid_disks, chunk_size, level, layout,
1, &storefd,
+ start, length, buf);
+ if (rv != 0) {
+ fprintf(stderr,
+ "test_stripe: save_stripes returned %d\n", rv);
+ exit(1);
+ }
+ } else if (save == 2) {
+ int rv = test_stripes(fds, offsets,
+ raid_disks, chunk_size, level, layout,
start, length);
if (rv != 0) {
- fprintf(stderr, "test_stripe: save_stripes returned %d\n", rv);
+ fprintf(stderr,
+ "test_stripe: test_stripes returned %d\n", rv);
exit(1);
}
} else {
storefd, 0ULL,
start, length);
if (rv != 0) {
- fprintf(stderr, "test_stripe: restore_stripes returned %d\n", rv);
+ fprintf(stderr,
+ "test_stripe: restore_stripes returned %d\n",
+ rv);
exit(1);
}
}