Hi
I encountered issues with spare disks and disks that differed slightly in
size when using raidreconf from raidtools 1.00.3
I noticed a few mentions of this in the list but don't see a fix posted
before.
the patch below fixes the behaviour (for me).
the spare disks were being counted as raid disks during the reconf, then
the kernel believed what the superblock said. The data was scattered after
the reconf and more so after the kernel resync. This could be avoided by
hotremoving spares, reconfing with only raid-disks in raidtabs and
hotadding the spares afterwards.
in some circumstances, different size disks composing a raid 5 array cause
raidreconf to bail. the total size of the array was calculated as
disk0.chunks+disk1.chunks+disk2.chunks etc. i've changed this to
smallest.chunks * (disks - 1). The problem manifested as a seek beyond end
of device (of the smallest device funnily enough)
anyway, the patch.
diff -ur raidtools-1.00.3/raidreconf.c raidtools-1.00.3+fixes/raidreconf.c
--- raidtools-1.00.3/raidreconf.c Wed Jan 15 08:58:25 2003
+++ raidtools-1.00.3+fixes/raidreconf.c Wed Dec 3 02:11:26 2003
@@ -93,7 +93,7 @@
}
if (version) {
- printf ("raidreconf 0.1.2 for mkraid version %d.%d.%d\n",
+ printf ("raidreconf 0.1.2-f for mkraid version %d.%d.%d\n",
MKRAID_MAJOR_VERSION, MKRAID_MINOR_VERSION,
MKRAID_PATCHLEVEL_VERSION);
return EXIT_VERSION;
@@ -175,7 +175,7 @@
fprintf (stderr, "Parsing %s\n", old_raidtab);
if (parse_config (fp)) {
fprintf (stderr,
- "Cannot parse old-array configuration
file");
+ "Cannot parse old-array configuration
file\n");
return EXIT_FAILURE;
}
@@ -218,7 +218,7 @@
fprintf (stderr, "Parsing %s\n", new_raidtab);
if (parse_config (fp)) {
fprintf (stderr,
- "Cannot parse new-array configuration
file.\n");
+ "Cannot parse new-array configuration
file\n");
return EXIT_FAILURE;
}
@@ -351,7 +351,7 @@
int i;
*sum = 0;
- for (i = 0; i != cfg->array.param.nr_disks; i++) {
+ for (i = 0; i != cfg->array.param.raid_disks; i++) {
int dnum = cfg->array.disks[i].raid_disk;
char *devname = cfg->device_name[dnum];
int fd = open (devname, O_RDONLY);
@@ -410,10 +410,10 @@
/* allocate fdsets and open files */
*oldarry =
(rrc_disk_t *) malloc (sizeof (rrc_disk_t) *
- oldcfg->array.param.nr_disks);
+ oldcfg->array.param.raid_disks);
*newarry =
(rrc_disk_t *) malloc (sizeof (rrc_disk_t) *
- newcfg->array.param.nr_disks);
+ newcfg->array.param.raid_disks);
if (!*oldarry || !*newarry) {
fprintf (stderr, "rrc_disk_t array malloc failure!\n");
return 1;
@@ -422,8 +422,8 @@
didnames =
(char **) malloc (sizeof (char *) *
- (oldcfg->array.param.nr_disks +
- newcfg->array.param.nr_disks));
+ (oldcfg->array.param.raid_disks +
+ newcfg->array.param.raid_disks));
if (!didnames) {
fprintf (stderr, "error allocating didnames array\n");
return 1;
@@ -433,7 +433,7 @@
didnames[cid] =
oldcfg->device_name[oldcfg->array.disks[0].raid_disk];
- for (i = 0; i != oldcfg->array.param.nr_disks; i++) {
+ for (i = 0; i != oldcfg->array.param.raid_disks; i++) {
int dnum = oldcfg->array.disks[i].raid_disk;
char *devname = oldcfg->device_name[dnum];
@@ -458,7 +458,7 @@
}
(*oldarry)[i].disk_id = cid;
}
- for (i = 0; i != newcfg->array.param.nr_disks; i++) {
+ for (i = 0; i != newcfg->array.param.raid_disks; i++) {
int dnum = newcfg->array.disks[i].raid_disk;
char *devname = newcfg->device_name[dnum];
int idi;
diff -ur raidtools-1.00.3/reconfiguration.c
raidtools-1.00.3+fixes/reconfiguration.c
--- raidtools-1.00.3/reconfiguration.c Wed Jan 15 08:58:25 2003
+++ raidtools-1.00.3+fixes/reconfiguration.c Wed Dec 3 02:11:26 2003
@@ -94,11 +94,11 @@
source_disk_free_map =
(char **) malloc (sizeof (char *) *
- old_md_cfg->array.param.nr_disks);
+ old_md_cfg->array.param.raid_disks);
if (!source_disk_free_map)
return "Cannot allocate source_disk_free_map";
- for (sd = 0; sd != old_md_cfg->array.param.nr_disks; sd++) {
+ for (sd = 0; sd != old_md_cfg->array.param.raid_disks; sd++) {
/* initialize .reconf_blocks in rrc structure */
old_rrc_cfg[sd].reconf_blocks = old_rrc_cfg[sd].blocks;
#if 0
@@ -121,7 +121,7 @@
}
/* Initialize the sink disk .reconf_blocks as well */
- for (sd = 0; sd != new_md_cfg->array.param.nr_disks; sd++) {
+ for (sd = 0; sd != new_md_cfg->array.param.raid_disks; sd++) {
new_rrc_cfg[sd].reconf_blocks = new_rrc_cfg[sd].blocks;
#if 0
new_rrc_cfg[sd].reconf_blocks = new_rrc_cfg[sd].chunks
@@ -247,7 +247,7 @@
if (
(ret =
- setup_free_blocks (old_md_cfg->array.param.nr_disks,
+ setup_free_blocks (old_md_cfg->array.param.raid_disks,
old_rrc_cfg))) return ret;
if ((ret = initialize_unique_disks ()))
diff -ur raidtools-1.00.3/rrc_common.c raidtools-1.00.3+fixes/rrc_common.c
--- raidtools-1.00.3/rrc_common.c Wed Jan 15 08:58:25 2003
+++ raidtools-1.00.3+fixes/rrc_common.c Wed Dec 3 02:11:26 2003
@@ -434,7 +434,7 @@
/*
* This may be a request to free a block from the source disks,
where the
- * corrosponding location on the sink disks is not yet free.
+ * corresponding location on the sink disks is not yet free.
* The common reader will detect this case and bring two gifts
from only
* one wish. So, worst case, we will end up with twice as many gifts
* as we had wishes.
@@ -457,7 +457,7 @@
unsigned long ret = 0;
int d;
- for (d = 0; d != old_md_cfg->array.param.nr_disks; d++) {
+ for (d = 0; d != old_md_cfg->array.param.raid_disks; d++) {
ret += nr_free_disk_blocks[d];
/* fprintf(stderr, "\nFree blocks in disk %i = %lu\n", d,
nr_free_disk_blocks[d]); */
}
@@ -470,7 +470,7 @@
{
int d;
- for (d = 0; d != old_md_cfg->array.param.nr_disks; d++)
+ for (d = 0; d != old_md_cfg->array.param.raid_disks; d++)
if (old_rrc_cfg[d].disk_id == diskid)
return 1;
return 0;
@@ -482,10 +482,10 @@
int disk;
unsigned long blk, ofs;
- for (disk = 0; disk != old_md_cfg->array.param.nr_disks; disk++)
+ for (disk = 0; disk != old_md_cfg->array.param.raid_disks; disk++)
if (old_rrc_cfg[disk].disk_id == diskid)
break;
- if (disk == old_md_cfg->array.param.nr_disks) {
+ if (disk == old_md_cfg->array.param.raid_disks) {
/* disk is not in source, so block is free per definition */
return 1;
}
@@ -517,10 +517,10 @@
int disk;
unsigned long blk, ofs;
- for (disk = 0; disk != old_md_cfg->array.param.nr_disks; disk++)
+ for (disk = 0; disk != old_md_cfg->array.param.raid_disks; disk++)
if (old_rrc_cfg[disk].disk_id == diskid)
break;
- if (disk == old_md_cfg->array.param.nr_disks) {
+ if (disk == old_md_cfg->array.param.raid_disks) {
/* disk is not in source, so block is free per definition */
fprintf (stderr, "\ndisk_id %i not in source\n", diskid);
return;
@@ -582,10 +582,10 @@
int disk;
unsigned long blk, ofs;
- for (disk = 0; disk != old_md_cfg->array.param.nr_disks; disk++)
+ for (disk = 0; disk != old_md_cfg->array.param.raid_disks; disk++)
if (old_rrc_cfg[disk].disk_id == diskid)
break;
- if (disk == old_md_cfg->array.param.nr_disks) {
+ if (disk == old_md_cfg->array.param.raid_disks) {
/* disk is not in source, so block cannot be unfreed
* This is an acceptable error, (in order to keep the
unfree_all_blocks() routines simple
*/
@@ -626,7 +626,7 @@
{
int disk;
- for (disk = 0; disk != old_md_cfg->array.param.nr_disks; disk++) {
+ for (disk = 0; disk != old_md_cfg->array.param.raid_disks; disk++) {
unsigned long blk;
fprintf (stderr, "\nUnfree blocks on disk %i\n", disk);
@@ -673,7 +673,7 @@
const char *ret = 0;
/* Traverse sink disks */
- for (dsk = 0; dsk != new_md_cfg->array.param.nr_disks; dsk++) {
+ for (dsk = 0; dsk != new_md_cfg->array.param.raid_disks; dsk++) {
fulfilled_t *gift =
gift_list_sink_diskid (new_rrc_cfg[dsk].disk_id);
@@ -737,7 +737,7 @@
* to know that the write has completed, if we will be doing more
* sophisticated requesting in the future
*/
- for (dsk = 0; dsk != new_md_cfg->array.param.nr_disks; dsk++)
+ for (dsk = 0; dsk != new_md_cfg->array.param.raid_disks; dsk++)
fsync (new_rrc_cfg[dsk].fd);
return ret;
@@ -902,11 +902,11 @@
cdepth++;
/* Calculate block and disk numbers */
source_block = block * reconf_block_size;
- for (dsk = 0; dsk != old_md_cfg->array.param.nr_disks;
+ for (dsk = 0; dsk != old_md_cfg->array.param.raid_disks;
dsk++)
if (old_rrc_cfg[dsk].disk_id == diskid)
break;
- if (dsk == old_md_cfg->array.param.nr_disks) {
+ if (dsk == old_md_cfg->array.param.raid_disks) {
fprintf (stderr,
"Diskid %i is not in source, when freeing
blocks and their friends...\n",
diskid);
diff -ur raidtools-1.00.3/rrc_raid5.c raidtools-1.00.3+fixes/rrc_raid5.c
--- raidtools-1.00.3/rrc_raid5.c Wed Jan 15 08:58:25 2003
+++ raidtools-1.00.3+fixes/rrc_raid5.c Wed Dec 3 02:11:26 2003
@@ -39,10 +39,10 @@
if (cfg->array.param.level != 5)
return ("Wrong level for RAID-5 Driver");
- if (cfg->array.param.nr_disks < 3)
+ if (cfg->array.param.raid_disks < 3)
return "Too few disks for RAID5";
- ctx->raid_disks = cfg->array.param.nr_disks;
+ ctx->raid_disks = cfg->array.param.raid_disks;
ctx->data_disks = ctx->raid_disks-1;
ctx->chunk_size = cfg->array.param.chunk_size;
ctx->algorithm = cfg->array.param.layout;
@@ -51,11 +51,16 @@
/* sum the total blocks in the array */
ctx->tot_chunks = 0;
- for (d = 0; d < ctx->raid_disks; ++d)
- ctx->tot_chunks += ctx->disks[d].chunks;
+ for (d = 0; d < cfg->array.param.nr_disks; ++d) {
+ if ((!ctx->tot_chunks) || (ctx->disks[d].chunks <
ctx->tot_chunks)) {
+ ctx->tot_chunks = ctx->disks[d].chunks;
+ }
+ }
+ fprintf(stderr, "smallest %lu\n", ctx->tot_chunks);
/* RAID5 uses one disk's worth of data for parity, so subtract it */
- ctx->tot_chunks -= ctx->disks[0].chunks;
+ ctx->tot_chunks *= (ctx->raid_disks - 1);
+ fprintf(stderr, "array total %lu\n", ctx->tot_chunks);
/* set up the disk_id -> disks[] mapping */
for (d = 0; d < ctx->raid_disks; ++d)
@@ -84,8 +89,9 @@
for (d = 0; d < ctx->raid_disks; ++d)
ctx->disk_total_blocks[d] =
- cfgdisks[d].chunks * (ctx->chunk_size / MD_BLK_SIZ) /
- reconf_block_size;
+ (ctx->tot_chunks / (ctx->raid_disks - 1)) *
+ (ctx->chunk_size / MD_BLK_SIZ) /
+ reconf_block_size;
return NULL;
}
there are a couple of debug lines still in there, but they only dump
smallest disk found + total raid blocks to stderr.
regards
Phil.
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
