Inconsistancies between 1.0.0.rc14 and 1.0.0.rc15 and Intel Software RAID metadata.
Luke Yelavich
luke.yelavich at canonical.com
Mon Jan 12 01:20:43 UTC 2009
When doing some testing with dmraid 1.0.0.rc14 and 1.0.0.rc15, I discovered some odd behavior. My testing involved creating a RAID5 array on 3 250GB SATA hard disks, connected to the Intel SATA
controller on a Gigabyte EP35-DS3R., which is an ICH9R chip. (See below for lspci output for this device.) The board is not running the latest BIOS, however newer BIOS revisions do not have changes
relating to the Intel Matrix Software RAID configuration. This BIOS has the Intel Matrix Option ROM Manager v7.5.0.1017, at least any such changes are not listed in the changelog on the Gigabyte
website.
So I created the array, 3 250GB Samsung hard disks, RAID5, 64K stripe size. Dmraid 1.0.0.rc14 managed to identify, and activate the array, however dmraid 1.0.0.rc15 was unable to do so. When
attempting to run dmraid -r with rc15, I got the following output:
ERROR: isw: Could not find disk /dev/sdd in the metadata
ERROR: isw: Could not find disk /dev/sdc in the metadata
ERROR: isw: Could not find disk /dev/sdb in the metadata
no raid disks
NOTE: I have a 4th drive on sda, with Linux installed.
The testing was performed on an Ubuntu intrepid system, with somewhat patched isw code. (See two attached patches). 1.0.0.rc15 has no patched isw code. I have not yet tried rc14 with the 2 attached
patches removed, that is my next step. The patches were originally taken from Mandriva's dmraid rc14 package.
I also performed a metadata dump with rc14, (see attached tarball), and whiped the metadata from the disks, and replaced it using the for loop in the manpage. It seems that rc14 did not get a
complete dump, as when I attempted to run dmraid -r, I got the no disks error.
Finally, I created isw metadata using dmraid rc15 on the same 3 disks, and my BIOS was unable to recognise the metadata.
lspci output: 00:1f.2 SATA controller: Intel Corporation 82801IR/IO/IH (ICH9R/DO/DH) 6 port SATA AHCI Controller (rev 02)
I think this is a big regression in rc15.
Further debugging suggestions welcome, so the issue can be fixed.
Luke
Ubuntu dmraid maintainer.
-------------- next part --------------
#! /bin/sh /usr/share/dpatch/dpatch-run
## 06_isw-metadata-fixes.dpatch by Luke Yelavich <themuso at ubuntu.com>
##
## All lines beginning with `## DP:' are a description of the patch.
## DP: Intel Software RAID metadata fixes.
@DPATCH@
diff -urNad dmraid-1.0.0.rc14-1ubuntu1~/1.0.0.rc14/lib/format/ataraid/isw.c dmraid-1.0.0.rc14-1ubuntu1/1.0.0.rc14/lib/format/ataraid/isw.c
--- dmraid-1.0.0.rc14-1ubuntu1~/1.0.0.rc14/lib/format/ataraid/isw.c 2006-09-23 04:11:37.000000000 +1000
+++ dmraid-1.0.0.rc14-1ubuntu1/1.0.0.rc14/lib/format/ataraid/isw.c 2008-07-08 14:15:28.000000000 +1000
@@ -364,21 +364,26 @@
{
int ret;
struct isw *isw = META(rd, isw);
+ int large = div_up(isw->mpb_size, ISW_DISK_BLOCK_SIZE) > 1;
to_disk(isw, FULL);
- /*
- * Copy 1st metadata sector to after the extended ones
- * and increment metadata area pointer by one block, so
- * that the metadata is filed in the proper sequence.
- */
- memcpy((void*) isw + rd->meta_areas->size, isw, ISW_DISK_BLOCK_SIZE);
- rd->meta_areas->area += ISW_DISK_BLOCK_SIZE;
+ if (large) {
+ /*
+ * Copy 1st metadata sector to after the extended ones
+ * and increment metadata area pointer by one block, so
+ * that the metadata is filed in the proper sequence.
+ */
+ memcpy((void*) isw + rd->meta_areas->size, isw,
+ ISW_DISK_BLOCK_SIZE);
+ rd->meta_areas->area += ISW_DISK_BLOCK_SIZE;
+ }
ret = write_metadata(lc, handler, rd, -1, erase);
/* Correct metadata area pointer. */
- rd->meta_areas->area -= ISW_DISK_BLOCK_SIZE;
+ if (large)
+ rd->meta_areas->area -= ISW_DISK_BLOCK_SIZE;
to_cpu(isw, FULL);
-------------- next part --------------
#! /bin/sh /usr/share/dpatch/dpatch-run
## 07_isw-raid10-nested.dpatch by Luke Yelavich <themuso at ubuntu.com>
##
## All lines beginning with `## DP:' are a description of the patch.
## DP: Add RAID10(0+1) nested RAID level support for Intel Software RAID.
@DPATCH@
diff -urNad dmraid-1.0.0.rc14-1ubuntu1~/1.0.0.rc14/lib/format/ataraid/isw.c dmraid-1.0.0.rc14-1ubuntu1/1.0.0.rc14/lib/format/ataraid/isw.c
--- dmraid-1.0.0.rc14-1ubuntu1~/1.0.0.rc14/lib/format/ataraid/isw.c 2008-07-08 14:18:03.000000000 +1000
+++ dmraid-1.0.0.rc14-1ubuntu1/1.0.0.rc14/lib/format/ataraid/isw.c 2008-07-08 14:18:23.000000000 +1000
@@ -25,30 +25,59 @@
#endif
static const char *handler = HANDLER;
+static struct isw_disk *_get_disk(struct isw *isw, struct dev_info *di);
+int is_raid10(struct isw *isw);
-/*
- * Make up RAID set name from family_num and volume name.
- */
-static size_t _name(struct isw *isw, struct isw_dev *dev,
- char *str, size_t len)
+static size_t _name(struct isw *isw, char *str, size_t len, int subset,
+ int num, struct isw_dev *dev)
{
- return snprintf(str, len, dev ? "isw_%u_%s" : "isw_%u",
+ switch(subset) {
+ case 2:
+ return snprintf(str, len, "isw_%u", isw->family_num);
+ case 1:
+ if(!is_raid10(isw))
+ return snprintf(str, len, "isw_%u_%s",
isw->family_num, (char*) dev->volume);
+ else
+ return snprintf(str, len, "isw_%u_%s-%u",
+ isw->family_num, (char*) dev->volume, num);
+ case 0:
+ return snprintf(str, len, "isw_%u_%s", isw->family_num,
+ (char*) dev->volume);
+ }
+ return 0;
}
-static char *name(struct lib_context *lc, struct isw *isw, struct isw_dev *dev)
+static char *name(struct lib_context *lc, struct raid_dev *rd,
+ unsigned int subset)
{
- size_t len;
- char *ret;
+ size_t len;
+ char *ret = NULL;
+ int id = 0;
+ struct isw *isw = META(rd, isw);
+ struct isw_disk *disk = isw->disk;
+ struct isw_dev *dev = (struct isw_dev*) (isw->disk + isw->num_disks);
- if ((ret = dbg_malloc((len = _name(isw, dev, NULL, 0) + 1)))) {
- _name(isw, dev, ret, len);
- mk_alpha(lc, ret + HANDLER_LEN, len - HANDLER_LEN -
- (dev ? strlen((char*) dev->volume) - 2 : 1));
- } else
+ if((subset == 1) && (is_raid10(isw))) {
+ if ((disk = _get_disk(isw, rd->di))) {
+ if(disk == isw->disk) id = 0;
+ else if(disk == isw->disk + 1) id = 1;
+ else if(disk == isw->disk + 2) id = 2;
+ else if(disk == isw->disk + 3) id = 3;
+ else return ret;
+ }
+ id = id % 2;
+ }
+
+ if ((ret = dbg_malloc((len = _name(isw, ret, 0, subset, id,
+ dev) + 1)))) {
+ _name(isw, ret, len, subset, id, dev);
+ mk_alpha(lc, ret + HANDLER_LEN, snprintf(ret, 0, "%u",
+ isw->family_num));
+ } else
log_alloc_err(lc, handler);
- return ret;
+ return ret;
}
/* Find a disk table slot by serial number. */
@@ -59,7 +88,7 @@
do {
if (!strncmp(di->serial, (const char*) disk->serial,
- MAX_RAID_SERIAL_LEN))
+ MAX_RAID_SERIAL_LEN))
return disk;
} while (++disk < isw->disk + isw->num_disks);
}
@@ -102,21 +131,38 @@
}
/* Neutralize disk type. */
-static enum type type(struct raid_dev *rd)
+static enum type type(struct isw *isw)
{
/* Mapping of Intel types to generic types. */
static struct types types[] = {
{ ISW_T_RAID0, t_raid0},
{ ISW_T_RAID1, t_raid1},
{ ISW_T_RAID5, t_raid5_la},
+ { ISW_T_RAID10, t_raid0},
{ 0, t_undef},
};
- struct isw_dev *dev = rd->private.ptr;
+
+ struct isw_dev *dev = (struct isw_dev*) (isw->disk + isw->num_disks);
+
+ if(is_raid10(isw))
+ dev->vol.map.raid_level = ISW_T_RAID10;
return dev ? rd_type(types, (unsigned int) dev->vol.map.raid_level) :
t_group;
}
+int is_raid10(struct isw *isw)
+{
+ int ret;
+ struct isw_dev *dev = (struct isw_dev*) (isw->disk + isw->num_disks);
+
+ ret = ((dev->vol.map.raid_level == ISW_T_RAID10) ||
+ (dev->vol.map.raid_level == ISW_T_RAID1 &&
+ isw->num_disks > 3));
+
+ return ret;
+}
+
/*
* Generate checksum of Raid metadata for mpb_size/sizeof(u32) words
* (checksum field itself ignored for this calculation).
@@ -298,7 +344,8 @@
}
/* Check for RAID disk ok. */
-static int disk_ok(struct lib_context *lc, struct dev_info *di, struct isw *isw)
+static int disk_ok(struct lib_context *lc, struct dev_info *di,
+ struct isw *isw)
{
struct isw_disk *disk = get_disk(lc, di, isw);
@@ -306,15 +353,15 @@
}
static void *isw_read_metadata(struct lib_context *lc, struct dev_info *di,
- size_t *sz, uint64_t *offset,
- union read_info *info)
+ size_t *sz, uint64_t *offset,
+ union read_info *info)
{
size_t size = ISW_DISK_BLOCK_SIZE;
uint64_t isw_sboffset = ISW_CONFIGOFFSET;
struct isw *isw;
if (!(isw = alloc_private_and_read(lc, handler, size,
- di->path, isw_sboffset)))
+ di->path, isw_sboffset)))
goto out;
/*
@@ -396,7 +443,7 @@
*/
/* Check state if isw device map. */
static int _check_map_state(struct lib_context *lc, struct raid_dev *rd,
- struct isw_dev *dev)
+ struct isw_dev *dev)
{
/* FIXME: FAILED_MAP etc. */
switch (dev->vol.map.map_state) {
@@ -420,20 +467,27 @@
struct raid_dev *r;
if (!_check_map_state(lc, rd, dev) ||
- !(r = alloc_raid_dev(lc, handler)))
+ !(r = alloc_raid_dev(lc, handler)))
return NULL;
if (!(r->private.ptr = alloc_private(lc, handler, sizeof(*dev))))
goto free;
-
memcpy(r->private.ptr, dev, sizeof(*dev));
- if ((r->type = type(r)) == t_undef) {
+
+ if (!(r->meta_areas = alloc_meta_areas(lc, rd, handler, 1)))
+ return 0;
+
+ r->meta_areas->offset = rd->meta_areas->offset;
+ r->meta_areas->size = rd->meta_areas->size;
+ r->meta_areas->area = rd->meta_areas->area;
+
+ if ((r->type = type(isw)) == t_undef) {
log_err(lc, "%s: RAID type %u not supported",
handler, (unsigned int) dev->vol.map.raid_level);
goto free;
}
- if (!(r->name = name(lc, isw, dev)))
+ if (!(r->name = name(lc, rd, 1)))
goto free;
r->di = rd->di;
@@ -465,43 +519,73 @@
return _get_disk(isw, RD(new)->di) < _get_disk(isw, RD(pos)->di);
}
+static void super_created(struct raid_set *super, void *private)
+{
+ super->type = t_raid1;
+ super->stride = ((struct isw_dev*) private)->vol.map.blocks_per_strip;
+}
+
+static int set_sort(struct list_head *pos, struct list_head *new)
+{
+ return 0;
+}
+
/*
* rs_group contains the top-level group RAID set (type: t_group) on entry
* and shall be returned on success (or NULL on error).
*/
static struct raid_set *group_rd(struct lib_context *lc,
- struct raid_set *rs_group,
- struct raid_dev *rd_meta)
+ struct raid_set *rs_group,
+ struct raid_dev *rd_meta)
{
unsigned int d;
void *private;
struct isw *isw = META(rd_meta, isw);
struct isw_dev *dev;
struct raid_dev *rd;
- struct raid_set *rs;
-
+ struct raid_set *rs, *ss;
+ char *nm = NULL;
+
/* Loop the device/volume table. */
for (d = 0; d < isw->num_raid_devs; d++) {
dev = raiddev(isw, d);
-
+
if (!(rd = _create_rd(lc, rd_meta, isw, dev)))
return NULL;
- if (!(rs = find_or_alloc_raid_set(lc, rd->name, FIND_ALL,
- rd, &rs_group->sets,
- create_rs, dev))) {
- free_raid_dev(lc, &rd);
- return NULL;
+ if(is_raid10(isw)) {
+ nm = name(lc, rd, 0);
+ ss = find_or_alloc_raid_set(lc, nm, FIND_TOP, rd,
+ LC_RS(lc), super_created, dev);
+ if (!(rs = find_or_alloc_raid_set(lc, rd->name,
+ FIND_ALL, rd, &ss->sets, create_rs,
+ dev))) {
+ free_raid_dev(lc, &rd);
+ return NULL;
+ }
+ } else {
+ if (!(rs = find_or_alloc_raid_set(lc, rd->name,
+ FIND_ALL, rd, &rs_group->sets,
+ create_rs, dev))) {
+ free_raid_dev(lc, &rd);
+ return NULL;
+ }
}
+ rs->status = s_ok;
+
/* Save and set to enable dev_sort(). */
private = rd->private.ptr;
rd->private.ptr = isw;
-
list_add_sorted(lc, &rs->devs, &rd->devs, dev_sort);
-
/* Restore. */
rd->private.ptr = private;
+
+ if(is_raid10(isw)) {
+ ss = join_superset(lc, name, super_created, set_sort,
+ rs, rd);
+ return ss;
+ }
}
return rs_group;
@@ -509,31 +593,35 @@
/* Add an Intel SW RAID device to a set */
static struct raid_set *isw_group(struct lib_context *lc,
- struct raid_dev *rd_meta)
+ struct raid_dev *rd_meta)
{
- struct raid_set *rs_group;
+ struct raid_set *rs_group = NULL;
+ struct isw *isw = META(rd_meta, isw);
if (T_SPARE(rd_meta))
return NULL;
- /*
- * Once we get here, an Intel SW RAID disk containing a metadata area
- * with a volume table has been discovered by isw_read.
- */
- /* Check if a top level group RAID set already exists. */
- if (!(rs_group = find_or_alloc_raid_set(lc, rd_meta->name, FIND_TOP,
- rd_meta, LC_RS(lc),
- NO_CREATE, NO_CREATE_ARG)))
- return NULL;
+ if(!is_raid10(isw)) {
+ /*
+ * Once we get here, an Intel SW RAID disk containing a metadata
+ * area with a volume table has been discovered by isw_read.
+ */
+ /* Check if a top level group RAID set already exists. */
+ if (!(rs_group = find_or_alloc_raid_set(lc, rd_meta->name,
+ FIND_TOP, rd_meta, LC_RS(lc),
+ NO_CREATE, NO_CREATE_ARG)))
+ return NULL;
- /*
- * Add the whole underlying (meta) RAID device to the group set.
- * Sorting is no problem here, because RAID sets and devices will
- * be created for all the Volumes of an ISW set and those need sorting.
- */
- rd_meta->private.ptr = rd_meta->meta_areas->area;
- list_add_sorted(lc, &rs_group->devs, &rd_meta->devs, dev_sort);
- rd_meta->private.ptr = NULL;
+ /*
+ * Add the whole underlying (meta) RAID device to the group set.
+ * Sorting is no problem here, because RAID sets and devices
+ * will be created for all the Volumes of an ISW set and those
+ * need sorting.
+ */
+ rd_meta->private.ptr = rd_meta->meta_areas->area;
+ list_add_sorted(lc, &rs_group->devs, &rd_meta->devs, dev_sort);
+ rd_meta->private.ptr = NULL;
+ }
/*
* We need to run through the volume table and create a RAID set and
@@ -557,17 +645,16 @@
}
static int check_rd(struct lib_context *lc, struct raid_set *rs,
- struct raid_dev *rd, void *context)
+ struct raid_dev *rd, void *context)
{
struct isw_dev *dev = rd->private.ptr;
/* FIXME: more status checks ? */
if (dev->status)
LOG_ERR(lc, 0, "%s device for volume \"%s\" broken on %s "
- "in RAID set \"%s\"",
- handler, dev->volume, rd->di->path, rs->name);
-
- return 1;
+ "in RAID set \"%s\"", handler, dev->volume,
+ rd->di->path, rs->name);
+ return 1;
}
static int _isw_check(struct lib_context *lc, struct raid_set *rs)
@@ -577,9 +664,14 @@
static int isw_check(struct lib_context *lc, struct raid_set *rs)
{
- return T_GROUP(rs) ? _isw_check(lc, rs) : 0;
+ /* If it is a stacked set like RAID10 */
+ if((!T_GROUP(rs)) && SETS(rs)) {
+ return check_raid_set(lc, rs, devices, NULL,
+ NO_CHECK_RD, NULL, handler);
+ } else {
+ return T_GROUP(rs) ? _isw_check(lc, rs) : 0;
+ }
}
-
/*
* IO error event handler.
*/
@@ -590,7 +682,7 @@
struct isw_disk *disk;
if (!(disk = get_disk(lc, rd->di, isw)))
- LOG_ERR(lc, 0, "%s: disk", handler);
+ LOG_ERR(lc, 0, "%s: disk", handler);
/* Avoid write trashing. */
if (S_BROKEN(status(lc, rd)))
@@ -634,7 +726,7 @@
for (i = 0; i < ISW_FILLERS; i++) {
if (isw->filler[i])
P("filler[%i]: %u", isw,
- isw->filler[i], i, isw->filler[i]);
+ isw->filler[i], i, isw->filler[i]);
}
/* Disk table. */
@@ -749,7 +841,7 @@
static struct dmraid_format isw_format = {
.name = HANDLER,
.descr = "Intel Software RAID",
- .caps = "0,1",
+ .caps = "0,1,10",
.format = FMT_RAID,
.read = isw_read,
.write = isw_write,
@@ -802,5 +894,7 @@
rd->status = status(lc, rd);
rd->type = t_group;
- return (rd->name = name(lc, isw, NULL)) ? 1 : 0;
+ return (rd->name = name(lc, rd, 2)) ? 1 : 0;
}
+
+
diff -urNad dmraid-1.0.0.rc14-1ubuntu1~/1.0.0.rc14/lib/format/ataraid/isw.h dmraid-1.0.0.rc14-1ubuntu1/1.0.0.rc14/lib/format/ataraid/isw.h
--- dmraid-1.0.0.rc14-1ubuntu1~/1.0.0.rc14/lib/format/ataraid/isw.h 2006-01-25 02:02:54.000000000 +1100
+++ dmraid-1.0.0.rc14-1ubuntu1/1.0.0.rc14/lib/format/ataraid/isw.h 2008-07-08 14:18:28.000000000 +1000
@@ -68,6 +68,7 @@
uint8_t raid_level;
#define ISW_T_RAID0 0
#define ISW_T_RAID1 1
+#define ISW_T_RAID10 2
#define ISW_T_RAID5 5 // since metadata version 1.2.02 ?
uint8_t num_members; // number of member disks
uint8_t reserved[3];
-------------- next part --------------
A non-text attachment was scrubbed...
Name: isw-metadata.tar.gz
Type: application/octet-stream
Size: 445 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/ataraid-list/attachments/20090112/9b170718/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: Digital signature
URL: <http://listman.redhat.com/archives/ataraid-list/attachments/20090112/9b170718/attachment.sig>
More information about the Ataraid-list
mailing list