2004-06-12 21:16:10 +00:00
|
|
|
/*-
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
* Copyright (c) 2004, 2007 Lukas Ertl
|
|
|
|
* Copyright (c) 2007, 2009 Ulf Lilleengen
|
2004-06-12 21:16:10 +00:00
|
|
|
* Copyright (c) 1997, 1998, 1999
|
|
|
|
* Nan Yang Computer Services Limited. All rights reserved.
|
|
|
|
*
|
|
|
|
* Parts written by Greg Lehey
|
|
|
|
*
|
|
|
|
* This software is distributed under the so-called ``Berkeley
|
|
|
|
* License'':
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
* 3. All advertising materials mentioning features or use of this software
|
|
|
|
* must display the following acknowledgement:
|
|
|
|
* This product includes software developed by Nan Yang Computer
|
|
|
|
* Services Limited.
|
|
|
|
* 4. Neither the name of the Company nor the names of its contributors
|
|
|
|
* may be used to endorse or promote products derived from this software
|
|
|
|
* without specific prior written permission.
|
|
|
|
*
|
|
|
|
* This software is provided ``as is'', and any express or implied
|
|
|
|
* warranties, including, but not limited to, the implied warranties of
|
|
|
|
* merchantability and fitness for a particular purpose are disclaimed.
|
|
|
|
* In no event shall the company or contributors be liable for any
|
|
|
|
* direct, indirect, incidental, special, exemplary, or consequential
|
|
|
|
* damages (including, but not limited to, procurement of substitute
|
|
|
|
* goods or services; loss of use, data, or profits; or business
|
|
|
|
* interruption) however caused and on any theory of liability, whether
|
|
|
|
* in contract, strict liability, or tort (including negligence or
|
|
|
|
* otherwise) arising in any way out of the use of this software, even if
|
|
|
|
* advised of the possibility of such damage.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
|
|
|
#include <sys/param.h>
|
|
|
|
#include <sys/malloc.h>
|
|
|
|
#include <sys/systm.h>
|
|
|
|
|
|
|
|
#include <geom/geom.h>
|
|
|
|
#include <geom/vinum/geom_vinum_var.h>
|
|
|
|
#include <geom/vinum/geom_vinum.h>
|
|
|
|
#include <geom/vinum/geom_vinum_share.h>
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
int gv_drive_is_newer(struct gv_softc *, struct gv_drive *);
|
|
|
|
static off_t gv_plex_smallest_sd(struct gv_plex *);
|
2007-04-12 17:54:35 +00:00
|
|
|
|
2004-06-12 21:16:10 +00:00
|
|
|
void
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_parse_config(struct gv_softc *sc, char *buf, struct gv_drive *d)
|
2004-06-12 21:16:10 +00:00
|
|
|
{
|
|
|
|
char *aptr, *bptr, *cptr;
|
|
|
|
struct gv_volume *v, *v2;
|
|
|
|
struct gv_plex *p, *p2;
|
|
|
|
struct gv_sd *s, *s2;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
int error, is_newer, tokens;
|
2004-06-12 21:16:10 +00:00
|
|
|
char *token[GV_MAXARGS];
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
is_newer = gv_drive_is_newer(sc, d);
|
2004-06-12 21:16:10 +00:00
|
|
|
|
|
|
|
/* Until the end of the string *buf. */
|
|
|
|
for (aptr = buf; *aptr != '\0'; aptr = bptr) {
|
|
|
|
bptr = aptr;
|
|
|
|
cptr = aptr;
|
|
|
|
|
|
|
|
/* Seperate input lines. */
|
|
|
|
while (*bptr != '\n')
|
|
|
|
bptr++;
|
|
|
|
*bptr = '\0';
|
|
|
|
bptr++;
|
|
|
|
|
|
|
|
tokens = gv_tokenize(cptr, token, GV_MAXARGS);
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (tokens <= 0)
|
|
|
|
continue;
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (!strcmp(token[0], "volume")) {
|
|
|
|
v = gv_new_volume(tokens, token);
|
|
|
|
if (v == NULL) {
|
|
|
|
G_VINUM_DEBUG(0, "config parse failed volume");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
v2 = gv_find_vol(sc, v->name);
|
|
|
|
if (v2 != NULL) {
|
|
|
|
if (is_newer) {
|
|
|
|
v2->state = v->state;
|
|
|
|
G_VINUM_DEBUG(2, "newer volume found!");
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
g_free(v);
|
|
|
|
continue;
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_create_volume(sc, v);
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
} else if (!strcmp(token[0], "plex")) {
|
|
|
|
p = gv_new_plex(tokens, token);
|
|
|
|
if (p == NULL) {
|
|
|
|
G_VINUM_DEBUG(0, "config parse failed plex");
|
|
|
|
break;
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
p2 = gv_find_plex(sc, p->name);
|
|
|
|
if (p2 != NULL) {
|
|
|
|
/* XXX */
|
|
|
|
if (is_newer) {
|
|
|
|
p2->state = p->state;
|
|
|
|
G_VINUM_DEBUG(2, "newer plex found!");
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
g_free(p);
|
|
|
|
continue;
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
error = gv_create_plex(sc, p);
|
|
|
|
if (error)
|
|
|
|
continue;
|
|
|
|
/*
|
|
|
|
* These flags were set in gv_create_plex() and are not
|
|
|
|
* needed here (on-disk config parsing).
|
|
|
|
*/
|
|
|
|
p->flags &= ~GV_PLEX_ADDED;
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
} else if (!strcmp(token[0], "sd")) {
|
|
|
|
s = gv_new_sd(tokens, token);
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (s == NULL) {
|
|
|
|
G_VINUM_DEBUG(0, "config parse failed subdisk");
|
|
|
|
break;
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
s2 = gv_find_sd(sc, s->name);
|
|
|
|
if (s2 != NULL) {
|
|
|
|
/* XXX */
|
|
|
|
if (is_newer) {
|
|
|
|
s2->state = s->state;
|
|
|
|
G_VINUM_DEBUG(2, "newer subdisk found!");
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
g_free(s);
|
|
|
|
continue;
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Signal that this subdisk was tasted, and could
|
|
|
|
* possibly reference a drive that isn't in our config
|
|
|
|
* yet.
|
|
|
|
*/
|
|
|
|
s->flags |= GV_SD_TASTED;
|
|
|
|
|
|
|
|
if (s->state == GV_SD_UP)
|
|
|
|
s->flags |= GV_SD_CANGOUP;
|
|
|
|
|
|
|
|
error = gv_create_sd(sc, s);
|
|
|
|
if (error)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This flag was set in gv_create_sd() and is not
|
|
|
|
* needed here (on-disk config parsing).
|
|
|
|
*/
|
|
|
|
s->flags &= ~GV_SD_NEWBORN;
|
|
|
|
s->flags &= ~GV_SD_GROW;
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Format the vinum configuration properly. If ondisk is non-zero then the
|
|
|
|
* configuration is intended to be written to disk later.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
gv_format_config(struct gv_softc *sc, struct sbuf *sb, int ondisk, char *prefix)
|
|
|
|
{
|
|
|
|
struct gv_drive *d;
|
|
|
|
struct gv_sd *s;
|
|
|
|
struct gv_plex *p;
|
|
|
|
struct gv_volume *v;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We don't need the drive configuration if we're not writing the
|
|
|
|
* config to disk.
|
|
|
|
*/
|
|
|
|
if (!ondisk) {
|
|
|
|
LIST_FOREACH(d, &sc->drives, drive) {
|
2004-10-02 20:12:20 +00:00
|
|
|
sbuf_printf(sb, "%sdrive %s device /dev/%s\n", prefix,
|
2004-06-12 21:16:10 +00:00
|
|
|
d->name, d->device);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(v, &sc->volumes, volume) {
|
|
|
|
if (!ondisk)
|
|
|
|
sbuf_printf(sb, "%s", prefix);
|
|
|
|
sbuf_printf(sb, "volume %s", v->name);
|
|
|
|
if (ondisk)
|
|
|
|
sbuf_printf(sb, " state %s", gv_volstate(v->state));
|
|
|
|
sbuf_printf(sb, "\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(p, &sc->plexes, plex) {
|
|
|
|
if (!ondisk)
|
|
|
|
sbuf_printf(sb, "%s", prefix);
|
|
|
|
sbuf_printf(sb, "plex name %s org %s ", p->name,
|
|
|
|
gv_plexorg(p->org));
|
|
|
|
if (gv_is_striped(p))
|
|
|
|
sbuf_printf(sb, "%ds ", p->stripesize / 512);
|
|
|
|
if (p->vol_sc != NULL)
|
|
|
|
sbuf_printf(sb, "vol %s", p->volume);
|
|
|
|
if (ondisk)
|
|
|
|
sbuf_printf(sb, " state %s", gv_plexstate(p->state));
|
|
|
|
sbuf_printf(sb, "\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(s, &sc->subdisks, sd) {
|
|
|
|
if (!ondisk)
|
|
|
|
sbuf_printf(sb, "%s", prefix);
|
|
|
|
sbuf_printf(sb, "sd name %s drive %s len %jds driveoffset "
|
|
|
|
"%jds", s->name, s->drive, s->size / 512,
|
|
|
|
s->drive_offset / 512);
|
|
|
|
if (s->plex_sc != NULL) {
|
|
|
|
sbuf_printf(sb, " plex %s plexoffset %jds", s->plex,
|
|
|
|
s->plex_offset / 512);
|
|
|
|
}
|
|
|
|
if (ondisk)
|
|
|
|
sbuf_printf(sb, " state %s", gv_sdstate(s->state));
|
|
|
|
sbuf_printf(sb, "\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-04-12 17:54:35 +00:00
|
|
|
static off_t
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_plex_smallest_sd(struct gv_plex *p)
|
2007-04-12 17:54:35 +00:00
|
|
|
{
|
|
|
|
struct gv_sd *s;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
off_t smallest;
|
2007-04-12 17:54:35 +00:00
|
|
|
|
|
|
|
KASSERT(p != NULL, ("gv_plex_smallest_sd: NULL p"));
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
s = LIST_FIRST(&p->subdisks);
|
|
|
|
if (s == NULL)
|
|
|
|
return (-1);
|
|
|
|
smallest = s->size;
|
2007-04-12 17:54:35 +00:00
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex) {
|
|
|
|
if (s->size < smallest)
|
|
|
|
smallest = s->size;
|
|
|
|
}
|
|
|
|
return (smallest);
|
|
|
|
}
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Walk over plexes in a volume and count how many are down. */
|
2004-06-12 21:16:10 +00:00
|
|
|
int
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_plexdown(struct gv_volume *v)
|
2004-06-12 21:16:10 +00:00
|
|
|
{
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
int plexdown;
|
|
|
|
struct gv_plex *p;
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
KASSERT(v != NULL, ("gv_plexdown: NULL v"));
|
|
|
|
|
|
|
|
plexdown = 0;
|
|
|
|
|
|
|
|
LIST_FOREACH(p, &v->plexes, plex) {
|
|
|
|
if (p->state == GV_PLEX_DOWN)
|
|
|
|
plexdown++;
|
|
|
|
}
|
|
|
|
return (plexdown);
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
gv_sd_to_plex(struct gv_sd *s, struct gv_plex *p)
|
|
|
|
{
|
|
|
|
struct gv_sd *s2;
|
|
|
|
off_t psizeorig, remainder, smallest;
|
2004-06-12 21:16:10 +00:00
|
|
|
|
|
|
|
/* If this subdisk was already given to this plex, do nothing. */
|
|
|
|
if (s->plex_sc == p)
|
|
|
|
return (0);
|
|
|
|
|
2006-03-30 14:01:25 +00:00
|
|
|
/* Check correct size of this subdisk. */
|
|
|
|
s2 = LIST_FIRST(&p->subdisks);
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Adjust the subdisk-size if necessary. */
|
|
|
|
if (s2 != NULL && gv_is_striped(p)) {
|
|
|
|
/* First adjust to the stripesize. */
|
|
|
|
remainder = s->size % p->stripesize;
|
|
|
|
|
|
|
|
if (remainder) {
|
|
|
|
G_VINUM_DEBUG(1, "size of sd %s is not a "
|
|
|
|
"multiple of plex stripesize, taking off "
|
|
|
|
"%jd bytes", s->name,
|
|
|
|
(intmax_t)remainder);
|
|
|
|
gv_adjust_freespace(s, remainder);
|
|
|
|
}
|
|
|
|
|
|
|
|
smallest = gv_plex_smallest_sd(p);
|
|
|
|
/* Then take off extra if other subdisks are smaller. */
|
|
|
|
remainder = s->size - smallest;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Don't allow a remainder below zero for running plexes, it's too
|
|
|
|
* painful, and if someone were to accidentally do this, the
|
|
|
|
* resulting array might be smaller than the original... not god
|
|
|
|
*/
|
|
|
|
if (remainder < 0) {
|
|
|
|
if (!(p->flags & GV_PLEX_NEWBORN)) {
|
|
|
|
G_VINUM_DEBUG(0, "sd %s too small for plex %s!",
|
|
|
|
s->name, p->name);
|
|
|
|
return (GV_ERR_BADSIZE);
|
|
|
|
}
|
|
|
|
/* Adjust other subdisks. */
|
|
|
|
LIST_FOREACH(s2, &p->subdisks, in_plex) {
|
|
|
|
G_VINUM_DEBUG(1, "size of sd %s is to big, "
|
|
|
|
"taking off %jd bytes", s->name,
|
|
|
|
(intmax_t)remainder);
|
|
|
|
gv_adjust_freespace(s2, (remainder * -1));
|
|
|
|
}
|
|
|
|
} else if (remainder > 0) {
|
|
|
|
G_VINUM_DEBUG(1, "size of sd %s is to big, "
|
|
|
|
"taking off %jd bytes", s->name,
|
|
|
|
(intmax_t)remainder);
|
|
|
|
gv_adjust_freespace(s, remainder);
|
|
|
|
}
|
2006-03-30 14:01:25 +00:00
|
|
|
}
|
|
|
|
|
2004-06-12 21:16:10 +00:00
|
|
|
/* Find the correct plex offset for this subdisk, if needed. */
|
|
|
|
if (s->plex_offset == -1) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/*
|
|
|
|
* First set it to 0 to catch the case where we had a detached
|
|
|
|
* subdisk that didn't get any good offset.
|
|
|
|
*/
|
|
|
|
s->plex_offset = 0;
|
2004-06-12 21:16:10 +00:00
|
|
|
if (p->sdcount) {
|
|
|
|
LIST_FOREACH(s2, &p->subdisks, in_plex) {
|
|
|
|
if (gv_is_striped(p))
|
|
|
|
s->plex_offset = p->sdcount *
|
|
|
|
p->stripesize;
|
|
|
|
else
|
|
|
|
s->plex_offset = s2->plex_offset +
|
|
|
|
s2->size;
|
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* There are no subdisks for this plex yet, just insert it. */
|
|
|
|
if (LIST_EMPTY(&p->subdisks)) {
|
|
|
|
LIST_INSERT_HEAD(&p->subdisks, s, in_plex);
|
|
|
|
|
|
|
|
/* Insert in correct order, depending on plex_offset. */
|
|
|
|
} else {
|
|
|
|
LIST_FOREACH(s2, &p->subdisks, in_plex) {
|
|
|
|
if (s->plex_offset < s2->plex_offset) {
|
|
|
|
LIST_INSERT_BEFORE(s2, s, in_plex);
|
|
|
|
break;
|
|
|
|
} else if (LIST_NEXT(s2, in_plex) == NULL) {
|
|
|
|
LIST_INSERT_AFTER(s2, s, in_plex);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
s->plex_sc = p;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Adjust the size of our plex. We check if the plex misses a subdisk,
|
|
|
|
* so we don't make the plex smaller than it actually should be.
|
|
|
|
*/
|
|
|
|
psizeorig = p->size;
|
|
|
|
p->size = gv_plex_size(p);
|
|
|
|
/* Make sure the size is not changed. */
|
|
|
|
if (p->sddetached > 0) {
|
|
|
|
if (p->size < psizeorig) {
|
|
|
|
p->size = psizeorig;
|
|
|
|
/* We make sure wee need another subdisk. */
|
|
|
|
if (p->sddetached == 1)
|
|
|
|
p->sddetached++;
|
|
|
|
}
|
|
|
|
p->sddetached--;
|
|
|
|
} else {
|
|
|
|
if ((p->org == GV_PLEX_RAID5 ||
|
|
|
|
p->org == GV_PLEX_STRIPED) &&
|
|
|
|
!(p->flags & GV_PLEX_NEWBORN) &&
|
2009-05-06 19:18:19 +00:00
|
|
|
p->state == GV_PLEX_UP) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
s->flags |= GV_SD_GROW;
|
|
|
|
}
|
|
|
|
p->sdcount++;
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
2004-07-31 21:34:21 +00:00
|
|
|
void
|
|
|
|
gv_update_vol_size(struct gv_volume *v, off_t size)
|
|
|
|
{
|
|
|
|
if (v == NULL)
|
|
|
|
return;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (v->provider != NULL) {
|
|
|
|
g_topology_lock();
|
|
|
|
v->provider->mediasize = size;
|
|
|
|
g_topology_unlock();
|
|
|
|
}
|
|
|
|
v->size = size;
|
|
|
|
}
|
2004-07-31 21:34:21 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Return how many subdisks that constitute the original plex. */
|
|
|
|
int
|
|
|
|
gv_sdcount(struct gv_plex *p, int growing)
|
|
|
|
{
|
|
|
|
struct gv_sd *s;
|
|
|
|
int sdcount;
|
2004-07-31 21:34:21 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
sdcount = p->sdcount;
|
|
|
|
if (growing) {
|
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex) {
|
|
|
|
if (s->flags & GV_SD_GROW)
|
|
|
|
sdcount--;
|
|
|
|
}
|
2004-07-31 21:34:21 +00:00
|
|
|
}
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
return (sdcount);
|
2004-07-31 21:34:21 +00:00
|
|
|
}
|
|
|
|
|
2007-04-12 17:54:35 +00:00
|
|
|
/* Calculates the plex size. */
|
|
|
|
off_t
|
|
|
|
gv_plex_size(struct gv_plex *p)
|
|
|
|
{
|
|
|
|
struct gv_sd *s;
|
|
|
|
off_t size;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
int sdcount;
|
2007-04-12 17:54:35 +00:00
|
|
|
|
|
|
|
KASSERT(p != NULL, ("gv_plex_size: NULL p"));
|
|
|
|
|
|
|
|
/* Adjust the size of our plex. */
|
|
|
|
size = 0;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
sdcount = gv_sdcount(p, 1);
|
2007-04-12 17:54:35 +00:00
|
|
|
switch (p->org) {
|
|
|
|
case GV_PLEX_CONCAT:
|
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex)
|
|
|
|
size += s->size;
|
|
|
|
break;
|
|
|
|
case GV_PLEX_STRIPED:
|
|
|
|
s = LIST_FIRST(&p->subdisks);
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
size = ((s != NULL) ? (sdcount * s->size) : 0);
|
2007-04-12 17:54:35 +00:00
|
|
|
break;
|
|
|
|
case GV_PLEX_RAID5:
|
|
|
|
s = LIST_FIRST(&p->subdisks);
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
size = ((s != NULL) ? ((sdcount - 1) * s->size) : 0);
|
2007-04-12 17:54:35 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return (size);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Returns the size of a volume. */
|
|
|
|
off_t
|
|
|
|
gv_vol_size(struct gv_volume *v)
|
|
|
|
{
|
|
|
|
struct gv_plex *p;
|
|
|
|
off_t minplexsize;
|
|
|
|
|
|
|
|
KASSERT(v != NULL, ("gv_vol_size: NULL v"));
|
|
|
|
|
|
|
|
p = LIST_FIRST(&v->plexes);
|
|
|
|
if (p == NULL)
|
|
|
|
return (0);
|
|
|
|
|
|
|
|
minplexsize = p->size;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
LIST_FOREACH(p, &v->plexes, in_volume) {
|
2007-04-12 17:54:35 +00:00
|
|
|
if (p->size < minplexsize) {
|
|
|
|
minplexsize = p->size;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return (minplexsize);
|
|
|
|
}
|
|
|
|
|
2004-06-12 21:16:10 +00:00
|
|
|
void
|
|
|
|
gv_update_plex_config(struct gv_plex *p)
|
|
|
|
{
|
|
|
|
struct gv_sd *s, *s2;
|
|
|
|
off_t remainder;
|
|
|
|
int required_sds, state;
|
|
|
|
|
|
|
|
KASSERT(p != NULL, ("gv_update_plex_config: NULL p"));
|
|
|
|
|
|
|
|
/* The plex was added to an already running volume. */
|
|
|
|
if (p->flags & GV_PLEX_ADDED)
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_set_plex_state(p, GV_PLEX_DOWN, GV_SETSTATE_FORCE);
|
2004-06-12 21:16:10 +00:00
|
|
|
|
|
|
|
switch (p->org) {
|
|
|
|
case GV_PLEX_STRIPED:
|
|
|
|
required_sds = 2;
|
|
|
|
break;
|
|
|
|
case GV_PLEX_RAID5:
|
|
|
|
required_sds = 3;
|
|
|
|
break;
|
|
|
|
case GV_PLEX_CONCAT:
|
|
|
|
default:
|
|
|
|
required_sds = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (required_sds) {
|
|
|
|
if (p->sdcount < required_sds) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_set_plex_state(p, GV_PLEX_DOWN, GV_SETSTATE_FORCE);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The subdisks in striped plexes must all have the same size.
|
|
|
|
*/
|
|
|
|
s = LIST_FIRST(&p->subdisks);
|
|
|
|
LIST_FOREACH(s2, &p->subdisks, in_plex) {
|
|
|
|
if (s->size != s2->size) {
|
2008-10-26 17:20:37 +00:00
|
|
|
G_VINUM_DEBUG(0, "subdisk size mismatch %s"
|
|
|
|
"(%jd) <> %s (%jd)", s->name, s->size,
|
2004-06-12 21:16:10 +00:00
|
|
|
s2->name, s2->size);
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_set_plex_state(p, GV_PLEX_DOWN,
|
|
|
|
GV_SETSTATE_FORCE);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Trim subdisk sizes to match the stripe size. */
|
2004-06-12 21:16:10 +00:00
|
|
|
remainder = s->size % p->stripesize;
|
|
|
|
if (remainder) {
|
2008-10-26 17:20:37 +00:00
|
|
|
G_VINUM_DEBUG(1, "size of sd %s is not a "
|
2004-06-12 21:16:10 +00:00
|
|
|
"multiple of plex stripesize, taking off "
|
2008-10-26 17:20:37 +00:00
|
|
|
"%jd bytes", s->name, (intmax_t)remainder);
|
2004-06-12 21:16:10 +00:00
|
|
|
gv_adjust_freespace(s, remainder);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
p->size = gv_plex_size(p);
|
2004-06-12 21:16:10 +00:00
|
|
|
if (p->sdcount == 0)
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_set_plex_state(p, GV_PLEX_DOWN, GV_SETSTATE_FORCE);
|
|
|
|
else if (p->org == GV_PLEX_RAID5 && p->flags & GV_PLEX_NEWBORN) {
|
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex)
|
|
|
|
gv_set_sd_state(s, GV_SD_UP, GV_SETSTATE_FORCE);
|
|
|
|
/* If added to a volume, we want the plex to be down. */
|
|
|
|
state = (p->flags & GV_PLEX_ADDED) ? GV_PLEX_DOWN : GV_PLEX_UP;
|
|
|
|
gv_set_plex_state(p, state, GV_SETSTATE_FORCE);
|
|
|
|
p->flags &= ~GV_PLEX_ADDED;
|
|
|
|
} else if (p->flags & GV_PLEX_ADDED) {
|
2004-06-12 21:16:10 +00:00
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex)
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_set_sd_state(s, GV_SD_STALE, GV_SETSTATE_FORCE);
|
|
|
|
gv_set_plex_state(p, GV_PLEX_DOWN, GV_SETSTATE_FORCE);
|
2004-06-12 21:16:10 +00:00
|
|
|
p->flags &= ~GV_PLEX_ADDED;
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
} else if (p->state == GV_PLEX_UP) {
|
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex) {
|
|
|
|
if (s->flags & GV_SD_GROW) {
|
|
|
|
gv_set_plex_state(p, GV_PLEX_GROWABLE,
|
|
|
|
GV_SETSTATE_FORCE);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Our plex is grown up now. */
|
|
|
|
p->flags &= ~GV_PLEX_NEWBORN;
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Give a subdisk to a drive, check and adjust several parameters, adjust
|
|
|
|
* freelist.
|
|
|
|
*/
|
|
|
|
int
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_sd_to_drive(struct gv_sd *s, struct gv_drive *d)
|
2004-06-12 21:16:10 +00:00
|
|
|
{
|
|
|
|
struct gv_sd *s2;
|
|
|
|
struct gv_freelist *fl, *fl2;
|
|
|
|
off_t tmp;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
fl2 = NULL;
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Shortcut for "referenced" drives. */
|
|
|
|
if (d->flags & GV_DRIVE_REFERENCED) {
|
|
|
|
s->drive_sc = d;
|
|
|
|
return (0);
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
|
|
|
|
/* Check if this subdisk was already given to this drive. */
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (s->drive_sc != NULL) {
|
|
|
|
if (s->drive_sc == d) {
|
|
|
|
if (!(s->flags & GV_SD_TASTED)) {
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
} else {
|
2009-10-05 08:44:31 +00:00
|
|
|
G_VINUM_DEBUG(0, "error giving subdisk '%s' to '%s' "
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
"(already on '%s')", s->name, d->name,
|
|
|
|
s->drive_sc->name);
|
|
|
|
return (GV_ERR_ISATTACHED);
|
|
|
|
}
|
|
|
|
}
|
2004-06-12 21:16:10 +00:00
|
|
|
|
|
|
|
/* Preliminary checks. */
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if ((s->size > d->avail) || (d->freelist_entries == 0)) {
|
|
|
|
G_VINUM_DEBUG(0, "not enough space on '%s' for '%s'", d->name,
|
|
|
|
s->name);
|
|
|
|
return (GV_ERR_NOSPACE);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* If no size was given for this subdisk, try to auto-size it... */
|
2004-06-12 21:16:10 +00:00
|
|
|
if (s->size == -1) {
|
|
|
|
/* Find the largest available slot. */
|
|
|
|
LIST_FOREACH(fl, &d->freelist, freelist) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (fl->size < s->size)
|
|
|
|
continue;
|
|
|
|
s->size = fl->size;
|
|
|
|
s->drive_offset = fl->offset;
|
|
|
|
fl2 = fl;
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* No good slot found? */
|
|
|
|
if (s->size == -1) {
|
2009-10-05 08:44:31 +00:00
|
|
|
G_VINUM_DEBUG(0, "unable to autosize '%s' on '%s'",
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
s->name, d->name);
|
|
|
|
return (GV_ERR_BADSIZE);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
* ... or check if we have a free slot that's large enough for the
|
|
|
|
* given size.
|
2004-06-12 21:16:10 +00:00
|
|
|
*/
|
|
|
|
} else {
|
|
|
|
i = 0;
|
|
|
|
LIST_FOREACH(fl, &d->freelist, freelist) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (fl->size < s->size)
|
|
|
|
continue;
|
|
|
|
/* Assign drive offset, if not given. */
|
|
|
|
if (s->drive_offset == -1)
|
|
|
|
s->drive_offset = fl->offset;
|
|
|
|
fl2 = fl;
|
|
|
|
i++;
|
|
|
|
break;
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Couldn't find a good free slot. */
|
|
|
|
if (i == 0) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
G_VINUM_DEBUG(0, "free slots to small for '%s' on '%s'",
|
|
|
|
s->name, d->name);
|
|
|
|
return (GV_ERR_NOSPACE);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* No drive offset given, try to calculate it. */
|
|
|
|
if (s->drive_offset == -1) {
|
|
|
|
|
|
|
|
/* Add offsets and sizes from other subdisks on this drive. */
|
|
|
|
LIST_FOREACH(s2, &d->subdisks, from_drive) {
|
|
|
|
s->drive_offset = s2->drive_offset + s2->size;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If there are no other subdisks yet, then set the default
|
|
|
|
* offset to GV_DATA_START.
|
|
|
|
*/
|
2004-06-15 20:56:25 +00:00
|
|
|
if (s->drive_offset == -1)
|
2004-06-12 21:16:10 +00:00
|
|
|
s->drive_offset = GV_DATA_START;
|
|
|
|
|
|
|
|
/* Check if we have a free slot at the given drive offset. */
|
|
|
|
} else {
|
|
|
|
i = 0;
|
|
|
|
LIST_FOREACH(fl, &d->freelist, freelist) {
|
|
|
|
/* Yes, this subdisk fits. */
|
|
|
|
if ((fl->offset <= s->drive_offset) &&
|
|
|
|
(fl->offset + fl->size >=
|
|
|
|
s->drive_offset + s->size)) {
|
|
|
|
i++;
|
|
|
|
fl2 = fl;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Couldn't find a good free slot. */
|
|
|
|
if (i == 0) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
G_VINUM_DEBUG(0, "given drive_offset for '%s' won't fit "
|
|
|
|
"on '%s'", s->name, d->name);
|
|
|
|
return (GV_ERR_NOSPACE);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now that all parameters are checked and set up, we can give the
|
|
|
|
* subdisk to the drive and adjust the freelist.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* First, adjust the freelist. */
|
|
|
|
LIST_FOREACH(fl, &d->freelist, freelist) {
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Look for the free slot that we have found before. */
|
|
|
|
if (fl != fl2)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* The subdisk starts at the beginning of the free slot. */
|
|
|
|
if (fl->offset == s->drive_offset) {
|
|
|
|
fl->offset += s->size;
|
|
|
|
fl->size -= s->size;
|
|
|
|
|
|
|
|
/* The subdisk uses the whole slot, so remove it. */
|
|
|
|
if (fl->size == 0) {
|
|
|
|
d->freelist_entries--;
|
|
|
|
LIST_REMOVE(fl, freelist);
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* The subdisk does not start at the beginning of the free
|
|
|
|
* slot.
|
|
|
|
*/
|
|
|
|
} else {
|
|
|
|
tmp = fl->offset + fl->size;
|
|
|
|
fl->size = s->drive_offset - fl->offset;
|
2004-06-12 21:16:10 +00:00
|
|
|
|
|
|
|
/*
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
* The subdisk didn't use the complete rest of the free
|
|
|
|
* slot, so we need to split it.
|
2004-06-12 21:16:10 +00:00
|
|
|
*/
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (s->drive_offset + s->size != tmp) {
|
|
|
|
fl2 = g_malloc(sizeof(*fl2), M_WAITOK | M_ZERO);
|
|
|
|
fl2->offset = s->drive_offset + s->size;
|
|
|
|
fl2->size = tmp - fl2->offset;
|
|
|
|
LIST_INSERT_AFTER(fl, fl2, freelist);
|
|
|
|
d->freelist_entries++;
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
break;
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the first subdisk on this drive, just insert it into the
|
|
|
|
* list.
|
|
|
|
*/
|
|
|
|
if (LIST_EMPTY(&d->subdisks)) {
|
|
|
|
LIST_INSERT_HEAD(&d->subdisks, s, from_drive);
|
|
|
|
|
|
|
|
/* There are other subdisks, so insert this one in correct order. */
|
|
|
|
} else {
|
|
|
|
LIST_FOREACH(s2, &d->subdisks, from_drive) {
|
|
|
|
if (s->drive_offset < s2->drive_offset) {
|
|
|
|
LIST_INSERT_BEFORE(s2, s, from_drive);
|
|
|
|
break;
|
|
|
|
} else if (LIST_NEXT(s2, from_drive) == NULL) {
|
|
|
|
LIST_INSERT_AFTER(s2, s, from_drive);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
d->sdcount++;
|
|
|
|
d->avail -= s->size;
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
s->flags &= ~GV_SD_TASTED;
|
|
|
|
|
2004-06-12 21:16:10 +00:00
|
|
|
/* Link back from the subdisk to this drive. */
|
|
|
|
s->drive_sc = d;
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
2006-03-30 14:01:25 +00:00
|
|
|
void
|
|
|
|
gv_free_sd(struct gv_sd *s)
|
|
|
|
{
|
|
|
|
struct gv_drive *d;
|
|
|
|
struct gv_freelist *fl, *fl2;
|
|
|
|
|
|
|
|
KASSERT(s != NULL, ("gv_free_sd: NULL s"));
|
|
|
|
|
|
|
|
d = s->drive_sc;
|
|
|
|
if (d == NULL)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* First, find the free slot that's immediately before or after this
|
|
|
|
* subdisk.
|
|
|
|
*/
|
|
|
|
fl = NULL;
|
|
|
|
LIST_FOREACH(fl, &d->freelist, freelist) {
|
|
|
|
if (fl->offset == s->drive_offset + s->size)
|
|
|
|
break;
|
|
|
|
if (fl->offset + fl->size == s->drive_offset)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* If there is no free slot behind this subdisk, so create one. */
|
|
|
|
if (fl == NULL) {
|
|
|
|
|
|
|
|
fl = g_malloc(sizeof(*fl), M_WAITOK | M_ZERO);
|
|
|
|
fl->size = s->size;
|
|
|
|
fl->offset = s->drive_offset;
|
|
|
|
|
|
|
|
if (d->freelist_entries == 0) {
|
|
|
|
LIST_INSERT_HEAD(&d->freelist, fl, freelist);
|
|
|
|
} else {
|
|
|
|
LIST_FOREACH(fl2, &d->freelist, freelist) {
|
|
|
|
if (fl->offset < fl2->offset) {
|
|
|
|
LIST_INSERT_BEFORE(fl2, fl, freelist);
|
|
|
|
break;
|
|
|
|
} else if (LIST_NEXT(fl2, freelist) == NULL) {
|
|
|
|
LIST_INSERT_AFTER(fl2, fl, freelist);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
d->freelist_entries++;
|
|
|
|
|
|
|
|
/* Expand the free slot we just found. */
|
|
|
|
} else {
|
|
|
|
fl->size += s->size;
|
|
|
|
if (fl->offset > s->drive_offset)
|
|
|
|
fl->offset = s->drive_offset;
|
|
|
|
}
|
|
|
|
|
|
|
|
d->avail += s->size;
|
2007-04-12 17:54:35 +00:00
|
|
|
d->sdcount--;
|
2006-03-30 14:01:25 +00:00
|
|
|
}
|
|
|
|
|
2004-06-12 21:16:10 +00:00
|
|
|
void
|
|
|
|
gv_adjust_freespace(struct gv_sd *s, off_t remainder)
|
|
|
|
{
|
|
|
|
struct gv_drive *d;
|
|
|
|
struct gv_freelist *fl, *fl2;
|
|
|
|
|
|
|
|
KASSERT(s != NULL, ("gv_adjust_freespace: NULL s"));
|
|
|
|
d = s->drive_sc;
|
|
|
|
KASSERT(d != NULL, ("gv_adjust_freespace: NULL d"));
|
|
|
|
|
|
|
|
/* First, find the free slot that's immediately after this subdisk. */
|
|
|
|
fl = NULL;
|
|
|
|
LIST_FOREACH(fl, &d->freelist, freelist) {
|
|
|
|
if (fl->offset == s->drive_offset + s->size)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* If there is no free slot behind this subdisk, so create one. */
|
|
|
|
if (fl == NULL) {
|
|
|
|
|
|
|
|
fl = g_malloc(sizeof(*fl), M_WAITOK | M_ZERO);
|
|
|
|
fl->size = remainder;
|
|
|
|
fl->offset = s->drive_offset + s->size - remainder;
|
|
|
|
|
|
|
|
if (d->freelist_entries == 0) {
|
|
|
|
LIST_INSERT_HEAD(&d->freelist, fl, freelist);
|
|
|
|
} else {
|
|
|
|
LIST_FOREACH(fl2, &d->freelist, freelist) {
|
|
|
|
if (fl->offset < fl2->offset) {
|
|
|
|
LIST_INSERT_BEFORE(fl2, fl, freelist);
|
|
|
|
break;
|
|
|
|
} else if (LIST_NEXT(fl2, freelist) == NULL) {
|
|
|
|
LIST_INSERT_AFTER(fl2, fl, freelist);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
d->freelist_entries++;
|
|
|
|
|
|
|
|
/* Expand the free slot we just found. */
|
|
|
|
} else {
|
|
|
|
fl->offset -= remainder;
|
|
|
|
fl->size += remainder;
|
|
|
|
}
|
|
|
|
|
|
|
|
s->size -= remainder;
|
|
|
|
d->avail += remainder;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Check if the given plex is a striped one. */
|
|
|
|
int
|
|
|
|
gv_is_striped(struct gv_plex *p)
|
|
|
|
{
|
|
|
|
KASSERT(p != NULL, ("gv_is_striped: NULL p"));
|
|
|
|
switch(p->org) {
|
|
|
|
case GV_PLEX_STRIPED:
|
|
|
|
case GV_PLEX_RAID5:
|
|
|
|
return (1);
|
|
|
|
default:
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find a volume by name. */
|
|
|
|
struct gv_volume *
|
|
|
|
gv_find_vol(struct gv_softc *sc, char *name)
|
|
|
|
{
|
|
|
|
struct gv_volume *v;
|
|
|
|
|
|
|
|
LIST_FOREACH(v, &sc->volumes, volume) {
|
|
|
|
if (!strncmp(v->name, name, GV_MAXVOLNAME))
|
|
|
|
return (v);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find a plex by name. */
|
|
|
|
struct gv_plex *
|
|
|
|
gv_find_plex(struct gv_softc *sc, char *name)
|
|
|
|
{
|
|
|
|
struct gv_plex *p;
|
|
|
|
|
|
|
|
LIST_FOREACH(p, &sc->plexes, plex) {
|
|
|
|
if (!strncmp(p->name, name, GV_MAXPLEXNAME))
|
|
|
|
return (p);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find a subdisk by name. */
|
|
|
|
struct gv_sd *
|
|
|
|
gv_find_sd(struct gv_softc *sc, char *name)
|
|
|
|
{
|
|
|
|
struct gv_sd *s;
|
|
|
|
|
|
|
|
LIST_FOREACH(s, &sc->subdisks, sd) {
|
|
|
|
if (!strncmp(s->name, name, GV_MAXSDNAME))
|
|
|
|
return (s);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find a drive by name. */
|
|
|
|
struct gv_drive *
|
|
|
|
gv_find_drive(struct gv_softc *sc, char *name)
|
|
|
|
{
|
|
|
|
struct gv_drive *d;
|
|
|
|
|
|
|
|
LIST_FOREACH(d, &sc->drives, drive) {
|
|
|
|
if (!strncmp(d->name, name, GV_MAXDRIVENAME))
|
|
|
|
return (d);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (NULL);
|
|
|
|
}
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* Find a drive given a device. */
|
|
|
|
struct gv_drive *
|
|
|
|
gv_find_drive_device(struct gv_softc *sc, char *device)
|
|
|
|
{
|
|
|
|
struct gv_drive *d;
|
|
|
|
|
|
|
|
LIST_FOREACH(d, &sc->drives, drive) {
|
|
|
|
if(!strcmp(d->device, device))
|
|
|
|
return (d);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (NULL);
|
|
|
|
}
|
|
|
|
|
2004-06-12 21:16:10 +00:00
|
|
|
/* Check if any consumer of the given geom is open. */
|
|
|
|
int
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_consumer_is_open(struct g_consumer *cp)
|
2004-06-12 21:16:10 +00:00
|
|
|
{
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (cp == NULL)
|
|
|
|
return (0);
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (cp->acr || cp->acw || cp->ace)
|
|
|
|
return (1);
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
gv_provider_is_open(struct g_provider *pp)
|
|
|
|
{
|
|
|
|
if (pp == NULL)
|
2004-06-16 14:41:04 +00:00
|
|
|
return (0);
|
2004-06-12 21:16:10 +00:00
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
if (pp->acr || pp->acw || pp->ace)
|
|
|
|
return (1);
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Compare the modification dates of the drives.
|
|
|
|
* Return 1 if a > b, 0 otherwise.
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
gv_drive_is_newer(struct gv_softc *sc, struct gv_drive *d)
|
|
|
|
{
|
|
|
|
struct gv_drive *d2;
|
|
|
|
struct timeval *a, *b;
|
|
|
|
|
|
|
|
KASSERT(!LIST_EMPTY(&sc->drives),
|
|
|
|
("gv_is_drive_newer: empty drive list"));
|
|
|
|
|
|
|
|
a = &d->hdr->label.last_update;
|
|
|
|
LIST_FOREACH(d2, &sc->drives, drive) {
|
|
|
|
if ((d == d2) || (d2->state != GV_DRIVE_UP) ||
|
|
|
|
(d2->hdr == NULL))
|
|
|
|
continue;
|
|
|
|
b = &d2->hdr->label.last_update;
|
|
|
|
if (timevalcmp(a, b, >))
|
2004-06-12 21:16:10 +00:00
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return the type of object identified by string 'name'. */
|
|
|
|
int
|
|
|
|
gv_object_type(struct gv_softc *sc, char *name)
|
|
|
|
{
|
|
|
|
struct gv_drive *d;
|
|
|
|
struct gv_plex *p;
|
|
|
|
struct gv_sd *s;
|
|
|
|
struct gv_volume *v;
|
|
|
|
|
|
|
|
LIST_FOREACH(v, &sc->volumes, volume) {
|
|
|
|
if (!strncmp(v->name, name, GV_MAXVOLNAME))
|
|
|
|
return (GV_TYPE_VOL);
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(p, &sc->plexes, plex) {
|
|
|
|
if (!strncmp(p->name, name, GV_MAXPLEXNAME))
|
|
|
|
return (GV_TYPE_PLEX);
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(s, &sc->subdisks, sd) {
|
|
|
|
if (!strncmp(s->name, name, GV_MAXSDNAME))
|
|
|
|
return (GV_TYPE_SD);
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(d, &sc->drives, drive) {
|
|
|
|
if (!strncmp(d->name, name, GV_MAXDRIVENAME))
|
|
|
|
return (GV_TYPE_DRIVE);
|
|
|
|
}
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
return (GV_ERR_NOTFOUND);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
|
|
|
|
2004-09-13 21:01:36 +00:00
|
|
|
void
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_setup_objects(struct gv_softc *sc)
|
2004-09-13 21:01:36 +00:00
|
|
|
{
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
struct g_provider *pp;
|
|
|
|
struct gv_volume *v;
|
|
|
|
struct gv_plex *p;
|
|
|
|
struct gv_sd *s;
|
|
|
|
struct gv_drive *d;
|
|
|
|
|
|
|
|
LIST_FOREACH(s, &sc->subdisks, sd) {
|
|
|
|
d = gv_find_drive(sc, s->drive);
|
|
|
|
if (d != NULL)
|
|
|
|
gv_sd_to_drive(s, d);
|
|
|
|
p = gv_find_plex(sc, s->plex);
|
|
|
|
if (p != NULL)
|
|
|
|
gv_sd_to_plex(s, p);
|
|
|
|
gv_update_sd_state(s);
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(p, &sc->plexes, plex) {
|
|
|
|
gv_update_plex_config(p);
|
|
|
|
v = gv_find_vol(sc, p->volume);
|
|
|
|
if (v != NULL && p->vol_sc != v) {
|
|
|
|
p->vol_sc = v;
|
|
|
|
v->plexcount++;
|
|
|
|
LIST_INSERT_HEAD(&v->plexes, p, in_volume);
|
|
|
|
}
|
|
|
|
gv_update_plex_config(p);
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_FOREACH(v, &sc->volumes, volume) {
|
|
|
|
v->size = gv_vol_size(v);
|
|
|
|
if (v->provider == NULL) {
|
|
|
|
g_topology_lock();
|
|
|
|
pp = g_new_providerf(sc->geom, "gvinum/%s", v->name);
|
|
|
|
pp->mediasize = v->size;
|
|
|
|
pp->sectorsize = 512; /* XXX */
|
|
|
|
g_error_provider(pp, 0);
|
|
|
|
v->provider = pp;
|
|
|
|
pp->private = v;
|
|
|
|
g_topology_unlock();
|
|
|
|
} else if (v->provider->mediasize != v->size) {
|
|
|
|
g_topology_lock();
|
|
|
|
v->provider->mediasize = v->size;
|
|
|
|
g_topology_unlock();
|
|
|
|
}
|
|
|
|
v->flags &= ~GV_VOL_NEWBORN;
|
|
|
|
gv_update_vol_state(v);
|
2004-09-13 21:01:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2004-06-12 21:16:10 +00:00
|
|
|
void
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
gv_cleanup(struct gv_softc *sc)
|
2004-06-12 21:16:10 +00:00
|
|
|
{
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
struct gv_volume *v, *v2;
|
|
|
|
struct gv_plex *p, *p2;
|
|
|
|
struct gv_sd *s, *s2;
|
|
|
|
struct gv_drive *d, *d2;
|
|
|
|
struct gv_freelist *fl, *fl2;
|
|
|
|
|
|
|
|
mtx_lock(&sc->config_mtx);
|
|
|
|
LIST_FOREACH_SAFE(v, &sc->volumes, volume, v2) {
|
|
|
|
LIST_REMOVE(v, volume);
|
|
|
|
g_free(v->wqueue);
|
|
|
|
g_free(v);
|
|
|
|
}
|
|
|
|
LIST_FOREACH_SAFE(p, &sc->plexes, plex, p2) {
|
|
|
|
LIST_REMOVE(p, plex);
|
2006-01-06 18:03:17 +00:00
|
|
|
g_free(p->bqueue);
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
g_free(p->rqueue);
|
2006-01-06 18:03:17 +00:00
|
|
|
g_free(p->wqueue);
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
g_free(p);
|
|
|
|
}
|
|
|
|
LIST_FOREACH_SAFE(s, &sc->subdisks, sd, s2) {
|
|
|
|
LIST_REMOVE(s, sd);
|
|
|
|
g_free(s);
|
2004-09-18 13:44:43 +00:00
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
LIST_FOREACH_SAFE(d, &sc->drives, drive, d2) {
|
|
|
|
LIST_FOREACH_SAFE(fl, &d->freelist, freelist, fl2) {
|
|
|
|
LIST_REMOVE(fl, freelist);
|
|
|
|
g_free(fl);
|
|
|
|
}
|
|
|
|
LIST_REMOVE(d, drive);
|
|
|
|
g_free(d->hdr);
|
|
|
|
g_free(d);
|
|
|
|
}
|
|
|
|
mtx_destroy(&sc->config_mtx);
|
2004-09-18 13:44:43 +00:00
|
|
|
}
|
|
|
|
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
/* General 'attach' routine. */
|
|
|
|
int
|
|
|
|
gv_attach_plex(struct gv_plex *p, struct gv_volume *v, int rename)
|
|
|
|
{
|
|
|
|
struct gv_sd *s;
|
|
|
|
struct gv_softc *sc;
|
|
|
|
|
|
|
|
g_topology_assert();
|
|
|
|
|
|
|
|
sc = p->vinumconf;
|
|
|
|
KASSERT(sc != NULL, ("NULL sc"));
|
|
|
|
|
|
|
|
if (p->vol_sc != NULL) {
|
|
|
|
G_VINUM_DEBUG(1, "unable to attach %s: already attached to %s",
|
|
|
|
p->name, p->volume);
|
|
|
|
return (GV_ERR_ISATTACHED);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Stale all subdisks of this plex. */
|
|
|
|
LIST_FOREACH(s, &p->subdisks, in_plex) {
|
|
|
|
if (s->state != GV_SD_STALE)
|
|
|
|
gv_set_sd_state(s, GV_SD_STALE, GV_SETSTATE_FORCE);
|
|
|
|
}
|
|
|
|
/* Attach to volume. Make sure volume is not up and running. */
|
|
|
|
if (gv_provider_is_open(v->provider)) {
|
|
|
|
G_VINUM_DEBUG(1, "unable to attach %s: volume %s is busy",
|
|
|
|
p->name, v->name);
|
|
|
|
return (GV_ERR_ISBUSY);
|
|
|
|
}
|
|
|
|
p->vol_sc = v;
|
|
|
|
strlcpy(p->volume, v->name, sizeof(p->volume));
|
|
|
|
v->plexcount++;
|
|
|
|
if (rename) {
|
|
|
|
snprintf(p->name, sizeof(p->name), "%s.p%d", v->name,
|
|
|
|
v->plexcount);
|
|
|
|
}
|
|
|
|
LIST_INSERT_HEAD(&v->plexes, p, in_volume);
|
|
|
|
|
|
|
|
/* Get plex up again. */
|
|
|
|
gv_update_vol_size(v, gv_vol_size(v));
|
|
|
|
gv_set_plex_state(p, GV_PLEX_UP, 0);
|
|
|
|
gv_save_config(p->vinumconf);
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
gv_attach_sd(struct gv_sd *s, struct gv_plex *p, off_t offset, int rename)
|
|
|
|
{
|
|
|
|
struct gv_sd *s2;
|
|
|
|
int error, sdcount;
|
|
|
|
|
|
|
|
g_topology_assert();
|
|
|
|
|
|
|
|
/* If subdisk is attached, don't do it. */
|
|
|
|
if (s->plex_sc != NULL) {
|
|
|
|
G_VINUM_DEBUG(1, "unable to attach %s: already attached to %s",
|
|
|
|
s->name, s->plex);
|
|
|
|
return (GV_ERR_ISATTACHED);
|
|
|
|
}
|
|
|
|
|
|
|
|
gv_set_sd_state(s, GV_SD_STALE, GV_SETSTATE_FORCE);
|
|
|
|
/* First check that this subdisk has a correct offset. If none other
|
|
|
|
* starts at the same, and it's correct module stripesize, it is */
|
|
|
|
if (offset != -1 && offset % p->stripesize != 0)
|
|
|
|
return (GV_ERR_BADOFFSET);
|
|
|
|
LIST_FOREACH(s2, &p->subdisks, in_plex) {
|
|
|
|
if (s2->plex_offset == offset)
|
|
|
|
return (GV_ERR_BADOFFSET);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Attach the subdisk to the plex at given offset. */
|
|
|
|
s->plex_offset = offset;
|
|
|
|
strlcpy(s->plex, p->name, sizeof(s->plex));
|
|
|
|
|
|
|
|
sdcount = p->sdcount;
|
|
|
|
error = gv_sd_to_plex(s, p);
|
|
|
|
if (error)
|
|
|
|
return (error);
|
|
|
|
gv_update_plex_config(p);
|
|
|
|
|
|
|
|
if (rename) {
|
|
|
|
snprintf(s->name, sizeof(s->name), "%s.s%d", s->plex,
|
|
|
|
p->sdcount);
|
|
|
|
}
|
|
|
|
if (p->vol_sc != NULL)
|
|
|
|
gv_update_vol_size(p->vol_sc, gv_vol_size(p->vol_sc));
|
|
|
|
gv_save_config(p->vinumconf);
|
|
|
|
/* We don't update the subdisk state since the user might have to
|
|
|
|
* initiate a rebuild/sync first. */
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Detach a plex from a volume. */
|
|
|
|
int
|
|
|
|
gv_detach_plex(struct gv_plex *p, int flags)
|
2004-09-18 13:44:43 +00:00
|
|
|
{
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
struct gv_volume *v;
|
|
|
|
|
|
|
|
g_topology_assert();
|
|
|
|
v = p->vol_sc;
|
|
|
|
|
|
|
|
if (v == NULL) {
|
|
|
|
G_VINUM_DEBUG(1, "unable to detach %s: already detached",
|
|
|
|
p->name);
|
|
|
|
return (0); /* Not an error. */
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Only proceed if forced or volume inactive.
|
|
|
|
*/
|
|
|
|
if (!(flags & GV_FLAG_F) && (gv_provider_is_open(v->provider) ||
|
|
|
|
p->state == GV_PLEX_UP)) {
|
|
|
|
G_VINUM_DEBUG(1, "unable to detach %s: volume %s is busy",
|
|
|
|
p->name, p->volume);
|
|
|
|
return (GV_ERR_ISBUSY);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|
Import the gvinum work that have been done during and after Summer of Code 2007.
The work have been under testing and fixing since then, and it is mature enough
to be put into HEAD for further testing.
A lot have changed in this time, and here are the most important:
- Gvinum now uses one single workerthread instead of one thread for each
volume and each plex. The reason for this is that the previous scheme was
very complex, and was the cause of many of the bugs discovered in gvinum.
Instead, gvinum now uses one worker thread with an event queue, quite
similar to what used in gmirror.
- The rebuild/grow/initialize/parity check routines no longer runs in
separate threads, but are run as regular I/O requests with special flags.
This made it easier to support mounted growing and parity rebuild.
- Support for growing striped and raid5-plexes, meaning that one can extend the
volumes for these plex types in addition to the concat type. Also works while
the volume is mounted.
- Implementation of many of the missing commands from the old vinum:
attach/detach, start (was partially implemented), stop (was partially
implemented), concat, mirror, stripe, raid5 (shortcuts for creating volumes
with one plex of these organizations).
- The parity check and rebuild no longer goes between userland/kernel, meaning
that the gvinum command will not stay and wait forever for the rebuild to
finish. You can instead watch the status with the list command.
- Many problems with gvinum have been reported since 5.x, and some has been hard
to fix due to the complicated architecture. Hopefully, it should be more
stable and better handle edge cases that previously made gvinum crash.
- Failed drives no longer disappears entirely, but now leave behind a dummy
drive that makes sure the original state is not forgotten in case the system
is rebooted between drive failures/swaps.
- Update manpage to reflect new commands and extend it with some examples.
Sponsored by: Google Summer of Code 2007
Mentored by: le
Tested by: Rick C. Petty <rick-freebsd2008 -at- kiwi-computer.com>
2009-03-28 17:20:08 +00:00
|
|
|
v->plexcount--;
|
|
|
|
/* Make sure someone don't read us when gone. */
|
|
|
|
v->last_read_plex = NULL;
|
|
|
|
LIST_REMOVE(p, in_volume);
|
|
|
|
p->vol_sc = NULL;
|
|
|
|
memset(p->volume, 0, GV_MAXVOLNAME);
|
|
|
|
gv_update_vol_size(v, gv_vol_size(v));
|
|
|
|
gv_save_config(p->vinumconf);
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Detach a subdisk from a plex. */
|
|
|
|
int
|
|
|
|
gv_detach_sd(struct gv_sd *s, int flags)
|
|
|
|
{
|
|
|
|
struct gv_plex *p;
|
|
|
|
|
|
|
|
g_topology_assert();
|
|
|
|
p = s->plex_sc;
|
|
|
|
|
|
|
|
if (p == NULL) {
|
|
|
|
G_VINUM_DEBUG(1, "unable to detach %s: already detached",
|
|
|
|
s->name);
|
|
|
|
return (0); /* Not an error. */
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Don't proceed if we're not forcing, and the plex is up, or degraded
|
|
|
|
* with this subdisk up.
|
|
|
|
*/
|
|
|
|
if (!(flags & GV_FLAG_F) && ((p->state > GV_PLEX_DEGRADED) ||
|
|
|
|
((p->state == GV_PLEX_DEGRADED) && (s->state == GV_SD_UP)))) {
|
|
|
|
G_VINUM_DEBUG(1, "unable to detach %s: plex %s is busy",
|
|
|
|
s->name, s->plex);
|
|
|
|
return (GV_ERR_ISBUSY);
|
|
|
|
}
|
|
|
|
|
|
|
|
LIST_REMOVE(s, in_plex);
|
|
|
|
s->plex_sc = NULL;
|
|
|
|
memset(s->plex, 0, GV_MAXPLEXNAME);
|
|
|
|
p->sddetached++;
|
|
|
|
gv_save_config(s->vinumconf);
|
|
|
|
return (0);
|
2004-06-12 21:16:10 +00:00
|
|
|
}
|