Mike Snitzer [Wed, 13 Jan 2010 01:44:37 +0000 (01:44 +0000)]
Merging device is loaded with "-cow" suffix and with base name of the
origin. This is needed so that "-cow" device can be found and removed
when lvremove is performed.
Mike Snitzer [Wed, 13 Jan 2010 01:39:44 +0000 (01:39 +0000)]
Add support for "snapshot-merge" target.
Introduces new libdevmapper function dm_tree_node_add_snapshot_merge_target
Verifies that the kernel (dm-snapshot) provides the 'snapshot-merge'
target.
Activate origin LV as snapshot-merge target. Using snapshot-origin
target would be pointless because the origin contains volatile data
while a merge is in progress.
Because snapshot-merge target is activated in place of the
snapshot-origin target it must be resumed after all other snapshots
(just like snapshot-origin does) --- otherwise small window for data
corruption would exist.
Ideally the merging snapshot would not be activated at all but if it is
to be activated (because snapshot was already active) it _must_ be done
after the snapshot-merge. This insures that DM's snapshot-merge target
will perform exception handover in the proper order (new->resume before
old->resume). DM's snapshot-merge does support handover if the reverse
sequence is used (old->resume before new->resume) but DM will fail to
resume the old snapshot; leaving it suspended.
To insure the proper activation sequence dm_tree_activate_children() was
updated to accommodate an additional 'activation_priority' level. All
regular snapshots are 0, snapshot-merge is 1, and merging snapshot is 2.
Alasdair Kergon [Tue, 12 Jan 2010 20:53:20 +0000 (20:53 +0000)]
Fix allocation code not to stop at the first area of a PV that fits.
This spurious 'break' has been here since this code was first committed
in June 2005 and stopped the algorithm behaving as described in the
comment above it and rendered the variable 'already_found_one' useless.
Testsuite updates and fixes for recently added features.
1. Found bug in 'redundant log' implementation that caused
problems when converting a linear that spanned multiple
devices to a mirror (wasn't checking for NULL value of
provided parameter in _alloc_parallel_area)
2. Testsuite was failing to perform tests when 'not' modifier
was used. This allowed a couple issues to slip through.
Added a 'not_sh' modifier that negates tests performed by
functions defined in the shell source file.
3. Was initializing a variable to far down, which cause
previously set value to be overridden. (This was the
result of the collision of the "redundant log" and
lvconvert fix patches.)
Mike Snitzer [Mon, 11 Jan 2010 19:08:18 +0000 (19:08 +0000)]
Reset _vgs_locked in lvmcache_init()
Upon successful fork(), _become_daemon() must assert that the locks that
are currently held belong to the parent, not the child. All of the
child's internal state saying 'this process holds a lock' has to be
reset.
A proper lvmcache_locking_reset() should follow later.
date: 2010/01/07 20:42:55; author: jbrassow; state: Exp; lines: +11 -0
The patch fixes some lvconvert issues (WRT mirror <-> mirror).
1) 'exisiting_mirrors' and 'lp->mirrors' where taken to be in 'n-1'
notation (i.e a 2-way mirror is '1' and a linear is '0'), but the
variables were in 'n' notation.
2) After adding the redundant mirror log support, I was calculating
log_count by looking at the mirror log LV, but didn't take into
account the fact that there could be no mirror log!
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Peter Rajnoha [Mon, 11 Jan 2010 15:36:24 +0000 (15:36 +0000)]
Add support to disable udev checking: DM_UDEV_DISABLE_CHECKING=1 env. variable.
Sometimes it is really needed to switch off udev checking and the warnings we show when
we detect that udev has not done its job right - the messages like "Udev should have done
this and that. Falling back to direct node creation/removal. " etc.
This would be especially handy while setting DM_DEV_DIR env var that could be set to a
different location than standard /dev (udev can't create nodes/symlinks out of that one
directory that is configured into udevd). The exact same situation happens while we're
running our tests.
Add the new mirror log type "redundant". The options are now:
--mirrorlog core: in-memory log
--mirrorlog disk: persistent log
--mirrorlog redundant: redundant persistent log
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
This patch adds the capability to split off a mirror legs.
It is pretty much the same as reducing the number of
mirror legs, but we just don't delete them afterwards.
The following command line interface is enforced:
prompt> lvconvert --splitmirror <n> -n <name> <VG>/<LV>
where 'n' is the number of images to split off, and
where 'name' is the name of the newly split off logical volume.
If more than one leg is split off, a new mirror will be the
result. The newly split off mirror will have a 'core' log.
Example:
[root@bp-01 LVM2]# !lvs
lvs -a -o name,copy_percent,devices
LV Copy% Devices
lv 100.00 lv_mimage_0(0),lv_mimage_1(0),lv_mimage_2(0),lv_mimage_3(0)
[lv_mimage_0] /dev/sdb1(0)
[lv_mimage_1] /dev/sdc1(0)
[lv_mimage_2] /dev/sdd1(0)
[lv_mimage_3] /dev/sde1(0)
[lv_mlog] /dev/sdi1(0)
[root@bp-01 LVM2]# lvconvert --splitmirrors 2 --name split vg/lv /dev/sd[ce]1
Logical volume lv converted.
[root@bp-01 LVM2]# !lvs
lvs -a -o name,copy_percent,devices
LV Copy% Devices
lv 100.00 lv_mimage_0(0),lv_mimage_2(0)
[lv_mimage_0] /dev/sdb1(0)
[lv_mimage_2] /dev/sdd1(0)
[lv_mlog] /dev/sdi1(0)
split 100.00 split_mimage_0(0),split_mimage_1(0)
[split_mimage_0] /dev/sde1(0)
[split_mimage_1] /dev/sdc1(0)
It can be seen that '--splitmirror <n>' is exactly the same
as '--mirrors -<n>' (note the minus sign), except there is the
additional notion to keep the image being detached from the
mirror instead of just throwing it away.
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Petr Rockai [Fri, 8 Jan 2010 13:04:10 +0000 (13:04 +0000)]
In lvconvert --repair --use-policies, for the allocate policies, return success
even if allocation fails, as long as the downconversion or corelog conversion
succeeded.
The patch fixes some lvconvert issues (WRT mirror <-> mirror).
The default log option for a mirror is 'disk'. If the log
type is not explicitly stated on the command line when
converting from an X-way mirror to a Y-way mirror, 'disk'
is chosen. So, if you have a 'core' log mirror and you
convert, your result will contain a 'disk' log.
This patch remembers what the old log type was. If the
user is merely trying to switch the number of mirror
images, the log type is now kept the same.
There is one historical behaviour I left in place...
If you have a 2-way, core-log mirror and you use lvconvert to
specify you want a 2-way mirror - without specifying the
log type - you will get a 2-way, disk-log mirror.
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com> Informal-IRC-ACK-by: agk
Milan Broz [Wed, 6 Jan 2010 13:26:21 +0000 (13:26 +0000)]
Remove empty "repaired" devices if empty in lvconvert.
The logic was that lvconvert repair volumes, marking
PV as MISSING and following vgreduce --removemissing
removes these missing devices.
Previously dmeventd mirror DSO removed all LV and PV
from VG by simply relying on
vgreduce --removemissing --force.
Now, there are two subsequent calls:
lvconvert --repair --use-policies
vgreduce --removemissing
So the VG is locked twice, opening space for all races
between other running lvm processes. If the PV reappears
with old metadata on it (so the winner performs autorepair,
if locking VG for update) the situation is even worse.
Patch simply adds removemissing PV functionality into
lvconcert BUT ONLY if running with --repair and --use-policies
and removing only these empty missing PVs which are
involved in repair.
(This combination is expected to run only from dmeventd.)
Mike Snitzer [Tue, 5 Jan 2010 21:14:04 +0000 (21:14 +0000)]
Use snapshot metadata usage to determine if snapshot is empty
Version >= 1.8.0 of the DM snapshot target appends metadata sectors used
to a snapshot's status. This patch allows LVM2 to accurately determine
if the snapshot store is empty. Knowing when a snapshot store is empty
is important in the context of snapshot-merge (means merge is complete).
Also update LVM2 to be aware of the possibility for "Merge failed" in
the snapshot-merge target's status.
Mike Snitzer [Tue, 5 Jan 2010 20:56:51 +0000 (20:56 +0000)]
Add a [--poll {y|n}] flag to vgchange and lvchange to control whether
the background polldaemon is allowed to start. It can be used
standalone or in conjunction with --refresh or --available y.
Control over when the background polldaemon starts will be particularly
important for snapshot-merge of a root filesystem.
Dracut will be updated to activate all LVs with: --poll n
The lvm2-monitor initscript will start polling with: --poll y
NOTE: Because we currently have no way of knowing if a background
polldaemon is active for a given LV the following limitations exist and
have been deemed acceptable:
1) it is not possible to stop an active polldaemon; so the lvm2-monitor
initscript doesn't stop running polldaemon(s)
2) redundant polldaemon instances will be started for all specified LVs
if vgchange or lvchange are repeatedly used with '--poll y'
Milan Broz [Tue, 5 Jan 2010 15:58:11 +0000 (15:58 +0000)]
Resume volumes in reverse order to preserve memlock pairing.
If renaming snapshot with virtual origin, the origin is renamed too.
But the code must resume LVs in reverse order to properly
pair memlock (in cluster locking).
(The resume of snapshot resumes origin too and later resume
is ignored otherwise.)
Milan Broz [Fri, 18 Dec 2009 12:44:20 +0000 (12:44 +0000)]
Remove missing flag if PV reappeared and is empty.
When PV device reappears with old metadata, it is
always updated to new version byt atutomatic metadata
repair.
Remove missing flag if device is empty.
If device contains allocated extents, issue warning that
user must remove volumes and re-add this PV before
manipulating with this volume.
This partially solves bug 547842 when one PV (log) is failed,
dmeventd removes that device and later this device reappears and
is wrongly added into VG marked missing.
Milan Broz [Wed, 9 Dec 2009 19:53:39 +0000 (19:53 +0000)]
Call explicitly suspend for temporary mirror layer.
The memlock_inc() fix is wrong, memlock count is not
propagated to long living process (clvmd) and just
it underflow there.
Also suspend is needed to pre-load precommited metadata
on other nodes (remapping to error taget in this case).
With explicit suspend we generate lock request and code
can update memlock count.
(Infinitely "locked" memory caused that fs_unlock() was not
called properly and on cluster nodes remains
old links in /dev/mapper for not active devices.)
(N.B. failing of suspend call here is not handled as fatal
error - the LV is going to be removed later anyway.)
Milan Broz [Wed, 9 Dec 2009 19:29:04 +0000 (19:29 +0000)]
Allow manipulation with precommited metadata even when a PV is missing.
The new recovery code first tries to repair LV and then removes failed PV
from VG. It means that during operation there can be VG with PV missing,
and vg_read code handles it like not consistent VG.
We already allows returning "inconsistent" commited metadata,
for mirror repair we need this for precommited too.
(The suspend call prepares precommited metadata to inactive table on
other cluster nodes.)
"Inconsistent" here means - correct metadata, just with some metadata areas
not found (obviously on missing or failed PVs).
Milan Broz [Wed, 9 Dec 2009 19:00:16 +0000 (19:00 +0000)]
Never ever use distributed lock for LV in non-clustered VG.
The LV locks make sense only for clustered LVs.
Properly check cluster flag and never issue cluster lock here.
There are several places in code, where it is already checked, this
patch add this check to all needed calls.
In previous code the lock behaviour was inconsistent,
for example, the pre/post callback can take lock even for local volume,
but deactivate call do not released this lock and it remains held forever.
The local LV lock request now just let run the underlying activation code
on local node, the same process like in local locking.
(Again, this is important for new mirror repair calls, here for local
mirrors but with cluster locking enabled.)
Milan Broz [Wed, 9 Dec 2009 18:42:02 +0000 (18:42 +0000)]
Get rid of magic masks in cluster locking code - clvmd part.
- do_command and lock_vg expect flags (no change here)
Bug fixes:
- lock_vg should check for NONBLOCK on lock_cmd, flags have this bit masked-out
- do_pre/post_command expect do not mask flag at all, this causes that
the code inside is never run! (see following patches, these functions
expect plain command without flags)
Milan Broz [Wed, 9 Dec 2009 18:28:27 +0000 (18:28 +0000)]
Get rid of magic masks in cluster locking code.
Patch should not cause any problems, only real change is
removing LCK_LOCAL bit from lock type flag, it is never used there.
(LCK_LOCAL is part arg[1] bits anyway.)
Milan Broz [Wed, 9 Dec 2009 18:09:52 +0000 (18:09 +0000)]
Remove newly created log volume if initial deactivation fails.
If there is problem deactivate LV and
_init_mirror_log is called with remove_on_failure = 1,
remove the newly created log LV from metadata.
(This can happen if there is active device with the same name
but different UUID.)
The main reason for this "workaround" patch is to
- do not keep _mlog volume in metadata, so user can repeat the action
- print better error message describing the real problem
# lvcreate -m 2 -n lv1 -l 1 --nosync vg_bar
WARNING: New mirror won't be synchronised. Don't read what you didn't write!
/dev/vg_bar/lv1_mlog: not found: device not cleared
Aborting. Failed to wipe mirror log.
Error locking on node bar-01: Input/output error
Unable to deactivate mirror log LV. Manual intervention required.
Failed to create mirror log.
# lvcreate -m 2 -n lv1 -l 1 --nosync vg_bar
WARNING: New mirror won't be synchronised. Don't read what you didn't write!
Aborting. Unable to deactivate mirror log.
Failed to initialise mirror log.
I see "Deactivated" message when I activate and "Activated" message when
I deactivate. The code uses "activate" as boolean but it can be any one
of the enum values from CHANGE_AY, CHANGE_AN, CHANGE_AE, etc.
Signed-off-by: Malahal Naineni <malahal@us.ibm.com> Acked-by: Dave Wysochanski <dwysocha@redhat.com>
Peter Rajnoha [Mon, 7 Dec 2009 12:03:47 +0000 (12:03 +0000)]
Disable udev rules on change event with DISK_RO=1.
There's a new change udev event generated since kernel 2.6.32 that
notifies userspace about a change in read-only attribute for block
devices (with DISK_RO=1 environment variable set).
We need to detect this and disable the rule application so the
meaning of this change event is not interchanged with the regular
change event used while resuming/renaming DM devices.
If there's anybody awaiting this notification in foreign rules,
he can still check for this env var and do the appropriate actions
separately.
Update a few more uint64_t's related to the 64-bit status change.
At this point they probably do not matter but going forward they
may - depends on future patches for replicator, etc. I think
these probably got missed because they were 'flags' so I changed
the name to 'status' to be consistent. So the on-disk
things 'flags' and the in structure 'status' (bits).
NOTE: WHATS_NEW already has entry for this in current release.
Signed-off-by: Dave Wysochanski <dwysocha@redhat.com> Acked-by: Mike Snitzer <snitzer@redhat.com>