]> sourceware.org Git - lvm2.git/log
lvm2.git
13 years agoCritical section
Zdenek Kabelac [Fri, 18 Feb 2011 14:16:11 +0000 (14:16 +0000)]
Critical section

New strategy for memory locking to decrease the number of call to
to un/lock memory when processing critical lvm functions.

Introducing functions for critical section.

Inside the critical section - memory is always locked.
When leaving the critical section, the memory stays locked
until memlock_unlock() is called - this happens with
sync_local_dev_names() and sync_dev_names() function call.

memlock_reset() is needed to reset locking numbers after fork
(polldaemon).

The patch itself is mostly rename:

memlock_inc  -> critical_section_inc
memlock_dec  -> critical_section_dec
memlock      -> critical_section

Daemons (clmvd, dmevent) are using memlock_daemon_inc&dec
(mlockall()) thus they will never release or relock memory they've
already locked memory.

Macros sync_local_dev_names() and sync_dev_names() are functions.
It's better for debugging - and also we do not need to add memlock.h
to locking.h header (for memlock_unlock() prototyp).

13 years agoReplace PV_MIN_SIZE with function pv_min_size()
Zdenek Kabelac [Fri, 18 Feb 2011 14:11:22 +0000 (14:11 +0000)]
Replace PV_MIN_SIZE with function pv_min_size()

Add configurable option to define minimal size of
of block device usable as a PV.

pv_min_size() is added to lvm-globals and it's being
initialized through _process_config.

Macro PV_MIN_SIZE is unused and removed.

New define DEFAULT_PV_MIN_SIZE_KB is added to lvm-global
and unlike PV_MIN_SIZE it uses KB units.

Should help users with various slow devices attached to the system,
which cannot be easily filtered out (like FDD on /dev/sdX):
https://bugzilla.redhat.com/show_bug.cgi?id=644578

13 years agoSupport 64bit ints in config
Zdenek Kabelac [Fri, 18 Feb 2011 14:08:22 +0000 (14:08 +0000)]
Support 64bit ints in config

13 years agoFix for bug 677739: removing final exclusive cmirror snapshot,
Jonathan Earl Brassow [Fri, 18 Feb 2011 00:36:04 +0000 (00:36 +0000)]
Fix for bug 677739: removing final exclusive cmirror snapshot,
                    results in clvmd deadlock

When a logical volume is activated exclusively in a cluster, the
local (non-cluster-aware) target is used.  However, when creating
a snapshot on the exclusive LV, the resulting suspend/resume fails
to load the appropriate device-mapper table - instead loading the
cluster-aware target.

This patch adds an 'exclusive' parameter to the pertinent resume
functions to allow for the right target type to be loaded.

13 years agoFix an lv_postorder bug where it failed to clear temporary flags, making it
Petr Rockai [Mon, 14 Feb 2011 19:27:05 +0000 (19:27 +0000)]
Fix an lv_postorder bug where it failed to clear temporary flags, making it
impossible to use twice with the same LV(s). Discovered by Milan.

13 years agoAvoid flooding syslog with redundant messages when a snapshot is invalidated
Petr Rockai [Mon, 14 Feb 2011 14:26:36 +0000 (14:26 +0000)]
Avoid flooding syslog with redundant messages when a snapshot is invalidated
(reported by Corey).

13 years agobug 659264: more examples for pvmove
Jonathan Earl Brassow [Wed, 9 Feb 2011 22:24:55 +0000 (22:24 +0000)]
bug 659264: more examples for pvmove

13 years agoAllow overriding the DM_DEV_DIR used by the test suite (fixes BZ 672796). Use
Petr Rockai [Wed, 9 Feb 2011 17:57:14 +0000 (17:57 +0000)]
Allow overriding the DM_DEV_DIR used by the test suite (fixes BZ 672796). Use
like this:

# LVM_TEST_DEVDIR=/tmp/dev make check

13 years agoFix a typo in pvmove.8 (fixes BZ 673618, spotted by John Bradshaw).
Petr Rockai [Wed, 9 Feb 2011 17:42:15 +0000 (17:42 +0000)]
Fix a typo in pvmove.8 (fixes BZ 673618, spotted by John Bradshaw).

13 years agopost-release
Alasdair Kergon [Wed, 9 Feb 2011 12:11:21 +0000 (12:11 +0000)]
post-release

13 years agopre-release
Alasdair Kergon [Wed, 9 Feb 2011 12:04:39 +0000 (12:04 +0000)]
pre-release

13 years agoorginal WHATS_NEW entry was changed to be less in-line with change it
Jonathan Earl Brassow [Tue, 8 Feb 2011 22:35:11 +0000 (22:35 +0000)]
orginal WHATS_NEW entry was changed to be less in-line with change it
was associated with.  Update entry to be clearer than both.

13 years agoRevert DEBUG_MEM cleanup commit
Zdenek Kabelac [Tue, 8 Feb 2011 15:52:00 +0000 (15:52 +0000)]
Revert DEBUG_MEM cleanup commit

As functions compiled within this define are apparently stil part of the public API,
(though lvm2 code is never using them unless this define is used for compilation),
keep functions available in the code for now -> revert.

13 years agoFix CRC32 calculation on big endian CPU
Zdenek Kabelac [Tue, 8 Feb 2011 12:41:08 +0000 (12:41 +0000)]
Fix CRC32 calculation on big endian CPU

Fix regresion from 2.02.75 speedup - so currently crc32 is a little bit
more complicated on big-endian CPU as the uint32_t needs to be shifted
on here.

13 years agopost-release
Alasdair Kergon [Fri, 4 Feb 2011 22:46:18 +0000 (22:46 +0000)]
post-release

13 years agopre-release
Alasdair Kergon [Fri, 4 Feb 2011 22:35:29 +0000 (22:35 +0000)]
pre-release

13 years agoAdd configure option --with-device-nodes-on
Zdenek Kabelac [Fri, 4 Feb 2011 22:17:54 +0000 (22:17 +0000)]
Add configure option --with-device-nodes-on

Make configurable default behaviour how to deal with device node creates.
With udev system natural options should be  'resume'.
For older systems where user expect there is node in /dev/mapper immediately
after  dmsetup create --notable  -  use 'create'

FIXME:
Code needs fixing passing this flag through udev cookie.

13 years agopre-release
Alasdair Kergon [Fri, 4 Feb 2011 22:07:43 +0000 (22:07 +0000)]
pre-release

13 years agoWarn if secure data flag requested but not supported by kernel.
Alasdair Kergon [Fri, 4 Feb 2011 21:26:33 +0000 (21:26 +0000)]
Warn if secure data flag requested but not supported by kernel.

13 years agoPlace back some common options
Zdenek Kabelac [Fri, 4 Feb 2011 21:19:03 +0000 (21:19 +0000)]
Place back some common options

Until man pages are generated keep some common options with all commands.

13 years agoAllow snapshots in a cluster as long as they are exclusively
Jonathan Earl Brassow [Fri, 4 Feb 2011 20:30:17 +0000 (20:30 +0000)]
Allow snapshots in a cluster as long as they are exclusively
activated.

In order to achieve this, we need to be able to query whether
the origin is active exclusively (a condition of being able to
add an exclusive snapshot).

Once we are able to query the exclusive activation of an LV, we
can safely create/activate the snapshot.

A change to 'hold_lock' was also made so that a request to aquire
a WRITE lock did not replace an EX lock, which is already a form
of write lock.

13 years agoAdd --addnodeonresume, --addnodeoncreate
Zdenek Kabelac [Fri, 4 Feb 2011 19:33:53 +0000 (19:33 +0000)]
Add --addnodeonresume, --addnodeoncreate

Add new function dm_task_set_add_node() to select between 2 types
of node creation in device directory.

DM_ADD_NODE_ON_RESUME is now default and ensures node is created on
resume. Old original behavior is accessible with DM_ADD_NODE_ON_CREATE.
In this case node would be created during dmsetup create --notable.

For the user 2 new options for dmsetup create are added:
[{--addnodeonresume | --addnodeoncreate }]

Properly working node creation on resume is needed for proper operation
stacking and ability to correctly check in which state the device should
after whole udev transation.

13 years agoRemove extra sync calls.
Zdenek Kabelac [Fri, 4 Feb 2011 19:21:47 +0000 (19:21 +0000)]
Remove extra sync calls.

Remove temporaly added fs_unlock() calls to fix clmvd usablity.
Now when the message passing is properly working - they are no longer needed.
Simplify no_locking  check for VG unlock - as message is always send
for all targets - clustered & non-clustered.

13 years agoUse cluster-wide message to request device name sync
Zdenek Kabelac [Fri, 4 Feb 2011 19:18:16 +0000 (19:18 +0000)]
Use cluster-wide message to request device name sync

Thanks to CLVMD_CMD_SYNC_NAMES propagation fix the message passing started
to work. So starts to send a message before the VG is unlocked.
Removing also implicit sync in VG unlock from clmvd as now the message
is delievered and processed in do_command().
Also add support for this new message into external locking
and mask this event from further processing.

13 years agoFix operation node stacking for consecutive dm ops
Zdenek Kabelac [Fri, 4 Feb 2011 19:14:39 +0000 (19:14 +0000)]
Fix operation node stacking for consecutive dm ops

With the ability to stack many operations in one udev transaction -
in same cases we are adding and removing same device at the same time
(i.e. deactivate followed by activate).

This leads to a problem of checking stacked operations:
i.e. remove /dev/node1 followed by create /dev/node1

If the node creation is handled with udev - there is a problem as
stacked operation gives warning about existing node1 and will try to
remove it - while next operation needs to recreate it.

Current code removes all previous stacked operation if the fs op is
FS_DEL - patch adds similar behavior for FS_ADD - it will try to
remove any 'delete' operation if udev is in use.

For FS_RENAME operation it seems to be more complex. But as we
are always stacking FS_READ_AHEAD after FS_ADD operation -
should be safe to remove all previous operation on the node
when udev is running.

Code does same checking for stacking libdm and liblvm operations.

As a very simple optimization counters were added for each stacked ops
type to avoid unneeded list scans if some operation does not exists in
the list.

Enable skipping of fs_unlock() (udev sync) if only DEL operations are staked.
as we do not use lv_info for already deleted nodes.

13 years agoSuport DM_SECURE_DATA_FLAG.
Milan Broz [Fri, 4 Feb 2011 16:08:11 +0000 (16:08 +0000)]
Suport DM_SECURE_DATA_FLAG.

It will be user for cryptsetup to ensure buffers are properly
wiped when sending sensitive data (key).

13 years agoIncrease hash table size to 1024 lv names and 64 pv uuids
Zdenek Kabelac [Thu, 3 Feb 2011 16:03:13 +0000 (16:03 +0000)]
Increase hash table size to 1024 lv names and 64 pv uuids

13 years agoRemove fs_unlock from lv_resume path
Zdenek Kabelac [Thu, 3 Feb 2011 01:58:20 +0000 (01:58 +0000)]
Remove fs_unlock from lv_resume path

Keep it within clvmd until message for SYNC starts to work.

13 years agoFix wipe size when seting up mda.
Zdenek Kabelac [Thu, 3 Feb 2011 01:41:03 +0000 (01:41 +0000)]
Fix wipe size when seting up mda.

13 years agoDo not check for open_count when not needed.
Zdenek Kabelac [Thu, 3 Feb 2011 01:24:46 +0000 (01:24 +0000)]
Do not check for open_count when not needed.

Disable open_count checking in lv_info it it's not used.

Fix previous commit (comment out unsable code for now).

13 years agoSynchronize with udev for lv_info
Zdenek Kabelac [Thu, 3 Feb 2011 01:16:35 +0000 (01:16 +0000)]
Synchronize with udev for lv_info

In case the open_count is requested via lv_info - check if there
are any udev operations in-progress - and wait for them
before checking for lv_info

13 years agoa few more comments
Alasdair Kergon [Wed, 2 Feb 2011 23:57:48 +0000 (23:57 +0000)]
a few more comments

13 years agoAllow CLVMD_CMD_SYNC_NAMES to be propagated around the cluster if requested.
Alasdair Kergon [Wed, 2 Feb 2011 23:39:39 +0000 (23:39 +0000)]
Allow CLVMD_CMD_SYNC_NAMES to be propagated around the cluster if requested.

13 years agoBetter fix for no-locking udev sync and clvmd
Zdenek Kabelac [Wed, 2 Feb 2011 20:04:39 +0000 (20:04 +0000)]
Better fix for no-locking udev sync and clvmd

This is better way how to fix clustered synchronization with udev.
As the code for message passing needs fixed - put currently
fs_unlock() after every active/deactive command in clvmd to
ensure nodes are properly created in time.

13 years agoRevert wrong fix for nolock locking missing fs_unlock
Zdenek Kabelac [Wed, 2 Feb 2011 13:34:00 +0000 (13:34 +0000)]
Revert wrong fix for nolock locking missing fs_unlock

Patch was wrond and introduced recursive lock_vol
Reverting it.

13 years agoUpdate WHATS_NEW for lvmdump change.
Dave Wysochanski [Tue, 1 Feb 2011 21:41:31 +0000 (21:41 +0000)]
Update WHATS_NEW for lvmdump change.

13 years agoAdd "dmsetup ls --tree" output to lvmdump.
Dave Wysochanski [Tue, 1 Feb 2011 21:39:15 +0000 (21:39 +0000)]
Add "dmsetup ls --tree" output to lvmdump.

It would be most useful to add "dmsetup ls --tree" to the commands run.
This command helps in answering the question "which devices are actually
underneath a given LV?"

Although the info is available with other existing dmsetup commands,
adding this command gives a much clearer summary of complex setups.

Here's an example of an LVM mirror, with mirror images on partitions
created on top of multipath devices.  The multipath devices are on
simple block devices.  As you can see, it is easy to see the stacking
from the "dmsetup ls --tree" output:

vgmpathtest-lvmpathmir (253:14)
 ├─vgmpathtest-lvmpathmir_mimage_1 (253:13)
 │  └─mpath5p1 (253:5)
 │     └─mpath5 (253:2)
 │        ├─ (8:16)
 │        └─ (8:0)
 ├─vgmpathtest-lvmpathmir_mimage_0 (253:12)
 │  └─mpath6p1 (253:6)
 │     └─mpath6 (253:3)
 │        ├─ (8:48)
 │        └─ (8:32)
 └─vgmpathtest-lvmpathmir_mlog (253:11)
    └─mpath7 (253:4)
       ├─ (8:80)
       └─ (8:64)
VolGroup00-LogVol01 (253:1)
 └─ (202:2)
vgtest-lvmir (253:10)
 ├─vgtest-lvmir_mimage_1 (253:9)
 │  └─ (7:1)
 ├─vgtest-lvmir_mimage_0 (253:8)
 │  └─ (7:0)
 └─vgtest-lvmir_mlog (253:7)
    └─ (7:3)
VolGroup00-LogVol00 (253:0)
 └─ (202:2)

But it is much harder to see the stacking with only the commands today
("dmsetup info", "dmsetup status", and "dmsetup table").  We could
piece together the stacking from "dmsetup table" but it requires
further processing (take output from "dmsetup info to get
map name to major/minor, then parse "dmsetup table", etc).

13 years agofix bad 'strcmp's in 'decode_lock_type' - missing !'s
Jonathan Earl Brassow [Tue, 1 Feb 2011 17:31:40 +0000 (17:31 +0000)]
fix bad 'strcmp's in 'decode_lock_type' - missing !'s

There was no effect from having this wrong yet, because the
tree of callers only ever cared about the answer to the first
condition (!response), which determines whether a lock is
held or not.  Correct responses, however, are needed soon.

13 years agoSkip sysinit test for cluster
Zdenek Kabelac [Mon, 31 Jan 2011 22:05:30 +0000 (22:05 +0000)]
Skip sysinit test for cluster
Add #bin/bash to utils.sh
and update fsadm test to current version of aux script

13 years agoFix udev synchronization for no-locking mode
Zdenek Kabelac [Mon, 31 Jan 2011 19:52:40 +0000 (19:52 +0000)]
Fix udev synchronization for no-locking mode

Instead of implicitly syncing udev operation in clustered and
file locking code -  call synchronization directly in lock_vol() when
the operation unlocks VG

The problem is missing implicit fs_unlock() in the no_locking code.
This is used with --sysinit on read-only filesystem locking dir.
In this case vgchange -ay could exit before all udev nodes are properly
synchronised and may cause problems with accessing such node right after
vgchange --sysinint command is finished.

Add test case for vgchange --sysinit.

13 years agoAdd a debug message when uevent is not generated and we call udev_complete internally.
Peter Rajnoha [Mon, 31 Jan 2011 11:54:55 +0000 (11:54 +0000)]
Add a debug message when uevent is not generated and we call udev_complete internally.

13 years agoLower device size of pvmove test
Zdenek Kabelac [Fri, 28 Jan 2011 16:15:08 +0000 (16:15 +0000)]
Lower device size of pvmove test

Lower the size of devices in this test for speedup.
And check only once that  LVs are prepared properly.

13 years agoLower disk space usage of inconsistent mda test
Zdenek Kabelac [Fri, 28 Jan 2011 16:13:39 +0000 (16:13 +0000)]
Lower disk space usage of inconsistent mda test

Smaller size of devices speedups this test.

13 years agoReplacei sleep wait with udev settle
Zdenek Kabelac [Fri, 28 Jan 2011 16:12:45 +0000 (16:12 +0000)]
Replacei sleep wait with udev settle

Use new udev_wait command instead of unpredictable sleep waiting.
As with more devices in the system, udev processing is slower.

13 years agoQuery only test related vg name
Zdenek Kabelac [Fri, 28 Jan 2011 16:11:14 +0000 (16:11 +0000)]
Query only test related vg name

13 years agoAdd command to wait for udevadm settle
Zdenek Kabelac [Fri, 28 Jan 2011 16:10:21 +0000 (16:10 +0000)]
Add command to wait for udevadm settle

13 years agoDo not break pipes
Zdenek Kabelac [Fri, 28 Jan 2011 16:08:39 +0000 (16:08 +0000)]
Do not break pipes

As the option 'set -e -o pipefail' is very sensite on pipe breaking
stop using '-q' for grep commands.

Otherwise this command (with large enough table) would fail:

 dmsetup table | egrep -q

with exit code 141 (128 + SIGPIPE)

As Peter suggested, he prefers to keep '-o pipefail' - so make sure all
piped commands will read the whole output and will not exit too early.

13 years agoAccelerate mirror tests
Zdenek Kabelac [Fri, 28 Jan 2011 16:07:04 +0000 (16:07 +0000)]
Accelerate mirror tests

Instead of recreation of whole device set for each test round,
just cleanup created LVs to empty VG.
Lower the size of PV devices to 16MB

13 years agoDisplay duration of tests when it passed.
Zdenek Kabelac [Fri, 28 Jan 2011 16:05:38 +0000 (16:05 +0000)]
Display duration of tests when it passed.

Enhance output with info about the test duration.
Cleanup few declarations in the code.

13 years agoUpdating man pages
Zdenek Kabelac [Fri, 28 Jan 2011 16:03:38 +0000 (16:03 +0000)]
Updating man pages

13 years agoCompile code for memory debuging only with DEBUG_MEM
Zdenek Kabelac [Fri, 28 Jan 2011 16:01:32 +0000 (16:01 +0000)]
Compile code for memory debuging only with DEBUG_MEM

When it's not in use - do not compile this code.
Improves lcov code coverage results for this code a lot :)

13 years agoSet DM_UDEV_DISABLE_OTHER_RULES_FLAG for suspended DM devices in udev rules.
Peter Rajnoha [Fri, 28 Jan 2011 11:41:51 +0000 (11:41 +0000)]
Set DM_UDEV_DISABLE_OTHER_RULES_FLAG for suspended DM devices in udev rules.

This is to avoid any scanning and processing of DM devices while they are in
suspended state (e.g. a rename while the device is suspended - a CHANGE event
is generated!). Otherwise, any scanning in the rules could end up with locking
the calling process until the device is resumed and so we don't receive a
notification about udev rules completion until then (and that effectively
locks out the process awaiting the notification!).

However, we still keep 'disk' and any 'subsystem' related udev rules running.
We trust these and these should check themselves whether a device is suspended
or not, not trying to run any scanning if it is.

13 years agoUse memcpy and add error message
Zdenek Kabelac [Fri, 28 Jan 2011 10:19:00 +0000 (10:19 +0000)]
Use memcpy and add error message

strncpy (which check each byte for \0) is not need as we always copy
the length size - so using memcpy is a bit cheaper.

Add missing log_error message for failed allocation.

13 years agoSkip NULL check before dm_free
Zdenek Kabelac [Fri, 28 Jan 2011 10:16:04 +0000 (10:16 +0000)]
Skip NULL check before dm_free

dm_free checks for NULL itself.

13 years agoAvoid rebuilding of uuid validation table
Zdenek Kabelac [Fri, 28 Jan 2011 10:14:08 +0000 (10:14 +0000)]
Avoid rebuilding of uuid validation table

Small CPU relax...

13 years agoImprove lvcreate "insufficient extents" errors to "insufficient free space".
Mike Snitzer [Fri, 28 Jan 2011 02:58:00 +0000 (02:58 +0000)]
Improve lvcreate "insufficient extents" errors to "insufficient free space".

13 years agoUse O_DIRECT when reading block devices.
Alasdair Kergon [Thu, 27 Jan 2011 00:21:37 +0000 (00:21 +0000)]
Use O_DIRECT when reading block devices.

13 years agoAdd missing lvm_quit in test
Zdenek Kabelac [Wed, 26 Jan 2011 21:21:56 +0000 (21:21 +0000)]
Add missing lvm_quit in test

Fix occasional confusing memory leak report in testing.

13 years agoInitialize pool object for each row
Zdenek Kabelac [Tue, 25 Jan 2011 21:51:30 +0000 (21:51 +0000)]
Initialize pool object for each row

Fix assert abort of dmsetup (when compiled with pool debug)
dmsetup splitname --nameprefixes --noheadings --rows gvg-a2

Move pool begin in the inner loop - otherwise it would using
already 'ended' pool object.

13 years agopost-release
Alasdair Kergon [Mon, 24 Jan 2011 23:34:46 +0000 (23:34 +0000)]
post-release

13 years agopre-release
Alasdair Kergon [Mon, 24 Jan 2011 23:24:06 +0000 (23:24 +0000)]
pre-release

13 years agoBring lvscan man page up-to-date.
Alasdair Kergon [Mon, 24 Jan 2011 20:02:07 +0000 (20:02 +0000)]
Bring lvscan man page up-to-date.

13 years agoFix lvchange --test to exit cleanly.
Alasdair Kergon [Mon, 24 Jan 2011 14:19:05 +0000 (14:19 +0000)]
Fix lvchange --test to exit cleanly.

13 years agoAdd change_tag to toollib.
Alasdair Kergon [Mon, 24 Jan 2011 13:38:31 +0000 (13:38 +0000)]
Add change_tag to toollib.
Allow multiple pvchange command line options to be specified together.

13 years agoIf other process finishes (or aborts) pvmove operation and
Milan Broz [Wed, 19 Jan 2011 23:11:39 +0000 (23:11 +0000)]
If other process finishes (or aborts) pvmove operation and
polling function cannot find any lv with PVMOVE flag, return
success and do not print  "aborting" message.

Fixes https://bugzilla.redhat.com/show_bug.cgi?id=602389

13 years agoIn some versions (RHEL6) dlm_create_lockspace() always
Milan Broz [Wed, 19 Jan 2011 23:09:31 +0000 (23:09 +0000)]
In some versions (RHEL6) dlm_create_lockspace() always
return lockspace reference (even if lockspace already exists)
and thus increases DLM lockspace count. It means that after
clvmd restart the lockspace is still in use.

(The only way to clean environment to enable clean cluster
shutdown is call "dlm_tool leave clvmd" several times.)

Because only one clvmd can run in time, we can use simpler logic,
try to open lockspace with dlm_open_lockspace() and only if it fails
try to create new one. This way the lockspace reference doesn not
increase.

Very easily reproducible with  "clvmd -S" command.

Patch also fixes return code when clvmd_restart fails and fixes
double free if debug option was specified during restart.

Fixes https://bugzilla.redhat.com/show_bug.cgi?id=612862

13 years agoRemove duplicate statement (pasted twice by patch at some point?)
Jonathan Earl Brassow [Wed, 19 Jan 2011 19:24:07 +0000 (19:24 +0000)]
Remove duplicate statement (pasted twice by patch at some point?)

13 years agoAdd test to make sure that a log device is retained when a
Jonathan Earl Brassow [Wed, 19 Jan 2011 19:22:07 +0000 (19:22 +0000)]
Add test to make sure that a log device is retained when a
mirror image fails and the policies are set to:
 mirror_image_fault_policy = "replace"
 mirror_log_fault_policy = "remove"

13 years agoRemove DEBUGLOG from dmeventd.
Milan Broz [Mon, 17 Jan 2011 23:14:05 +0000 (23:14 +0000)]
Remove DEBUGLOG from dmeventd.

13 years agoAdd -f (don't fork) option to clvmd and fix clvmd -d<num> description.
Milan Broz [Mon, 17 Jan 2011 23:13:14 +0000 (23:13 +0000)]
Add -f (don't fork) option to clvmd and fix clvmd -d<num> description.

13 years agoremove unused definitions
Alasdair Kergon [Mon, 17 Jan 2011 19:02:44 +0000 (19:02 +0000)]
remove unused definitions

13 years agopost-release
Alasdair Kergon [Mon, 17 Jan 2011 18:16:18 +0000 (18:16 +0000)]
post-release

13 years agopre-release
Alasdair Kergon [Mon, 17 Jan 2011 16:46:50 +0000 (16:46 +0000)]
pre-release

13 years agoRemove -f from list of targets for cleanup
Zdenek Kabelac [Mon, 17 Jan 2011 15:18:10 +0000 (15:18 +0000)]
Remove -f from list of targets for cleanup

rm -f is already in used for this cleanup.

13 years agoDo not scan devices unnecessarily for reseting error counter
Zdenek Kabelac [Mon, 17 Jan 2011 15:16:55 +0000 (15:16 +0000)]
Do not scan devices unnecessarily for reseting error counter

For reseting error counter use directly btree cached elements and do not
create whole dev_iterator.

13 years agoSet default value to counter_last
Zdenek Kabelac [Thu, 13 Jan 2011 15:03:28 +0000 (15:03 +0000)]
Set default value to counter_last

Missed in previous commit and could be using some random stack value.

13 years agoDo not log backtrace for correctly processed command
Zdenek Kabelac [Thu, 13 Jan 2011 15:00:29 +0000 (15:00 +0000)]
Do not log backtrace for correctly processed command

13 years agoInclude default make.tmpl
Zdenek Kabelac [Thu, 13 Jan 2011 14:58:59 +0000 (14:58 +0000)]
Include default make.tmpl

Use default clean targets for this Makefile
Use also same build flags for compiling testing tools like harness.

13 years agoEnable monitoring for cluster tests
Zdenek Kabelac [Thu, 13 Jan 2011 14:57:18 +0000 (14:57 +0000)]
Enable monitoring for cluster tests

13 years agoSkip unnecessary lock_vol() call after volume deactivation
Zdenek Kabelac [Thu, 13 Jan 2011 14:56:17 +0000 (14:56 +0000)]
Skip unnecessary lock_vol() call after volume deactivation

Improve condition within lock_vol so we are not calling extra unlock
if the volume just has been deactivated.

Patch uses lck_type and replaces negative 'and' condition to more
readable 'or' condition.
Few missing strace traces added.

13 years agoAdd exec_cmd paramater sync_needed
Zdenek Kabelac [Thu, 13 Jan 2011 14:51:32 +0000 (14:51 +0000)]
Add exec_cmd paramater  sync_needed

As sync_local_dev_names() cannot be called within activation context,
add new parametr which allows to select if the sync call is needed
before executing new command.

13 years agoCheck for file with clvmd pid
Zdenek Kabelac [Thu, 13 Jan 2011 11:05:27 +0000 (11:05 +0000)]
Check for file with clvmd pid

Instead of checking  $LOCAL_CLVMD set during some 'aux' execution which
doesn't seem to be propagated to this shell - check for existance of pid
file of clvmd process - so this test is skipped in singlenode cluster test.

13 years agoHotfix to stop harness busylooping
Zdenek Kabelac [Thu, 13 Jan 2011 11:02:55 +0000 (11:02 +0000)]
Hotfix to stop harness busylooping

Harness seems to be able to busyloop in while cycle and not moving forward
for certain buffer - so check whethere there was some progress.
This fix allows to continue after failed cluster test.

Fix gcc warning for hiding global variable 's' -> sig.

13 years agoReplace fs_unlock by sync_local_dev_names to notify local clvmd. (2.02.80)
Alasdair Kergon [Wed, 12 Jan 2011 20:42:50 +0000 (20:42 +0000)]
Replace fs_unlock by sync_local_dev_names to notify local clvmd. (2.02.80)
Introduce sync_local_dev_names and CLVMD_CMD_SYNC_NAMES to issue fs_unlock.

13 years agoKill clvmd properly after locking_type 3 tests (using clvmd/singlenode).
Petr Rockai [Wed, 12 Jan 2011 16:07:55 +0000 (16:07 +0000)]
Kill clvmd properly after locking_type 3 tests (using clvmd/singlenode).

13 years agoadd fio
Alasdair Kergon [Wed, 12 Jan 2011 15:28:33 +0000 (15:28 +0000)]
add fio

13 years agos/log_verbose/log_error/ - Increase log level on error message.
Jonathan Earl Brassow [Tue, 11 Jan 2011 17:21:01 +0000 (17:21 +0000)]
s/log_verbose/log_error/ - Increase log level on error message.

13 years agoAdd disk to mirrored log type conversion.
Jonathan Earl Brassow [Tue, 11 Jan 2011 17:05:08 +0000 (17:05 +0000)]
Add disk to mirrored log type conversion.

13 years agopost-release
Alasdair Kergon [Mon, 10 Jan 2011 21:12:54 +0000 (21:12 +0000)]
post-release

13 years agoreissue 2.02.80
Alasdair Kergon [Mon, 10 Jan 2011 21:12:18 +0000 (21:12 +0000)]
reissue 2.02.80

13 years agoFix missing declaration for fs_unlock
Zdenek Kabelac [Mon, 10 Jan 2011 19:49:42 +0000 (19:49 +0000)]
Fix missing declaration for fs_unlock

13 years agoAvoid cookie sharing between forked processes
Zdenek Kabelac [Mon, 10 Jan 2011 19:31:02 +0000 (19:31 +0000)]
Avoid cookie sharing between forked processes

Before fork, ensure cookie is reset so it's not shared between processes.

13 years agopost-release
Alasdair Kergon [Mon, 10 Jan 2011 14:51:33 +0000 (14:51 +0000)]
post-release

13 years agopre-release
Alasdair Kergon [Mon, 10 Jan 2011 14:28:17 +0000 (14:28 +0000)]
pre-release

13 years agoSpeedup consequent activation calls
Zdenek Kabelac [Mon, 10 Jan 2011 14:02:30 +0000 (14:02 +0000)]
Speedup consequent activation calls

Stop calling fs_unlock() from lv_de/activate().
Start using internal lvm fs cookie for dm_tree.
Stop directly calling dm_udev_wait() and
dm_tree_set/get_cookie() from activate code -
it's now called through fs_unlock() function.

Add lvm_do_fs_unlock()

Call fs_unlock() when unlocking vg where implicit unlock solves the
problem also for cluster - thus no extra command for clustering
environment is required - only lvm_do_fs_unlock() function is added
to call lvm's fs_unlock() while holding lvm_lock mutex in clvmd.

Add fs_unlock() also to set_lv() so the command waits until devices
are ready for regular open (i.e. wiping its begining).

Move fs_unlock() prototype to activation.h to keep fs.h private
in lib/activate dir and not expose other functions from this header.

13 years agoAdd internal fs cookie
Zdenek Kabelac [Mon, 10 Jan 2011 13:44:39 +0000 (13:44 +0000)]
Add internal fs cookie

Add functions for handling internal lvm cookie used for
all dm_tree operations until fs_unlock is called.

13 years agoDefine DM_COOKIE_AUTO_CREATE
Zdenek Kabelac [Mon, 10 Jan 2011 13:42:31 +0000 (13:42 +0000)]
Define DM_COOKIE_AUTO_CREATE

TODO: Use it also for already written code.

13 years agoUse strndup as GNU extension
Zdenek Kabelac [Mon, 10 Jan 2011 13:25:22 +0000 (13:25 +0000)]
Use strndup as GNU extension

strndup needs _GNU_SOURCE defined as it's GNU extension.
Remove duplicate string.h header added in previous harness commit.

13 years agoCache config_tree
Zdenek Kabelac [Mon, 10 Jan 2011 13:15:57 +0000 (13:15 +0000)]
Cache config_tree

Start to use config_tree for cache just like vgmetadata.
When vgmetadata are erased destroy its cached config tree.

13 years agoChange import_vg_from_buffer to use config_tree
Zdenek Kabelac [Mon, 10 Jan 2011 13:13:42 +0000 (13:13 +0000)]
Change import_vg_from_buffer to use config_tree

Change function import_vg_from_buffer() to import_vg_from_config_tree().
Instead of creating config tree inside the function allow config tree to
be passed as parameter - usable later for caching.

This page took 0.071105 seconds and 5 git commands to generate.