]> sourceware.org Git - lvm2.git/log
lvm2.git
8 years agopvscan: restore polling in autoactivation handler
Ondrej Kozina [Thu, 7 Jan 2016 14:17:08 +0000 (15:17 +0100)]
pvscan: restore polling in autoactivation handler

This commit fixes regression in auto-activation code introduced
in commit: c26d81d6e6939906729d91fae83cd8bbdd743bb7.

- resolves rhbz1295562

8 years agoreport: use brackets to signify LVs which are not visible when reporting lv_parent
Peter Rajnoha [Mon, 11 Jan 2016 14:34:35 +0000 (15:34 +0100)]
report: use brackets to signify LVs which are not visible when reporting lv_parent

Use common _lvname_disp to report lv_parent. The _lvname_disp
takes care of properly marking LVs which are not visible - such
LVs are always enclosed in brackets when reported within any
other field.

For example, thin pool over RAID.

Before:

$ lvs -a -o name,lv_parent,data_lv,metadata_lv vg
  LV                          Parent           Data               Meta
  cache_pool                                   [cache_pool_tdata] [cache_pool_tmeta]
  [cache_pool_tdata]          cache_pool
  [cache_pool_tdata_rimage_0] cache_pool_tdata
  [cache_pool_tdata_rimage_1] cache_pool_tdata
  [cache_pool_tdata_rmeta_0]  cache_pool_tdata
  [cache_pool_tdata_rmeta_1]  cache_pool_tdata
  [cache_pool_tmeta]          cache_pool
  [cache_pool_tmeta_rimage_0] cache_pool_tmeta
  [cache_pool_tmeta_rimage_1] cache_pool_tmeta
  [cache_pool_tmeta_rmeta_0]  cache_pool_tmeta
  [cache_pool_tmeta_rmeta_1]  cache_pool_tmeta
  [lvol0_pmspare]

With this patch applied:

$ lvs -a -o name,lv_parent,data_lv,metadata_lv vg
  LV                          Parent             Data               Meta
  cache_pool                                     [cache_pool_tdata] [cache_pool_tmeta]
  [cache_pool_tdata]          cache_pool
  [cache_pool_tdata_rimage_0] [cache_pool_tdata]
  [cache_pool_tdata_rimage_1] [cache_pool_tdata]
  [cache_pool_tdata_rmeta_0]  [cache_pool_tdata]
  [cache_pool_tdata_rmeta_1]  [cache_pool_tdata]
  [cache_pool_tmeta]          cache_pool
  [cache_pool_tmeta_rimage_0] [cache_pool_tmeta]
  [cache_pool_tmeta_rimage_1] [cache_pool_tmeta]
  [cache_pool_tmeta_rmeta_0]  [cache_pool_tmeta]
  [cache_pool_tmeta_rmeta_1]  [cache_pool_tmeta]
  [lvol0_pmspare]

8 years agocleanup: use _field_set_value and _string_disp consistently in report.c
Peter Rajnoha [Mon, 11 Jan 2016 14:01:35 +0000 (15:01 +0100)]
cleanup: use _field_set_value and _string_disp consistently in report.c

Do not mix dm_report_field_set_value and _field_set_value and
use single function call throughout for clarity. The same applies
for dm_report_field_string and _string_disp.

8 years agoreport: fix invalid memory read when reporting cache LV policy name
Peter Rajnoha [Mon, 11 Jan 2016 11:51:08 +0000 (12:51 +0100)]
report: fix invalid memory read when reporting cache LV policy name

Fix regression caused by commit c2d4330f27277717bc3b684b702189079b257b77
which removed the dm_pool_strdup for the cache policy name in
_cache_policy_disp report function.

This regression was hit with buffered reporting only (which is
used by default). The reason is that for buffered reporting, we're
iterating over LVs in VG (process_each_lv) while gathering
all the information that is needed for the report. In this case,
the LV's cache policy name has not been duped, but only the pointer
to the original VG buffer was stored. When the LV iteration finished,
the VG buffer was freed and any report to output called later
(dm_report_output call) accessed already freed VG data.

This didn't appear if unbuffered reporting was used (--unbuffered)
because in this case, the data were reported to output as
soon as they were processed, hence it was reported to output
before the VG data was freed.

8 years agopost-release
Alasdair G Kergon [Fri, 8 Jan 2016 18:51:08 +0000 (18:51 +0000)]
post-release

8 years agopre-release v2_02_139
Alasdair G Kergon [Fri, 8 Jan 2016 18:46:41 +0000 (18:46 +0000)]
pre-release

8 years agodocument commits since last release
David Teigland [Fri, 8 Jan 2016 15:53:58 +0000 (09:53 -0600)]
document commits since last release

8 years agoman: lvs: document F,D and M thin pool health status chars for lv_attr in lvs man...
Peter Rajnoha [Fri, 8 Jan 2016 14:47:01 +0000 (15:47 +0100)]
man: lvs: document F,D and M thin pool health status chars for lv_attr in lvs man page

8 years agolvmdump: also add lvm2-activation{-early,-net}.service systemd status for lvmdump -s
Peter Rajnoha [Mon, 4 Jan 2016 14:10:07 +0000 (15:10 +0100)]
lvmdump: also add lvm2-activation{-early,-net}.service systemd status for lvmdump -s

The lvm2-activation{-early,-net}.service systemd unit statuses were missing
in dump gathered by lvmdump -s. These are quite important when debugging
scenarios with systemd environment and where lvmetad is not used.

8 years agolvmlockd: update VG lock version earlier
David Teigland [Tue, 15 Dec 2015 22:14:49 +0000 (16:14 -0600)]
lvmlockd: update VG lock version earlier

Have commands send lvmlockd the update message
in vg_write instead of vg_commit, so that it's
not done while LVs are suspended.  If the vg_write
is not committed, and the seqno sent to lvmlockd
is not used, then lvmlockd can detect this when
the next update uses the same seqno.

8 years agovgrename: use process_each_vg
David Teigland [Tue, 1 Dec 2015 20:09:01 +0000 (14:09 -0600)]
vgrename: use process_each_vg

Use process_each_vg() to lock and read the old VG,
and then call the main vgrename code.

When real VG names are used (not a UUID in place of the
old name), the command still pre-locks the new name
(when strcmp wants it locked first), before calling
process_each_vg on the old name.

In the case where the old name is replaced with a UUID,
process_each_vg now translates that UUID into the real
VG name, which it locks and reads.  In this case, we
cannot do pre-locking to maintain lock ordering because
the old name is unknown.  So, in this case the strcmp
based lock ordering is suppressed and the old name is
always locked first.  This opens a remote chance for
lock ordering conflict between racing vgrenames between
two names where one or both commands use the UUID.

8 years agopvscan: Remove duplicate filter wipe.
Alasdair G Kergon [Mon, 14 Dec 2015 20:14:59 +0000 (20:14 +0000)]
pvscan: Remove duplicate filter wipe.

Also always clear the internal lvmcache after rescanning, and
reinstate a test for --trustcache so that 'pvs --trustcache'
(for example) avoids rescanning.

8 years agoprocess_each_pv: do full scan earlier to find new devices
David Teigland [Fri, 11 Dec 2015 20:02:36 +0000 (14:02 -0600)]
process_each_pv: do full scan earlier to find new devices

Before commit c1f246fedfc349c25749da501e68a7f70bd122b0,
_get_all_devices() did a full device scan before
get_vgnameids() was called.  The full scan in
_get_all_devices() is from calling dev_iter_create(f, 1).
The '1' arg forces a full scan.

By doing a full scan in _get_all_devices(), new devices
were added to dev-cache before get_vgnameids() began
scanning labels.  So, labels would be read from new devices.
(e.g. by the first 'pvs' command after the new device appeared.)

After that commit, _get_all_devices() was called
after get_vgnameids() was finished scanning labels.
So, new devices would be missed while scanning labels.
When _get_all_devices() saw the new devices (after
labels were scanned), those devices were added to
the .cache file.  This meant that the second 'pvs'
command would see the devices because they would be
in .cache.

Now, the full device scan is factored out of
_get_all_devices() and called by itself at the
start of the command so that new devices will
be known before get_vgnameids() scans labels.

8 years agopost-release
Alasdair G Kergon [Mon, 14 Dec 2015 12:25:48 +0000 (12:25 +0000)]
post-release

8 years agopre-release v2_02_138
Alasdair G Kergon [Mon, 14 Dec 2015 12:24:21 +0000 (12:24 +0000)]
pre-release

8 years agoChange messages from verbose to debug
David Teigland [Fri, 11 Dec 2015 21:28:46 +0000 (15:28 -0600)]
Change messages from verbose to debug

These messages about outdated PVs should not
be verbose because they always appear, even
when there are no outdated PVs.

8 years agocleanup: add missing WHATS_NEW
Zdenek Kabelac [Fri, 11 Dec 2015 19:13:20 +0000 (20:13 +0100)]
cleanup: add missing WHATS_NEW

8 years agotests: check lvrename of stacked cache pool
Zdenek Kabelac [Fri, 11 Dec 2015 19:11:11 +0000 (20:11 +0100)]
tests: check lvrename of stacked cache pool

8 years agofix static linking
Riku Voipio [Tue, 8 Dec 2015 14:40:08 +0000 (16:40 +0200)]
fix static linking

Static linking fails currently, as -lm and -lpthread are missing:

gcc -O2  -fPIC -O2  -L../libdm -L../lib -L../libdaemon/client -static
-L../libdm/ioctl \
      -o dmsetup.static dmsetup.o -ldevmapper    -lrt
../libdm/ioctl/libdevmapper.a(libdm-stats.o): In function
`dm_stats_create_region':
libdm-stats.c:(.text+0x2d69): undefined reference to `log10'
libdm-stats.c:(.text+0x2d6e): undefined reference to `lround'
../libdm/ioctl/libdevmapper.a(pool.o): In function `dm_pool_create':
pool.c:(.text+0x134): undefined reference to `pthread_mutex_lock'
pool.c:(.text+0x14f): undefined reference to `pthread_mutex_unlock'

Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
8 years agotests: fix logging
Zdenek Kabelac [Thu, 10 Dec 2015 19:47:30 +0000 (20:47 +0100)]
tests: fix logging

Actually  file redirection must be before stderr redir.

8 years agolvrename: always allow to rename pools
Zdenek Kabelac [Wed, 9 Dec 2015 12:52:47 +0000 (13:52 +0100)]
lvrename: always allow to rename pools

Since we mark cache-pool as 'hidden/private' while it is in-use,
we may still allow user to change it's name.

It should not cause any harm and user may prefer better naming
for a cache-pool in use.

8 years agocache: support stacked rename
Zdenek Kabelac [Mon, 7 Dec 2015 12:53:00 +0000 (13:53 +0100)]
cache: support stacked rename

Preserve skip_pool flag when running for_each_sub_lv() so
lvrename continues to work when thin-pool is using cached
data LV.

8 years agolvmlockd: reconnect to lvmetad if it's restarted
David Teigland [Thu, 10 Dec 2015 16:50:19 +0000 (10:50 -0600)]
lvmlockd: reconnect to lvmetad if it's restarted

If lvmetad is restarted after lvmlockd has connected
to it, then lvmlockd should reconnect.

8 years agolvrename: move the lvmlockd LV lock
David Teigland [Wed, 9 Dec 2015 17:51:25 +0000 (11:51 -0600)]
lvrename: move the lvmlockd LV lock

The function it was in is used for various
internal renaming of hidden LVs where a lock
from lvmlockd does not apply.

8 years agodmeventd: Don't trust fifo with wrong attrs.
Alasdair G Kergon [Tue, 8 Dec 2015 01:48:17 +0000 (01:48 +0000)]
dmeventd: Don't trust fifo with wrong attrs.

If an existing fifo has the wrong attributes it cannot be trusted
so we must unlink it and recreate it correctly.
(Replaces 2c8d6f5c90d5be62b48ba2881f2a6631091dc5af: if the other end of
the fifo already got opened while its mode was insecure, delaying the
chmod isn't going to make any difference!)

8 years agodmeventd: Extend checks on client socket.
Alasdair G Kergon [Tue, 8 Dec 2015 00:54:32 +0000 (00:54 +0000)]
dmeventd: Extend checks on client socket.

Reinstate and extend checks removed by e1b111b02accb4145b82b8b47ce57ed93b1a7184.

The code has always assumed that only root has access to the directory
containing the fifos and that they are under the complete control of
dmeventd code.  If anything is found not to be as expected, then open()
should certainly not be attempted!

8 years agopost-release
Alasdair G Kergon [Sat, 5 Dec 2015 15:36:22 +0000 (15:36 +0000)]
post-release

8 years agopre-release v2_02_137
Alasdair G Kergon [Sat, 5 Dec 2015 15:33:19 +0000 (15:33 +0000)]
pre-release

8 years agotests: extend test
Zdenek Kabelac [Fri, 4 Dec 2015 21:09:05 +0000 (22:09 +0100)]
tests: extend test

8 years agoarchiver: fix reporting for check_current_backup
Zdenek Kabelac [Fri, 4 Dec 2015 21:04:14 +0000 (22:04 +0100)]
archiver: fix reporting for check_current_backup

It's getting a bit more complex here.

Basic idea behind is - check_current_backup() should not
log error when a user is using a read-only filesystem,
so e.g.  vgscan  will not report any error when it tries
to take missing backup.

We still have cases when error could be reported though,
e.g. the backup this would be a symbolic link, but these
are rather misconfiguration and unexpected case.

8 years agodebug: correct stack tracing
Zdenek Kabelac [Fri, 4 Dec 2015 21:02:24 +0000 (22:02 +0100)]
debug: correct stack tracing

Here the 'goto' is correct path, as  !device_is_usable
is traceable with <backtrace>.

Keep the 'stack' for unusable device.

8 years agolibdm: add some doc for mirror status
Zdenek Kabelac [Fri, 4 Dec 2015 19:24:47 +0000 (20:24 +0100)]
libdm: add some doc for mirror status

Comment content of struct for mirror status.

8 years agolvconvert: Reinstate raid merge after splitmirror.
Alasdair G Kergon [Thu, 3 Dec 2015 17:40:10 +0000 (17:40 +0000)]
lvconvert: Reinstate raid merge after splitmirror.

After commit 46c8d6bb8ae91ee67c8633496cc6c3d92bb4f3ce
(lvconvert: Improve message for raid without -m)

8 years agotests: check read-only backup archive
Zdenek Kabelac [Thu, 3 Dec 2015 17:16:07 +0000 (18:16 +0100)]
tests: check read-only backup archive

8 years agovgextend: reinstantiate archiving
Zdenek Kabelac [Thu, 3 Dec 2015 17:15:02 +0000 (18:15 +0100)]
vgextend: reinstantiate archiving

Since commit f5d06efbab4c7c77298b18b95c23111575a8db1a we lost archiving.
Restore it now with process_each_vg.

8 years agocleanup: drop log_suppress(2) usage
Zdenek Kabelac [Thu, 3 Dec 2015 16:51:43 +0000 (17:51 +0100)]
cleanup: drop log_suppress(2) usage

No longer need to use  log_suppress(2) instance so dropped.

8 years agoarchiver: inital change toward proper logging
Zdenek Kabelac [Thu, 3 Dec 2015 16:32:47 +0000 (17:32 +0100)]
archiver: inital change toward proper logging

We have to modes of  'archive()' usage -
1. compulsory - fail stops command and user may try '-An' option
to do a command.

2. non-compulsory - some fails in archiving are ignorable (i.e.
read-only filesystem where archive dir is located).

Those 2 cases needs to be properly handle - i.e. the non-compulsory
logging should not be tampering  error logging message production.

So more work here is needed

8 years agolog: use full buffer size for printf
Zdenek Kabelac [Thu, 3 Dec 2015 14:28:01 +0000 (15:28 +0100)]
log: use full buffer size for printf

Pass full buffer size to printf() function - no reason to make
buffer 1 char smaller.

Also rename locn buffer to message buffer directly since it's
not used for anything else.

TODO: we may use same buffer also for 'buf[]' since there is
no collision - so may safe 1K on stack usage.

8 years agocleanup: use try_id_read_format
Zdenek Kabelac [Thu, 3 Dec 2015 14:26:44 +0000 (15:26 +0100)]
cleanup: use try_id_read_format

Better then using log_suppress in this case.

8 years agouuid: add id_read_format_try
Zdenek Kabelac [Thu, 3 Dec 2015 14:25:09 +0000 (15:25 +0100)]
uuid: add id_read_format_try

Provide id_read_format() functionality without log_error
when ID is not valid one - it will just return 0.

Does not need to use log_suppress() then.

8 years agolvconvert: Reinstate mirror to raid conversions.
Alasdair G Kergon [Thu, 3 Dec 2015 14:40:14 +0000 (14:40 +0000)]
lvconvert: Reinstate mirror to raid conversions.

Reinstate conversions from mirror to raid stopped
by commit 46c8d6bb8ae91ee67c8633496cc6c3d92bb4f3ce
(lvconvert: Improve message for raid without -m).

8 years agoman lvm: add section about unique VG names
David Teigland [Wed, 2 Dec 2015 16:37:25 +0000 (10:37 -0600)]
man lvm: add section about unique VG names

8 years agotoollib: only interpret vgname arg as uuid for vgrename
David Teigland [Tue, 1 Dec 2015 21:50:14 +0000 (15:50 -0600)]
toollib: only interpret vgname arg as uuid for vgrename

In general, --select should be used to specify a VG by UUID,
but vgrename already allows a uuid to be substituted for
the name, so continue to allow it in that case.

8 years agotoollib: allow VG UUID to be used in place of VG name
David Teigland [Tue, 1 Dec 2015 18:00:57 +0000 (12:00 -0600)]
toollib: allow VG UUID to be used in place of VG name

If the VG arg from the command line does not match the
name of any known VGs, then check if the arg looks like
a UUID.  If it's a valid UUID, then compare it to the
UUID of known VGs.  If it matches the UUID of a known VG,
then process that VG.

8 years agolvresize: use process_each_vg
David Teigland [Tue, 1 Dec 2015 17:01:26 +0000 (11:01 -0600)]
lvresize: use process_each_vg

No functional change.

8 years agolvrename: use process_each_vg
David Teigland [Tue, 1 Dec 2015 15:54:33 +0000 (09:54 -0600)]
lvrename: use process_each_vg

No functional change.

8 years agolvcreate: use process_each_vg
David Teigland [Mon, 30 Nov 2015 23:32:49 +0000 (17:32 -0600)]
lvcreate: use process_each_vg

No functional change.

8 years agovgextend: pass single vgname as process_each_vg arg
David Teigland [Mon, 30 Nov 2015 23:00:26 +0000 (17:00 -0600)]
vgextend: pass single vgname as process_each_vg arg

Pass the single vgname as a new process_each_vg arg
instead of setting a cmd flag to tell process_each_vg
to take only the first vgname arg from argv.

Other commands with different argv formats will be
able to use it this way.

8 years agolvmcache: change duplicate VG name warnings to verbose
David Teigland [Mon, 30 Nov 2015 21:28:22 +0000 (15:28 -0600)]
lvmcache: change duplicate VG name warnings to verbose

When two different VGs with the same name exist,
they are both stored in lvmcache using the vginfo->next
list.  Previously, the code would print warnings (sometimes)
when adding VGs to this list.  Now the duplicate VG names
are handled by higher level code, so this list no longer
needs to print warnings about duplicate VG names being found.

8 years agovg_read: look up vgid from name
David Teigland [Mon, 30 Nov 2015 21:12:01 +0000 (15:12 -0600)]
vg_read: look up vgid from name

After recent changes to process_each, vg_read() is usually
given both the vgname and vgid for the intended VG.

However, in some cases vg_read() is given a vgid with
no vgname, or is given a vgname with no vgid.

When given a vgid with no vgname, vg_read() uses lvmcache
to look up the vgname using the vgid.  If the vgname is
not found, vg_read() fails.

When given a vgname with no vgid, vg_read() should also
use lvmcache to look up the vgid using the vgname.
If the vgid is not found, vg_read() fails.

If the lvmcache lookup finds multiple vgids for the
vgname, then the lookup fails, causing vg_read() to fail
because the intended VG is uncertain.

Usually, both vgname and vgid for the intended VG are passed
to vg_read(), which means the lvmcache translations
between vgname and vgid are not done.

8 years agotoollib: remove unused function
David Teigland [Mon, 30 Nov 2015 20:17:33 +0000 (14:17 -0600)]
toollib: remove unused function

8 years agoprocess_each: resolve duplicate VG names
David Teigland [Mon, 30 Nov 2015 18:11:01 +0000 (12:11 -0600)]
process_each: resolve duplicate VG names

If two different VGs with the same name exist on the system,
a command that just specifies that ambiguous name will fail
with a new error:

$ vgs -o name,uuid
  ...
  foo qyUS65-vn32-TuKs-a8yF-wfeQ-7DkF-Fds0uf
  foo vfhKCP-mpc7-KLLL-Uh08-4xPG-zLNR-4cnxJX

$ lvs foo
  Multiple VGs found with the same name: foo
  Use the --select option with VG UUID (vg_uuid).

$ vgremove foo
  Multiple VGs found with the same name: foo
  Use the --select option with VG UUID (vg_uuid).

$ lvs -S vg_uuid=qyUS65-vn32-TuKs-a8yF-wfeQ-7DkF-Fds0uf
  lv1 foo ...

This is implemented for process_each_vg/lv, and works
with or without lvmetad.  It does not work for commands
that do not use process_each.

This change includes one exception to the behavior shown
above.  If one of the VGs is foreign, and the other is not,
then the command assumes that the intended VG is the local
one and uses it.

8 years agoprocess_each: always use list of vgnames on system
David Teigland [Mon, 30 Nov 2015 18:51:44 +0000 (12:51 -0600)]
process_each: always use list of vgnames on system

This makes process_each_vg/lv always use the list of
vgnames on the system.  When specific VGs are named on
the command line, the corresponding entries from
vgnameids_on_system are moved to vgnameids_to_process.

Previously, when specific VGs were named on the command
line, the vgnameids_on_system list was not created, and
vgnameids_to_process was created from the arg_vgnames
list (which is only names, without vgids).

Now, vgnameids_on_system is always created, and entries
are moved from that list to vgnameids_to_process -- either
some (when arg_vgnames specifies only some), or all (when
the command is processing all VGs, or needs to look at
all VGs for checking tags/selection).

This change adds one new lvmetad lookup (vg_list) to a
command that specifies VG names.  It adds no new work
for other commands, e.g. non-lvmetad commands, or
commands that look at all VGs.

When using lvmetad, 'lvs foo' previously sent one
request to lvmetad: 'vg_lookup foo'.
Now, 'lvs foo' sends two requests to lvmetad:
'vg_list' and 'vg_lookup foo <uuid>'.

(The lookup can now always include the uuid in the request
because the initial vg_list contains name/vgid pairs.)

8 years agomirror: fix condition
Zdenek Kabelac [Tue, 1 Dec 2015 12:59:20 +0000 (13:59 +0100)]
mirror: fix condition

Recent patch tested wrong condition for error

8 years agocleanup: use dm_get_status_mirror
Zdenek Kabelac [Mon, 30 Nov 2015 20:17:25 +0000 (21:17 +0100)]
cleanup: use dm_get_status_mirror

Use libdm function to parse mirror status report.

8 years agolib: pass mem pool to check_transient_status
Zdenek Kabelac [Mon, 30 Nov 2015 20:24:31 +0000 (21:24 +0100)]
lib: pass mem pool to check_transient_status

check_transient_status() may need to allocate some memory,
so pass in already existing mem pool.

8 years agotests: unit test for mirror status
Zdenek Kabelac [Mon, 30 Nov 2015 19:54:52 +0000 (20:54 +0100)]
tests: unit test for mirror status

8 years agolibdm: introduce dm_get_status_mirror
Zdenek Kabelac [Mon, 30 Nov 2015 20:16:43 +0000 (21:16 +0100)]
libdm: introduce  dm_get_status_mirror

Add missing function to parse mirror status.

8 years agolvconvert: Improve message for raid without -m.
Alasdair G Kergon [Mon, 30 Nov 2015 22:36:05 +0000 (22:36 +0000)]
lvconvert: Improve message for raid without -m.

8 years agolvmcache: new function to check if VG is foreign
David Teigland [Mon, 30 Nov 2015 17:54:56 +0000 (11:54 -0600)]
lvmcache: new function to check if VG is foreign

When not using lvmetad, this uses the system_id field in
the cached vginfo structs that are populated during a scan.

When using lvmetad, this requests the VG from lvmetad, and
checks the system_id field in the returned metadata.

8 years agosystem_id: refactor check for allowed system_id
David Teigland [Mon, 30 Nov 2015 17:46:55 +0000 (11:46 -0600)]
system_id: refactor check for allowed system_id

Refactor the code that checks for an allowable system_id
so that it can be used from other places.

8 years agolvmcache: include system_id in vginfo cache
David Teigland [Mon, 30 Nov 2015 17:32:17 +0000 (11:32 -0600)]
lvmcache: include system_id in vginfo cache

Save system_id just like creation_host and lock_type
strings in vginfo cache.

8 years agolvmetad: include both vgid and vgname in lookup request
David Teigland [Tue, 24 Nov 2015 20:47:53 +0000 (14:47 -0600)]
lvmetad: include both vgid and vgname in lookup request

When the command already knows both the vgid and vgname,
it should send both to lvmetad for a more exact request,
and it can save lvmetad the work of a name lookup.

8 years agopost-release
Alasdair G Kergon [Sat, 28 Nov 2015 01:29:00 +0000 (01:29 +0000)]
post-release

8 years agopre-release v2_02_136
Alasdair G Kergon [Sat, 28 Nov 2015 01:25:53 +0000 (01:25 +0000)]
pre-release

8 years agotests: make unit testing usable again
Zdenek Kabelac [Fri, 27 Nov 2015 09:54:57 +0000 (10:54 +0100)]
tests: make unit testing usable again

Make unit tests usable/compilable with newer header files.
Add 'initial' dmlist_t  for list tests.
More will come...

8 years agocleanup: clean gcc shadow declaration of version warning
Zdenek Kabelac [Thu, 26 Nov 2015 20:52:22 +0000 (21:52 +0100)]
cleanup: clean gcc shadow declaration of version warning

8 years agodebug: show LV name where dlid creation failed
Zdenek Kabelac [Thu, 26 Nov 2015 20:52:05 +0000 (21:52 +0100)]
debug: show LV name where dlid creation failed

8 years agocleanup: using display_lvname
Zdenek Kabelac [Wed, 25 Nov 2015 15:06:31 +0000 (16:06 +0100)]
cleanup: using display_lvname

Use for showing vgname/lvname in messages.
No functional change.

8 years agocleanup: remove unused code
Zdenek Kabelac [Wed, 25 Nov 2015 14:45:19 +0000 (15:45 +0100)]
cleanup: remove unused code

Remove long outstand unused code lines, which were already
been obsoleted by other code.

Statuses and snapshot tree creation is already handled differently.

Also drop some 'extra' log_error() and use only stack;
since error has already been reported.

8 years agocleanup: avoid allocation for vg_name
Zdenek Kabelac [Wed, 25 Nov 2015 15:04:27 +0000 (16:04 +0100)]
cleanup: avoid allocation for vg_name

Since we do not use dev_manager in a way we would have destroyed VG
content while  in-use - we could safely keep just pointer.
So dropping strdup.

Also it seems we actually no longer use vg_name for anything
so it may possibly go away completely unless it would be useful
for debugging...

8 years agolibdm: const raid params and error for unsupported type
Zdenek Kabelac [Thu, 26 Nov 2015 08:18:46 +0000 (09:18 +0100)]
libdm: const raid params and error for unsupported type

Accept const struct with raid params (No API change).
Also add extra error message when raid type is unsupported.

8 years agolvmconfig: add --sinceversion for --type new
Peter Rajnoha [Wed, 25 Nov 2015 13:09:20 +0000 (14:09 +0100)]
lvmconfig: add --sinceversion for --type new

Just for convenience to display all new configuration settings
introduced since given version (before, there was only --atversion
to display settings introduced in concrete version).

For example:
$ lvmconfig --type new --sinceversion 2.2.120
allocation {
# cache_mode="writethrough"
# cache_settings {
# }
}
global {
use_lvmlockd=0
# lvmlockd_lock_retries=3
# sanlock_lv_extend=256
use_lvmpolld=1
}
activation {
}
# report {
# compact_output_cols=""
# time_format="%Y-%m-%d %T %z"
# }
local {
# host_id=0
}

8 years agocleanup: rename lv_ondisk to lv_committed
Zdenek Kabelac [Wed, 25 Nov 2015 10:10:32 +0000 (11:10 +0100)]
cleanup: rename lv_ondisk to lv_committed

Patch has no functional change.

8 years agocleanup: change ondisk committed
Zdenek Kabelac [Wed, 25 Nov 2015 09:52:22 +0000 (10:52 +0100)]
cleanup: change ondisk committed

Patch has no functional change.

8 years agocleanup: rename vg_ondisk to vg_committed
Zdenek Kabelac [Tue, 24 Nov 2015 22:29:18 +0000 (23:29 +0100)]
cleanup: rename vg_ondisk to vg_committed

Unifying terminology.

Since all the metadata in-use are ALWAYS on disk - switch
to terminology  committed and precommitted.

Patch has no functional change inside.

8 years agothin: fix previous update of partial tree building
Zdenek Kabelac [Tue, 24 Nov 2015 22:24:11 +0000 (23:24 +0100)]
thin: fix previous update of partial tree building

We do want to preserve 'active' thin-pool,
so add this 'fake' layer only when activating.

TODO:  think how to use thin-pool without fake LV layer.

8 years agotests: improve teardown
Zdenek Kabelac [Mon, 23 Nov 2015 22:17:54 +0000 (23:17 +0100)]
tests: improve teardown

Do not try to execute vgremove, when test has left suspended devices.

8 years agotests: data correctness after thin-pool resize
Zdenek Kabelac [Fri, 20 Nov 2015 21:28:21 +0000 (22:28 +0100)]
tests: data correctness after thin-pool resize

8 years agocleanup: use display_lvname
Zdenek Kabelac [Sun, 22 Nov 2015 22:03:01 +0000 (23:03 +0100)]
cleanup: use display_lvname

Switch debug msg to use display_lvname.
Link to VG early, so we have access to VG from LV.

8 years agothin: skip detach preload from pools
Zdenek Kabelac [Sun, 22 Nov 2015 22:04:11 +0000 (23:04 +0100)]
thin: skip detach preload from pools

lv preload for detached LVs started to be used also
for various other types which just happens to pass through
weak if() condition.

TODO: find here better solution to rather explicitly check
for types we really need to preload.

8 years agocleanup: move towards using direct LV pointers
Zdenek Kabelac [Sat, 21 Nov 2015 22:31:44 +0000 (23:31 +0100)]
cleanup: move towards using direct LV pointers

We do not won't to 'expose'  internals of VG struct.
ATM we use lists to keep all LVs - we may want to switch
to better struct for quicker 'search'.

Since we do not need 'lists' but always actual LV,
switch find_lv_in_vg_by_lvid() to return LV,
and replaces some use case of  find_lv_in_vg()
with 'better' working find_lv() which already
returns LV.

8 years agothin: work with active thin-pool
Zdenek Kabelac [Mon, 23 Nov 2015 18:53:43 +0000 (19:53 +0100)]
thin: work with active thin-pool

When 'lvextend -L+XX vg/thinpool'  do not leave inactive table
loaded for 'wrapping' LV on top of resized thin-pool
(ATM we use linear  LV for this with same size as thin-pool).

8 years agomemlock: add more libs on ignore list
Zdenek Kabelac [Sat, 21 Nov 2015 23:26:06 +0000 (00:26 +0100)]
memlock: add more libs on ignore list

Udev recently start to 'link-in' major amount of useless libs.
(Seem to be faulty 'systemd' link-in all issue)
Anyway - avoid locking those libs in RAM.

8 years agolibdm: enhance thin-pool preload
Zdenek Kabelac [Mon, 23 Nov 2015 22:15:04 +0000 (23:15 +0100)]
libdm: enhance thin-pool preload

When preloading thin-pool device node for already
existing/running thin-pool do not resume such thin-pool.

This allows to properly schedule commit point for metadata,
when thin-pool data or metadata volume is resized.

8 years agolibdm: drop extra space from cache target line
Zdenek Kabelac [Mon, 23 Nov 2015 22:12:43 +0000 (23:12 +0100)]
libdm: drop extra space from cache target line

Extra space between 'cache' target and metadata device caused
string comparation being not equal and thus always causing
table reload even when uneeded.

8 years agoman lvmcache: include chunk size
David Teigland [Mon, 23 Nov 2015 17:57:41 +0000 (11:57 -0600)]
man lvmcache: include chunk size

8 years agopost-release
Alasdair G Kergon [Mon, 23 Nov 2015 03:40:34 +0000 (03:40 +0000)]
post-release

8 years agopre-release v2_02_135
Alasdair G Kergon [Mon, 23 Nov 2015 03:33:43 +0000 (03:33 +0000)]
pre-release

8 years agotests: more cache conversion checks
Zdenek Kabelac [Thu, 19 Nov 2015 13:31:58 +0000 (14:31 +0100)]
tests: more cache conversion checks

8 years agocache: lvconvert repairs only thin pools
Zdenek Kabelac [Thu, 19 Nov 2015 12:32:03 +0000 (13:32 +0100)]
cache: lvconvert repairs only thin pools

Avoid internal error message where thin pool repair code tries to
fix cache pool - was catched later in code stack, so rather
catch this early and make the repair function exlusive
to thin pools.

So far we have no code for repairing cache pools
(other then the automatic during activation/deactivation).

8 years agocache: enable raid conversion for _cdata and _cmeta
Zdenek Kabelac [Thu, 19 Nov 2015 12:30:55 +0000 (13:30 +0100)]
cache: enable raid conversion for _cdata and _cmeta

Since thin-pool supports convertion of data and metadata LV,
enable the same for cache data and metadata LV.

8 years agotests: stacktrace on skip if message is empty
Marian Csontos [Thu, 19 Nov 2015 11:11:58 +0000 (12:11 +0100)]
tests: stacktrace on skip if message is empty

8 years agotests: add missing --skip option and S env.variable
Marian Csontos [Thu, 19 Nov 2015 09:48:17 +0000 (10:48 +0100)]
tests: add missing --skip option and S env.variable

8 years agocleanup: cast resulting value explicitely
Zdenek Kabelac [Thu, 19 Nov 2015 10:59:19 +0000 (11:59 +0100)]
cleanup: cast resulting value explicitely

8 years agocleanup: add stack traces
Zdenek Kabelac [Wed, 18 Nov 2015 08:15:45 +0000 (09:15 +0100)]
cleanup: add stack traces

8 years agocleanup: use display_lvname in pmspare
Zdenek Kabelac [Wed, 18 Nov 2015 08:14:24 +0000 (09:14 +0100)]
cleanup: use display_lvname in pmspare

Just switch to use display_lvname().
Also squeeze possibly failing strncpy into INTERNAL_ERROR
as lvname always should fit.

8 years agolibdm: report: consolidate use of string list selection structures
Peter Rajnoha [Wed, 18 Nov 2015 09:54:09 +0000 (10:54 +0100)]
libdm: report: consolidate use of string list selection structures

8 years agotests: update lvmcache-exercise
David Teigland [Tue, 17 Nov 2015 18:05:22 +0000 (12:05 -0600)]
tests: update lvmcache-exercise

To handle multiple VGs with the same name.
Simply using the VG name is ambiguous, and
lvmetad requires the VG uuid be used to
specify which one is meant.

8 years agocoverity: trying different model
Zdenek Kabelac [Tue, 17 Nov 2015 17:58:15 +0000 (18:58 +0100)]
coverity: trying different model

Model reseting of given set, but it still makes
FD_SET to think it's accessing uninitialized value.

Maybe model also FD_SET.

This page took 0.077114 seconds and 5 git commands to generate.