IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
When a command does a sequence of
vg_write + vg_commit + vg_write + vg_commit,
initialization of non-PV devices happens during the
first vg_write, and does not need to be repeated by
the second vg_write.
When creating a lockd VG, this sequence occurs because
the VG is first created, then the lockd data is created,
then the lockd data is then written to the VG metadata.
Certain stacks of cached LVs may have unexpected consequences.
So add a warning function called when LV is cached to detect
such caces and WARN user about them - the best we could do ATM.
When we insert layer we also move status flag-bits for certain LV types,
so internal volume_group structure remains consistent.
(Perhaps it's misuse of 'insert_layer' function and we should have
another similar function for this.)
Basically we aim to maintain the same state as after reading fresh
metadata out of volume group.
Currently we when i.e. cache 'raid' LV - this should transfer 'raidLV' flag
to _corigin LV and cache is no longer a raid.
TODO: bits for stacked devices needs more exact rules.
Move code which runtime detects settings for cache_policy
out of config dir to cache seg handling code.
Also mark cache_mode as command profilable setting.
Revert back to already existing behavior which has been slightly
modified by a900d150e4.
At the end however it seem to be equal to change TID right with first
metadata write.
Existing code missed handling for 'unused' thin-pool which would
require to also check empty message list for TID==0.
So with the fix we now again preserve 'active' thin-pool volume
when first thin volume is created - this property was lost and caused
problems in cluster, where the lock was hold, but volume was no longer
active on the node.
Another missing part was the proper support for already increased,
but unfinished TID change.
So going back here with existing logic -
TID is increased with first MDA update.
Code allows start with either same TID or (TID-1).
If there are messages, TID must be lower by 1 for sending,
otherwise messages were already posted.
Change logic and naming of some internal API functions.
cache_set_mode() and cache_set_policy() both take segment.
cache mode is now correctly 'masked-in'.
If the passed segment is 'cache' segment - it will automatically
try to find 'defaults' according to profiles if the are NOT
specified on command line or they are NOT already set for cache-pool.
These defaults are never set for cache-pool.
Add code to detect available cache features.
Support policy_mq & policy_smq features which might be disabled.
Introduce global_cache_disabled_features_CFG.
lvrename should not be done if the LV is active on another host.
This check was mistakenly removed when the code was changed to
use LV uuids in locks rather than LV names.
The vgchange/lvchange activation commands read the VG, and
don't write it, so they acquire a shared VG lock from lvmlockd.
When other commands fail to acquire a shared VG lock from
lvmlockd, a warning is printed and they continue without it.
(Without it, the VG metadata they display from lvmetad may
not be up to date.)
vgchange/lvchange -a shouldn't continue without the shared
lock for a couple reasons:
. Usually they will just continue on and fail to acquire the
LV locks for activation, so continuing is pointless.
. More importantly, without the sh VG lock, the VG metadata
used by the command may be stale, and the LV locks shown
in the VG metadata may no longer be current. In the
case of sanlock, this would result in odd, unpredictable
errors when lvmlockd doesn't find the expected lock on
disk. In the case of dlm, the invalid LV lock could be
granted for the non-existing LV.
The solution is to not continue after the shared lock fails,
in the same way that a command fails if an exclusive lock fails.
A segfault was reported when extending an LV with a smaller number of
stripes than originally used. Under unusual circumstances, the cling
detection code could successfully find a match against the excess
stripe positions and think it had finished prematurely leading to an
allocation being pursued with a length of zero.
Rename ix_offset to num_positional_areas and move it to struct
alloc_state so that _is_condition() can obtain access to it.
In _is_condition(), areas_size can no longer be assumed to match the
number of positional slots being filled so check this newly-exposed
num_positional_areas directly instead. If the slot is outside the
range we are trying to fill, just ignore the match for now.
(Also note that the code still only performs cling detection against
the first segment of the LV.)
Keep policy name separate from policy settings and avoid
to mangling and demangling this string from same config tree.
Ensure policy_name is always defined.
There are two different failure conditions detected in
access_vg_lock_type() that should have different error
messages. This adds another failure flag so the two
cases can be distinguished to avoid printing a misleading
error message.
There are at least a couple instances where
the lock_args check does not work correctly,
(listed in the comment), so disable the
NULL check for lock_args until those are
resolved.
Existing messaging intarface for thin-pool has a few 'weak' points:
* Message were posted with each 'resume' operation, thus not allowing
activation of thin-pool with the existing state.
* Acceleration skipped suspend step has not worked in cluster,
since clvmd resumes only nodes which are suspended (have proper lock
state).
* Resume may fail and code is not really designed to 'fail' in this
phase (generic rule here is resume DOES NOT fail unless something serious
is wrong and lvm2 tool usually doesn't handle recovery path in this case.)
* Full thin-pool suspend happened, when taken a thin-volume snapshot.
With this patch the new method relocates message passing into suspend
state.
This has a few drawbacks with current API, but overal it performs
better and gives are more posibilities to deal with errors.
Patch introduces a new logic for 'origin-only' suspend of thin-pool and
this also relates to thin-volume when taking snapshot.
When suspend_origin_only operation is invoked on a pool with
queued messages then only those messages are posted to thin-pool and
actual suspend of thin pool and data and metadata volume is skipped.
This makes taking a snapshot of thin-volume lighter operation and
avoids blocking of other unrelated active thin volumes.
Also fail now happens in 'suspend' state where the 'Fail' is more expected
and it is better handled through error paths.
Activation of thin-pool is now not sending any message and leaves upto a tool
to decided later how to finish unfinished double-commit transaction.
Problem which needs some API improvements relates to the lvm2 tree
construction. For the suspend tree we do not add target table line
into the tree, but only a device is inserted into a tree.
Current mechanism to attach messages for thin-pool requires the libdm
to know about thin-pool target, so lvm2 currently takes assumption, node
is really a thin-pool and fills in the table line for this node (which
should be ensured by the PRELOAD phase, but it's a misuse of internal API)
we would possibly need to be able to attach message to 'any' node.
Other thing to notice - current messaging interface in thin-pool
target requires to suspend thin volume origin first and then send
a create message, but this could not have any 'nice' solution on lvm2
side and IMHO we should introduce something like 'create_after_resume'
message.
Patch also changes the moment, where lvm2 transaction id is increased.
Now it happens only after successful finish of kernel transaction id
change. This change was needed to handle properly activation of pool,
which is in the middle of unfinished transaction, and also this corrects
usage of thin-pool by external apps like Docker.
Make it possible to define format for time that is displayed.
The way the format is defined is equal to the way that is used
for strftime function, although not all formatting options as
used in strftime are available for LVM2 - the set is restricted
(e.g. we do not allow newline to be printed). The lvm.conf
comments contain the whole list that LVM2 accepts for time format
together with brief description (copied from strftime man page).
For example:
(defaults used - the format is the same as used before this patch)
$ lvs -o+time vg/lvol0 vg/lvol1
LV VG Attr LSize Time
lvol0 vg -wi-a----- 4.00m 2015-06-25 16:18:34 +0200
lvol1 vg -wi-a----- 4.00m 2015-06-29 09:17:11 +0200
(using 'time_format = "@%s"' in lvm.conf - number of seconds
since the Epoch)
$ lvs -o+time vg/lvol0 vg/lvol1
LV VG Attr LSize Time
lvol0 vg -wi-a----- 4.00m @1435241914
lvol1 vg -wi-a----- 4.00m @1435562231
Synchronize with udev logic before reusing device as snapshot.
This patch tries to fix the problem with udev, where we manage
to 'active' LV for clearing, then we deactivate such device and
active again as member of 'origin&snapshot' tree all in 1 step.
There needs to be a sync point where udev has time to remove all links,
otherwise we race with scans and we may end-up with mysterious 'free'
links in the system pointing to wrong dm names.
This patch tries to fix failing topology cluster tests..
With thin-pool kernel target module 1.13 it's now support usage of
external origin with sizes which are not 'alligned' with chunk size
of thin-pool.
Enable lvm2 support for this and also fix reporting of data_percent
usage for case sizes are not alligned.
Just as 'e' means activation with an exclusive lock,
add an 's' to mean activation with a shared lock.
This allows the existing but implicit behavior of '-ay'
of clvm LVs to be specified explicitly. For local VGs,
asy simply means ay, just like aey means ay.
For local VGs, ay == aey == asy
For clvm VGs, ay == asy, aey == aey, asy == asy
In process_each_{vg,lv,pv} when no vgname args are given,
the first step is to get a list of all vgid/vgname on the
system. This is exactly what lvmetad returns from a
vg_list request. The current code is doing a vg_lookup
on each VG after the vg_list and populating lvmcache with
the info for each VG. These preliminary vg_lookup's are
unnecessary, because they will be done again when the
processing functions call vg_read. This patch eliminates
the initial round of vg_lookup's, which can roughly cut in
half the number of lvmetad requests and save a lot of extra work.
When kernel target reports sync status as 0% it might as well mean
it's 100% in sync, just the target is in some race inconsistent
state - so reread once again and take a more optimistic value ;)
Patch tries to work around:
https://bugzilla.redhat.com/show_bug.cgi?id=1210637
When performing initial allocation (so there is nothing yet to
cling to), use the list of tags in allocation/cling_tag_list to
partition the PVs. We implement this by maintaining a list of
tags that have been "used up" as we proceed and ignoring further
devices that have a tag on the list.
https://bugzilla.redhat.com/983600
Add A_PARTITION_BY_TAGS set when allocated areas should not share tags
with each other and allow _match_pv_tags to accept an alternative list
of tags. (Not used yet.)
Do not keep dangling LVs if they're removed from the vg->lvs list and
move them to vg->removed_lvs instead (this is actually similar to already
existing vg->removed_pvs list, just it's for LVs now).
Once we have this vg->removed_lvs list indexed so it's possible to
do lookups for LVs quickly, we can remove the LV_REMOVED flag as
that one won't be needed anymore - instead of checking the flag,
we can directly check the vg->removed_lvs list if the LV is present
there or not and to say if the LV is removed or not then. For now,
we don't have this index, but it may be implemented in the future.
This avoids a problem in which we're using selection on LV list - we
need to do the selection on initial state and not on any intermediary
state as we process LVs one by one - some of the relations among LVs
can be gone during this processing.
For example, processing one LV can cause the other LVs to lose the
relation to this LV and hence they're not selectable anymore with
the original selection criteria as it would be if we did selection
on inital state. A perfect example is with thin snapshots:
$ lvs -o lv_name,origin,layout,role vg
LV Origin Layout Role
lvol1 thin,sparse public,origin,thinorigin,multithinorigin
lvol2 lvol1 thin,sparse public,snapshot,thinsnapshot
lvol3 lvol1 thin,sparse public,snapshot,thinsnapshot
pool thin,pool private
$ lvremove -ff -S 'lv_name=lvol1 || origin=lvol1'
Logical volume "lvol1" successfully removed
The lvremove command above was supposed to remove lvol1 as well as
all its snapshots which have origin=lvol1. It failed to do so, because
once we removed the origin lvol1, the lvol2 and lvol3 which were
snapshots before are not snapshots anymore - the relations change
as we're processing these LVs one by one.
If we do the selection first and then execute any concrete actions on
these LVs (which is what this patch does), the behaviour is correct
then - the selection is done on the *initial state*:
$ lvremove -ff -S 'lv_name=lvol1 || origin=lvol1'
Logical volume "lvol1" successfully removed
Logical volume "lvol2" successfully removed
Logical volume "lvol3" successfully removed
Similarly for all the other situations in which relations among
LVs are being changed by processing the LVs one by one.
This patch also introduces LV_REMOVED internal LV status flag
to mark removed LVs so they're not processed further when we
iterate over collected list of LVs to be processed.
Previously, when we iterated directly over vg->lvs list to
process the LVs, we relied on the fact that once the LV is removed,
it is also removed from the vg->lvs list we're iterating over.
But that was incorrect as we shouldn't remove LVs from the list
during one iteration while we're iterating over that exact list
(dm_list_iterate_items safe can handle only one removal at
one iteration anyway, so it can't be used here).
Refactor the recent metadata-reading optimisation patches.
Remove the recently-added cache fields from struct labeller
and struct format_instance.
Instead, introduce struct lvmcache_vgsummary to wrap the VG information
that lvmcache holds and add the metadata size and checksum to it.
Allow this VG summary information to be looked up by metadata size +
checksum. Adjust the debug log messages to make it clear when this
shortcut has been successful.
(This changes the optimisation slightly, and might be extendable
further.)
Add struct cached_vg_fmtdata to format-specific vg_read calls to
preserve state alongside the VG across separate calls and indicate
if the details supplied match, avoiding the need to read and
process the VG metadata again.
Since we take a lock inside vg_lock_newname() and we do a full
detection of presence of vgname inside all scanned labels,
there is no point to do this for second time to be sure
there is no such vg.
The only side-effect of such call would be a full validation of
some already exising VG metadata - but that's not the task for
vgcreate when create a new VG.
This call noticable reduces number of scans during 'vgcreate'.
When reading VG mda from multiple PVs - do all the validation only
when mda is seen for the first time and when mda checksum and length
is same just return already existing VG pointer.
(i.e. using 300PVs for a VG would lead to create and destroy 300 config trees....)
Previous versions of lvm will not obey the restrictions
imposed by the new system_id, and would allow such a VG
to be written. So, a VG with a new system_id is further
changed to force previous lvm versions to treat it as
read-only. This is done by removing the WRITE flag from
the metadata status line of these VGs, and putting a new
WRITE_LOCKED flag in the flags line of the metadata.
Versions of lvm that recognize WRITE_LOCKED, also obey the
new system_id. For these lvm versions, WRITE_LOCKED is
identical to WRITE, and the rules associated with matching
system_id's are imposed.
A new VG lock_type field is also added that causes the same
WRITE/WRITE_LOCKED transformation when set. A previous
version of lvm will also see a VG with lock_type as read-only.
Versions of lvm that recognize WRITE_LOCKED, must also obey
the lock_type setting. Until the lock_type feature is added,
lvm will fail to read any VG with lock_type set and report an
error about an unsupported lock_type. Once the lock_type
feature is added, lvm will allow VGs with lock_type to be
used according to the rules imposed by the lock_type.
When both system_id and lock_type settings are removed, a VG
is written with the old WRITE status flag, and without the
new WRITE_LOCKED flag. This allows old versions of lvm to
use the VG as before.
The seg_monitor did not display monitored status for thick snapshots
and mirrors (with mirror log *not* mirrored). The seg monitor did work
correctly even before for other segtypes - thins and raids.
Before (mirrors and snapshots, only mirrors with mirrored log properly displayed monitoring status):
[0] f21/~ # lvs -a -o lv_name,lv_layout,lv_role,seg_monitor vg
LV Layout Role Monitor
mirror mirror public
[mirror_mimage_0] linear private,mirror,image
[mirror_mimage_1] linear private,mirror,image
[mirror_mlog] linear private,mirror,log
mirror_with_mirror_log mirror public monitored
[mirror_with_mirror_log_mimage_0] linear private,mirror,image
[mirror_with_mirror_log_mimage_1] linear private,mirror,image
[mirror_with_mirror_log_mlog] mirror private,mirror,log monitored
[mirror_with_mirror_log_mlog_mimage_0] linear private,mirror,image
[mirror_with_mirror_log_mlog_mimage_1] linear private,mirror,image
thick_origin linear public,origin,thickorigin
thick_snapshot linear public,snapshot,thicksnapshot
With this patch applied (monitoring status displayed for all mirrors and snapshots):
[0] f21/~ # lvs -a -o lv_name,lv_layout,lv_role,seg_monitor vg
LV Layout Role Monitor
mirror mirror public monitored
[mirror_mimage_0] linear private,mirror,image
[mirror_mimage_1] linear private,mirror,image
[mirror_mlog] linear private,mirror,log
mirror_with_mirror_log mirror public monitored
[mirror_with_mirror_log_mimage_0] linear private,mirror,image
[mirror_with_mirror_log_mimage_1] linear private,mirror,image
[mirror_with_mirror_log_mlog] mirror private,mirror,log monitored
[mirror_with_mirror_log_mlog_mimage_0] linear private,mirror,image
[mirror_with_mirror_log_mlog_mimage_1] linear private,mirror,image
thick_origin linear public,origin,thickorigin
thick_snapshot linear public,snapshot,thicksnapshot monitored
Set ACCESS_NEEDS_SYSTEM_ID VG status flag whenever there is
a non-lvm1 system_id set. Prevents concurrent access from
older LVM2 versions.
Not set on VGs that bear a system_id only due to conversion
from lvm1 metadata.
format_text processes both lvm2 on-disk metadata and metadata read
from other sources such as backup files. Add original_fmt field
to retain the format type of the original metadata.
Before this patch, /etc/lvm/archives would contain backups of
lvm1 metadata with format = "lvm2" unless the source was lvm1 on-disk
metadata.
The vg->lvm1_systemd_id needs to be initialized as all the code around
counts with that. Just like we initialize lvm1_system_id in vg_create
(no matter if it's actually LVM1 or LVM2 format), this patch adds this
init in alloc_vg as well so the rest of the code does not segfaul
when trying to access vg->lvm1_system_id.
In log messages refer to it as system ID (not System ID).
Do not put quotes around the system_id string when printing.
On the command line use systemid.
In code, metadata, and config files use system_id.
In lvmsystemid refer to the concept/entity as system_id.
The only realistic way for a host to have active LVs in a
foreign VG is if the host's system_id (or system_id_source)
is changed while LVs are active.
In this case, the active LVs produce an warning, and access
to the VG is implicitly allowed (without requiring --foreign.)
This allows the active LVs to be deactivated.
In this case, rescanning PVs for the VG offers no benefit.
It is not possible that rescanning would reveal an LV that
is active but wasn't previously in the VG metadata.
cmirror uses the CPG library to pass messages around the cluster and maintain
its bitmaps. When a cluster mirror starts-up, it must send the current state
to any joining members - a checkpoint. When mirrors are large (or the region
size is small), the bitmap size can exceed the message limit of the CPG
library. When this happens, the CPG library returns CPG_ERR_TRY_AGAIN.
(This is also a bug in CPG, since the message will never be successfully sent.)
There is an outstanding bug (bug 682771) that is meant to lift this message
length restriction in CPG, but for now we work around the issue by increasing
the mirror region size. This limits the size of the bitmap and avoids any
issues we would otherwise have around checkpointing.
Since this issue only affects cluster mirrors, the region size adjustments
are only made on cluster mirrors. This patch handles cluster mirror issues
involving pvmove, lvconvert (from linear to mirror), and lvcreate. It also
ensures that when users convert a VG from single-machine to clustered, any
mirrors with too many regions (i.e. a bitmap that would be too large to
properly checkpoint) are trapped.
A foreign VG should be silently ignored by a reporting/display
command like 'vgs'. If the reporting/display command specifies
a foreign VG by name on the command line, it should produce an
error message.
Scanning commands pvscan/vgscan/lvscan are always allowed to
read and update caches from all PVs, including those that belong
to foreign VGs.
Other non-report/display/scan commands always ignore a foreign
VG, or report an error if they attempt to use a foreign VG.
vgimport should always invalidate the lvmetad cache because
lvmetad likely holds a pre-vgexported copy of the VG.
(This is unrelated to using foreign VGs; the pre-vgexported
VG may have had no system_id at all.)
When checking whether the system ID permits access to a VG, check for
each permitted situation first, and only then issue the appropriate
error message. Always issue a message for now. (We'll try to
suppress some of those later when the VG concerned wasn't explicitly
requested.)
Add more messages to try to ensure every return code is checked and
every error path (and only an error path) contains a log_error().
Add self-correction to vgchange -c to deal with situations where
the cluster state and system ID state are out-of-sync (e.g. if
old tools were used).
Move the lvm1 sys ID into vg->lvm1_system_id and reenable the #if 0
LVM1 code. Still display the new-style system ID in the same
reporting field, though, as only one can be set.
Add a format feature flag FMT_SYSTEM_ON_PVS for LVM1 and disallow
access to LVM1 VGs if a new-style system ID has been set.
Treat the new vg->system_id as const.
Dop unused value assignments.
Unknown is detected via other combination
(!linear && !striped).
Also change the log_error() message into a warning,
since the function is not really returning error,
but still keep the INTERNAL_ERROR.
Ret value is always set later.
The dev ext source must be reset for the dev_cache_get call
(which evaluates filters), not lvmcache_label_scan - so fix
original commit 727c7ff85d.
Also, add comments in _pvcreate_check fn explaining why
refresh filter and rescan is needed and exactly in which
situations.
Before, we refreshed filters and we did full rescan of devices if
we passed through wiping (wipe_known_signatures fn call). However,
this fn returns success even if no signatures were found and so
nothing was wiped. In this case, it's not necessary to do the
filter refresh/rescan of devices as nothing changed clearly.
This patch exports number of wiped signatures from all the
wiping functions below. The caller (_pvcreate_check) then checks
whether any wiping was done at all and if not, no refresh/rescan
is done, saving some time and resources.
pvcreate code path executes signature wiping if there are any signatures
found on device to prepare the device for PV. When the signature is wiped,
the WATCH udev rule triggers the event which then updates udev database
with fresh info, clearing the old record about previous signature.
However, when we're using udev db as dev-ext source, we'd need to wait
for this WATCH-triggered event. But we can't synchronize against such
events (at least not at this moment). Without this sync, if the code
continues, the device could still be marked as containing the old
signature if reading udev db. This may end up even with the device
to be still filtered, though the signature is already wiped.
This problem is then exposed as (an example with md components):
$ mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sda /dev/sdb --run
$ mdadm -S /dev/md0
$ pvcreate -y /dev/sda
Wiping linux_raid_member signature on /dev/sda.
/dev/sda: Couldn't find device. Check your filters?
$ echo $?
5
So we need to temporarily switch off "udev" dev-ext source here
in this part of pvcreate code until we find a way how to sync
with WATCH events.
(This problem does not occur with signature wiping which we do
on newly created LVs since we already handle this properly with
our udev flags - the LV_NOSCAN/LV_TEMPORARY flag. But we can't use
this technique for non-dm devices to keep WATCH rule under control.)
for_each_sub_lv() now scans in depth also pools, however for
rename we actually do want to skip pools.
So add a new for_each_sub_lv_except_pools() to be used by rename,
every other user of for_each_sub_lv() scans every sub LV with pools
included.
This is i.e. necessary for properly working preload of pools
that are using raid arrays.
This is a regression from v115 where some of the fields/properties
were converted to using the common "struct lvinfo" and
"struct lv_seg_status" so we don't need to issue info and status
ioctl several times per one reported line. Not all fields are
converted yet, but one that *is* converted is the lv_attr field
with the lv_attr_dup counterpart used in lvm_lv_get_attr lvm2app fn.
These changes were introduced with e34b004422
and later - this patch introduced the "info_ok" field in the
lv_with_info_and_seg_status structure which encapsulates the lvinfo
and lv_seg_status struct.
For the lv_attr_dup, the lv_attr_dup code missed the
assignment for the "info_ok" flag which saves the result of the
lv_info_with_seg_status call. Hence such info was marked
as unusable - unknown and it was returned as such via lvm_lv_get_attr
lvm2app fn.
When raid leg is extracted, now the preload code handles this state
correctly and put proper new table entry into dm tree,
so the activation of extracted leg and removed metadata works
after commit.
Rename original lv_error_when_full field to lv_when_full and also
convert it from binary field to string field displaying three
possible values: "error", "queueu" or "" (blank for undefined).
$ lvs vg/pool vg/pool1 vg/linear_lv -o+lv_when_full
LV VG Attr LSize Data% Meta% WhenFull
linear_lv vg -wi-a----- 4.00m
pool vg twi-aotz-- 4.00m 0.00 0.98 queue
pool1 vg twi-a-tz-- 4.00m 0.00 0.88 error
For -S|--select these synonyms are recognized:
"error" -> "error when full", "error if no space"
"queue" -> "queue when full", "queue if no space"
"" -> "undefined"
Recently the single 'status' code has been used for number of cache
features.
Extend the API a little bit to allow usage also for lv_attr_dup.
As the function itself is used in lvm2api - add a new function:
lv_attr_dup_with_info_and_seg_status() that is able to use
grabbed info & status information.
report_init() is now using directly passed lvdm struct pointer
which holds the infomation whether lv_info() was correctly obtained or
there was some error when trying to read it.
Move 'healt' attribute to status.
TODO convert raid function to use the already known status.
The previous patch felt short WRT disabling allocation on PVs holding other
legs of the RAID LV persistently; this patch introduces an internal,
transient PV flag PV_ALLOCATION_PROHIBITED to address this very problem.
General problem description for completeness:
An 'lvconvert --repair $RAID_LV" to replace a failed leg of a multi-segment
RAID10/4/5/6 logical volume can lead to allocation of (parts of) the replacement
image component pair on the physical volume of another image component
(e.g. image 0 allocated on the same PV as image 1 silently impeding resilience).
Patch fixes this severe resilince issue by prohibiting allocation on PVs
already holding other legs of the RAID set. It allows to allocate free space
on any operational PV already holding parts of the image component pair.
Support error_if_no_space feature for thin pools.
Report more info about thinpool status:
(out_of_data (D), metadata_read_only (M), failed (F) also as health
attribute.)
An 'lvconvert --repair $RAID_LV" to replace a failed leg of a multi-segment
RAID10/4/5/6 logical volume can lead to allocation of (parts of) the replacement
image component pair on the physical volume of another image component
(e.g. image 0 allocated on the same PV as image 1 silently impeding resilience).
Patch fixes this severe resilince issue by prohibiting allocation on PVs
already holding other legs of the RAID set. It allows to allocate free space
on any operational PV already holding parts of the image component pair.
Better than previous patch which changed log_warn to log_error -
we can have multiple MDAs and if one of them fails to be written,
we can still continue with other MDAs if we're in a mode where
we can handle missing PVs - so keep the log_warn for single
failed MDA write as it was before.
However, add log_error with "Failed to write VG <vg_name>." in
case we're not handling missing PVs or no MDA was written at all
during VG write process. This also prevents an internal error in
which the vg_write fails and we're not issuing any other log_error
in vg_write caller or above, so we end up with:
"Internal error: Failed command did not use log_error".
$ lvcreate -l1 -m1 --type mirror vg
Logical volume "lvol0" created.
$ lvconvert --type raid1 vg/lvol0
Before:
$ lvs -a vg
LV VG Active Attr LSize Cpy%Sync Layout Role
lvol0 vg active rwi-a-r--- 4.00m 100.00 raid,raid1 public
[lvol0_mimage_0_rimage_0] vg active iwi-aor--- 4.00m linear private,raid,image
[lvol0_mimage_1_rimage_1] vg active iwi-aor--- 4.00m linear private,raid,image
[lvol0_rmeta_0] vg active ewi-aor--- 4.00m linear private,raid,metadata
[lvol0_rmeta_1] vg active ewi-aor--- 4.00m linear private,raid,metadata
Incorrect name: lvol0_mimage_0_rimage_0
With this patch applied:
$ lvs -a vg
LV VG Active Attr LSize Cpy%Sync Layout Role
lvol0 vg active rwi-a-r--- 4.00m 100.00 raid,raid1 public
[lvol0_rimage_0] vg active iwi-aor--- 4.00m linear private,raid,image
[lvol0_rimage_1] vg active iwi-aor--- 4.00m linear private,raid,image
[lvol0_rmeta_0] vg active ewi-aor--- 4.00m linear private,raid,metadata
[lvol0_rmeta_1] vg active ewi-aor--- 4.00m linear private,raid,metadata
Proper name: lvol0_rimage_0
When mirror has missing PVs and there are mirror images on those missing
PVs, we delete the images and during this delete operation, we also
reactivate the LV. But if we're trying to reactivate the LV in cluster
which is not active and at the same time cmirrord is not running (which
is OK since we may have created the mirror LV as inactive), we end up
with:
"Error locking on node <node_name>: Shared cluster mirrors are not available."
That is because we're trying to activate the mirror LV without cmirrord.
However, there's no need to do this reactivation if the mirror LV (and
hence it's sub LVs) were not activated before.
This issue caused failure in mirror-vgreduce-removemissing.sh test
recently with this sequence (excerpt from the test script):
prepare_lvs_
lvcreate -an -Zn -l2 --type mirror -m1 --nosync -n $lv1 $vg "$dev1" $dev2" "$dev3":$BLOCKS
mimages_are_on_ $lv1 "$dev1" "$dev2"
mirrorlog_is_on_ $lv1 "$dev3"
aux disable_dev "$dev2"
vgreduce --removemissing --force $vg
The important thing about that test is that we're not running cmirrord,
we're activating the mirror with "-an" so it's inactive and then
vgreduce --removemissing tries to reactivate the mirror images
as part of the _delete_lv function call inside and since cmirrord
is not running, we end up with the "Shared cluster mirrors are not
available." error.
When creating cluster mirrors while they're not supposed to be activated
immediately after creation, we don't need to check for cmirrord availability.
We can just create these mirrors and let the check to be done on activation
later on. This is addendum for commit cba6186325.
When creating/activating clustered mirrors, we should have cmirrord
available and running. If it's not, we ended up with rather cryptic
errors like:
$ lvcreate -l1 -m1 --type mirror vg
Error locking on node 1: device-mapper: reload ioctl on failed: Invalid argument
Failed to activate new LV.
$ vgchange -ay vg
Error locking on node node 1: device-mapper: reload ioctl on failed: Invalid argument
This patch adds check for cmirror availability and it errors out
properly, also giving a more precise error messge so users are able
to identify the source of the problem easily:
$ lvcreate -l1 -m1 --type mirror vg
Shared cluster mirrors are not available.
$ vgchange -ay vg
Error locking on node 1: Shared cluster mirrors are not available.
Exclusively activated cluster mirror LVs are OK even without cmirrord:
$ vgchange -aey vg
1 logical volume(s) in volume group "vg" now active
When we split leg from raid - we take a proper new lock for a new LV.
However for now activation checks only 'existince' of device UUID,
but it's not validating device has a proper name.
As a quick fix call suspend()/resume() to rename after split mirror.
When chunk size needs to be estimated, the code missed to round
to proper 64kb boundaries (or power of 2 for older thin pool driver).
So for some data and metadata size (i.e. 10GB and 4MB) it resulted
in incorrect chunk size (not being a multiple of 64KB)
Fix it by adding proper rounding and also use 1 routine for 2 places
where the same calculation is made.
Fix also incorrect printed warning that has used 'ffs()'
(which returns first 'least significant' bit in word)
and it was not really giving any useful size info and replace it
with properly estimated chunk size.
Use log_warn when we are effectively not creating an error -
we 'allowed' inconsistent read for a reason - so it's just warning
level we process inconsistent VG - it's upto caller later to decide
error level of command return value and in case of error it needs
to use log_error then.
Failed recovery provides different (NULL) VG then FAILED_INCONSISTENT.
Mark it with different failure bit - since FAILED_INCONSISTENT is
supposed to contain something 'usable' (thought inconsistent).
Since we support device stack of pools over pool
(thin-pool with cache data volume) the existing code
is no longer able to detect orphan _pmspare.
So instead do a _pmspare check after volume removal,
and remove spare afterwards.
This would be in case the pool segment was not found.
LVM2.2.02.112/lib/metadata/pool_manip.c:238:36: warning: Access to field 'segtype' results in a dereference of a null pointer (loaded from variable 'pool_seg')
LVM2.2.02.112/lib/metadata/cache_manip.c:73: overflow_before_widen: Potentially overflowing expression "*pool_metadata_extents *vg->extent_size" with type "unsigned int" (32 bits, unsigned) is evaluated using 32-bit arithmetic, and then used in a context that expects an expression of type "uint64_t" (64 bits, unsigned).
LVM2.2.02.112/lib/activate/dev_manager.c:217: overflow_before_widen: Potentially overflowing expression "seg_status->seg->len * extent_size" with type "unsigned int" (32 bits, unsigned) is evaluated using 32-bit arithmetic, and then used in a context that expects an expression of type "uint64_t" (64 bits, unsigned).
LVM2.2.02.112/lib/activate/dev_manager.c:217: overflow_before_widen: Potentially overflowing expression "seg_status->seg->le * extent_size" with type "unsigned int" (32 bits, unsigned) is evaluated using 32-bit arithmetic, and then used in a context that expects an expression of type "uint64_t" (64 bits, unsigned).
Do not use 'any' policy name as a value in config tree - so we stick
with 'policy_settings' and extra 'policy_name' for libdm params.
Update lvm2 API as well.
Example of supported metadata:
policy = "mq"
policy_settings {
migration_threshold = 2048
sequential_threshold = 512
random_threshold = 4
read_promote_adjustment = 10
}
More efficient spare volume creation. Save 1 extra commit
and properly activate this volume according to our cluster
activation rules (using lv_active_change() for this).
Since we 'layer' for cache origin which and we support dropping
cache layer - we need to restore origin name in case
the origin LV is more complex target - i.e. raid.
Drop _corig from name
Cleanup and rename parent -> parent_lv.
Revert part of commit 51a29e6056,
it's probably bad idea to continue with any recovery, when
vg_write() or vg_commit() fail - so it's better to leave it as it is.
Let's use this function for more activations in the code.
'needs_exlusive' will enforce exlusive type for any given LV.
We may want to activate LV in exlusive mode, even when we know
the LV (as is) supports non-exlusive activation as well.
lvcreate -ay -> exclusive & local
lvcreate -aay -> exclusive & local
lvcreate -aly -> exclusive & local
lvcreate -aey -> exclusive (might be on any node).
Call check_new_thin_pool() to detect in-use thin-pool.
Save extra reactivation of thin-pool when thin pool is not active.
(it's now a bit more expensive to invoke thin_check for new pools.)
For new pools:
We now active locally exclusively thin-pool as 'public' LV.
Validate transaction_id is till 0.
Deactive.
Prepare create message for thin-pool and exclusively active pool.
Active new thin LV.
And deactivate thin pool if it used to be inactive.
Replace lv_cache_block_info() and lv_cache_policy_info()
with lv_cache_status() which directly returns
dm_status_cache structure together with some calculated
values.
After use mem pool stored inside lv_status_cache structure
needs to be destroyed.
Add API call to calculate extents from percentage value.
Size is based in DM_PERCENT_1 units.
(Supporting decimal point number).
This commit is preparing functionality for more global
usage of % with i.e. --size option.
Unlike with thin-pool - with cache we support all args also
directly when create cache volume.
So the result of 'separate' cache-pool creation and setting its
options should give same result as specifying those args
during cache creation.
Cache-pool values are used as defaults if the params are
not specified with cache creation.
Move code for creation of thin volume into a single place
out of lv_extend(). This allows to drop extra pool arg
for alloc_lv_segment() && lv_extend() and makes code
more easier to read and follow.
When we create volumes with chunk size bigger then extent size
we try to round up to some nearest chunk boundary.
Until now we did this for thins - use same logic for
cache volumes.
Refactor lvcreate code.
Prefer to use arg_outside_list_is_set() so we get automatic 'white-list'
validation of supported options with different segment types.
Drop used lp->cache, lp->cache and use seg_is_cache(), seg_is_thin()
Draw clear border where is the last moment we could change create
segment type.
When segment type is given with --type - do not allow it to be changed
later.
Put together tests related to individual segment types.
Finish cache conversion at proper part of lv_manip code after
the vg_metadata are written - so we could correcly clean-up created
stripe LV for cache volume.
Move test for size of new LV names in front before
any creation of LV.
Properly check striped segtype kernel presence,
since passed 'segtype' is already tested.
Keep deactivation error path local to wiping part of the function.
Create metadata with temporary flag (it's activated, zeroed
and deactivated).
Introduce new option to specify pool data size.
This will be user to create i.e. cache & cachepool at once.
And possible for thin external origin snapshot.
This is only very basic patch to enable options, the
real working code will come later.
We want to print smarter warning message only when
the zeroing was not provided on the first zeroable segment
of newly created LV.
Put warning within _should_wipe_lv function to avoid reevaluation
of same conditions twice.
Hide creation of temporary LVs and print them only in verbose mode.
e.g. hides confusing message about creation of _pmspare
device during creation of pool.
Ask for lock the proper LV.
Use the top-most LV to query for locally exclusive lock.
The rest of operations are then using 'lv_info()'
TODO:
Check all devices are reloaded from proper level.
In general any query on lv_is_active is supposed to be running
ona lv_lock_holder() volume.
Instead of segtype->ops->name() introduce lvseg_name().
This also allows us to leave name() function 'empty' for default
return of segtype->name.
TODO: add functions for rest of ops->
When we are given an existing LV name - it needs to be allowed
to pass in even restricted name as the LV could have existed
long before we introduced some new restriction on prefix/suffix.i
Fix the regression on name limits and drop restriction to be applied
on any existing LVs - only the new created LV names have to be
complient with current name restrictions.
FIXME: we are currently using restricted names incorrectly in few
other places - device_is_usable() skips restricted names,
and udev flags are also incorrectly set for restricted names
so these LVs are not getting links properly.
find_pv_in_vg fn iterates over the list of PVs covered by the VG and
each PV's pvl->pv->dev is compared with device acquired from device
cache. However, in case pvl->pv->dev is NULL as well as device cache
returns NULL (e.g. when device is filtered), we'll get incorrect match
and the code calling find_pv_in_vg uses incorrect PV (as it thinks
it's the exact PV with the pv_name). The INTERNAL_ERROR covers this
situation and errors out immediately.
The warnings arg was used to enable logging of warnings
when reading a PV. This arg is turned into a set of flags
with the WARN_PV_READ flag matching the existing behavior.
A new flag WARN_INCONSISTENT is added that will cause
vg_read_internal() to log the "VG is not consistent"
warning so the various callers do not need to log
this warning themselves.
A new vg_read flag READ_WARN_INCONSISTENT is used from
reporting to enable the WARN_INCONSISTENT flag in
vg_read_internal.
[Committed by agk with cosmetic changes and tweaks.]
Process PVs by iterating through VGs, then iterating through
devices if the command needs to process non-PV devices.
The process_single function can always use the VG and PV args.
[Committed by agk with cosmetic changes and tweaks.]
Introduce pool function for validation of chunk size.
It's good idea to be able to reject invalid chunk size
when entered on command line before we open VG.
Move code to better locations.
Improve test and remove invalid ones
(i.e. no reason to require cache size to be >= then origin).
Correctly comment where the code is doing actual conversion
of other existing volume - we do already a similar thing with
external origins.
Lots of new command line options and combinations is now supported.
Hopefully older syntax still works as well.
lvcreate --cache --cachepool vg/pool -l1
lvcreate --type cache --cachepool vg/pool -l1
lvcreate --type cache-pool vg/pool -l1
lvcreate --type cache-pool --name pool vg -l1
... and many many more ...
Since _pmspare is internal volume move it to
lv_remove_single - so it's automatically removed with
last remove thin-pool.
lv_remove_with_dependencies() is not always used for pool removal.
The cache mode of a new cache pool is always explicitly
included in the vg metadata. If a cache mode is not
specified on the command line, the cache mode is taken
from lvm.conf allocation/cache_pool_cachemode, which
defaults to "writethrough".
The cache mode can be displayed with lvs -o+cachemode.
There are actually three filter chains if lvmetad is used:
- cmd->lvmetad_filter used when when scanning devices for lvmetad
- cmd->filter used when processing lvmetad responses
- cmd->full_fiilter (which is just cmd->lvmetad_filter + cmd->filter chained together) used
for remaining situations
This patch adds the third one - "cmd->full_filter" - currently this is
used if device processing does not fall into any of the groups before,
for example, devices which does not have the PV label yet and we're just
creating a new one or we're processing the devices where the list of the
devices (PVs) is not returned by lvmetad initially.
Currently, the cmd->full_filter is used exactly in these functions:
- lvmcache_label_scan
- _pvcreate_check
- pvcreate_vol
- lvmdiskscan
- pvscan
- _process_each_label
If lvmetad is used, then simply cmd->full_filter == cmd->filter because
cmd->lvmetad_filter is NULL in this case.
We are not using already defined segement type names where we could.
There is a lot of other places in device-mapper and LVM2 we have those
hardcoded so we should better finally have a common interface in
libdevmapper to avoid this.
Use of lv_info() internally in lv_check_not_in_use(),
so it always could use with_open_count properly.
Skip sysfs() testing in open_count == 0 case.
Accept just 'lv' pointer like other functions.
The function has 'built-in' lv_is_active_locally check,
which however is not what we need to check in many place.
For now at least remotely active snapshot merge is
detected and for this case merge on next activation is scheduled.
We use adjusted_mirror_region_size() in two different contexts.
Either on command line -
here we do want to inform user about reduction of size.
Or in pvmove activation context -
here we should only use 'verbose' info.
When requesting to reload an LV imrove this API to
automatically reload its lock holding LV as in cluster
only top-level LVs are addressable with lock.
When vg_ondisk is NULL we do not need to search
through the whole VG to find out the same LV.
NOTE: as of now - VG locking is not enabled as some code parts
are breaking memory locking rules (lvm2app).
Once we enforce VG locking for read-only commands the effect
will be much better for larger VGs.
