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.
Avoid double pool allocation call by placing string
directly after the node/value structure end.
It would be likely better to not copy the string at all
and derefence it from the original string however that
is much larger rework - and also for unescaping we would
need to allocate anyway.
This code is faster when calculating crc32 checksum for larger
block areas. There is also SIMD variant present in the code,
however ATM the influence on performance of lvm2 is not that big..
One vg struct is copied to another vg struct, rather than
importing vg metadata text to create the new vg struct.
Enable with global/vg_copy_internal="binary"
Detect when we have mixed dos partition with gpt's PMBR partition.
This is not a sane configuration, but detect it anyway, just in case
someone configures such partition layout manually and forcefully and
incorrectly defines one of the partition types to be the GPT's PMBR.
For example:
❯ fdisk -l /dev/sdc
Device Boot Start End Sectors Size Id Type
/dev/sdc1 2048 67583 65536 32M 83 Linux
/dev/sdc2 67584 262143 194560 95M ee GPT
Before:
(The partition filter passes even though there's real existing dos
partition - the empty GPT PMBR overrides it.)
❯ pvcreate /dev/sdc
WARNING: PMBR signature detected on /dev/sdc at offset 510. Wipe it? [y/n]:
Wiping PMBR signature on /dev/sdc.
Physical volume "/dev/sdc" successfully created.
With this patch applied:
(The GPT PMBR does not override the existence of the dos partition.)
❯ pvcreate /dev/sdc
Cannot use /dev/sdc: device is partitioned
This provides better hints when trying to resize the fs on top of an LV.
Also needs a3f6d2f593 for proper operation.
❯ lvs -o name,size vg/swap
lv_name lv_size
swap 60.00m
Before:
❯ lvextend -L72m vg/swap
Size of logical volume vg/swap changed from 60.00 MiB (15 extents) to 72.00 MiB (18 extents).
Logical volume vg/swap successfully resized.
❯ lvreduce -L60m vg/swap
File system swap found on vg/swap.
File system device usage is not available from libblkid.
❯ lvreduce -L50m vg/swap
Rounding size to boundary between physical extents: 52.00 MiB.
File system swap found on vg/swap.
File system device usage is not available from libblkid.
After:
❯ lvextend -L72m vg/swap
Size of logical volume vg/swap changed from 60.00 MiB (15 extents) to 72.00 MiB (18 extents).
Logical volume vg/swap successfully resized.
❯ lvreduce -L60m vg/swap
File system swap found on vg/swap.
File system size (60.00 MiB) is equal to the requested size (60.00 MiB).
File system reduce is not needed, skipping.
Size of logical volume vg/swap changed from 72.00 MiB (18 extents) to 60.00 MiB (15 extents).
Logical volume vg/swap successfully resized.
❯ lvreduce -L50m vg/swap
Rounding size to boundary between physical extents: 52.00 MiB.
File system swap found on vg/swap.
File system size (60.00 MiB) is larger than the requested size (52.00 MiB).
File system reduce is required and not supported (swap).
blkid does not report FSLASTBLOCK for a swap device. However, blkid
does report FSSIZE for swap devices, so use this field (and including
the header size which is of FSBLOCKSIZE for the swap) instead to
set the "filesystem last block" which is used subsequently for
further calculations and conditions.
We already detect msdos partition table. If it is empty, that is, there
is just the partition header and no actual partitions defined, then the
filter-partitioned passes, otherwise not.
Do the same for GPT partition table.
New config setting sanlock_align_size can be used to configure
the sanlock lease size that lvmlockd will use on 4K disks.
By default, lvmlockd and sanlock use 8MiB align_size (lease size)
on 4K disks, which supports up to 2000 hosts (and max host_id.)
This can be reduced to 1, 2 or 4 (in MiB), to reduce lease i/o.
The reduced sizes correspond to smaller max hosts/host_id:
1 MiB = 250 hosts
2 MiB = 500 hosts
4 MiB = 1000 hosts
8 MiB = 2000 hosts (default)
(Disks with 512 byte sectors always use 1MiB leases and support
2000 hosts/host_id, and are not affected by this.)
In cases user is sure he is not using his 'rootfs' or 'swap' on LVs
managed with his command - it possible to completely bypass pinning
process to RAM which may eventually slightly speedup command execution,
(however at the risk the process can be eventually delayed by swapping).
Basicaly use this only at your risk...
TODO: add some dmeventd support for this.
Previously, lvmlockd detected the end of the lvmlock LV
by doing i/o to it until an i/o error was returned.
This triggered sanlock warning messages, so use the LV
size to avoid accessing beyond the end of the device.
Previously, every lvcreate would refresh the lvmlock LV
in case another machine had extended it. This involves
a lot of unnecessary work in most cases, so now compare
the LV size and device size to detect when a refresh is
needed.
lvremove of a thin lv while the pool is inactive would
leave the pool locked but inactive.
lvcreate of a thin snapshot while the pool is inactive
would leave the pool locked but inactive.
lvcreate of a thin lv could activate the pool to check
a threshold before the pool lock was acquired in lvmlockd.
The lv_hash wasn't being passed to the seg-specific text import
functions, so they were doing many find_lv() calls which consumes
a lot of time when there are many LVs in the metadata.
While performing udev sync semaphore's inc/dec operation, we use the
result from GETVAL semctl just to print a debug message with current
value of that sempahore, nothing else.
If the GETVAL fails for whetever reason while the actual inc/dec
completes successfully, just log a warning message about the GETVAL
(and print the debug messages without the actual semaphore value)
and return success for the inc/dec operation as a whole.
Clean up udev sync semaphore on fail path during its creation, otherwise
the caller will have no handle returned to clean it up itself and the
semaphore will keep staying in the system. The only way to clean it up
would be to call `dmsetup udevcomplete_all` which would destroy all
udev sync semaphores, not just the failed one, which we don't want.
The same message is printed while performing create/inc/dec operation and
the GETVAL semctl fails. Add a prefix so we know exactly in which of
these functions the issue actually happened.
LV with pvmove_ prefix is not allowed to be created by user
so bigger chance our selected name will never exist.
TODO: probably add code to get generic unused LV name...
Older gcc doesn't really like complex types (buffer, struct) to be
initialized without extra {} around such type.
So pick any other 'single type' var from a struct and set it to 0,
rest will do the compiler without emitting a warning.
Revert 373372c8ab and instead update
our validation code to handle LVs with empty segment - currently
we should need this only for pvmove operation, thus such LV should
have name 'pvmove%u'.
This fixes a problem where user tried i.e. pvmove on a VG with single
PV - as reported: https://github.com/lvmteam/lvm2/issues/148
Reported-by: bob@redhat.com
The .cache and compile_commands.json is used by popular source crawling and
indexing clang tools which in turn may be integrated with source code editors.
We may reuse the .cache directory for for other caches and temporary
files.
The /doc/.ikiwiki and /public are related to the ikiwiki.
Avoid possible udev race - since dmsetup create is
not using the same cookie logic as lvm2 commands,
try to avoid racing on some systems with udev scanning.
The option can be used in multiple ways (like --cachesettings):
--integritysettings key=val
--integritysettings 'key1=val1 key2=val2'
--integritysettings key1=val1 --integritysettings key2=val2
Use with lvcreate or lvconvert when integrity is first enabled
to configure:
journal_sectors
journal_watermark
commit_time
bitmap_flush_interval
allow_discards
Use with lvchange to configure (only while inactive):
journal_watermark
commit_time
bitmap_flush_interval
allow_discards
lvchange --integritysettings "" clears any previously configured
settings, so dm-integrity will use its own defaults.
lvs -a -o integritysettings displays configured settings.
log/command_log_report config setting defaults to 1 now if json or json_std
output format is used (either by setting report/output_format config
setting or using --reportformat cmd line arg).
This means that if we use json/json_std output format, the command log
messages are then part of the json output too, not interleaved as
unstructured text mixed with the json output.
If log/command_log_report is set explicitly in the config, then we still
respect that, no matter what output format is used currently. In this
case, users can still separate and redirect the output by using
LVM_OUT_FD, LVM_ERR_FD and LVM_REPORT_FD so that the different types
do not interleave with the json/json_std output.
In case of different PV sizes in a VG, the lvm2 allocator falls short
to define extended segments resiliently asked for 100%FREE RaidLV extension
and a RAID distinct allocation check fails. Fix is to release a memory pool
on the resulting error path.
Until the lvm2 allocator gets enhanced (WIP) to do such complex (and other)
allocations proper, a workaround is to extend a RaidLV to any free space on
its already allocated PVs by defining those PVs on the lvextend command line
then iteratively run further such lvextend commands to extend it to its
final intended size. Mind, this may be a non-trivial extension interation.
The cmd struct is now required in many more functions, and
it's added as a function arg for most direct dev-cache function
calls. The cmd struct is added to struct device (dev->cmd) so
that it can be accessed in many other cases where dev-cache
functions are being called from places where getting the cmd
struct is too difficult.
The dm devs cache is separate from the ordinary dev cache,
so give the function names distinct prefixes, using
"dm_devs_cache" to prefix dm devs cache functions.
When a PV is stacked on an LV, the PV needs to be
dropped from bcache before the LV is processed.
The LV can be found in dev-cache using its name
rather than the devno.
The list of dm devs was in the cmd struct and had a
different lifetime than the radix trees referencing
those dm devs. Now the list and radix trees are
created and destroyed together.
In the context of dm, 'device' refers to a dm device, but
in the context of lvm, 'device' refers to struct device.
Change some lvm function names to make that difference clearer.
dev_manager_get_device_list() -> dev_manager_get_dm_active_devices()
get_device_list() -> get_dm_active_devices()
device_get_uuid() -> dev_dm_uuid(), devno_dm_uuid()
The comment explained that the ex global lock was just
used to trigger global cache invalidation, which is no
longer needed. This extra locking can cause problems
with LVM-activate when local and shared VGs are mixed
(and the incorrect exit code for errors was causing
problems.)
vgchange -an vg is permitted when the vg lockspace
is not available, because LVs could still be active
for some reason, and they should be inactive when not
properly locked. In case lvmlockd was not running, or
the lockspace was not started, the command was
unnecessarily trying and failing to unlock every LV,
printing errors for every LV. We can skip this when
the lockspace is known to not be available.
vgchange --lockstop will fail with EBUSY if orphan locks in the
lock manager prevent stopping the lockspace. The orphan locks
can then be adopted and released, and the lockspace then stopped
cleanly.
Lock adoption is not part of standard command behavior, but can
be used for manual recovery or cleanup from unexpected failure
cases. Like other lockopt values, they are hidden options for
--lockopt. Different lock managers will behave differently.
Adopting locks with lvmlockd -A1 is more accurate and automatic.
--lockopt adoptls
. for vgchange --lockstart
. adopt existing ls, or fail if no existing lockspace is found
--lockopt adoptgl | adoptvg | adoptlv
. for commands using lvmlockd locks
. adopt orphan gl/vg/lv lock, or fail the lock request if
no orphan lock is found
. will fail if orphan lock exists with a different lock mode
. command may still continue with a failed shared lock request
--lockopt adopt
. for lockstart or any command using lvmlockd locks
. adopt existing lockspace, or start lockspace if none exists
. adopt orphan gl/vg/lv lock, or acquire new lock if no orphan found
. will fail if orphan lock exists with a different lock mode
. command may still continue with a failed shared lock request
. with dlm this option only works for ls
Stop printing "Skipping global lock: lockspace not found or started"
for vgchange --lockstart, since it's generally an inherent limitation
that the global lock isn't available until after locking is started.
Update the start delay warning to "a few seconds".
The lvb is used to hold lock versions, but lock verions are
no longer used (since the removal of lvmetad), so the lvb
is not actually useful. Disable their use for sanlock to
avoid the extra i/o required to maintain the lvb.
vgremove with --lockopt force should skip lvmlockd-related
steps and allow a forced vg cleanup, in addition to using
--nolocking to skip normal locking calls.
Previously, a command would call lockd_vg() for a local VG,
which would go to lvmlockd, which would send back ENOLS,
and the command would not care when it saw the VG was local.
The pointless back-and-forth to lvmlockd for local VGs can
be avoided by checking the VG lock_type in lvmcache (which
label_scan now saves there; this wasn't the case back when
the original lockd_vg logic was added.) If the lock_type
saved during label_scan indicates a local VG, then the
lockd_vg step is skipped.
The idea in the patch 6e6d4c62b for handling -suffix as
indication of private device needs to be disabled.
Some problematic cases are currently not resolvable and some
more thinking is needed.
Once fixed, we can revert this patch.
For large device sets our dm_hash can produce larger amounf of mapping
collision and we would need to further increase our has size.
So instead use the radix_tree code which is immune agains growing size
of devices and uses memory more effiecently to store all the paths.
Instead of less efficient 'btree' switch dev_cache to use
radix_tree, that is generating more efficient tree mapping.
Some direct use of btree iteration replace with our dev_iter code.
Convert the persisten filter to use more memory compact radix_tree as
dm_hash is bound to preallocated number of slots and stores whole
key together with value.
When DM uuid cache is available, we can possibly avoid unnecessary
status ioctl() when we check the device for 'usable' uuid.
If this test passes the existing code will got through the full check.
Move the code around caching active dm device devno, name and uuid
from device_mapper/libdm-iface to dev_cache file - as libdm layer
cares about 'decoding' ioctl data from kernel and caching for use by
lvm stays within lvm.
Introduce:
dev_cache_update_dm_devs
dev_cache_get_dm_dev_by_devno
dev_cache_get_dm_dev_by_uuid
Use radix_tree for searching.
Do not attach layer suffix to the UUID when activating component LV.
In this case we want to see allow this LV to be public, thus
such LV should not be using -layer suffix in its UUID.
This also requires that our 'cached' access will check for
both UUID (with & without suffix) which was unnoticed issue before.
This change is now necesssary since our udev rule automatically expects
any LV with -layer suffix is private and will prevent generaring
any systemd unit even when there are no 'DM' flags bits passed via
cookie mechanism while creating such LV.
In order to free SubLVs after a stripe removing reshape, lvconvert has
to be run without layout changes. Prevent a layout changing request
in case any such freed SubLVs exist and inform the user about the fact
requesting to release them first.
Calculates expected size before/after reshapes adding/removing stripes
to/from RaidLVs with levels 4/5/6/10 and compares it with the actual
one the block layer shows. Stripes reshaped to are listed in the
tst_stripes variable. mkfs/fsck/resize2fs the respective RaidLVs
to confirm ext4 can be resized accordingly without issues.
We automatically ignore these devs, when lvm2 create devs,
whoever when lvm2 database is dropped or someone just
created these devs with such formated UUID, there is no
other informantion then to check DM UUID.
SUSE kernels distribution enables CONFIG_PRINTK_CALLER by default.
One line of cat /dev/kmsg is like:
6,904,9506214456,-,caller=T24012;loop8: detected capacity change from 0
to 354304
If CONFIG_PRINTK_CALLER is off:
6,721,53563833,-;loop0: detected capacity change from 0 to 354304
',caller=T24012' is the redundant part needed to be handled.
Otherwise pos will be lager than buf size causing sz underflowed.
Then constructor of std::string will throw a exception to break
tests:
$ make check_local T=shell/000-basic.sh
VERBOSE=0 ./lib/runner \
--testdir . --outdir results \
--flavours ndev-vanilla --only shell/000-basic.sh --skip @
running 1 tests
running: [ndev-vanilla] shell/000-basic.sh
0:00.000Exception: basic_string::_M_create
make[1]: *** [Makefile:148: check_local] Error 1
make[1]: Leaving directory '/root/lvm2/test'
make: *** [Makefile:89: check_local] Error 2
Fix it with strchr for ';' as delimiter to get pos.
Reported-by: Su Yue <glass.su@suse.com>
To more easily adopt radix_tree over existing code base, add
abstraction over 'radix_tree_iterate' which basically builds
an array of all traversed values, and then it's just easy to
go over all array elements.
TODO: code should be converted to use radix_tree_iterate()
directly as it's more efficient.
Note: it can be possibly to rewrite recursive _iterate() usage
to linear travesal, not sure whether it's worth the effort
as conversion to 'radix_itree_iterator' is relatively simple.
Add simple 'wrapper' inline functions to insert or return ptr lookup value.
(So the user doesn't need to deal with 'union radix_value' locally and
also it makes easier to translage some lvm2 functions to use radix_tree).
Note: If the stored 'value' would NULL, it cannot be recognized
from a case of 'not found'. So usable only when 'values' stored with
tree are not NULL.
Some updates to _dump() debugging function so the printed result
can be more easily examined by human.
Also print 'prefix' as string when all chars are printable.
Add 'simple' variant of _dump also to 'simple' version of radix tree
(which is however normally not compiled).
Instead of using 'key state & key end' uint8_t* switch to use
void* key, & size_t keylen. This allows easier adaptation with
lvm code base with way too much casting with every use of function.
Also correctly mark const buffers to avoid compiled warnings and
casting.
Adapt the only bcache user ATM for API change.
Adapt unit test to match changed API.
Fix commit b65a2c3f3a "vgimportdevices: skip lvmlockd locking"
which intended to disable lvmlockd locking, but the lockd_gl_disable
flag was mistakenly set after lock_global() so it wasn't effective.
This caused vgimportdevices to fail unless locking was started.
Enums are single 'values' so not a proper type for bitfields.
(Probably better to use such values as defines).
Although here 'daemon_talk()' is part of library API, it's hidden
non-public API call - and moreover 'enum' and 'unsigned' are
using the same size, so linker shouldn't have any issue with
this symbol usage.
For this reason there are no 'versioning' tricks applied.
Size of these hashes was quite small, so raise the size of
hashed entries to reduce amount of hash collistion.
Select some unique/unused number for hash_create below 8192.
Replace call to get_dm_uuid_from_sysfs() with use of
device_get_uuid() which gets the same information,
but instead of several syscalls it need either 1 or even 0
when the information is cached with newer kernels.
We've got cached DM list before grabbing lock, so there
is some chance, that DM table has changed and we would
need to refresh this info.
TODO: benchmark, whether it would even make sense to refresh cache
and keep it content instead of using individual ioctl() for tree build.
Function that is working with DM target is located within
lib/activate directory.
This function is able to use cached dm_device_list when possible
to quickly resolve checks for device's UUID.
Function can fully replace get_dm_uuid_from_sysfs() and instead
of syscalls for open/read/close get the UUID with single ioctl.
When there is cached dm devs list, we can get many UUID from
a single syscall.
For better use of cached data located within cmd_context,
pass this structure from the top level function.
Also add missing '_' for static _dev_cache_index_devs.
No other change here.
Introduce function to find device's name and uuid for
a given major:minor.
This information is cached with dm_device_list which reads all the
info from single ioctl(DM_DEVICE_LIST).
Lvm keeps major:minor name & uuid for active devices in the system.
Skip scan and stat() for dirs and nodes within known /dev/ paths,
where no block devices are located.
Also strlen(_cache.dev_dir) just once.
TODO: add more dirs to _no_scan (configurable via lvm.conf ?)
Just like with _VAL strings, also _ARG strings do not need to
be present - as we can easily check for LONG opt version just
by adding attribute.
With attribute ARG_LONG_OPT string arg name[] becomes unused
and can be safely removed.
Also within _find_command_id_function() we do not need to handle
'command_enum == CMD_NONE' as separate case and just use single loop.
Previous commit 82617852a4
introduce bug in complession - as the rl_completion_matches()
needs to always advance to next element where the index
is held in static variable.
Add comment about this usage.
When PVs are created on LVs, remove the devices file entries
for the PVs when the LVs are removed. In general, the devices
file entries should be removed with lvmdevices --deldev when
the LVs are removed (lvremove is the equivalent of detaching
a device from the system when layering PVs on LVs.)
This change is effectively an automatic lvmdevices --deldev
command that is built into lvremove when the LV has a PV on it.
An OS installer can create system.devices for the system and
disks, but an OS image cannot create the system-specific
system.devices. The OS image can instead configure the
image so that lvm will create system.devices on first boot.
Image preparation steps to enable auto creation of system.devices:
- create empty file /etc/lvm/devices/auto-import-rootvg
- remove any existing /etc/lvm/devices/system.devices
- enable lvm-devices-import.path
- enable lvm-devices-import.service
On first boot of the prepared image:
- udev triggers vgchange -aay --autoactivation event <rootvg>
- vgchange activates LVs in the root VG
- vgchange finds the file /etc/lvm/devices/auto-import-rootvg,
and no /etc/lvm/devices/system.devices, so it creates
/run/lvm/lvm-devices-import
- lvm-devices-import.path is run when /run/lvm/lvm-devices-import
appears, and triggers lvm-devices-import.service
- lvm-devices-import.service runs vgimportdevices --rootvg --auto
- vgimportdevices finds /etc/lvm/devices/auto-import-rootvg,
and no system.devices, so it creates system.devices containing
PVs in the root VG, and removes /etc/lvm/devices/auto-import-rootvg
and /run/lvm/lvm-devices-import
Run directly, vgimportdevices --rootvg (without --auto), will create
a new system.devices for the root VG, or will add devices for the
root VG to an existing system.devices.
lvm2 for a while already optimizes 'vgremove' operation to
a single commit when possible if all LVs can be
easily deactivated.
So the number of LVs doesn't matter much - but the tested
case 'test_delete_non_complete_job' seems to be taking
some time anyway to capture the exception.
So just reducing the running time of the test significatanly
as we don't need to create 64LVs for 4 'execution mode' runs.
Order is used for man page generation (although not completely).
So place 'zero & error' to the end of list.
Keep linear,striped,snapshot in front.
For the rest use alphabetic order.
Move part of the 'inner' loop which is would be otherwise
always production same results for all 'opt_enum' values
out of the loop, so it can be evaluated just once.
Instead of resolving and storing 'command_fn'
withing 'struct command' use just funtion enum
and resolve function pointer just in place,
where it is really needed - first try to resolve
'new style' and fallback to 'old style' named.
Instead of storing command_id as string, direcly
translate string to enum index and use 'command_enum()'
to get string when needed for printing.
This way we can easily detect error in the structure
while parsing it - and we can later avoid separate
'translation' loop.
We can more efficiently use command_name struct to
lookup for lvm_command_enum and avoid many repeated
command name searches since we already know
the enum index that is now stored in 'struct command'.
Constity members in cmdline_context, would be nice, to replace
this static struct with couple function calls.
Also replace some 'while' loops with for loops, so code
is more readable.
Split struct command_name to the constant part (keep the name)
and new 'struct command_name_args' which holds runtime computed
info. To get to the _args part - we can easily use
lvm_command_enum as equivalent index.
Constified part 'struct command_name' is now fully stored
in .data.rel.ro segment, while command_name_args part goes
to .bss segment.
Code will be further reduced with next refactoring.
Refactor code to not allocate memory for rule->opts,
instead use uint16_t array of MAX_RULE_OPTS within cmd_rule.
Also print more info if array would not be enough (>= 8).
Add slight delay for waiting until 'started' dmeventd is
responsing to other 'dmeventd -i' command.
TODO: race observed here is somewhat unclear, might need some more
details
There is no reason to wait for 10sec when removing 'vg' at test
exit - we just need to kill 'sleep 10' process forked from flock.
(Not using 'flock -F' as this flag is not across all machines used
for testing)
This reverts commit d16a8f80e9.
So the correction was OK, however we missed to fix also
cut&paste bug here - as the second check should be
actually checking field->type.
Update definitions to add support for some more VDO options
when converting volumes to be used as thin-pool with vdo data volume.
Split some option in existing OO_LVCONVERT_VDO to OO_LVCONVERT_VDO_POOL
and reused then with OO_LVCONVERT_THINPOOL.
When convert already existing vdopool to be used as
thin-pool backend and user is passinng option for VDO configuration
process them - as we know converted LV is offline, we can do such
change easily instead of telling user to run separate lvchange later.
When converting volumes to be used for thin-pool with VDO, allow
users to control wipesingature behaviour.
By default volumes should be checked against signature, and if
they are present, we promt user whether he wants to process with
conversion and lose i.e. filesystem present on such volume.
Users that want to bypass prompt in script can use either --yes
or they can disable wipe signature -Wn.
Resolve reported leak by renaming existing pckk() to pvck_mf()
and wrapping pvck() over this. This function can easily
free allocated buffer within the subfunction.
When we switch supported_reserved_types_with_range to const
gcc repots this problem:
warning: ‘and’ of mutually exclusive equal-tests is always 0
!(iter->type & supported_reserved_types_with_range))) {
It's not clear from the history what was the actual intention of this
internal error test, let's assume the check wanted to make sure
that when DM_REPORT_FIELD_RESERVED_VALUE_RANGE is set,
some other fields from supported_reserved_types_with_range set
are also selected.
This was rather API mistake - the internal of libdm
do handle suffixes list as const string, just the API
was only using 'const char **'.
So the user may pass safely casted 'const char * const`.
man-generate --check actually noticed 2 same definitions
for snapshot create with 'lvreate -T [--snapshot]'
and 'lvcreate --snapshot [-T]'.
So drop the '-T' from second alternative variant as
thin type is already implied here.
Refactor code so the definitions may become 'static const'
and with configure_command_option_values() we update options
val_enum for actually running command option when used.
Also update _update_relative_opt() which is used for
generating man pages and command help.
Introduce enumeration for lvm2 commands - so we may
use enum cmd_COMMAND instead of string checking.
So running command now does not modified opt_names.
Function to _allocate_memory() was not compiled-in when lvm2 was
build with support for better tracking memory pool with valgrind.
Instead now correctly avoid this function only when running
withing valgrind environment.
Do not modify flags field from 'strcut command_name' and
instead control this via cmd_context get_vgname_from_options.
Flag GET_VGNAME_FROM_OPTIONS is currently used only by lvconvert.
Create leading dirs for lvmdbusd lock file, otherwise it fails to start:
| systemd[1]: Starting LVM2 D-Bus service...
| lvmdbusd[1602]: [1602]: Error during creation of lock file(/var/lock/lvm/lvmdbusd): errno(2), exiting!
Signed-off-by: Kai Kang <kai.kang@windriver.com>
In the past it was common for a single command to run
multiple lvmcache_label_scan, and this code was a way
to make each call select the same duplicate pvs. Now
commands run a single lvmcache_label_scan, so this is
not needed.
When reseting stream buffer - check for being run within valgrind
and only in this case skip this code.
Define VALGRIND_POOL was incorrectly used for this logic.
We can run and build this code just fine with VALGRIND_POOL
and just rely on runtime detection - so we do not close
descriptors also within running valgrind session.
Old system do not work well with -l% in findstring,
so use a different strategy to recognize whether -lreadline needs
another library (-ltinfo, -lncourses...)
(So we don't need to solve this via 'configure')
Also for now comment out -Wl,-z,pack-relative-relocs and leave it
for the package maintainer whether they want to use.
Possibly add some 'configure' autodetection for usable switch,
as it's relatively new feature..
Add some -Wl flags separatly and avoid their duplication.
Also add --as-needed only when system is using 'newer' readline
library - on these older system the usage of '--as-needed'
seems to be causing some hard to solve problem - so avoid it.
Attach -Wl,-z,relro,-z,now,-z,pack-relative-relocs,--as-needed
to LDFLAGS, but only if LDFLAGS already doesn't contain 'relro'
(so it's not given repeatedly).
Also start to use -z,now linkage also when building libraries
with default compilation - this avoid calling symbol resolver
while library function are using function needing resolving.
Note: Fedora or RHEL rpm building is using:
CFLAGS=$(rpm --eval %{build_cflags})
LDFLAGS=$(rpm --eval %{build_ldflags})
Also split -DUSE_SD_NOTIFY into DEFS from CFLAGS.
Use LDFLAGS separately with every use of CLDFLAGS and leave
this flag only for handling versioning.
This will reflect any LDFLAGS setting use during make.
Look not only for whole 64byte sequence,
but seek also 32byte, 16byte and 8byte parts of the
key.
Currently to pass memcpy ZMM problems add possible
workaround in the form of GLIBC_TUNABLES setting.
Avoid using 'lvs' from 'get' shell - as that would wait until
whole group of processes is finished.
TODO: rethink what would be the point of starting 'dmeventd' with lvs.
It seems to break some rules.
Refactor _start_daemon and add synchronization delay waiting
untill new forked dmeventd actually open fifos and is ready
to process messages.
This closes some race window in testing.
When no lock manager for the global lock had been set yet,
and the first global lock request was received, and both
dlm and sanlock were running, lvmlockd would assume it
should use the dlm for the global lock, and would start
the "lvm_global" dlm lockspace automatically. That's
not always correct, so don't assume the dlm should be used,
fail the lock request, and wait until a VG with a specific
lock type is started to determine the lock manager to use.
Detect also presence of 'vdoformat' tool.
Fallback to 'kvdo' modprobe only when dm-vdo fails
(removed ugly error message from log)
Also add extra check for scsi model being present
so the test can wait a bit if 'scsi_debug' takes longer time.
No need for 'aux' within aux.
Refactor existing code from tools/lvmcmdline.c to
libdaemon/server/daemon-stray.h daemon_close_stray_fds()
used to close stray descriptors above some specified Fd.
This is code parses content of /proc dir to minimize 'blind' closing
of all possible descriptors within rlimits range.
As we have the same code in few other places in it's more 'trivial'
version - these were actually sensitive to high amount of descriptors,
which might be configured on some system.
With this patch we effectively resolve this reported gitlab issue:
https://gitlab.com/lvmteam/lvm2/-/issues/5
TODO: Current placement might not be ideal - however considering
existing code base constrains it's not so simple.
ATM it uses lib/misc/lvm-file.h for custom_fds declaration
and rest of functinality is included in daemon header file.
Let's move proc into include/configure.h so this define can
be easily used across the source base.
Configure defines it - but ATM we do not provide any configure
option to change it - there should be no need to ever change it.
Handle the case, where we 'kill -9' running dmeventd,
and restarting 'dmeventd -R' manages to still contact this
'zombie' damone and manages to get list of monitored devices
out of this and eventually 'run' new and in this case
unexpected instance of dmeventd.
Some test do expect event_activation to be set.
So add explicit configuring of this setting in tests,
but also add new default which kind of does it globally
as it's expected default (yet our testing rpms might
be create with disabled event_activation)
By adding this to each test individually - it's now easy
to locate such tests...
Move static to main() local vars.
Initilize libdm logging also before starting _restart_dmeventd()
so function there can be also logged.
Move call of _info_dmevent() out of option processing - so some
options like -V are processed with higher priority.
Use _restart_dmeventd() with return values 0,1,2.
Also let's use already created fifos struct.
Make sure the _systemd_activation variable is properly initialized
from _systemd_handover before calling _restart_dmeventd() as
this variable is used there to decide where 1 or 2 should
be returned - aka either letting systemd to initilize
or restart dmeventd itself.
This is initial test how to disable event_activation on
machines, where lvm2-testsuite packages are installed
with its lvm.conf file.
TODO: find more elegant mechanism
Occasionally this test fails as soemtimes UUID actually
may constain LV[d] string causing it to grep mismatch
UUID and LV name and eventually fail test for wrong reason.
As a simple workaround print the LV name first and
check the name is followed by a space character.
This code is somewhat complex and involves recursion and pointer
shuffling which confuses coverity here.
Let's add masking comment for this warning as there is no double
free in this code.
Let's stop Coverity thinking here we are using freed FILE*
for anything else then comparing numbers.
For this use the original source of old_stream pointer.
Add new configurable option for building lvm2 with enable/disable
default autoactivation setting.
Might be useful for building i.e. rpms for systems where
this event_activation is not desired.
Parse timestamps included in kernel kmsg line and add current system
time to the messsage as well - this makes it easier to look over,
when chasing some messages in journal after some time...
Before initiation fifo communication, check whether there is
running dmeventd - in case there is no running instance, this
would be just blocked for 5 seconds trying to open fifo.
Also awk got annoyed by this \# char sequence which is reported
as incorrect, however older rpm builder were failing without this.
So let's just try other variant.
Thin-pool and cache-pool targets got already quite stable so let's try
to remove checking of pools when using lvremove or vgremove commands.
This skips checking pools when they are going to be removed, but it
also when removing thin volume that was the only user of a thin-pool.
In this case thin-pool will be still there and could be activated
again with another thin volume and thin_check will be executed
in this moment. In this case it can delay discovery of metadata damage.
Shortens processing of 'lvcreate -L -V -T' command and
avoid deactivation and its activation with thin_check of the empty
created thin-pool that will be used for the new thin volume
made with a single lvcreate command.
Shuffle code to parsing VDO message also for lvs segment status
so it can report correctly data usage for VDO LVs.
For this change move code and also change its API to use just mempool.
Fixes usage with upstream 6.9 vdo target driver.
When using message API for parsing VDO stats info, 0 was wrongly
used for fallback for trying the old sysfs API.
Switch to use ULLONG_MAX for values that could not have been obtained
through the message call.
Fixes lvdisplay info for freshly created VDO volume with 0 used data
blocks.
Make a seperate function to decode which ID should be user
for cvol meta or data volume - also avoids duplication of code.
As a result it's now also easier to see how the lvid is build.
Fix cutting away 1 character via incorrect usage of dm_strncpy
introduced in last batch of commits and use sizeof(buffer) to
get proper sizes.
In case of use strcpy_name_len() the replacement was invalid,
so we need to restore this case since sanlock uses buffer without
nul ending, so we would strip one more character from the buffer.
Also start to use dm_strncpy() without (void) for unchecked usage.
Add internal inline function wrapper for dm_strncpy().
Use it for calls where we test the result.
Avoids emitting warnings in Coverity for unchecked usage.
This sanity check actually confused in some way Coverity
giving it some assumption about array access.
Since these two checks basically validated compiler's capability
to add and then substract the number from char pointer we likely
don't really need them - as if this base math would not work,
compiler would be having far more troubles...
So drop them - and get rid of report:
Event overrun-call: Overrunning callee's array of size 513 by...
Reduce #lv and intervals.
We are trying to ensure that the daemon stops while it's busy processing
its internal queues. Decrease the number of LVs and intervals to allow
this test to complete in less time.
Add a global timeout value to be used for the threads to end waiting for
whatever it is they are blocked on. The values varied from 2-5 seconds,
which is way longer than needed. Value of 0.5 shows no CPU load when
service is running and is idle.
Shuffle code to avoid using static variable to pass parsed option.
Code is now easier to follow and also number of coverity reports
will go away.
There should be no functional change.
Add debug tracing for syscall failures.
Also switch some log_error to log_warn when command does not exit
with 'error' result and only warns user.
Easier error path handling.
Initialize some vars at declaration time.
Here we actually need to slowdown only $dev2 - since repair operation
is only reading data from this device and compares it with origin $dev1,
and if they match there is no write...
Move memset() to the initialization function define_commands().
There is also not much point in clearing memory on command's exit
so drop zeroing of ~2M of RAM.
Use \0 as EOL in compiled-in syntax description to avoid
unnecessary line copy that just replaced \n with \0.
Also use already splitted lines when possible.
Handle mismatch of reported 'dm raid' status, where the active
raid LV can be actually showing higher numebr of raid leg devices,
that the number of shown status characters.
This can happen if the raid leg is dropped during initial
resynchronization.
Setting db_persist is required for dm devices so that their properties
are carried over on switch-root from the initrd to the rootfs. This
logic has always lived in dracut
(https://github.com/dracutdevs/dracut/blob/master/modules.d/90dm/11-dm.rules).
However, this means that other initramfs generators each have to
implement and maintain the same rule which leads to unnecessary
duplication.
Instead, let's make the rule part of the upstream lvm rules, which
will ensure that generated initramfses will just work if they make
sure the lvm udev rules are installed, without having to figure out
that they have to add an extra rule themselves on top.
Identical rule in Arch Linux's lvm2 package: https://gitlab.archlinux.org/archlinux/packaging/packages/lvm2/-/blob/main/11-dm-initramfs.rules?ref_type=heads
For DM devices, the add/change/remove can appear as action for genuine
udev events.
However, there are more action types (bind, unbind, move, online, offline)
which never appear as actions for genuine DM udev events, but they can
still be synthesized (e.g. by writing "<action>" to "/sys/.../uevent" file
or by calling "udevadm trigger --action=<action>").
Let's also process these extra action types so that the udev-related content
is not lost completely, keeping all the symlinks and udev db entries just like
this was a synthetic udev event with "change" action.
Related to https://gitlab.com/lvmteam/lvm2/-/issues/4.
Bump the rules version in order to indicate that upper level rules
should consume DM_UDEV_DISABLE_OTHER_RULES_FLAG rather than DM_NOSCAN
and DM_SUSPENDED.
Also update the comments at the top of the file that describe the
exported properties, and add a note about internal device-mapper
properties.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Peter Rajnoha <prajnoha@redhat.com>
DM_NOSCAN is not an official API any more and doesn't have to be
restored from the udev db. Rename it to .DM_NOSCAN.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Peter Rajnoha <prajnoha@redhat.com>
DM_SUSPENDED is a device-mapper internal flag, which is not intended to be
used by other rules, and which is determined by 10-dm.rules from sysfs for
every uevent. Rename it to ".DM_SUSPENDED", so that it won't be saved in the
udev database.
Known consumers of DM_SUSPENDED are 66-kpartx.rules (from multipath-tools) and
99-systemd.rules (from systemd). These will have to be adapted.
11-dm-mpath.rules will be changed to use .DM_SUSPENDED.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Peter Rajnoha <prajnoha@redhat.com>
DM_UDEV_DISABLE_OTHER_RULES_FLAG is used as the "output" flag of the
device-mapper rules, to be consumed by non-dm rules. It is a logical OR of
several conditions that might make dm devices inaccessible. 10-dm.rules
calculates it for every uevent, whether it's genuine or spurious.
DM_SUBSYSTEM_UDEV_FLAG0 is just another flag that needs to be or'd in. We
don't need to restore the previous state of DM_UDEV_DISABLE_OTHER_RULES_FLAG.
Actually, doing so is wrong if the flag has previously been set because the
device was suspended, and the device isn't suspended anymore.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Peter Rajnoha <prajnoha@redhat.com>
We use DM_UDEV_DISABLE_OTHER_RULES_FLAG to tell upper non-DM layers
to keep their hands off the device in question, for any reason.
One possible reason is that the device is supended; another is that
the cookie carries the flag of the same name.
DM_SUSPENDED is not restored from the db, but evaluated anew for every
uevent. Therefore DM_UDEV_DISABLE_OTHER_RULES_FLAG shouldn't be
restored, either. Use a new variable DM_COOKIE_DISABLE_OTHER_RULES_FLAG
to save and restore the original value from the cookie.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Peter Rajnoha <prajnoha@redhat.com>
DISK_RO is set in the environment of a block-device uevent if and only if
the read-only (ro) attribute of the device just changed (the kernel
function set_disk_ro() was called). It is not synoymous with the "ro" sysfs
attribute; the device could very well be write-protected if DISK_RO is not
set. Device mapper-level probing is possible for DISK_RO events, but it makes
little sense, because the device propreties haven't changed as far as dm is
concerned. But we should import possible previously set device properties
to avoid confusing follow-up rules. We should do this for both DISK_RO=1
and DISK_RO=0 events.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Peter Rajnoha <prajnoha@redhat.com>
ID_FS_TYPE is the most important udev property for most follow-up
rules. It must be imported from the udev db if blkid can't be run.
Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Peter Rajnoha <prajnoha@redhat.com>
When using cached LV with cachevols (so not with cachepool),
the loaded table could have been using more then one mapping line
for sub devices - resulting into data corruption in some cases
when i.e. taking snapshot of such cached LV with and instead of
single line - 2 lines were generated into DM table as the code
skipped protection again repeated addition.
vg-fast_cvol-cdata: 0 16384 linear 253:2 16384
vg-fast_cvol-cdata: 16384 16384 linear 253:2 16384
New code is also refactoring to use _add_new_cvol_subdev_to_dtree
(similar _add_cvol_subdev.. ) and also the addition of subdev has
been moved after check for already processed node.
Also the cachevol sub devices are now added with the insertion
of cachevol with cached LV.
Document usage of chaining external origins as it is now possibly to
create thinA LV in poolA which use thinB LV in poolB as it external
origin and user could create chain of such LV.
When using swap_lv_identifiers() we were basiclly exchanging 'names'
and only according to the caller some more data were 'transfered'.
However in most cases we should swap properties like 'hostname' as
the creation information should be preserved.
So let's do the function more universal.
Improve support for building DM tree when there is a chain
of external origins used for LV.
For this we cannot use track_external_lv_deps as this works
only for LV with just one external origin in its device tree.
Instead add directly 'dev' to the instead of add whole LV.
This avoid possibly recurive endless loop, however we may eventally
have some problems with undiscovered/missing devices in DM tree.
When we have some existing LV and this LV is being converted to
external origin - during the DM table manipulation there is a short
moment when the LV is being 'resumed' as 'read-only' volume
while still being live as 'rw' volume i.e. we could have had
a single thin LV active twice.
To avoid such weird scenarios of dual access to a same volume, we
just postpone a resume until a moment, where the existing volume
is already suspended thus no I/O can be in flight to such device.
Note: however there is slight catch - that we now have basically
a different 'risk' case where a resume of such i.e. new external
origin LV might fail and we are already in suspend tree state -
resolving error path in this situation is untrivial as well...
Fix/support creation and usage of the external origin
across thin-pools - so thin LV can use thin LV from
some other thin-pool as external origin (read-only).
When creating external origin via 'lvcreate --type thin'
add the validation for LV being usable as external origin
since certain LVs cannot be really used this way.
Also call this function early during lvcreate cmdline arg
validation se we do not need to do unecesary operation.
Over the time the code for preloading detached LVs got unnecessarily
complicate. But actually we need to preload only LVs that
were previously non-toplevel (invisible) LVs and became visible
toplevel LVs in the precommitted metadata.
If there would be needed some other rule, it would likely be a bug in
conversion code forgetting to set visibility flag on detached LV.
This reduces number of unnecessary repeated DM tree preloading.
When building in other dir, ensure srcdir is used to find helper
script (via vpath) through $< usage.
Add also missing [INSTALL] prints for some installed files.
External origins for thin volumes can be also used at the same time
as old(thick) snapshot origins. However in this case it's possible
the LV is only active as being 'external' origin, but old snapshot LVs
are not active. For this case before handling these
LVs for un/monitoring check the active state of origin LV.
This should prevent warnings of monitoring failures.
Make recursive directory path creation reusable via
dir_create_recursive.
While we already have dm_create_dir() - it's not taking mode arg,
so let's make lvm's internal file helper function.
Instead of parsing the whole /proc/kallsyms use faster variant
of using modprobe tool logic.
lvm2 here wants to know whether the particular DM cache policy is
present in the kernel - however since the cache policy does not have
any kernel module parameters and it can be built-in to a kernel
there is no /sys/modules directory in such case and we would need to call
modprobe everytime we want detect such case.
The old solution tried to look for particular kernel symbol
(and like not the right way, as smq_exit might be actually ommitted).
New version checks MODULES_PATH/`uname -r`/modules.builtin for
whether is present cache policy module instead of CPU expensive parsing
of kallsyms.
If lvm.conf has use_devicesfile=0 and /etc/lvm/device/system.devices
exists, then rename it to system.devices-unused.YYYYMMDD.HHMMSS.
This prevents an old, incorrect system.devices from being used in
the future if lvm.conf is changed to use_devicesfile=1.
Create backup copies of system.devices in /etc/lvm/devices/backup
named system.devices-YYYYMMDD.HHMMSS.NNNN. NNNN is the version
counter from the file.
Each time that an lvm command writes a new system.devices file,
it also writes the same file in the backup directory.
A new comment line is added to system.devices with HASH=<num>
where <num> is a crc calculated from the uncommented lines in
system.devices. This lets lvm detect if the file has been
modified outside of lvm itself.
If system.devices is edited directly, the next time a command
reads the file, the crc will not match the HASH value. The
command will then rewrite system.devices with the correct HASH
value, and create a backup reflecting the edits.
A default limit of 50 backup files is kept, configurable by
lvm.conf devicesfile_backup_limit (set to 0 to disable backups.)
If LVM LVs happen to contain PVs, they are passed to the lvm udev
rule for processing, where they should be ignored. PVs on LVs
most likely belong to VM images, and don't belong to the host
which sees the LV. It's unsafe for the host to use these PVs.
Without this change, the LV would be processed by pvscan which
would generally ignore it, either because of the devices file,
or because of the default lvm policy to not consider LVs as
potential PVs. This change makes the udev rule consistent
with that policy and avoids the unnecessary system messages
produced when pvscan ignores the LV.
Since understanding the reason for choosing the appmachineid over the
direct use of machineid is not easily found, I extended to help text to
clarify this a bit.
Function to recalc chunk_size according to dev hints needs to be
used after chunk_size is being set to thin pool segment - correct
this ordering mistake introduced in previous refactoring commit.
Previous patch that introduced support for thinpool with vdo
not correctly handled header size - as this part is not fully usable
yet. We are going to try to use the 0, but current state of code is not
yet compliant to this logic so keep vdo_header_size during conversion
and alos correctly pass through virtual_extents to vdo formating.
Add code to handle creation of thin-pool with VDO data backend
which can be seen as compressed deduplicated thin-pool.
To avoid need of changing to many internal APIs, pass the conversion
parameters for create thin-pool data volume via cmd_context.
Introduce struct vdo_convert_params {} to pass-in all the parameters
needed for the conversion of an LV to a vdopool + vdo LV.
Function convert_vdo_lv() is also able to create a new LV and swap
segments, so the passed in LV can be later on use for futher
conversion so this refactoring makes it ready for more enhanced
usage.
Introduce vdo_convert_params and use vdo_params from this structure
also with lvcreate_params.
Later we will use this for convertion of thin-pool data volume to VDO.
If a RaidLV mapping is required to be refreshed as a result of temporarily failed
and recurred RAID leg device (pairs) caused by writes to the LV during failure,
the requirement is reported by volume health character r' in position 9 of the
LV's attribute field (see 'man lvs' about additional volume health characters).
As this character can be overlooked, this patch adds messages to the top
of the lvs command output informing the user explicitely about the fact.
Fix typos in previous commit 3589e515d.
Both commands default [raid_](min|max)recoveryrate to 0 but ensure
min_recovery_rate is not larger than max_recoveryrate. This results
in command failure without requesting the user to also define
max_recovery_rate >= min_recovery_rate.
Fix both commands by defining max_recovery_rate = min_recoveryrate
in case "lvcreate/lvchange --minrecoveryrate Size ..." requests a
larger value than current maxrecoveryrate without also giving option
Both commands default [raid_](min|max)recoveryrate to 0 but ensure
min_recovery_rate is not larger than max_recoveryrate. This results
in command failure without requestinng the user to also define
max_recovery_rate >= min_recovery_rate.
Fix both commands by defining max_recovery_rate = min_recoveryrate
in case "lvcreate/lvchange --minrecoveryrate Size ..." requests a
larger value than current maxrecoveryrate without also giving option
"--maxrecoveryrate Size ..." with a size greater or equal than min.
The first lv_attr flag is 'i' or 'I' for a raid image.
(i: raid image, I: out of sync raid image)
For integrity raid images (_iorig), the flag was not being set.
pvs -A|--allpvs
Show PVs that would otherwise be excluded by the devices file.
pvscan -A|--allpvs
Show PVs that would otherwise be excluded by the devices file.
For those devices that are included by the devices file,
their device ID is displayed in place of the usual "lvm2"
format and size.
(pvs -a|--all is unchanged, and shows devices not formatted as PVs.)
A pvid string read from system.devices could be less
then ID_LEN since system.devices fields can be edited.
Ensure the pvid buffer is ID_LEN+1 even if the string
read from the file is shorter.
Include info in the temp file to confirm that it should be used.
The temp file is meant to suppress repeated, identical searches
for the same PVIDs on the same set of devices. Write to the file
a count and hash of the missing PVIDs and a count and hash of the
devices to search. A subsequent command will ignore and remove
the temp file if any of these values differ. We don't want to
suppress a search if a change has occured, and a missing PV could
be found by scanning devices.
Problematic scenario:
. the device for a PV has no wwid, so it's identified in system.devices
with IDTYPE=devname IDNAME=/dev/foo
. user adds/enables a wwid for the device
. on reboot, the device name changes, e.g. now /dev/bar
. the code that searches for the new device name includes an
optimization to skip looking on devs that have a wwid, on
the basis that a device with a wwid won't have IDTYPE=devname
. this optimization causes lvm to not look for the PV on /dev/bar
since that device now has a wwid, so the PV is not found
. the optimization is enabled by search_for_devnames="auto"
. change the default to search_for_devnames="all" which does not
use the problematic optimization
- add new comparison between old and new entries, and use this
as the basis for new dedicated output for check and update
- add new --refresh option to search for missing PVIDs on all
devices, and possibly update the device ID
- internally, only use the term "refresh" for cases where a
new device ID may be found and assigned for a missing PVID
During vdo testing with smaller block devices the test needs to determine
which PVs in a VG are unused. This was leveraging LvCommon.Devices, but
that isn't populated when a LV is resides on a LV pool instead of a PV.
This is a known limitation of the code at this time. For now we will walk
all the PVs in the VG looking for ones that don't have any associated LVs
and use them instead.
Signed-off-by: Tony Asleson <tasleson@redhat.com>
When getting raid status from some older kernels, we may get an 'a'
for 'A' leg when doing initial synchronization.
This may prohibit removal of newly synchronized leg until synchronization
is finished.
So in this case change the status to look like being reported
from a newer kernel version.
Incorrectly matching a dev to a devname id (due to changing devnames)
before matching the dev to a proper device id, can result in the
dev not being matched to the real id.
When reporting in JSON format, we need to be able to find the 'last
displayed row', not just 'last row' as we did before. This is used
to decide whether to put the JSON_SEPARATOR (the ',' character)
between the lines when reporting in JSON format.
This is mainly important in case we use a combination of JSON format
and a report marked with DM_REPORT_OUTPUT_MULTIPLE_TIMES flag.
Such report may be reused several times with different selection
criteria each time. In that case, the report always contains all lines
in memory, even though some of them do not need to pass the selection
criteria that are currently used.
Without DM_REPORT_OUTPUT_MULTIPLE_TIMES flag, the report only contains
the lines that have passed the selection criteria, so the this wasn't
an issue in this case.
Fix suggested by Lars Ellenberg and reported here: https://github.com/lvmteam/lvm2/issues/130
It looks like there is some kernel bug/limitation
that may cause invalid table load processing:
dmsetup load LVMTEST-LV1
device-mapper: reload ioctl on LVMTEST-LV1 failed: Invalid argument
md/raid:mdX: reshape_position too early for auto-recovery - aborting.
md: pers->run() failed ...
device-mapper: table: 253:38: raid: Failed to run raid array (-EINVAL)
device-mapper: ioctl: error adding target to table
However ATM there is not much we can do then make delays bigger.
TODO: fixing md core...
With commit d7e922480e
lvconvert -m may fail if we try to remove 1st. leg that
is out-of-sync while other leg is in-sync.
Hot fix allows to proceed with such down conversion.
Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Fixes commit 63b469c160
"device_id: fix hints with device ids"
It's not correct for internal filtering to be a factor
in validating the set of devs that are the basis for hints.
Instead, look for inconsistencies between a hint entry and
the corresponding devices file entry.
Extend the test a bit futher so we can keep logic of resize
working similarly well for older and newer systems.
Test uses new 'aux have_fsinfo'function to regnize compiled
version of lvm.
Apply the same logic for 'lvreduce' which exists for newer
systems (compiled with HAVE_BLKID_SUBLKS_FSINFO)
also for older systems for one very common practical case where
the active LV does not have any blkid known signature/filesystem.
New variant recognized this situation and allowed to proceed
without requesting a prompt, while the older variant always
requested confirmation prompt.
With this patch command now works equily for both variants
for 'active LV' without signature and allows to reduce LV
without prompting.
Simplify configuration with enabling options:
- For --enable-dmeventd automatically --enable-cmdlib.
- For --enable-dbus-service auto --enable-notify-dbus.
Fix check linux/fiemap.h for dmfilemapd.
Before checking seg_type of the first area, check there is
some existing area.
Since we now support error and zero segtypes, these do not have
any PV area present.
When new app links with current dm_task_run() that propagates ioctl()
errno - ensure it will not be usable with older version without this
supported feature.
Introduce couple configure options:
--without-blkid
--without-systemd
--without-udev
These can be useful for build with --disable-shared
where those libraries are often distibuted without
their static libraries.
However such compiled static binaries then have limitted usage.
Since LVM 2.02 it appears that compiling LVM on a platform which
lacks a shared library linker/loader will fail to produce any
binaries (dmsetup, lvm2, etc).
Adds support for the standard --disable-shared flag, and
use it to disable attempts at building shared libraries.
Modified-by: zkabelac
Fix a bug in _stats_set_aux() that causes bogus data to appear
in the 'userdata' field of stats reports when previously grouped
regions are ungrouped:
/var/tmp/File With Spaces: Created new group with 1 region(s) as group ID 0.
Removed group ID 0 on fedora-root
Name GrpID RgID ObjType RgStart RgSize #Areas ArSize ProgID UserData
fedora-root - 0 region 6.39g 100.00m 1 100.00m dmstats #-
^^
This is the aux_data separator character followed by empty user data.
The _stats_set_aux() function should only emit the separator if
there is a valid group descriptor for the region.
Creating a group with a name that contains whitespace causes an error in
the @stats_set_aux message:
device-mapper: message ioctl on (253:0) failed: Invalid argument
Could not create regions from file /var/tmp/File With Spaces.
Fix this by quoting the group alias and backslash escaping any
embedded space characters.
Refactor the function that parses regions from @stats_list data to
separate the parsing of variable string data from the fixed parts
of the @stats_list response.
Fix a double free in the error path from _stats_create_group() by
clearing the group struct embedded in the dm_stats handle before
returning:
device-mapper: message ioctl on (253:0) failed: Invalid argument
Could not create regions from file /var/tmp/File With Spaces.
free(): double free detected in tcache 2
Aborted (core dumped)
When setting up dm-verity devices with signed root hashes it is very
useful to have a recognizable error code when a key is not present in
the kernel keyring. Turns out the kernel actually returns ENOKEY in that
case, but this gets lost in libdevmapper.
This fixes this: in _create_and_load_v4() it copies the error code from
the ioctl from the sub-tasks back to the main task field on failure.
This is not enough to make libcryptsetup actually propagate the ENOKEY
correctly, that also needs a patch to libcryptsetup, but this is part of
the puzzle.
Fix some interactions between device IDs and hints. Hints
may limit the scanned devices which should not always trigger
a search for the PVs that were intentionally not scanned.
Hints should also be invalidated if they contain a device
that's become excluded by an internal filter such as the
device_id filter.
Search for a PV on other devices if it's a devname entry
and the name doesn't exist on the system. This restores
code that should not have been removed in commit 1901a47df
"device_id: fix conditions for device_ids_refresh"
throtling mirror device is becoming useless with faster CPUS,
as way to many data can be transferred before throttling steps-in.
So prefer using dm-delay for test and keep throttling as fallback.
Fix commit 847f1dd99c
"device_id: rewrite validation of devname entries"
which began calling device_ids_refresh() in cases where it
was unnecessary, leading to extra PV searches and warnings.
Specifically, a command like "lvs <vg>" would use the hints
file to scan only devices for the named VG. This means that
scanning other PVs would be skipped, and device IDs of those
PVs could not be validated because there are no PVID values
to verify. This missing info would cause messages about
the missing info, and would cause device_ids_refresh to
search for the PVs that had been intentionally skipped.
Detection of how the command is supposed to behave actually depends on
the configure.h compilation and whether binary is compiled with
HAVE_BLKID_SUBLKS_FSINFO.
This makes it somewhat complicated in a way how to recognize which
behavior is expected.
Currently we can eventually recognize by checking error output
of some 'random' lvresize command and see if the --fs checksize is
actually recognized and rejected. If this changes - test needs
to be updated.
After umout we may race with system udevd rule - so
just retry once again after 1s sleep - that should be
enough - otherwise we would need some loop here...
Multi-line echo command are problemat across variety of bash version
and may have produce shorter results.
Convert to stable heredoc string with 'tab' skipping <<- for better
formating.
Since we now push more data into journal, parser reading this file
for --continue mode need to be adapted.
Also properly align batch mode with '.' for max test length name.
This is mainly useful in internal testing - but keep sysfs dir also
passed to filter.
Also drop use of static variable within sysfs filter and base whole
config at creation time.
Restore fsync() call For more accurate tracking by buildbot.
Try different rather tricky way of static_cast to use
already opened FD instead of seperate open(),fsync(),close().
It's pretty strange there is no way to enforce fsync() for
C++ iostreams. Flush() is actully not equal.
Add Timespec class to increase time resolution to miliseconds
(can switch to microseconds if ever needed).
Use more const and const_interators and pass by reference.
Output rusage also to list result file.
Reduce inlining of C++ constructors.
If the system changes, locate PVs that appear on different devices,
and update the device IDs in the devices file. A system change is
detected by saving the DMI product_uuid or hostname in the devices
file, and comparing it to the current system value. If a root PV
is restored or copied to a new system with different devices, then
the product_uuid or hostname should change, and trigger lvm to
locate PVIDs from system.devices on new devices.
When exit on file is present in a system and term/break signal is
catched, them dmeventd is no longger refusing to exit.
For the correct shutdown, there should be ideally unmonitoring call,
however in some case it's very hard to implement this correct procedure.
With this 'exit on' file dmeventd at least avoid 'blocking' shutdown,
before systemd kills use with -9 anyway possibly even in some unwanted
stated of internal dmeventd processing (i.e. in the middle of some lvm
command processing).
To quickly get info about some internal dmeventd status,
implment 'dmeventd -i' support.
Reported messages are some 'raw' internal informations mainly
useful to developers.
Instead of just exiting in the middle of monitoring,
unregisted all monitored devices first and then exit.
To speedup this path, all send internal SIGINT when thread
unregiters itself, to wakup-up main sleeping loop.
Make a local copy of the 'idx' string to avoid
overlapping during the rebuild of name.
This fixes cases where users specified raid
component LVs for moving.
Reported-by: kotarou3@github.com
No write outside of $LVM_TEST_DIR (removed /test access).
Use 'aux prepare_scsi_debug_dev' for automated scsi_debug handling
Properly use "" around shell vars.
Smarter read of PVS values.
Relax requirement to only work with real /dev dir.
Handle the case of device teardown where the first pass
could have only a single, but opened device, for removal.
In such case we want to at least once go through
the udev_wait and retry removal again.
TODO: maybe a sleep .1 might be usable as well with udev_wait
Instead of relying on 'pvs' output - check directly system
configuation and use lvmdevice accrording to use_devicesfile setting.
Also drop use of --fs ignore for filesystem extension for better
backward compatibility with older lvm version.
Shuffle code a bit so the '--no-snapshot' path does not execute
a few unnecessary commands.
Work also with devices that may have ':' inside their generated
/dev/disk/by-id
Ensure there is no race with systems' auto activation while using
the snapshot for conversion.
Update system's vdoconf.yml after the use of snapshot for conversion.
Skip unnecesary prompt for 'convert' while using snapshot and query only
for final snaphot merge.
Prohibit conversion for a device with the PV header.
Enhance 'trap' protection for more signals.
Improve clean() recovery path.
Replace bash 'test' command with [].
Correct some output message to print $TOOL.
Support also options without '-' in the middle i.e. --nosnapshot.
For shellcheck predefine all variables extracted from vdoconf.yml.
Since lvmdbusd testing tends to do its own logging,
try for now to disable very generic logging mechnanism
of the test suite and see the result.
Some lvmdbusd test seems to rely on some log/file logic
which is modified with the use of these shell vars.
Require VDO version 6.2.3.
Skip the part of the test that needs vdo wrapper and 2 different
versions of vdoprepareforlvm to prepare shifted VDO header
at the 2MiB offset.
Hunt also for devices with LVMTEST prefix in UUID.
Call teardown_devs_prefixed - so if they hold RAM or SCSI
they are closed before trying to remove kernel modules.
When creating thin pool or check pool there is allocated LV
for metadata and for such allocation user should be able to
specify list of PVs on cmdline.
Also fix unused passed list of PV for thick to thin conversion,
where the code was using whole PV set from a VG (but since it's
been not enabled on cmdline, user could not hit this issue).
Also remove unneeded initialization of use_pvh.
When the import is used on a system, that uses devices file,
the final activation was impossible for the case the converted
volume was not present in devices file.
Currently add volume automatically in such case.
Also add some more debugging output from the script.
TODO: Consider enhnacing lvconvert with extending devices file.
VDO is using specific path for some device paths.
i.e. for /dev/sda it could be /dev/disk/by-id/scsi-xxxxx.
This used to be not a problem before lvm2 started to use snapshot,
but now it needs to replace matching device path.
So switch to the path naming used in vdoconf.yml file.
Add '--headings none|abbrev|full|0|1|2' command line option to select
the heading type.
none|0 - no headings
abbrev|1 - column name abbreviations
full|2 - full column names (column names are equal to exact names
that -o|--options also accepts to set report output)
Reuse existing report/headings config setting to make it possible to
change the type of headings to display:
0 - no headings
1 - column name abbreviations (default and original functionality)
2 - full column names (column names are equal to exact names that
-o|--options also accepts to set report output)
Also, add '--headings none|abbrev|full|0|1|2' command line option
so we are able to select the heading type for each LVM reporting
command directly.
Add new DM_REPORT_OUTPUT_FIELD_IDS_IN_HEADINGS report output flag.
If enabled, column IDs are reported instead of column names in report
headings.
The 'column IDs' are IDs as found in 'const struct dm_report_field_type *fields'
array that is passed during report initialization (that is, a call to
dm_report_init/dm_report_init_with_selection). In this case, the 'id'
dm_report_field_type member is used instead of the 'heading' member.
While create new LV for pool volume, use name from 'pool_metadata%d' naming
sequence. This LV is later on renamed to pool_t/cmeta, but if there
is any error in the middle, we may evenutally leave some 'volume',
With this name it can be slightly more obvious how it got there,
but also when we handle _pmspare name - we get slightly more predictible
name used there for it.
However for a standard usage this commit shall no visible impact as the
name is used temporarily just for cleaning LV.
Set the lock_args string in addition to doing initialization.
lvconvert calls lockd_init_lv_args() directly, skipping
the normal lockd_init_lv() which usually sets lock_args.
lvmlockd locking for swapmetadata adjusted for commit
ac36153e99 lvconvert: preserve UUID for swapped metadata
Now that the LV uuid is swapped between LVs, the lvmlockd lock can
simply be moved between them, and the same lock can continue to be
used for the LV outside of the pool.
Once lvm2 repairs pool's metadata LV and preserves the original metadata LV
with unmodified metadata, for such LV in VG use newly created UUID for new
_pmspare and actually preserve UUID for this hidden _pmspare (if it
exists).
The lockd lock needs to be freed for the LV that is becoming
the new metadata LV, and a new lockd lock needs to be created
for the old metadata LV that is becoming an independent LV.
Fixes b3e45219c2
We incorrectly marked pv_major and pv_minor fields as being of string
type, even though the values were already correctly handled as integers
internally. This confused -S|--select that tried to compare string
values instead of integers.
Reported here: https://github.com/lvmteam/lvm2/issues/122
Previous fix was invalid (after some in-place shuffling)
'dd' copied goes to 'stderr' so we need to catch all output.
Grep needs to check output of tee tool.
Ensure 'C' locales are in use with 'dd'.
Use cachepool name for create name for metadata backup LV.
(so we do not generate 2 'sequences' of metadata filenames.)
Move path preparion before handling _pmspare.
Also drop extra call to sync_local_dev_names() as it's
already got in sync with call of exec_cmd().
When repair thinpool or cachepool, lvm2 leaves original metadata
volume backup. To avoid potential damage of those data, mark such
volume as 'read-only' and also allow user to use --setactivationskip
option for this volume.
TODO: likely better default would be to automatically skip, but
that might need some more thinking about recovery reporting doc.
Replace the use of internal /dev/mapper names with the use of
public LV names /dev/vg/lv for use with repair tools.
For this make the activation of _pmspare LV to be handled as
a component activation with public name.
Metadata is already atomatically activated this way (as readonly).
So if there is any 'error' happening, we leave public LVs in
system.
Pass more args with some 'aux' commands:
wipefs_a, enable_dev, disable_dev
(so it's a bit more efficient using single udev_wait call).
Use prepare_vg instead of prepare_pvs.
Keep using test directory for created files.
Trap errors and remove brd in this case.
Use some shell builtins to reduce fork count.
Use "$VAR".
Run 'pvs' with devlist (so not acceing other system devices).
New dmpd tools return version string in different format,
so update code to understand both variant.
Also hide some shell var setting to local functions.
lvresize --usepolicy requires resized LVs to be active.
(So it's not only required for shared VG).
The test for active pool needs to use lv_info to query 'layer'
otherwise the pool is considered inactive if it was not activated
explicitely - thun 'implicit' activation with VDO or ThinLV was
not managed by --usepolicy option.
Some linkers do need libdevmapper-event-lvm2.so.2.03,
so add also this symlink to the tests /lib dir.
This fixes the need to use previous LD_LIBRARY_PATH complex
setting and now works with much shorter list.
Last commit c38b668fc3 was pushed
with type 'scanf' instead of 'sscanf' needed for buffer reading.
Interestingly this caused scanning from 'stdin' descriptor and
thus failures in lvm shell usage.
The code in init_log_file relies on the process name (COMM) to not
contain whitespaces. This change fixes it by looking up the right-most
parenthesis to safely jump past COMM.
For more context see:
https://www.openwall.com/lists/oss-security/2022/12/21/6
Code is only used with testing, so it should have no impact on regular
users.
Reported-by: Hugues Evrard <hevrard@google.com>
mm
With 3596558861 it's been introduced
a more fine grained description.
However 'disabled' might be actually more confusing then empty field,
so keep only the info about 'not enabled'aka dmevend is not allowed
to monitor LV which otherwise could be monitored.
Update pool conversion function to handle also conversion of
thick LV to thin LV by moving thick LV into thin pool data LV
and creating fully provissioned thin LV on top of this volume.
Reworking existing conversion to use insert_layer_for_lv co
the uuid is now kept with thin-pool - this should however not
really matter as we are doing full deactivation & activation cycle.
With conversion to thin LV user can use same set of arguments
to set chunk-size.
TODO: add some smart code to decide best values for chunks sizes.
For proper functionality of insert_layer_for_lv we need to
move more bits to layerd LV.
Add some missing new types and correct usage of caller,
so the new LV type is set after the movement.
Validate cache origin in front of the prompt.
Also add some rules to command description file.
TODO:
more validation needed also for lvcreate,
more complex rules with "OR" seems to be needed.
Enable support to cache thin/error/zero virtual targets.
Use can now select whether he want to cache whole thin-pool,
or an individual thin volume out of whole thin-pool.
Support using zero and error LVs as data volume for
cachepool is possibly useful for benchmarking, not much
can be expected from such setup.
Avoid activation when going to skip zeroing of 'error' segtype
(so it's not erroring out).
Also skip zeroing for 'zero' segtype LV (already being zero).
When lvm2 calculates the maximal usable COW size and crops the user
requested size to this value, don't return the error result from
the 'lvextend' operation.
We already apply the same logic when resizing thin-pool beyond
the supported maximal size.
FIXME: The return code error logic here is somewhat fuzzy.
Output from vdo manager may actually indent output with spaces,
so trim leading and ending space.
Also add support for verbosity flag for vdo conversion tool.
This vdo parameter existed in the early stage of integration of vdo into lvm2,
but later it's been removed from vdoformat tool - so actually if
there would be any non-zero value it would cause error on lvcreate.
Option was not stored on disk in lvm2 metadata.
Remove this vdo parameter from lvm2 sources.
(Although this vdo parameter will be still accepted on cmdline through
--vdosettings option, but it will be ignored.)
Fix the testing logic.
With raid5 device the layout of files on a filesystem does not define
which leg will actually contain the block we try to damage.
So test will now figure out which device has damaged block.
Use 'check' functionality and also drop unneeded random write as we
now can identify easily in other way.
Compilation may configure it's own /etc path so ensure the test
has a defined location for access to this dir during testing.
Also prepare machine_id filei (with the use of uuidgen tool)
for the test.
Ensure the volume after conversion has the properly aligned size to the
volume group extent size. This would be visible when using virtual size,
that cannot be divided by extent size.
Before the user had to manually adjust the size after conversion to get
access to all data stored on VDO volume.
The command "lvcreate --type thin --snapshot ..." to create a thin
snapshot would fail.
commit d651b340e6 removed the optional
"--type thin" from the command definition "lvcreate --snapshot LV_thin",
and added --type thin as AUTOTYPE. This was correct and should not have
changed anything if all the command defs were correct, but it broke
the "lvcreate --type thin --snapshot" case. It reveals a problem in a
different command definintion: "lvcreate --type thin LV_thin" that was
missing --snapshot in its OO list.
If we want to have const structure - use a const pointer to avoid,
however making individual members const make it only assingle
in construct time and we already violate this logic since
we memcpy into the structure - so drop these unnecessary consts...
Add support for usage of 'dm-snapshot' for doing whole device conversion
in a snapshot which could be merged once the whole conversion has been
made.
This helps with cases where there would be any unexpected failure in the
middle of conversion process and user can continue using original
device until problem in conversion is fixed.
Import tool now uses 'truncate' tool to create a small backend file (20M) for loop device
to hold COW exception store.
Option --vdo-config has been added to allow specifing location of vdo
configuration file.
Option --no-snapshot allows to use import tool without creation of
snapshot, however recovery after finished VDO conversion and unfinished
lvm2 conversion is then very difficult.
Option --uuid-prefix allow to specify DM UUID prefix for snapshot
device.
Use read with -r.
Improve metadata parser to handle volume_geometry bio_offset,
which needs to be substracted from 'region' start_block when present.
This bio_offset block is non-zero i.e. with converted VDO volumes.
Also fix some converted structure value (but they are not in use).
Fix in the code that matches devices to system.devices entries when
the devices have the same serial number. A non-PV device in
system.devices has no pvid value, and the code was segfaulting
when checking the null pvid value.
If we are executing 'report_for_selection' to do an internal report
just for the selection itself (not for display), we can call it more
than once. In that case, we are reusing the same selection handle
(e.g. inside 'process_each_lv_in_vg').
The selection handle has 'report_type' field which is a union of all
report types needed for the report based on selection fields we actually
use.
The 'report_type' is further clarified based on checks and rules inside
'_get_final_report_type' which 'report_for_selection' calls. Then, this
final report type unambiguously identifies proper branch to take in
'report_all_in_{pv,vg,lv}' that is called next.
If the 'report_for_selection' is called more than once with the same
selection handle, we need to make sure that we always restore the report
type to its initial value, so all the rules inside 'report_for_selection'
are applied correctly next time.
This patch fixes the missing restoration of the 'report_type' value in
'selection_handle' that is reused for recurring 'report_for_selection'
calls.
An example scenario where this failed was with selecting an LV for
removal with "lvremove --select" while using a field in the selection
that required extra DM info or DM status call for the LV (that is,
"Logical Volume Device Info Fields" and "Logical Volume Device Status Fields"
as visible in 'lvs -S help').
In previous lvm versions, trailing spaces at the end of a t10 wwid would
be replaced with underscores, so the IDNAME string in system.devices
would look something like "t10.123_". Current versions of lvm ignore
trailing spaces in a t10 wwid, so the IDNAME string used would be
"t10.123". The different values would cause lvm to not recognize a
device in system.devices with the trailing _. Fix this by ignoring
trailing underscores in the IDNAME string from system.devices.
The recent fix 05c2b10c5d ensures that raid LV images are not
using the same devices. This was happening in the lvextend commands
used by this test, so fix the test to use more devices to ensue
redundancy.
vgrename does not support -S|--select, so do not provide a hint about
using it. Instead, provide a hint about using VG uuid directly.
❯ vgs
WARNING: VG name vg1 is used by VGs DXjcSK-gWfu-5gLh-9Kbg-sG49-dtRr-GqXzGL and MVMfyM-sjOa-M2xV-AT4Y-JddR-h4SP-UO5Ttk.
Fix duplicate VG names with vgrename uuid, a device filter, or system IDs.
VG #PV #LV #SN Attr VSize VFree
vg1 1 0 0 wz--n- 124.00m 124.00m
vg1 1 0 0 wz--n- 124.00m 124.00m
(vgrename does not support -S|--select)
❯ vgrename vg1 vg2
WARNING: VG name vg1 is used by VGs DXjcSK-gWfu-5gLh-9Kbg-sG49-dtRr-GqXzGL and MVMfyM-sjOa-M2xV-AT4Y-JddR-h4SP-UO5Ttk.
Fix duplicate VG names with vgrename uuid, a device filter, or system IDs.
Multiple VGs found with the same name: skipping vg1
Use VG uuid in place of the VG name.
(vgchange does support -S|--select)
❯ vgchange --addtag a vg1
WARNING: VG name vg1 is used by VGs DXjcSK-gWfu-5gLh-9Kbg-sG49-dtRr-GqXzGL and MVMfyM-sjOa-M2xV-AT4Y-JddR-h4SP-UO5Ttk.
Fix duplicate VG names with vgrename uuid, a device filter, or system IDs.
Multiple VGs found with the same name: skipping vg1
Use --select vg_uuid=<uuid> in place of the VG name.
In case of e.g. 3 PVs, creating or extending a RaidLV causes SubLV
collocation thus putting segments of diffent rimage (and potentially
larger rmeta) SubLVs onto the same PV. For redundant RaidLVs this'll
compromise redundancy. Fix by detecting such bogus allocation on
lvcreate/lvextend and reject the request.
Check for running (possibly leftover) lvmdbusd running in the
system - as this daemon may interfere with this test as in this
case both be operating on same 'live' data in /run/lvm.
Make test faster by agregating sets of operation to work on a single
created filesystem yet checking all the variants of extension and reduction.
Split 'xfs' part into separate test and convert it for use of the
minimal size 300M nowdays required by mkfs.xfs.
lvreduce uses _lvseg_get_stripes() which was unable to get raid stripe
info with an integrity layer present. This caused lvreduce on a
raid+integrity LV to fail prematurely when checking stripe parameters.
An unhelpful error message about stripe size would be printed.
When lvmcache info is dropped because it's an md component,
then the lvmcache vginfo can also be dropped, but the list
iterator was still using the list head in vginfo, so break
from the loop earlier to avoid it.
Previous commit cause the pvmove could actually be started in unexpected
order - so make sure, we are not starting new pvmove in same VG until
the previous one is started.
Keep backups within test_dir instead of dev_dir (so it doesn't
leak large files there if the tests are run over real /dev dir).
Move restoring of dm_mirror throttling before test_dir is removed.
Not quite sure if this helps anything, some of testing
machines can't reliably remove scsi_debug, reporting
they are in use - but it's not easily reproducible...
aux wait_pvmove_lv_ready() now handles multiple pvmove LVs
at one go - which allows a bit fast checking - although
at some point we may need to switch to use delayed devs
since mirror throttling seems to be no longer working well,
as CPU are getting so fast, that most of data are already
pvmoved before throttling has any chance to do something...
When 'brd' device can be removed (is unused AKA not opened),
remove such device and use again for testing.
Let's assume user has no unused brd device left in the system.
When the 'tests' sometimes fail to cleanup devices, with this
change futher cleanup from some next test may evenually release
brd device and make it available for testing.
There is no easy way to detect, whether device supports zeroing,
and kernel also zeroes device when it's not directly supported,
but with extra message:
operation not supported error, dev X, sector Y op 0x9:(WRITE_ZEROES)...
So to avoid generating such message with every 'lvcreate', use for
zeroing of upto 8K just standard write of zeroed page.
(maybe we can go with even larger sizes).
Convert test to use only ext4 instead of 300M demanding XFS.
Shorten 'B' files to 4K and use 4K strip size with >raid1 arrays
so we do not risk spreading of the file across stripe.
Also use easier 'aux corrupt_dev()' method to introduce a bit
corruption into a block device with integrity.
TODO: shorten _wait_recalc (should't be needed).
Add function to corrupt some bytes in give file path presenting
a device. 1st. patern in just once replaced with 2nd. pattern.
Usable to simulate some bit corruption for integrity devices.
Convert test to use a single skeleton and only different pieces
keep in separate tests.
Lower raid disk usage to smaller size and switch to ext4
as way less demanding fileystem.
Instead of using size of 'empty header' in vdopool use fixed size 4K
for a 'wrappeing' vdo-pool device.
This fixes the issue when user tried to activate vdo-pool after
a conversion from vdo managed device with 'vgchange -ay' - where
this command activated all LVs with 'vdo-pool' wrapping device as well,
but this converted pool uses 0-length header.
This 4k size should usually prevent other tools like 'blkid' recognize
such device as anything - so it shouldn't cause any problems with
duplicate indentification of devices.
Correcting output results from configure script when
printing locking dir.
Shuffle dmfilemapd close to dmeventd option and tracing.
Couple indent changes.
syscall 186 is specific to x86 64bit. As this is different from arch
to arch and between same arch different arch size we will only grab
thread ID using built-in python support if it is supported.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2166931
When the daemon is starting we do an initial fetch of lvm state. If we
happened to get some type of failure with lvm during this time we would
exit. During error injection testing this happened enough that
the unit tests were unable to finish. Add retries to ensure we can get
started during error injection testing.
Previously we were injecting a missing key in the lv, vg, and pv.
Given the order of processing in lvmdbusd, this prevented us from
exercising all the error paths. Change to returning just 1 instead.
When we sort the LVs, we can stumble on a missing key, protect against
this as well.
Seen in error injection testing:
Traceback (most recent call last):
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/fetch.py", line 198, in update_thread
num_changes = load(*_load_args(queued_requests))
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/fetch.py", line 83, in load
rc = MThreadRunner(_main_thread_load, refresh, emit_signal).done()
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/utils.py", line 726, in done
raise self.exception
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/utils.py", line 732, in _run
self.rc = self.f(*self.args)
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/fetch.py", line 40, in _main_thread_load
(lv_changes, remove) = load_lvs(
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/lv.py", line 148, in load_lvs
return common(
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/loader.py", line 37, in common
objects = retrieve(search_keys, cache_refresh=False)
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/lv.py", line 72, in lvs_state_retrieve
lvs = sorted(cfg.db.fetch_lvs(selection), key=get_key)
File "/home/tasleson/projects/lvm2/daemons/lvmdbusd/lv.py", line 35, in get_key
pool = i['pool_lv']
KeyError: 'pool_lv'
Remove old code that became incorrect at some point.
It's probably a fragment of an old condition that was left
behind because it wasn't understood. We don't want to drop
the MISSING_PV flag just because the PV has no mda in use.
The device that was missing may have stale data, so the user
needs to decide if the device should be removed or restored.
Different sequences of steps that could be used to handle raid LVs
after VG takeover (what would happen in cluster failover) combined
with the loss of a disk.
When we are using -S|--select for non-reporting tools while using command log
reporting (log/report_command_log=1 setting), we need to create an internal
processing handle to handle the selection itself. In this case, the internal
processing handle to execute the selection (to process the -S|--select) has
a parent handle (that is processing the actual non-reporting command).
When this parent handle exists, we can't destroy the command log report
in destroy_processing_handle as there's still the parent processing to
finish. The parent processing may still generate logs which need to be
reported in the command log report. If the command log report was
destroyed prematurely together with destroying the internal processing
handle for -S|--select, then any subsequent log request from processing
the actual command (and hence an attermpt to access the command log report)
ended up with a segfault.
See also: https://bugzilla.redhat.com/show_bug.cgi?id=2175220
There is a window of time where the following can occur.
1. An API request is in process to the lvm shell, we have written some
command to the lvm shell and we are blocked on that thread waiting
2. A signal arrives to the daemon which causes us to exit. The signal
handling code path goes directly to the lvm shell and writes
"exit\n". This causes the lvm shell to simply exit.
3. The thread that was waiting for a response gets an EIO as the child
process has exited. This bubbles up a failure.
This is addressed by placing a lock in the lvm shell to prevent
concurrent access to the shell. We also gather additional debug data
when we get an error in the lvm shell read path. This should help if
the lvm shell exits/crashes on its own.
Replace spaces with \040 in directory paths from getmntent (mtab).
The recent commit 5374a44c57 compares mount point directory paths
from /etc/mtab and /proc/mounts, in order to detect when a mounted
LV has been renamed. The directory path comparison does not work
correctly when the path contains spaces because getmntent uses
ascii space chars and proc replaces spaces with \040.
Add some Gentoo based patches for better support of static linking.
This are not tested nor supported by upstream developers.
Usage requires presence of several libraries in their static form
which is however not commonly available.
Selinux modified by zkabelac to still work on older sofrware which
did not provided libselinux.pc at a time - see keep the old check
present and use pkg-config only when possible.
When user enables lockd libraris, code needs to fail, when then are
missing. Also when notify-dbus support if enabled, and libsystemd is
missing, abort configuration.
Check for pkg-config --libs libdlm_lt and test if the returned value
contains word 'pthread' - if so, it's likely a buggy result from
incorrect config file and use directly -ldlm_lt for this case.
There may be some unconventional methods of sharing disks with vgimport
where hints could cause confusion. Hints should be disabled when
sharing disks, or pvscan --cache used to regenerate hints as needed,
but invalidating hints from vgimport might help reduce confusion.
Convert lvmlockd to use configure _LIBS and _CFLAGS for
discovered libraries.
TODO: ATM we ignore discovered libdlm and use libdlm_lt instead.
Also libseagate_ilm is hard to find unicorn for testing.
Convert naming SYSTEMD_CFLAGS/LIB -> LIBSYSTEMD_CFLAGS/LIBS
to better fit library check for libsystemd.
Build lvmlockd with SD_NOTIFY when we have defined LIBSYSTEMD_LIBS.
Coverity is complaining about unchecked strcpy here, which is
irelevant as we preallocate buffer to fit in copied string,
however we could actually reuse these size and use just memcpy().
So lets make some simple conversions.
This was trying to warn about a misconfig where the root PV
was excluded by the devices file, but it wasn't covering all
cases. Also, it's not very useful because the root LV is
usually activated directly in the initrd.
With the recent use of DEVLINKS, there is no longer any real
point in checking the filter for symlink names. Removing
this check should not change behavior with or without symlinks
in the filter.
Commit 7a8b7b4add introducing lockidm
support missed to use AC_SUBST() for a variable and provide it only
in prebuilt configure, so with the next autoreconf this variable
was lost and IDM support was no longer compiled.
Fixes: https://github.com/lvmteam/lvm2/issues/98
Reported-by: Tom Prohofsky
Correct setting of migration_threshold and add clarify
how the user can restore default cache settings for cache policies.
Also document mq aliasing with smq for newer kernels.
When two parallel commands would have tried to create our
/dev/mapper/control node at the same time, one of them could
actually fail even if the 2nd. command actually mknod()
this control node correctly.
So for EEXIST case add detection if the control node is ok.
This may possibly help with some race case in early boot.
We still need to support build without any blkid present,
so use PKG_CHECK_EXISTS() instead of direct failure
from PKG_CHECK_MODULES for too old version.
Allow users to specify their path to systemd-run binary:
configure --with-systemd-run=/my/path/system-run
By defaults it autodetected in $PATH and fallbacks to:
/usr/bin/systemd-run.
Previous commit e67ebc5c40 used incorrect check
for blkid version.
Fix it by always checking for at least version 2.24,
as we can't use older version anyway.
Added BLKID_SUBLKS_FSINFO is present in newer version.
Reduce shellcheck warnings about missing {} for possible array
dereference.
Make sure we are not loosing error code when assigning local vars
and explicitely ignore 'errors' from standalone lines when needed.
Add some missing quotes.
Use $() instead of ancient ``
Avoid writing some temporary data into /tmp - test need to store
files within its own 'testdir' - so it can be properly discarded.
Addressing problem for user of system without default bash shell.
As reported "set -o pipefail" is a bashism that causes the build to
fail when /bin/sh does not point to bash.
For example, this has been observed causing build failures
on Gentoo when /bin/sh is symlinked to /bin/dash.
Rule has been reworked and we started to use .DELETE_ON_ERROR to
ensure that with any errors during file generation, such invalid
file is automatically removed.
Rules were reworked to avoid using pipe and instead use temporary
files when necessary.
Printing lines with echo was replaced with POSIX recomended 'printf'
as proposed by reporter since handling of escape characters and
the "-n" flag for "echo" aren't portable across shells.
Also some build deps has been added.
Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1822280
Gentoo-bug: https://bugs.gentoo.org/682404
Gentoo-bug: https://bugs.gentoo.org/716496
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1822280
Reported-by: Michael Orlitzky <michael@orlitzky.com>
TODO: investage if the temporary files could be handled via some
intermediate target solution - ATM I couldn't make it work equally
well as current solution use shell 'trap' to remove temp file.
Commit dropping lvmetad support 117160b27e
also removed cleaning of its generated files. However we need to keep
this functionality, otherwise we can leak them during various bisect.
Easier is to keep such rules forever.
Also commit c1ab9fb37f moved cmds.h to
include, so again keep it removed so it's not left there and in any
case could not misslead anyone.
Use configure detection instead of trying to directly include
blkid.h inside lvresize.c - where we do not pass in BLKID_CFLAGS
and since we actually do not need to use blkid there, introeduce
test variable HAVE_BLKID_SUBLKS_FSINFO and avoid trying to
use blkid.h in this place - this also fixes builing problem for
systems without blkid.h.
`AS_IF([...])` is more portable, as it respects macro expansions of
`AC_REQUIRE()`.
This is recommended Autoconf best practice, since in nested
conditionals, it is generally unknowable whether some macro invokes
`AC_REQUIRE()` deep down:
https://www.gnu.org/software/autoconf/manual/autoconf-2.71/html_node/Common-Shell-Constructs.html#index-AS_005fIF-1
As a result, the hacky `pkg_config_init` function is not needed
anymore, since any `PKG_*` invocation will ensure that
`PKG_PROG_PKG_CONFIG` will have been called, due to the fact that
`AC_REQUIRE()` will trickle up.
`[[ ... ]]` only works in bash, and not in POSIX sh, specifically
dash fails on this, which is a popular alternative to bash for running
configure scripts.
test's `-a` and `-o` options are considered obsolescent by POSIX,
because they interact badly with expressions and can become ambiguous
depending on string arguments:
https://pubs.opengroup.org/onlinepubs/9699919799/utilities/test.html
`==` only works in bash.
Fix hang of vgimportclone command when:
the PV(s) being imported are not actually clones/duplicates, and
the -n vgname arg is the same as the current vgname.
(Not the intended use of the command, but it should still work.)
In this case, the old and new vgnames ended up being the same, when
the code expected they would be different. A file lock on both the
old and new vgnames is used, so when both flocks are on the same
file, the second blocks indefinitely.
Fix the new vgname to be the old name plus a numeric suffix, which
is the expected result.
Follow-up for e10f67e917.
The commit e10f67e917 tries to keep device
node symlinks even if the device is in the suspended state. However,
necessary properties that may previously obtained by the blkid command
were not imported at least in the .rules file. So, unless ID_FS_xyz
properties are imported by another earlier .rules file, the device node
symlinks are still lost when event is processed in the suspended state.
Let's explicitly import the necessary properties.
RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=2158628
GHPR: https://github.com/lvmteam/lvm2/pull/105
* TODO: Add banner for Important News: Critical Bugs, Important Announcements,...
-->
<!--
* TODO: Add a feed for latest articles/release-notes on the right
-->
<!--
## About LVM2
-->
LVM aka LVM2 refers to the userspace toolset that provide logical volume
management facilities on linux.
<!--
It is reasonably backwards-compatible with the
original LVM1 toolset.
* TODO: Add information about LVM1 metadata format conversion!
-->
LVM offers more flexibility than using partitions, allowing one to
* grow and, where supported by filesystem, shrink volumes,
* create snapshots of existing volumes,
* mirror data on multiple disks including RAID levels 5 or 6,
* striping data on multiple disks,
* create a read or write cache.
To use LVM2 you need 3 things:
* [device-mapper](https://sourceware.org/dm/) in your kernel (upstream since long ago)
* the userspace device-mapper support library (*libdevmapper*) (part of lvm2)
* and the userspace LVM2 tools.
## Getting LVM
Most of linux distribution offer packaged LVM tools.
Depending on your distribution use
# RPM based distributions (Fedora):
yum install lvm2
# DEB based distributions (Debian, Ubuntu):
apt-get install lvm2
Tarballs of the userspace LVM2 source code releases are available from [sourceware.org](https://sourceware.org/pub/lvm2/) [ftp](ftp://sourceware.org/pub/lvm2/).
List of official [mirror sites](https://sourceware.org/mirrors.html) (including http and rsync protocols).
### LVM Releases
[[!inline pages="release-notes/2.03.* and !*/template and !*/Discussion and !tagged(draft) and !tagged(pending)" limit=2 rootpage="release-notes"]]
[[More releases|release-notes/index]]
## Getting Started
<!--
TODO: We are missing a lvm(7) man page explaining this, I think it would be a nic addition!
And perhaps so would be a lvmtroubleshooting(7) guide.
-->
Word of warning first! Even though LVM errs on the side of data safety it is a
tool with low level access and one may seriously harm their data when used
incorrectly!
* Physical Volume (PV) is underlying disk, local or remote, encrypted or even
a mdadm RAID volume. PV is divided into so called Physical Extents (PE) which
are a basic allocation unit.
List PVs using [pvs(8)](https://man7.org/linux/man-pages/man8/pvs.8.html) or
[[!meta title="Version 2.03.24 - Feature and Bug Fix Release"]]
Version 2.03.24
===============
* **Changes in udev rules, new version of systemd is recommended (256).**
* **When using LVs as PVs these are no longer auto activated!**
* Allow creating VDO device for thin data when creating thin pool.
* More devices can be used as external origin for thin snapshot.
* And [[!toggle text="more"]]
[[!toggleable text="""
Features
-----------------------
* Swap properties like hostname, date and time when swapping LVs.
* Allow thin snapshot taken of thin volume from another pool as external origin.
* Allow chaining of external origins.
### Changes to device handling
* **IMPORTANT:** When `devices/scan_lvs` is enabled found LVs are no longer auto
activated, must enable `LVM_PVSCAN_ON_LVS` in udev rules (*69-dm-lvm.rules*.)
* NOTE: It is not recommended to use LVs for PVs. If you have an use case for
this talk to us, please.
* Changes to devices file are now backed up in */etc/lvm/devices/backup/*.
Controlled by `devices/devicesfile_backup_limit` configuration option.
* When `devices/use_devicesfile` is set to `0` existing file is renamed to
*system.devices-unused.YYYYMMDD.HHMMSS* to prevent outdated file to be used
on reenabling.
### Use VDO device for thin pool's data
New option `--pooldatavdo` during lvcreate or lvconvert allows using VDO as backing device.
More options (`--compression {y|n}`, `--deduplication {y|n}`,
`--vdosettings STRING`) to control the options are supported by these commands.
Example of converting a *lvol1* to a thin pool with enabled compression and deduplication:
lvconvert --type thin-pool --pooldatavdo y vg/lvol1
Changes in command line
-----------------------
* Add *--wipesignature* option controlling signature wiping when converting volume to thin-pool.
* Allow *lvcreate --snapshot* without *{-T|--thin}* option when creating snapshot of a thin volume.
* Allow *--raidintegrity{,mode,blocksize}* options with implicit RAID1 (i.e. when *-m1* is used on command line)
./configure options
-------------------
These changes are for packagers:
* Add `--with-default-event-activation` option controlling default for
`global/event_activation`. Default value is unchanged. Useful when creating
builds for testing.
* Add `--with-modulesdir` option.
Changes in udev rules
---------------------
**IMPORTANT:** Rework of udev rules in cooperation with systemd is included in this release.
* `DM_SUSPENDED` and `DM_NOSCAN` are now entirely internal for DM and cannot be used outside of DM rules.
* Upper level rules should consume `DM_UDEV_DISABLE_OTHER_RULES_FLAG` rather than `DM_NOSCAN` and `DM_SUSPENDED`.
Also few more minor improvements:
* Better handling of `DISK_RO` events and suspended devices
"""]]
[[!tag]]
[[!meta date="Thu May 16 12:12:06 2024 +0200"]]
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
