IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
When pvmoving LV - the target for LV is a mirror so the validation
that checked the type is matching was incorrect.
While we need a more generic enhancment of LVS output for pvmoved LVs,
for now at least stop showing internal errors and 'X' symbols in attrs.
The last commit related to this was incomplete:
"Implement lock-override options without locking type"
This is further reworking and reduction of the locking.[ch]
layer which handled all clustering, but is now only used
for file locking. The "locking types" that this layer
implemented were removed previously, leaving only the
standard file locking. (Some cluster-related artifacts
remain to be cleared out later.)
Command options to override or modify locking behavior
are reimplemented here without using the locking types.
Also, deprecated locking_type values are recognized,
and implemented as if one of the equivalent override
options was set.
Options that override file locking are:
. --nolocking disables all file locking.
. --readonly grants read lock requests without actually
taking a file lock, and refuses write lock requests.
. --ignorelockingfailure tries to set up file locks and
uses them normally if possible. When not possible, it
behaves like --readonly, but allows activation.
. --sysinit is the same as ignorelockingfailure.
. global/metadata_read_only acquires actual read file
locks, and refuses write lock requests.
(Some of these options could probably be deprecated
because they were added as workarounds to various
locking_type behaviors that are now deprecated.)
The locking_type setting now has one valid value: 1 which
refers to standard file locking. Configs that contain
deprecated values are recognized and still work in
largely the same way:
. 0 disabled all locking, now implemented like --nolocking
is set. Allow the nolocking option in all commands.
. 1 is the normal file locking setting and is unchanged.
. 2 was for external locking which was not used, and
reverts to normal file locking.
. 3 was for cluster/clvm. This reverts to normal file
locking, and prints messages about lvmlockd.
. 4 was equivalent to readonly, now implemented like
--readonly is set.
. 5 disabled all locking, now implemented like
--nolocking is set.
The options: --nolocking, --readonly, --sysinit
override, or make exceptions to, the normal file locking
behavior. Implement these by just checking for the
options in the file locking path instead of using
special locking types.
Basic LV functions:
activate_lv(), deactivate_lv(),
suspend_lv(), resume_lv()
were routed through the locking infrastruture on the way to:
lv_activate_with_filter(), lv_deactivate(),
lv_suspend_if_active(), lv_resume_if_active()
This commit removes the locking infrastructure from the
middle and calls the later functions directly from the former.
There were a couple of ancillary steps that the locking
infrastructure added along the way which are still included:
- critical section inc/dec during suspend/resume
- checking for active component LVs during activate
The "activation" file lock (serializing activation) has not
been kept because activation commands have been changed to
take the VG file lock exclusively which makes the activation
lock unused and unnecessary.
Four commands lock two VGs at a time:
- vgsplit and vgmerge already have their own logic to
acquire the locks in the correct order.
- vgimportclone and vgrename disable this ordering check.
Different flavors of activate_lv() and lv_is_active()
which are meaningful in a clustered VG can be eliminated
and replaced with whatever that flavor already falls back
to in a local VG.
e.g. lv_is_active_exclusive_locally() is distinct from
lv_is_active() in a clustered VG, but in a local VG they
are equivalent. So, all instances of the variant are
replaced with the basic local equivalent.
For local VGs, the same behavior remains as before.
For shared VGs, lvmlockd was written with the explicit
requirement of local behavior from these functions
(lvmlockd requires locking_type 1), so the behavior
in shared VGs also remains the same.
Remove the io error message from bcache.c since it is not
very useful without the device path.
Make the io error messages from dev_read_bytes/dev_write_bytes
more user friendly.
"lvconvert --type {linear|striped|raid*} ..." on a striped/linear
LV provides convenience interim type to convert to the requested
final layout similar to the given raid* <-> raid* conveninece types.
Whilst on it, add missing raid5_n convenince type from raid5* to raid10.
Resolves: rhbz1439925
Resolves: rhbz1447809
Resolves: rhbz1573255
In this command, lvcreate creates a new LV and then combines
it with an existing cache pool, producing a cache LV. This
command was previously not allowed in in a shared VG.
When the lvmlockd lock is shared, upgrade it to ex
when repair (writing) is needed during vg_read.
Pass the lockd state through additional read-related
functions so the instances of repair scattered through
vg_read can be handled.
(Temporary solution until the ad hoc repairs can be
pulled out of vg_read into a top level, centralized
repair function.)
It's not an error if a command requests the global lock
when it has already acquired it. It shouldn't happen,
but there could be cases we've not found.
We have been warning about duplicate devices (and disabling lvmetad)
immediately when the dup was detected (during label_scan). Move the
warnings (and the disabling) to happen later, after label_scan is
finished.
This lets us avoid an unwanted warning message about duplicates
in the special case were md components are eliminated during the
duplicate device resolution.
This minor patch fixes grammar in a few messages which get
printed to users. It also fixes the same grammar mistake in
several comments.
Signed-off-by: Rick Elrod <relrod@redhat.com>
--
The device-mapper directory now holds a copy of libdm source. At
the moment this code is identical to libdm. Over time code will
migrate out to appropriate places (see doc/refactoring.txt).
The libdm directory still exists, and contains the source for the
libdevmapper shared library, which we will continue to ship (though
not neccessarily update).
All code using libdm should now use the version in device-mapper.
As we start refactoring the code to break dependencies (see doc/refactoring.txt),
I want us to use full paths in the includes (eg, #include "base/data-struct/list.h").
This makes it more obvious when we're breaking abstraction boundaries, eg, including a file in
metadata/ from base/
md devices using an older superblock version have
superblocks at the end of the md device. For commands
that skip reading the end of devices during filtering,
the md component devs will be scanned, and will appear
as duplicate PVs to the original md device. Remove
these md components from the list of unused duplicate
devices, so they are treated as if they had been
ignored during filtering. This avoids the restrictions
that are placed on using PVs with duplicates.
All these functions are now used as utilities,
e.g. for ioctl (not for io), and need to
open/close the device each time they are called.
(Many of the opens can probably be eliminated by
just using the bcache fd for the ioctl.)
with the --labelsector option. We probably don't
need all this code to support any value for this
option; it's unclear how, when, why it would be
used.
Filters are still applied before any device reading or
the label scan, but any filter checks that want to read
the device are skipped and the device is flagged.
After bcache is populated, but before lvm looks for
devices (i.e. before label scan), the filters are
reapplied to the devices that were flagged above.
The filters will then find the data they need in
bcache.
The clvmd saved_vg data is independent from the normal lvm
lvmcache vginfo data, so separate saved_vg from vginfo.
Normal lvm doesn't need to use save_vg at all, and in clvmd,
lvmcache changes on vginfo can be made without worrying
about unwanted effects on saved_vg.
To avoid the chance of freeing a saved vg while another
code path is using it, defer freeing saved vgs until
all the lvmcache content is dropped for the vg.
In case "lvconvert -mN RaidLV" was used on a degraded
raid1 LV, success was returned instead of an error.
Provide message to inform about the need to repair first
before changing number of mirrors and exit with error.
Add new lvconvert-m-raid1-degraded.sh test.
Resolves: rhbz1573960
I don't like having this in a common header because it means you end
up including too much and causing unneccessary dependencies. eg,
lib/misc/lib.h includes libdevmapper.h, internationalisation, and
logging stuff.
There are likely more bits of code that can be removed,
e.g. lvm1/pool-specific bits of code that were identified
using FMT flags.
The vgconvert command can likely be reduced further.
The lvm1-specific config settings should probably have
some other fields set for proper deprecation.
The mixed up vg repair code in vg_read was trying
to repair a vg when vg_read was called by clvmd.
The clvmd daemon isn't supposed to be repairing
or writing a vg.
(This is a temporary workaround; vg repair will soon
be pulled out of vg_read so it can be called in a
controlled way and consolidated instead of spread
around.)
Shift refresh of mirror table right into monitor_dev_for_events().
Use !vg_write_lock_held() to recognize use of lvchange/vgchange.
(this shall change if this would no longer work, but requires
futher some API changes).
With this patch dm mirror table is only refreshed when necassary.
Also update WARNING message about mirror usage without monitoring
and display LV name.
bcache_invalidate() now returns a bool to indicate success. If fails
if the block is currently held, or the block is dirty and writeback
fails.
Added a bunch of unit tests for the invalidate functions.
Fixed some bugs to do with invalidating errored blocks.
In some pvmove tests, clvmd uses the new (precommitted)
saved_vg, but then requests the old saved_vg, and
expects that the new saved_vg be returned instead of
the old. So, when returning the new saved_vg, forget
the old one so we don't return it again.
The filters save information about devices that should
be ignored, so if we need to repeat a scan (unusual,
but happens in clvmd), we need to update the filters.
When clvmd does a full label scan just prior to
calling _vg_read(), pass a new flag into _vg_read
to indicate that the normal rescan of VG devs is
not needed.
After reading a VG, stash it in lvmcache as "saved_vg".
Before reading the VG again, try to use the saved_vg.
The saved_vg is dropped on VG lock operations.
The copy of the VG which clvmd stashes in lvmcache should
not only be used between suspend and resume, but between
sequential LV operations in clvmd, so that clvmd does not
need to reread the VG for each one. Prepare for that by
renaming the stashed VG as "saved_vg".
For reporting commands (pvs,vgs,lvs,pvdisplay,vgdisplay,lvdisplay)
we do not need to repeat the label scan of devices in vg_read if
they all had matching metadata in the initial label scan. The
data read by label scan can just be reused for the vg_read.
This cuts the amount of device i/o in half, from two reads of
each device to one. We have to be careful to avoid repairing
the VG if we've skipped rescanning. (The VG repair code is very
poor, and will be redone soon.)
Don't allow writes in test mode. test mode should be
more sophisticated than just faking writes, and this
should be a last defense for cases where test mode is
not being checked correctly.
Recent changes allow some major simplification of the way
lvmcache works and is used. lvmcache_label_scan is now
called in a controlled fashion at the start of commands,
and not via various unpredictable side effects. Remove
various calls to it from other places. lvmcache_label_scan
should not be called from anywhere during a command, because
it produces an incorrect representation of PVs with no MDAs,
and misclassifies them as orphans. This has been a long
standing problem. The invalid flag and rescanning based on
that is no longer used and removed. The 'force' variation is
no longer needed and removed.
We can't let clvmd keep all scanned devs open,
which prevents them from being removed. So
drop the bcache data (and close fds) affter
doing a label scan.
Also set up bcache before the clvm-specific
vg_read (which needs to rescan the vg's devs
using bcache) and destroy the bcache after.
The error handling code wasn't working, but it
appears that just removing it is what we need.
The doesn't really need any different behavior
related to bcache blocks on an io error, it just
wants to know if there was an error.
In some odd cases (e.g. tests) there are very few devices
which results in creating too few blocks in bcache, so
create bcache with a minimum number of blocks.
Commands using lvmetad will not begin with a proper
label_scan which initializes bcache, but may later
decide they need to scan a set of devs, in which case
they'll need bcache set up at that point.
The improved detection of bad metadata when scanning
(where errors were ignored before) means we now have to
override some errors when forcibly erasing damaged metadata.
Drop an extra label scan in the recovery part
of vg_read. This is a temporary improvement
until the pending replacement for the broken
recovery code burried in vg_read.
This is a temporary hacky workaround to the problem of
reads going through bcache and writes not using bcache.
The write path wants to read parts of data that it is
incrementally writing to disk, but the reads (using
bcache) don't work because the writes are not in the
bcache. For now, add a dev to bcache before each attempt
to read it in case it's being used on the write path.
Create a new dev->bcache_fd that the scanning code owns
and is in charge of opening/closing. This prevents other
parts of lvm code (which do various open/close) from
interfering with the bcache fd. A number of dev_open
and dev_close are removed from the reading path since
the read path now uses the bcache.
With that in place, open(O_EXCL) for pvcreate/pvremove
can then be fixed. That wouldn't work previously because
of other open fds.
The copy of VG metadata stored in lvmcache was not being used
in general. It pretended to be a generic VG metadata cache,
but was not being used except for clvmd activation. There
it was used to avoid reading from disk while devices were
suspended, i.e. in resume.
This removes the code that attempted to make this look
like a generic metadata cache, and replaces with with
something narrowly targetted to what it's actually used for.
This is a way of passing the VG from suspend to resume in
clvmd. Since in the case of clvmd one caller can't simply
pass the same VG to both suspend and resume, suspend needs
to stash the VG somewhere that resume can grab it from.
(resume doesn't want to read it from disk since devices
are suspended.) The lvmcache vginfo struct is used as a
convenient place to stash the VG to pass it from suspend
to resume, even though it isn't related to the lvmcache
or vginfo. These suspended_vg* vginfo fields should
not be used or touched anywhere else, they are only to
be used for passing the VG data from suspend to resume
in clvmd. The VG data being passed between suspend and
resume is never modified, and will only exist in the
brief period between suspend and resume in clvmd.
suspend has both old (current) and new (precommitted)
copies of the VG metadata. It stashes both of these in
the vginfo prior to suspending devices. When vg_commit
is successful, it sets a flag in vginfo as before,
signaling the transition from old to new metadata.
resume grabs the VG stashed by suspend. If the vg_commit
happened, it grabs the new VG, and if the vg_commit didn't
happen it grabs the old VG. The VG is then used to resume
LVs.
This isolates clvmd-specific code and usage from the
normal lvm vg_read code, making the code simpler and
the behavior easier to verify.
Sequence of operations:
- lv_suspend() has both vg_old and vg_new
and stashes a copy of each onto the vginfo:
lvmcache_save_suspended_vg(vg_old);
lvmcache_save_suspended_vg(vg_new);
- vg_commit() happens, which causes all clvmd
instances to call lvmcache_commit_metadata(vg).
A flag is set in the vginfo indicating the
transition from the old to new VG:
vginfo->suspended_vg_committed = 1;
- lv_resume() needs either vg_old or vg_new
to use in resuming LVs. It doesn't want to
read the VG from disk since devices are
suspended, so it gets the VG stashed by
lv_suspend:
vg = lvmcache_get_suspended_vg(vgid);
If the vg_commit did not happen, suspended_vg_committed
will not be set, and in this case, lvmcache_get_suspended_vg()
will return the old VG instead of the new VG, and it will
resume LVs based on the old metadata.
When process_each_pv() calls vg_read() on the orphan VG, the
internal implementation was doing an unnecessary
lvmcache_label_scan() and two unnecessary label_read() calls
on each orphan. Some of those unnecessary label scans/reads
would sometimes be skipped due to caching, but the code was
always doing at least one unnecessary read on each orphan.
The common format_text case was also unecessarily calling into
the format-specific pv_read() function which actually did nothing.
By analyzing each case in which vg_read() was being called on
the orphan VG, we can say that all of the label scans/reads
in vg_read_orphans are unnecessary:
1. reporting commands: the information saved in lvmcache by
the original label scan can be reported. There is no advantage
to repeating the label scan on the orphans a second time before
reporting it.
2. pvcreate/vgcreate/vgextend: these all share a common
implementation in pvcreate_each_device(). That function
already rescans labels after acquiring the orphan VG lock,
which ensures that the command is using valid lvmcache
information.
When lvmlockd indicates that the lvmetad cache is out of
date because of changes by another node, lvmetad_pvscan_vg()
rescans the devices in the VG to update lvmetad. Use the
new label_scan in this function to use the common code and
take advantage of the new aio and reduced reads.
This fixes the use of lvmcache_label_rescan_vg() in the previous
commit for the special case of independent metadata areas.
label scan is about discovering VG name to device associations
using information from disks, but devices in VGs with
independent metadata areas have no information on disk, so
the label scan does nothing for these VGs/devices.
With independent metadata areas, only the VG metadata found
in files is used. This metadata is found and read in
vg_read in the processing phase.
lvmcache_label_rescan_vg() drops lvmcache info for the VG devices
before repeating the label scan on them. In the case of
independent metadata areas, there is no metadata on devices, so the
label scan of the devices will find nothing, so will not recreate
the necessary vginfo/info data in lvmcache for the VG. Fix this
by setting a flag in the lvmcache vginfo struct indicating that
the VG uses independent metadata areas, and label rescanning should
be skipped.
In the case of independent metadata areas, it is the metadata
processing in the vg_read phase that sets up the lvmcache
vginfo/info information, and label scan has no role.
Move the location of scans to make it clearer and avoid
unnecessary repeated scanning. There should be one scan
at the start of a command which is then used through the
rest of command processing.
Previously, the initial label scan was called as a side effect
from various utility functions. This would lead to it being called
unnecessarily. It is an expensive operation, and should only be
called when necessary. Also, this is a primary step in the
function of the command, and as such it should be called prominently
at the top level of command processing, not as a hidden side effect
of a utility function. lvm knows exactly where and when the
label scan needs to be done. Because of this, move the label scan
calls from the internal functions to the top level of processing.
Other specific instances of lvmcache_label_scan() are still called
unnecessarily or unclearly by specific commands that do not use
the common process_each functions. These will be improved in
future commits.
During the processing phase, rescanning labels for devices in a VG
needs to be done after the VG lock is acquired in case things have
changed since the initial label scan. This was being done by way
of rescanning devices that had the INVALID flag set in lvmcache.
This usually approximated the right set of devices, but it was not
exact, and obfuscated the real requirement. Correct this by using
a new function that rescans the devices in the VG:
lvmcache_label_rescan_vg().
Apart from being inexact, the rescanning was extremely well hidden.
_vg_read() would call ->create_instance(), _text_create_text_instance(),
_create_vg_text_instance() which would call lvmcache_label_scan()
which would call _scan_invalid() which repeats the label scan on
devices flagged INVALID. lvmcache_label_rescan_vg() is now called
prominently by _vg_read() directly.
To do label scanning, lvm code calls lvmcache_label_scan().
Change lvmcache_label_scan() to use the new label_scan()
based on bcache.
Also add lvmcache_label_rescan_vg() which calls the new
label_scan_devs() which does label scanning on only the
specified devices. This is for a subsequent commit and
is not yet used.
New label_scan function populates bcache for each device
on the system.
The two read paths are updated to get data from bcache.
The bcache is not yet used for writing. bcache blocks
for a device are invalidated when the device is written.
When user configured lvm2 to NOT user monitoring, activated mirror
actually hang upon error and it's quite unusable moment.
So instead Warn those 'brave' non-monitoring users about possible
problem and activation mirror without blocking error handling.
This also makes it a bit simpler for test suite to handle trouble
cases when test is running without dmeventd.
When adjusting region size for clustered VG it always needs to fit
2 full bitset into 1MB due to old limits of CPG.
This is relatively big amount of bits, but we have still limitation
for region size to fit into 32bits (0x8000000).
So for too big mirrors this operation needs to fail - so whenever
function returns now 0, it means we can't find matching region_size.
Since return 0 is now 'error' we need to also pass proper region_size
when creating pvmove mirror.
Since extent_size is no longer power_of_2 this max region size
evalution was rather producing random bitsize as a combination
of lowest bit from number of extents and extent size itself.
Correct calculation to use whole LV size and pick biggest
possible power of 2 value smaller then UINT32_MAX.
Drop mirrored mirror log limitation that applies only in very limited
use-case and actually mirrored mirror log is deprecated anyway.
So 'disk' mirror log is selecting the correct minimal size, and
bigger size is only enforced with real mirrored mirror log.
Also for mirrored mirror log we let use 'smalled' region size if needed
so if user uses 1G region size, we still keep small mirror log
with much smaller region size in this case when needed.
Also mirror log extent calculation is now properly detecting error
with too big mirrors where previosly trimmed uint32_t was applies
unintentionally.
Whenever we make visible LV out of previously invisible one,
reload it's table - the is mandator for proper udev rule
processing as well as ensure content of dm table is correct.
TODO: this new generic rule probably make extra raid rules unnecessary.
Only policy 'smq' is meant to be used with format version 2.
Code used to let pass 'mq' policy also with format 2. But 'mq'
is obsoloted wth smq and kernel currently matches it. But this
is incompatible with older original mq logic - so disallow creation
of this rather useless combination.
If the tools for checking thin_pool or cache metadata are missing,
issue rather just a WARNING, but let the operation of activation
continue.
This has the advantage, the if user is missing those tools,
but he already started to use thinpool or cacheing, he can
access these volumes with a WARNING.
Also if the user is using too old tools i.e. for CacheV2 format
dmpd tool 0.7 is required - provide informative WARNING and
skip failure from older tool version which can't understand
new format V2.
In case a newly created RaidLV is blacklisted using config
\"activation { volume list = [ ... ] }\" (i.e. its SubLVs stay inactive),
the metadata SubLVs can't get wiped thus failing the creation.
As a result, the RaidLV together with its SubLVs
is left behind in an inconsistent state.
Fix by removing the RaidLV and provide a hint about volume_list reasoning.
Resolves: rhbz1161347
While prioritized_section() based on raised priority works
nicely for standard lvm comman - separate counter is actually needed
when it's used in daemons like clvmd/dmeventd where priority
stays raised all the time.
Detect we are in prioritezed section instead of critical one,
since these operation were supposed to NOT be happining during
whole set of operation.
This patch fixes verification of udev operations.
Introduce prioritized_section() as a closer match to previous logic
of critical_section() that has been held over longer sequence of
ioctl commands - essentially it's matching operation on a single
cookie.
While 'critical_section()' now corresponds to locked memory - we hold
this memory only between suspend/resume thus notion of 'cookie' was
lost.
This patch restores some logic unintentionaly lost with dropping
memory locking for just activation/deactivation calls.
With these read errors it's useful to know the reason.
Also avoid to log error just once so we know exactly
how many times we did failing read.
On the other hand reduce repeated log_error() on code 'backtrace'
path and change severity of message to just log_debug() so the
actual read error is printed once for one read.
Just like lvm2 has internal devices like _tdata which is using UUID with
suffix, there is similar private type of device for crypto device where
they are using CRYPT-TEMP uuid prefix.
Also ignore stratis.
Some kernel version suffer from bad state transition where a device
steps into 'frozen' mode. Any application that tries to read such
raid gets unfortunatelly bloked.
As some sort of protection try to skip such raid device from being
scanned to minimize chances to block lvm2 command on such scan.
When such device is found, warning gets printed.
RaidLVs on read_only_volume_list have their SubLVs
activated readonly thus disabling metadata updates
or image resynchronization/recovery. Bug also causes
automatic repairs to fail.
Fix by always activating the RAID SubLVs readwrite.
Resolves: rhbz1208269
Just like with lvcreate, this lvconvert case also need to properly
check which LV actually holds lock for cached origin - as it might
be i.e. thin-pool tdata subLV.
When snapshot is created in read-only mode with 'lvcreate -s -pr...',
lvm2 still needs to be able to write to layered -cow volume
to store metadata and exceptions blocks.
TODO: in some case we might be able to do full tree with read-only
volume but this probably needs futher validation:
1. checking snapshot header already exist
2. origin & snapshot are both in read-only mode.
Occasionaly users may need to peek into 'component devices.
Normally lvm2 does not let users activation component.
This patch adds special mode where user can activate
component LV in a 'read-only' mode i.e.:
lvchange -ay vg/pool_tdata
All devices can be deactivated with:
lvchange -an vg | vgchange -an....
If componet devices could be activated alone, ensure they are not breaking
common commands.
TODO: mostly likely this is not a definite list of all needed checks
and more will come later.
This is the 'last' place where a LV is present in metadata.
Any removed device should not be left active in dm table.
So this check is an extra validation protection to capture any
forgotten deactivation (adding 1 extra ioctl into lvremove path)
Introduce:
lv_is_component() check is LV is actually a component device.
lv_component_is_active() checking if any component device is active.
lv_holder_is_active() is any component holding device is active.
Instead of checking with existing size of external origin LV,
use correctly the new 'wanted' size of this LV whether it fits
the limitiation requirements for older thin-pool target.
Otherwise code started to the the resize, updates metadata and
just fails during 'resize' in case the LV was active. For
inactive LV operation could have actually passed.
Checking here for cache_pool is not necessary and in effect
the check is not even right - since there are internal
states that do allow to active such LV.
Fix missing 'externalLV' traversing for thins with external origins.
Replace extra for_each_sub_lv_except_pools() with better
internal logic allowing selectively to cut of processed subLV tree.
Extend error code for function 'fn()' when it returns -1 it will
stop futher tree scan for given LV.
Also a bit simplify code to have only one place that
is calling 'fn()' and use level counter to know
depth of traversing.
Update renaming travering to skip trees for pools
and external origins.
While 'file-locking' code always dropped cached VG before
lock was taken - other locking types actually missed this.
So while the cache dropping has been implement for i.e. clvmd,
actually running command in cluster keept using cache even
when the lock has been i.e. dropped and taken again.
This rather 'hard-to-hit' error was noticable in some
tests running in cluster where content of PV has been
changed (metadata-balance.sh)
Fix the code by moving cache dropping directly lock_vol() function.
TODO: it's kind of strange we should ever need drop_cached_metadata()
used in several places - this all should happen automatically
this some futher thinking here is likely needed.
So this is a bit more complex and possibly worth futher checking.
ATM clvmd drops cmd->mem mempool AFTER refresh of cmd.
So anything allocating from cmd->mem during toolcontext init
will likely die at some point in time.
As a quick fix - just use regular malloc/free for 'dso' alloction.
It's worth to note - cmd->libmem seems to be often misused
causing hidden memleaking for clvmd.
Build dso plugin name during segtype initialisation and just
use the string during command life-time.
Also slightlt update message verbosity and make it very_verbose
when operation is going to be made and 'verbose' when it's done.
Avoid using same return code for reporting 2 different things
and stricly report error code by return value and add new
parameter for reporting monitoring status.
This makes easier to recognize which error we got from dm_event
and continue only with ENOENT.
With pthreaded daemons like 'dmeventd' using liblvm via plugin,
lvm2 actually should not 'play' with streams at all - as there
could be parallel outputs running.
As a current quick workaround just disable change for pthreaded
program (gettid() != getpid()).
TODO: it's possible the change of buffering actually doesn't serve us
any measurable benefit and could be dropped as whole later...
Meanwhile this patch is fixing this occasional valgrind race report:
Invalid read of size 4
at 0x571892C: vfprintf (in /usr/lib64/libc-2.26.9000.so)
by 0x57216B3: fprintf (in /usr/lib64/libc-2.26.9000.so)
by 0x5042886: dm_event_log (libdevmapper-event.c:925)
by 0x10B015: _dmeventd_log (dmeventd.c:125)
by 0x10D289: _unregister_for_event (dmeventd.c:1146)
by 0x10E52E: _handle_request (dmeventd.c:1583)
by 0x10E6D7: _do_process_request (dmeventd.c:1631)
by 0x10E7C6: _process_request (dmeventd.c:1660)
by 0x1101A4: main (dmeventd.c:2285)
Address 0x6264d30 is 192 bytes inside a block of size 552 free'd
at 0x4C2ED68: free (vg_replace_malloc.c:530)
by 0x573907D: fclose@@GLIBC_2.2.5 (in /usr/lib64/libc-2.26.9000.so)
by 0x6AC5C00: reopen_standard_stream (log.c:189)
by 0x6A8E62C: destroy_toolcontext (toolcontext.c:2271)
by 0x6BA5C22: lvm_fin (lvmcmdline.c:3339)
by 0x6BD5EF3: lvm2_exit (lvmcmdlib.c:123)
by 0x6856013: dmeventd_lvm2_exit (dmeventd_lvm.c:103)
by 0x66535B8: unregister_device (dmeventd_thin.c:432)
by 0x10CBBC: _do_unregister_device (dmeventd.c:926)
by 0x10CD74: _monitor_unregister (dmeventd.c:979)
by 0x10D094: _monitor_thread (dmeventd.c:1066)
by 0x54B35E0: start_thread (in /usr/lib64/libpthread-2.26.9000.so)
by 0x57C30EE: clone (in /usr/lib64/libc-2.26.9000.so)
Block was alloc'd at
at 0x4C2DBBB: malloc (vg_replace_malloc.c:299)
by 0x573932B: fdopen@@GLIBC_2.2.5 (in /usr/lib64/libc-2.26.9000.so)
by 0x6AC5DC2: reopen_standard_stream (log.c:200)
by 0x6A8D11D: create_toolcontext (toolcontext.c:1898)
by 0x6BA5B6B: init_lvm (lvmcmdline.c:3319)
by 0x6BD5BC8: cmdlib_lvm2_init (lvmcmdlib.c:34)
by 0x6BD5F04: lvm2_init (lvm2cmd.c:20)
by 0x6855EA7: dmeventd_lvm2_init (dmeventd_lvm.c:67)
by 0x665305F: register_device (dmeventd_thin.c:352)
by 0x10CB7A: _do_register_device (dmeventd.c:916)
by 0x10CEE4: _monitor_thread (dmeventd.c:1006)
by 0x54B35E0: start_thread (in /usr/lib64/libpthread-2.26.9000.so)
by 0x57C30EE: clone (in /usr/lib64/libc-2.26.9000.so)
....
Process terminating with default action of signal 6 (SIGABRT): dumping core
at 0x570016B: raise (in /usr/lib64/libc-2.26.9000.so)
by 0x5701520: abort (in /usr/lib64/libc-2.26.9000.so)
by 0x57437D8: __libc_message (in /usr/lib64/libc-2.26.9000.so)
by 0x5743831: __libc_fatal (in /usr/lib64/libc-2.26.9000.so)
by 0x5744056: _IO_vtable_check (in /usr/lib64/libc-2.26.9000.so)
by 0x574751C: __overflow (in /usr/lib64/libc-2.26.9000.so)
by 0x574191A: fputc (in /usr/lib64/libc-2.26.9000.so)
by 0x50428E3: dm_event_log (libdevmapper-event.c:934)
by 0x10B015: _dmeventd_log (dmeventd.c:125)
by 0x10D289: _unregister_for_event (dmeventd.c:1146)
by 0x10E52E: _handle_request (dmeventd.c:1583)
by 0x10E6D7: _do_process_request (dmeventd.c:1631)
by 0x10E7C6: _process_request (dmeventd.c:1660)
by 0x1101A4: main (dmeventd.c:2285)
In fact pvmove does support 'clustered-core' target for clustered
pvmove of LVs activated on multiple nodes.
This patch restores support for activation of pvmove on all nodes
for LVs that are also activate on all nodes.
Actually the removed code is necessary - since not all writes are
getting alligned buffer - older compilers seems to be not able
to create 4K aligned buffers on stack - this the aligning code still
need to be present for write path.
Add protectional internall error whenever we spot activation
of 'exclusive' only segments in 'non-exclusive' mode.
TODO: possibly the activation locking could be enhanced to handle
this fully behind the scene - as for now this works purely for
lvchange/vgchange activation.
Use properly exclusive activation when reactivating origin after
snapshot merge (since origin must have been previously also exlusively
activated).
Same applies when converting volumes to thin-pool or cache.
Previously used 'only' local activation incorrectly allowed local
activation of some targets (i.e. raid) - thus 'leaking' chance to
activate same device on another node - which can be a problem
for device types like raid.
No longer use the external 'result' pointer internally to set up the
cached label. The callback _set_label_read_result() is now given the
internal label pointer directly
Callers that don't need the result are no longer required to pass a
label pointer into label_read().
If the data being requested is present in last_[extra_]devbuf,
return that directly instead of reading it from disk again.
Typical LVM2 access patterns request data within two adjacent 4k blocks
so we eliminate some read() system calls by always reading at least 8k.
Callers that read larger amounts of data now get a pointer to read-only
data directly without copying it through an intermediate buffer. This
data is owned by the device layer so the callers no longer free it.
If it obtains the data, it passes it into the supplied callback function
and returns 1. Otherwise the callback receives failed = 1.
Updated config_file_read_fd to use this and similarly return the data
via a callback fn of its own.
Dedicated functions are now used to process each piece of data obtained,
so the refactoring in this file gives us one for the vgsummary and one
for the metadata header. This new type of function takes two parameters
(for now), the obtained data plus a single struct (that must not
reference any data on the stack) that wraps up the entire context needed
to process it.
Rename dev_read() to dev_read_buf() - the function that reads data
into a supplied buffer.
Introduce a new dev_read() that allocates the buffer it returns and
switch the important users over to this. No caller may change the
returned data. (For now, callers are responsible for freeing it after
use, but later the device layer will take full ownership.)
dev_read_buf() should only be used for tiny buffers or unimportant code
(such as the old disk formats).
The creation of wrapped around metadata - where the start of metadata is
written up to the end of the buffer and the remainder follows back at
the start of the buffer - is now restricted to cases where writing the
metadata in one piece wouldn't fit. This shouldn't happen in 'normal'
usage so let's begin treating the code for this as a special case that
can be ignored when optimising 'normal' cases.
If there is sufficient space in the metadata area, align the next
metadata to a disk offset that is a multiple of 4096 bytes and
don't write it circularly. If it doesn't all fit at the end
of the metadata area, go back to the start and write it all there
contiguously.
If there is insufficient space to use the new stricter rules, revert to
the original behaviour, aligning on 512-byte boundaries wrapping around
the circular buffer as required.
Even after writing some metadata encountered problems, some commands
continue (rightly or wrongly) and attempt to make further changes.
Once an mda is marked MDA_FAILED, don't try to use it again.
This also applies when reverting, where one loop already skips
failed mdas but the other doesn't.
This fixes some device open_count warnings on relevant failure paths.
Use new ALIGN_ABSOLUTE macro when calculating the start location
of new metadata and adjust the end of buffer detection so that
there is no longer an imposed gap between old and new metadata.
Currently both start and offset should always be divisible by alignment,
so this should have no effect, but a later patch will increase alignment
so these variables can no longer be optimised out.
Although it doesn't look like it can be a measurable problem
and costs some time to flip priorities outside of activation window.
So just like with memory locking preserve priority until call
memlock_unlock() appears.
(addition to commit c086dfadc3).
Expand out the metadata wrapping calculations to prepare
to support a larger alignment.
The current alignment is 512 bytes so
(mdac_area_start + rlocn->offset) % alignment is zero.
Mark the first metadata area on each text format PV as MDA_PRIMARY.
Pass this information down to the device layer so that when
there are two metadata areas on a block device, we can easily
distinguish two independent streams of I/O.
In case of failed legs, raid replaces those with
e.g. "vg-lv_rimage_0-missing_0_0" mapped to an error target.
Those errouneously remain on deactivation.
Fix by removing them on deactivation/removal of the RaidLV.
Introduce enum dev_io_reason to categorise block device I/O
in debug messages so it's obvious what it is for.
DEV_IO_SIGNATURES /* Scanning device signatures */
DEV_IO_LABEL /* LVM PV disk label */
DEV_IO_MDA_HEADER /* Text format metadata area header */
DEV_IO_MDA_CONTENT /* Text format metadata area content */
DEV_IO_FMT1 /* Original LVM1 metadata format */
DEV_IO_POOL /* Pool metadata format */
DEV_IO_LV /* Content written to an LV */
DEV_IO_LOG /* Logging messages */
If the recovery of the repleced leg(s) of a RaidLV created without
initial resynchronization (i.e. "lvcreate --nosync ...") got
interrupted, it can't be extended because of the < 100% sync rate.
In case caller passes in changed stripe size when reshaping raid4/5
to 1 stripe aiming to convert to raid1 and optionally to linear,
ignore it to prevent data corruption.
Use new 3rd. state of trace_pvmove_deps == 2.
In this state we know, we have already seen the node and can skip futher
testing. Remainging value 1 signals we want to track, and value 0
is for ignoring tracking, but node is still checking in this case.
Reduces large amount of duplicate ioctl queries.
Check also all snapshosts when resume is requested,
the origin volume is already resume, but possibly
some subLV or snapshot LV could be suspended if
we are still in critical_section.
When entering any critical section, lvm2 used to lock process memory
and raised task priority to avoid problem with page swapping and minimize
time of having non-resumed devices in table.
With this patch, memory locking which which is expensive is only used when
entering 'suspending' section as only in this section there is risk
lvm could be suspending a device which later can be needed for paging.
Raised priority is still kept for all section entrances as this is
low-cost operation and may accelerate table resumes - although the real
impact can be still considered later.
When pvmove is finished and metadata are updated, the code missed
to merge possible mergable segments - so add explicit merging
call after pvmoved volumes are unlocked.
This avoids weird results where i.e. lvs could have been reporting
non-matching segments as lvs upon metadata read is doing silent segment
merging while dm table left after pvmove was still preserving
non-merged segments.
ATM we want to support delayed resume purely in pvmove case.
So have libdm logic internal to recognize difference beween
pvmove and other targets that do use delayed resume.
This fixes problem introduced with commit aa68b898ff
for mirror-on-mirror or snapshot-on-mirror problem.
TODO: likely added new API call and let libdm user select
delayed nodes explicitely.
Use code which detectes handlers in a way, which is more
backward-compatible friendly.
Replace read of 'sysfs' uuid entry with dm ioctl call.
Use /sys/block/dm-X/holders path instead of
new path /sys/dev/block/major:minor/holders.
TODO:
There are few more occurencies of this logic around the code
so some abstract interface should be considered.
In some cases the message could be slightly misleading so use
here rather conditional.
TODO:
In future we may possibly further tune the message in case we are
certain the level of redundancy protection has not been reduced.
When pvmove is finished and does 'suspend/resume' on PVMOVE LV,
on resume path committed metadata are already showing 'standalone'
pvmove LV prepared just for removal.
However code should be able to 'resume' preloaded LV there were
participating in pvmove operation.
Previously this was all done in the 'tools' part of lvm2 code.
So the lvconvert upon pvmove finish had to explicitely call 'resume' on every such LV.
Now 'smarted' activation code is able to deduce and combine all information from
the active dm table and committed metadata so single call resolves
it all in one go.
Internally holders are detected by reading sysfs directory to capture
all needed UUID which are then looked in lvm2 metadata and all such
LVs are automatically collected into dmtree.
Replace complex code with standard lv_update_and_reload_origin().
Extra suspend should not be necessary.
(If they would be - dependency tree would have bug for fixing).
Only thin-pool with origin_only suspend is allowed to be not suspending anything.
In such case pairing resume will 'decrement' critical section counter.
Just like suspend handles preload for pvmove finish,
in similar way handle suspend of starting pvmove.
In this case the precommited metadata are checked for list of PVMOVEed
LVs and those are suspended in with committed metadata.
There is no need to differentiation between clustered VG and normal VG.
As the activation depends on locking type.
Use unconditionally locally exclusive activation for pvmove.
Whenever pvmove tree is going to be generated for suspend
and such LV has a user - use this 'using LV' to generate
correct dm tree holding all components.
LV is asked for resume, and its already resume and tool
is inside 'critical_section()' check if there is any suspended sub LV.
In that case 'resume' operation will not be skipped.
When activation of LVs fails prior pvmove start, try to deactivate
already activated LVs.
TODO: possibly remember which LVs where already activate and only those
take down - devices which are already in-use will stay active.
Only lv_committed() now uses vg->vg_committed and it appears redundant
if its contents match the enclosing VG so don't waste cycles creating it
when that's known to be true when no write lock is held so the struct
won't get modified.
- Use 'lvmcache' consistently instead of 'metadata cache'
- Always use 5 characters for source line number
- Remember to convert uuids into printable form
- Use <no name> rather than (null) when VG has no name.
If the suspend/resume sequence would leave some device in suspend
for possible later resume, backup cannot be takes (fs holding backups
could be still frozen in critical section())
Move check for presence of raid4 into the right place
so there is no way how to hit activation of any LV
with raid4 on kernel which does not support it.
Commit 763db8aab0 rejects 2-legged
conversions to striped/raid0 but different messages are displayed
for raid0 or striped. This commit provides the same rejection messages.
raid4/5 LVs may only be converted to striped or raid0/raid0_meta
in case they have at least 3 legs. 2-legged raid4/5 are a result
of either converting a raid1 to raid4/5 (takeover) or converting
a raid4/5 with more than 2 legs to raid1 with 2 legs (reshape).
The raid4/5 personalities map those as raid1,
thus reject conversion to striped/raid0.
Resolves: rhbz1511047
Since vg_validate() now rejects LVs without segments and
insert_layer_for_segments_on_pv() gets just created
'layer_lv' without segment, it needs to be hidden
from vg->lvs during processing of _align_segment_boundary_to_pe_range()
as this function calls lv_validate() and now requires
vg to be consistent. LV is then put back into vg->lvs.
Since 4fa5add6b1 ("pvcreate: Wipe cached
bootloaderarea when wiping label.") label_remove is responsible
for the lvmcache_del. (toollib and liblvm need fixing to share
the code.)
