Fabian Ebner [Tue, 26 Jan 2021 11:45:23 +0000 (12:45 +0100)]
Diskmanage: introduce usage helper
Note that this is a slight behavior change, because now the first
partition's usage which is not simply 'partition' will become the disk's
usage. Previously, if any partition was 'mounted', it would become the disk's
usage, then 'LVM', 'ZFS', etc.
A partitions usage defaults to 'partition' if nothing more specific can be
found, and is never treated as unused for now.
Fabian Ebner [Tue, 26 Jan 2021 11:45:19 +0000 (12:45 +0100)]
Diskmanage: refactor and rename get_parttype_info
in preparation to also query the file system type from lsblk. Note that the
result now also includes devices without a parttype, so a definedness check in
get_devices_by_partuuid is needed. This will be useful when the whole device
contains a filesystem.
Fabian Ebner [Tue, 26 Jan 2021 11:45:17 +0000 (12:45 +0100)]
Disks: return correct journal disk candidates
Previously any GPT initialized disk without an osdid (i.e. equal to -1) would
be included in the list of journal disk candidates, for example a ZFS disk. But
the OSD creation API call will fail for those. To fix it, re-use the condition
from the corresponding check in that API call (in PVE/API2/Ceph/OSD.pm).
Now, included disks are unused disks, those with usage 'partitions' and GPT, and
those with usage 'LVM'.
Fabian Ebner [Wed, 27 Jan 2021 12:57:36 +0000 (13:57 +0100)]
mark PBS storages as shared
Like this, the property will get added when parsing the storage configuration
and PBS storages will correctly show up as shared storages in API results.
AFAICT the only affected PBS operation is free_image via vdisk_free, which will
now be protected by a cluster-wide lock, and that shouldn't hurt.
Another issue this fixes, which is the reason this patch exists, was reported
in the forum[0]. The free space from PBS storages was counted once for each node
that had access to the storage.
Alwin Antreich [Wed, 16 Dec 2020 11:59:04 +0000 (12:59 +0100)]
fix: check connection for nfs v4 only server
the check_connection is done by querying the exports of the nfs server
in question. With nfs v4 those exports aren't listed anymore since nfs
v4 employs a pseudo-filesystem starting from root (/).
rpcinfo allows to query the existence of an nfs v4 service.
Dominik Csapak [Thu, 21 Jan 2021 15:47:59 +0000 (16:47 +0100)]
add workaround for zfs rollback bug
as described in the zfs bug https://github.com/openzfs/zfs/issues/10931
the kernel keeps around cached data from mmaps after a rollback, thus
having invalid data in files that were allegedly rolled back
to workaround this (until a real fix comes along), we unmount the subvol,
invalidating the kernel cache anyway
the dataset gets mounted on the next 'activate_volume' again
Fabian Ebner [Tue, 15 Dec 2020 10:59:29 +0000 (11:59 +0100)]
fix #3199: by fixing usage of strftime
In a very early version I wanted to parse the date from the backup
name, and when switching to using the ctime and localtime() instead,
I forgot to update the usage of strftime.
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com> Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Fabian Ebner [Mon, 14 Dec 2020 15:03:17 +0000 (15:03 +0000)]
prune mark: correctly keep track of already included backups
This needs to happen in a separate loop, because some time intervals are not
subsets of others, i.e. weeks and months. Previously, with a daily backup
schedule, having:
* a backup on Sun, 06 Dec 2020 kept by keep-daily
* a backup on Sun, 29 Nov 2020 kept by keep-weekly
would lead to the backup on Mon, 30 Nov 2020 to be selected for keep-monthly,
because the iteration did not yet reach the backup on Sun, 29 Nov 2020 that
would mark November as being covered.
reuse the one from DirPlugin by directing the call to it, but with
the actual $class. This should stay stable, as we provide an ABI and
try to always use $class->helpers.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Previous to this we did not called the plugins update_volume_notes at
all in the case where a user delted the textarea, which results to
passing a falsy value ('').
Also adapt the currently sole implementation to delete the notes field
in the undef or '' value case. This can be done safely, as we default
to returning an empty string if no notes file exists.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
pbs: activate storage: fully validate if storage config works
improves UX of on_update and on_add hooks *a lot*.
This is a bit more expensive than the TCP ping, or even just an
unauthenticated ping, but not as bad as a full datastore status - as
this only reads the datastore config file (which is normally in page
cache anyway).
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Fabian Ebner [Fri, 27 Nov 2020 09:35:44 +0000 (10:35 +0100)]
plugin: hooks: add explicit returns
to avoid returning something unexpected. Finish what afeda182566292be15413d9b874720876eac14c9 already started for all the other
plugins. At least for ZFS's on_add_hook this is necessary (adding a ZFS storage
currently fails as reported here [0]), but it cannot hurt
in the other places either as the only hooks we expect to return something
currently are PBS's on_add_hook and on_update_hook.
Dominik Csapak [Tue, 24 Nov 2020 09:09:32 +0000 (10:09 +0100)]
Storage/Plugin: add get/update_volume_comment and implement for dir
and add the appropriate api call to set and get the comment
we need to bump APIVER for this and can bump APIAGE, since
we only use it at this new call that can work with the default
implementation
Fabian Ebner [Mon, 23 Nov 2020 12:33:09 +0000 (13:33 +0100)]
convert maxfiles to prune_backups when reading the storage configuration
If there are already prune options configured, simply delete the maxfiles
setting. Having set both is invalid from vzdump's perspective anyways, and any
backup job on such a storage failed, meaning a user would've noticed.
If there are no prune options, translate the maxfiles value to keep-last,
except for maxfiles being zero (=unlimited), in which case we use keep-all.
If both are not set, don't set anything, so:
1. Storages don't suddenly have retention options set.
2. People relying on vzdump defaults can still use those.
Fabian Ebner [Mon, 23 Nov 2020 12:33:08 +0000 (13:33 +0100)]
prune: introduce keep-all option
useful to have an alternative to the old maxfiles = 0. There has to
be a way for vzdump to distinguish between:
1. use the /etc/vzdump.conf default (when no options are configured for the storage)
2. use no limit (when keep-all=1)
Dominic Jäger [Wed, 28 Oct 2020 10:04:54 +0000 (11:04 +0100)]
lvmthin: Match snapshot remove regex to allowed names
We allow snapshot names that match pve-configid but during qm destroy we have
not removed all snapshots that match pve-configid so far. For example, the name
x-y was allowed but the resulting snap_vm-105-disk-0_x-y was not removed.
Reported-by: Hannes Laimer <h.laimer@proxmox.com> Signed-off-by: Dominic Jäger <d.jaeger@proxmox.com>
Fabian Ebner [Fri, 13 Nov 2020 13:08:55 +0000 (14:08 +0100)]
prune: allow having all prune options zero/missing
This is basically necessary for the GUI's prune widget, because we want to
pass along all options equal to zero when all the number fields are cleared.
And it's more similar to how it's done in PBS now.
Bumped the APIAGE and APIVER, in case some external plugin needs to adapt to
the now less restrictive schema for 'prune-backups'.
Thomas Lamprecht [Thu, 12 Nov 2020 17:01:41 +0000 (18:01 +0100)]
api: storage create/update: return parts of the configuration
First, doing such things can make client work slightly easier, as the
submitted values do not need to be made available in any callback
handling the response.
But the actual reason for doing this now is, that this is a
preparatory step for allowing the user to download/print/.. an
autogenerated PBS client encryption key.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Thomas Lamprecht [Thu, 12 Nov 2020 16:31:16 +0000 (17:31 +0100)]
avoid unecessary try for over optimization
That was lots of code and hash map touching for the case where one
avoided a extra stat, which result probably was in the page cache
anyway, for the case that a backup has a comment.
A case which is rather be unlikely - comments are normally done for
the occasional explicit backup (e.g., before major upgrade, before a
configuration change in that guest, ...), at least not worth some
relatively complicated effort making that sub harder to read and
maintain.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Fabian Ebner [Fri, 6 Nov 2020 14:30:55 +0000 (15:30 +0100)]
fix #3030: always activate volumes in storage_migrate
AFAICT the snapshot activation is not necessary for our plugins at the moment,
but it doesn't really hurt and might be relevant in the future or for external
plugins.
Deactivating volumes is up to the caller, because for example, for replication
on a running guest, we obviously don't want to deactivate volumes.
Stoiko Ivanov [Fri, 6 Nov 2020 14:19:40 +0000 (15:19 +0100)]
add check for fsfreeze before snapshot
In order to take a snapshot of a container volume, which can be mounted
read-only with RBD, the volume needs to be frozen (fsfreeze (8)) before taking
the snapshot.
This commit adds helpers to determine if the FIFREEZE ioctl needs to be called
for the volume.
Jan-Jonas Sämann [Fri, 30 Oct 2020 03:57:22 +0000 (04:57 +0100)]
Diskmanage: Use S.M.A.R.T. attributes for SSDs wearout lookup
This replaces a locally maintained hardware map in
get_wear_leveling_info() by commonly used register names of
smartmontool. Smartmontool maintains a labeled register database that
contains a majority of drives (including versions). The current lookup
produces false estimates, this approach hopefully provides more reliable
data.
Jan-Jonas Sämann [Fri, 30 Oct 2020 03:57:21 +0000 (04:57 +0100)]
Update disk_tests/ssd_smart/sde data
Provides recent test data for disk_tests/ssd_smart/sde_smart. The
previous data was created using an older smartmontools version, which
did not yet support the drive and therefore had bogus attribute mapping.
Fabian Ebner [Thu, 1 Oct 2020 08:11:36 +0000 (10:11 +0200)]
fix #1452: also log stderr of remote command with insecure storage migration
Commit 8fe00d99449b7c80e81ab3c9826625a4fcd89aa4 already
introduced the necessary logging for the secure code path,
so presumably the bug was already fixed for most people.
Delay the potential die for the send command to be able to log
the ouput+error from the receive command. Like this we also see e.g.
'volume ... already exists' instead of just 'broken pipe'.
Fabian Ebner [Thu, 1 Oct 2020 08:11:35 +0000 (10:11 +0200)]
avoid output of zfs get command on volume import
quiet takes care of both the error and success case.
Without this, there are lines like:
myzpool/vm-4352-disk-0@__replicate_4352-7_1601538554__ name myzpool/vm-4352-disk-0@__replicate_4352-7_1601538554__ -
in the log if the dataset exists, and this information is
already present in more readable form.
Thomas Lamprecht [Tue, 27 Oct 2020 06:03:17 +0000 (07:03 +0100)]
fix #3097: cifs, nfs: increase connection check timeout to 10s
we already have the ZFS pool plugin as precedent to use 10s, at for
network with remote off-site storage one can get to 200 - 300ms
RTT latency, which means that for a protocol needing multiple rounds of
communication, one can easily get over 2s while not being in a broken
network.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Stoiko Ivanov [Mon, 12 Oct 2020 15:34:57 +0000 (17:34 +0200)]
LIO: untaint values read from remote config
The LIO backend for ZFS over iSCSI fetches the json-config periodically from
the target.
This patch reduces the stored config values to those which are actually used
and additonally untaints the values read from the remote host's config-file.
Since the LUN index is used in calls to targetcli on the remote host (via
run_command), untainting prevents the call to crash when run with '-T'.
Tested by creating a zfs over iscsi backed VM, starting it, adding disks,
resizing disks, removing disks, creating snapshots, rolling back to a snapshot.
Stoiko Ivanov [Fri, 9 Oct 2020 15:13:44 +0000 (17:13 +0200)]
ZFSPlugin: untaint lun number
ZFS over iSCSI fetches information about the disk-images via ssh, thus
the obtainted data is tainted (perlsec (1)).
Since pvedaemon runs with '-T' enabled trying to start a VM via GUI/API failed,
while it still worked via `qm` or `pvesh`.
The issue surfaced after commit cb9db10c1a9855cf40ff13e81f9dd97d6a9b2698 in
pve-common ('run_command: improve performance for logging and long lines'),
and results from concatenating the original (tainted) buffer to a variable,
instead of a captured subgroup.
Untainting the value in ZFSPlugin should not cause any regressiosn, since the
other 3 target providers already have a match on '\d+' for retrieving the
lun number.
reported via pve-user [0].
reproduced and tested by setting up a LIO-target (on top of a virtual PVE),
adding it as storage and trying to start a guest (with a disk on the
ZFS over iSCSI storage) with `perl -T /usr/sbin/qm start $vmid`
When creating a new ZFS storage, also instantiate an import-unit for the pool.
This should help mitigate the case where some pools don't get imported during
boot, because they are not listed in an existing zpool.cache file.
This patch needs the corresponding addition of 'zfs-import@.service' in
the zfsonlinux repository.
commit 815df2dd08ac4c7295135262e60d64fbb57b8f5c introduced a small issue
when activating linked clone volumes - the volname passed contains
basevol/subvol, which needs to be translated to subvol.
using the path method should be a robust way to get the actual path for
activation.
Found and tested by building the package as root (otherwise the zfs
regressiontests are skipped).
Reported-by: Thomas Lamprecht <t.lamprecht@proxmox.com> Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Makes it possible to clone and start a container whose
ZFS subvols are not yet mounted for some reason. If a
subvol cannot be mounted, there's a better error now:
zfs error: cannot mount '/myzpool/subvol-103-disk-0': directory is not empty
Previously, cloning would quietly do an "empty" clone,
and startup would fail with:
mount_autodev: 1074 Permission denied - Failed to create "/dev" directory
lxc_setup: 3238 Failed to mount "/dev"
do_start: 1224 Failed to setup container "103"
__sync_wait: 41 An error occurred in another process (expected sequence number 5)
projects / pve-storage.git / log