LVM replaces any dashes '-' in an LV or PV name with two '--' for the
created device node in /dev/mapper/ to distinguish the seperating
character between the PV and LV name.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
This lock is held during VM startup, so that multiple calls will not
start VMs twice. But this means that the timeout needs to incorporate
the time it might take a VM to boot, so increase it quite a bit.
This could previously lead to "interrupted system call" errors when
accessing backups with many disks.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
New kernel has stricter checks on tmpfs with stick-bit on directories, so some
commands (i.e. proxmox-tape changer status) fails when executed as root, because
permission checks fails when locking the drive.
This patch move the drive locks to /run/proxmox-backup/drive-lock.
Note: This is incompatible to old locking mechmanism, so users may not
run tape backups during update (or running backup can fail).
move key_derivation to pbs-datastore
pbs-api-types should only contain "basic" types which
* are usually required by clients
* don't depend on pbs-related code directly
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
These are mostly tokio specific "hacks" or "workarounds" we
only really need/want in our binaries without pulling it in
via our library crates.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
During startup most of the stuff is happening in milliseconds (or
less), so the timestamp granularity of seconds made it hard to tell
if the previous command required 990ms or 1ms, which is quite the
difference in the restore daemon context.
Using micros seems not to bring too much additional information, a
millisecond is already an ok lower time resolution for logging, so
switch only to millis for now.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
fixes file restore again.
The new Memcom tracking file lives in `/run/proxmox-backup` and is
always created on REST interaction, as CachedUserInfo uses it to
efficiently track config changes, and such a cache is used in each
REST handle_request.
Further, the Memcom infra expects the base run PBS dir to exists
already, which is an OK assumption to have, but in the file-restore
daemon we have a significantly more minimal environment, and the run
dir was simply not required there, even /run isn't a tmpfs yet.
Fixes fda19dcc6f ("fix CachedUserInfo by using a shared memory version counter")
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Parses JSON output from 'pvs' and 'lvs' LVM utils and does two passes:
one to scan for thinpools and create a device node for their
metadata_lv, and a second to load all LVs, thin-provisioned or not.
Should support every LV-type that LVM supports, as we only parse LVM
tools and use 'vgscan --mknodes' to create device nodes for us.
Produces a two-layer BucketComponent hierarchy with VGs followed by LVs,
PVs are mapped to their respective disk node.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Reviewed-By: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Prefix zpool mount paths to avoid clashing with other mount namespaces
(like LVM).
Also ignore "already-mounted" error and return it as success instead -
as we always assume that a mount path is unique, this is a safe
assumption, as nothing else could have been mounted here.
This fixes an issue where a mountpoint=legacy subvol might be available
on different disks, and thus have different Bucket instances that don't
share the mountpoint cache, which could lead to an error if the user
tried opening it multiple times on different disks.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Reviewed-By: Fabian Grünbichler <f.gruenbichler@proxmox.com>
otherwise the path ends in an array ["foo", "bar"] instead of "foo/bar"
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Reviewed-By: Fabian Grünbichler <f.gruenbichler@proxmox.com>
To support nested BucketComponents, it is necessary to dedup them, as
otherwise two components like:
/foo/bar
/foo/baz
will result in /foo being shown twice at the first hierarchy.
Also make the size property based on index and optional, as for example
/foo in the example above might not have a size, and bar/baz might have
differing sizes.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Reviewed-By: Fabian Grünbichler <f.gruenbichler@proxmox.com>
These require mounting using the regular 'mount' syscall.
Auto-generates an appropriate mount path.
Note that subvols with mountpoint=none cannot be mounted this way, and
would require setting the mountpoint property, which is not possible as
the zpools have to be imported with readonly=on.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Uses the ZFS utils to detect, import and mount zpools. These are
available as a new Bucket type 'zpool'.
Requires some minor changes to the existing disk and partiton detection
code, so the ZFS-specific part can use the information gathered in the
previous pass to associate drive names with their 'drive-xxxN.img.fidx'
node.
For detecting size, the zpool has to be imported. This is only done with
pools containing 5 or less disks, as anything else might take too long
(and should be seldomly found within VMs).
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Even through best efforts at keeping it small, including the ZFS tools
in the initramfs seems to have exhausted the small overhead we had left
- give it a bit more RAM to compensate.
Also disable the ZFS ARC, as it's no use in such a memory constrained
environment, and we cache on the QEMU/rust layer anyway.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
since the output:
Result: "<UPID>"
is not really interesting, show instead the task log while
the datastore is creating, since it is now run in a worker
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
admin/datastore reads linearly only, so no need for cache (capacity of 1
basically means no cache except for the currently active chunk).
mount can do random access too, so cache last 8 chunks for possibly a
mild performance improvement.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
when we remove a datastore via api/cli, the proxy
has sometimes leftover references to that datastore in its
DATASTORE_MAP which includes an open filehandle on the
'.lock' file
this prevents unmounting/exporting the datastore even after removal,
only a reload/restart of the proxy did help
add a command to our command socket, which removes all non
configured datastores from the map, dropping the open filehandle
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
This reverts commit 75f9f40922, which is
no longer needed now that we use tokio >= 1.6 which contains the proper
fix.
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
While the issue with vsock packets starving kernel memory is mostly
worked around by the '64k -> 4k buffer' patch in
'proxmox-backup-restore-image', let's be safe and also limit the number
of concurrent transfers. 8 downloads per VM seems like a fair value.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
The extract API call may be active for more than the watchdog timeout,
so a simple ping is not enough.
This adds an "inhibit" API, which will stop the watchdog from completing
as long as at least one WatchdogInhibitor instance is alive. Keep one in
the download task, so it will be dropped once it completes (or errors).
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
See this PR for more info: https://github.com/tokio-rs/tokio/pull/3756
As a workaround use a pair of connected unix sockets - this obviously
incurs some overhead, albeit not measureable on my machine. Once tokio
includes the fix we can go back to a DuplexStream for performance and
simplicity.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Used to specify a filesystem placed directly on a disk, without a
partition table inbetween. Detected by simply attempting to mount the
disk itself.
A helper "make_dev_node" is extracted to avoid code duplication.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
A bucket might contain multiple (or 0) layers of components in its path
specification, so allow a mapping between bucket type strings and
expected component depth. For partitions, this is 1, as there is only
the partition number layer below the "part" node.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
- refactor the combinators,
- make it take a `&T: Serialize` instead of a Value, and
allow sending the raw string via `send_raw_command`.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
we cannot add a plugin with an existing ID so this completion helper
is rather counterproductive...
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Set PBS_QEMU_DEBUG=1 on a command that starts a VM and then connect to
the debug root shell via:
minicom -D \unix#/run/proxmox-backup/file-restore-serial-10.sock
or similar.
Note that this requires 'proxmox-backup-restore-image-debug' to work,
the postinst script is updated to also generate the corresponding image.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
A PCI bus can only support up to 32 devices, so excluding built-in
devices that left us with a maximum of about 25 drives. By adding a new
PCI bridge every 32 devices (starting at bridge ID 2 to avoid conflicts
with automatic bridges), we can theoretically support up to 8096 drives.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
The guest kernel requires more memory depending on how many disks are
attached. 256 seems to be enough for basically any reasonable and
unreasonable amount of disks though.
For debug instance, make it 1G, as these are never started automatically
anyway, and need at least 512MB since the initramfs (especially when
including a debug build of the daemon) is substantially bigger.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Helps to clean up a VM that has crashed, is not responding to vsock API
calls, but still has a running QEMU instance.
We always check the process commandline to ensure we don't kill a random
process that took over the PID.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
otherwise, the kernel driver exposes file names as iso 8859-1,
but we want to have them as utf8.
This mapping should always work, since UTF16 can be cleanly converted
to UTF8.
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
else we sometimes forget to remove it from the 'params' variable
and use that further, running into 'invalid parameter' errors
found by giving 'output-format' paramter to proxmox-tape status
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
This allows mounting XFS partitons with 'dirty' states, like from a
running VM. Otherwise XFS tries to write recovery information, which
fails on a read-only mount.
Tested-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Tested-by: Dominik Csapak <d.csapak@proxmox.com>
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Drive serials have a character limit of 20, longer names like
"drive-virtio0.img.fidx" or "drive-efidisk0.img.fidx" would get cut off.
Fix this by removing the suffix, it is not necessary to uniquely
identify an image.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
With the vsock-pkt-buffer fix in proxmox-backup-restore-image, we can
use way less memory for the VM without risking any crashes. 128 MiB
seems to be the lowest it will go and still be fully reliable.
While at it, add the "panic=1" argument to the kernel command line, so
in case the kernel *does* run out of memory, it will at least restart
automatically.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Read image sizes (.pxar.fidx/.img.didx) from manifest and partition
sizes from /sys/...
Requires a change to ArchiveEntry, as DirEntryAttribute::Directory
does not have a size associated with it (and that's probably good).
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
to make the following cryptic error:
proxmox-file-restore failed: Error: Invalid byte 46, offset 5.
more understandable:
proxmox-file-restore failed: Error: Failed base64-decoding path '/root.pxar.didx' - Invalid byte 46, offset 5.
when a user passes in a non-base64 path but sets `--base64`.
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
same functionality as crypto_parameters, except it keeps the file
descriptor passed as "keyfd" open (and seeks to the beginning after
reading), if one is given.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
For the actual partitions and blockdevices in a backup, which the
user sees like folders in the file-restore ui
Encoded as "None", to avoid cluttering DirEntryAttribute, where it
wouldn't make any sense to have.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
These can't be entered or restored anyway, and cause issues with catalog
files for example.
Also a clippy fix.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
if a datastore or root is not used directly on the pool dir
(e.g. the installer creates 2 sub datasets ROOT/pbs-1), info in
/proc/self/mountinfo returns not the pool, but the path to the
dataset, which has no iostats itself in /proc/spl/kstat/zfs/
but only the pool itself
so instead of not gathering data at all, gather the info from the
underlying pool instead. if one has multiple datastores on the same
pool those rrd stats will be the same for all those datastores now
(instead of empty) similar to 'normal' directories
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
The data on the restore daemon is either encoded into a pxar archive, to
provide the most accurate data for local restore, or encoded directly
into a zip file (or written out unprocessed for files), depending on the
'pxar' argument to the 'extract' API call.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Allows listing files and directories on a block device snapshot.
Hierarchy displayed is:
/archive.img.fidx/bucket/component/<path>
e.g.
/drive-scsi0.img.fidx/part/2/etc/passwd
(corresponding to /etc/passwd on the second partition of drive-scsi0)
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Includes methods to start, stop and list QEMU file-restore VMs, as well
as CLI commands do the latter two (start is implicit).
The implementation is abstracted behind the concept of a
"BlockRestoreDriver", so other methods can be implemented later (e.g.
mapping directly to loop devices on the host, using other hypervisors
then QEMU, etc...).
Starting VMs is currently unused but will be needed for further changes.
The design for the QEMU driver uses a locked 'map' file
(/run/proxmox-backup/$UID/restore-vm-map.json) containing a JSON
encoding of currently running VMs. VMs are addressed by a 'name', which
is a systemd-unit encoded combination of repository and snapshot string,
thus uniquely identifying it.
Note that currently you need to run proxmox-file-restore as root to use
this method of restoring.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Includes functionality for scanning and referring to partitions on
attached disks (i.e. snapshot images).
Fairly modular structure, so adding ZFS/LVM/etc... support in the future
should be easy.
The path is encoded as "/disk/bucket/component/path/to/file", e.g.
"/drive-scsi0/part/0/etc/passwd". See the comments for further
explanations on the design.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Add a watchdog that will automatically shut down the VM after 10
minutes, if no API call is received.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Implements the base of a small daemon to run within a file-restore VM.
The binary spawns an API server on a virtio-vsock socket, listening for
connections from the host. This happens mostly manually via the standard
Unix socket API, since tokio/hyper do not have support for vsock built
in. Once we have the accept'ed file descriptor, we can create a
UnixStream and use our tower service implementation for that.
The binary is deliberately not installed in the usual $PATH location,
since it shouldn't be executed on the host by a user anyway.
For now, only the API calls 'status' and 'stop' are implemented, to
demonstrate and test proxmox::api functionality.
Authorization is provided via a custom ApiAuth only checking a header
value against a static /ticket file.
Since the REST server implementation uses the log!() macro, we can
redirect its output to stdout by registering env_logger as the logging
target. env_logger is already in our dependency tree via zstd/bindgen.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
This allows switching the base user identification/authentication method
in the rest server. Will initially be used for single file restore VMs,
where authentication is based on a ticket file, not the PBS user
backend (PAM/local).
To avoid putting generic types into the RestServer type for this, we
merge the two calls "extract_auth_data" and "check_auth" into a single
one, which can use whatever type it wants internally.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
For now it only supports 'list' and 'extract' commands for 'pxar.didx'
files. This should be the foundation for a general file-restore
interface that is shared with block-level snapshots.
This is packaged as a seperate .deb file, since for block level restore
it will need to depend on pve-qemu-kvm, which we want to seperate from
proxmox-backup-client.
[original code for proxmox-file-restore.rs]
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
[code cleanups/clippy, use helpers::list_dir_content/ArchiveEntry, no
/block subdir for .fidx files, seperate binary and package]
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Add a new module containing key-related functions and schemata from all
over, code moved is not changed as much as possible.
Requires adapting some 'use' statements across proxmox-backup-client and
putting the XDG helpers quite cozily into proxmox_client_tools/mod.rs
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
by changing the 'store' parameter of the restore api call to a
list of mappings (or a single default datastore)
for example giving:
a=b,c=d,e
would restore
datastore 'a' from tape to local datastore 'b'
datastore 'c' from tape to local datastore 'e'
all other datastores to 'e'
this way, only a single datastore can also be restored, by only
giving a single mapping, e.g. 'a=b'
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
in commit `asyncify pxar create_archive`, we changed from a
separate thread for creating a pxar to using async code, but the
StdChannelWriter used for both pxar and catalog can block, which
may block the tokio runtime for single (and probably dual) core
environments
this patch adds a wrapper struct for any writer that implements
'std::io::Write' and wraps the write calls with 'block_in_place'
so that if called in a tokio runtime, it knows that this code
potentially blocks
Fixes: 6afb60abf5 ("asyncify pxar create_archive")
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
so that the tape backup can be restored as any user, given
the current logged in user has the correct permission.
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
found and semi-manually replaced by using:
codespell -L mut -L crate -i 3 -w
Mostly in comments, but also email notification and two occurrences
of misspelled 'reserved' struct member, which where not used and
cargo build did not complain about the change, soo ...
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
...including common schemata, connect(), extract_*() and completion
functions.
For later use with proxmox-file-restore binary.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
...to take advantage of the aio::Encoder from the pxar create.
Rather straightforward conversion, but does require getting rid of
references in the Archiver struct, and thus has to be given the Mutex
for the catalog directly. The callback is boxed.
archive_dir_contents can call itself recursively, and thus needs to
return a boxed future.
Users are adjusted, namely PxarBackupStream is converted to use an
Abortable future instead of a thread so it supports async in its handler
function, and the pxar bin create_archive is converted to an async API
function. One test case is made to just use 'block_on'.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
in most uses, we want to remove the drive from the param afterwards
where we don't, we already overwrite it with the result of this function
this fixes some commands (like 'proxmox-tape read-label --drive foo')
that failed with:
parameter 'drive': duplicate parameter.
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
to avoid confusing messages about using encryption keys when restoring
plaintext backups, or about loading master keys when they are not
actually used for the current operation.
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
pull out the crypt-mode to logically group arms and make the whole mess
a bit more "human-parsable".
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
it's needed for PVE's LXC integration, and might be interesting for
other more special usage scenarios as well.
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Dietmar Maurer
Stefan Reiter
Wolfgang Bumiller
Thomas Lamprecht
Dominik Csapak
Hannes Laimer
Fabian Grünbichler
Dylan Whyte