The way _ostree_repo_import_object() is written, a hardlink copy is only
attempted if the source repo is trusted, so update the docs for
ostree_repo_import_object_from_with_trust() to reflect that.
Closes: #1777
Approved by: cgwalters
This allows specifying gpgpath as list of
paths that can point to a file or a directory. If a directory path
is given, paths to all regular files in the directory are added
to the remote as gpg ascii keys. If the path is not a directory,
the file is directly added (whether regular file, empty - errors
will be reported later when verifying gpg keys e.g. when pulling).
Adding the gpgkeypath property looks like:
ostree --repo=repo remote add --set=gpgpath="/path/key1.asc,/path/keys.d" R1 https://example.com/some/remote/ostree/repoCloses#773Closes: #1773
Approved by: cgwalters
When falling back to copying objects when importing them into a
bare-user repo, we only actually need to transfer over the
`user.ostreemeta` xattr.
This allows the destination repo to be on a separate filesystem that
might not even support `security.selinux`. (I hit this while importing
over virtio-9p).
Closes: #1771
Approved by: cgwalters
I found this useful while hacking on rpm-ostree but I think it might be
useful enough to upstream. This stat is really helpful for validating
that a pipeline is hitting the devino cache sweet spot.
Closes: #1772
Approved by: cgwalters
if a file ".wh..wh..opq" is present in a directory, delete anything
from lower layers that is already in that directory.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Closes: #1486
Approved by: cgwalters
This renames a config key to make its semantics more obvious. Despite
what the commit message says, it only applies when a set of repo finders
is not specified (either on the command line or in a library API call).
This also renames the corresponding ostree_repo_get function. We can do
this since it hasn't been released yet.
Closes: #1763
Approved by: pwithnall
Rather than manually starting the `ostree-finalize-staged.service` unit,
we can leverage systemd's path units for this. It fits quite nicely too,
given that we already have a path we drop iif we have a staged
deployment.
To give some time for the preset to make it to systems, we don't yet
drop the explicit call to `systemctl start`. Though we do make it
conditional based on a DEBUG env var so that we can actually test it in
CI for now. Once we're sure this has propagated, we can drop the
`systemctl start` path and the env var together.
Closes: #1740
Approved by: cgwalters
This commit disables searching on the local network for refs, unless
explicitly requested by the user either by changing the value of the
"core.repo-finders" config option, or by passing an OstreeRepoFinderAvahi to
ostree_repo_find_remotes_async() / ostree_repo_finder_resolve_async(),
or by specifying "lan" in the --finders option of the find-remotes
command.
The primary reason for this is that ostree_repo_find_remotes_async()
takes about 40% longer to complete with the LAN finder enabled, and that
API is used widely (e.g. in every flatpak operation). It's also probable
that some users don't want ostree doing potentially unexpected traffic
on the local network, even though everything pulled from a peer is GPG
verified.
Flathub will soon deploy collection IDs to everyone[1] so these code
paths will soon see a lot more use and that's why this change is being
made now.
Endless is the only potential user of the LAN updates feature, and we
can revert this patch on our fork of ostree. For it to be used outside
Endless OS we will need to upstream eos-updater-avahi and
eos-update-server into ostree.
[1] https://github.com/flathub/flathub/issues/676Closes: #1758
Approved by: cgwalters
Currently libostree essentially has two modes when it's pulling refs:
the "legacy" code paths pull only from the Internet, and the code paths
that are aware of collection IDs try to pull from the Internet, the
local network, and mounted filesystems (such as USB drives). The problem
is that while we eventually want to migrate everyone to using collection
IDs, we don't want to force checking LAN and USB sources if the user
just wants to pull from the Internet, since the LAN/USB code paths can
have privacy[1], security[2], and performance[3] implications.
So this commit implements a new repo config option called "repo-finders"
which can be configured to, for example, "config;lan;mount;" to check
all three sources or "config;mount;" to disable searching the LAN. The
set of values mirror those used for the --finders option of the
find-remotes command. This configuration affects pulls in three places:
1. the ostree_repo_find_remotes_async() API, regardless of whether or
not the user of the API provided a list of OstreeRepoFinders
2. the ostree_repo_finder_resolve_async() /
ostree_repo_finder_resolve_all_async() API
3. the find-remotes command
This feature is especially important right now since we soon want to
have Flathub publish a metadata key which will have Flatpak clients
update the remote config to add a collection ID.[4]
This effectively fixes https://github.com/flatpak/flatpak/issues/1863
but I'll patch Flatpak too, so it doesn't pass finders to libostree only
to then have them be removed.
[1] https://github.com/flatpak/flatpak/issues/1863#issuecomment-404128824
[2] https://github.com/ostreedev/ostree/issues/1527
[3] Based on how long the "ostree find-remotes" command takes to
complete, having the LAN finder enabled slows down that step of the
pull process by about 40%. See also
https://github.com/flatpak/flatpak/issues/1862
[4] https://github.com/flathub/flathub/issues/676Closes: #1758
Approved by: cgwalters
The idea is that if the process is running as root, it can change
ownership of newly written files to match the owner of the repo.
Unfortunately, it currently applies in the other direction, too - a
non-root user writing to a root owned repository. If the repo is
writable by the user but owned by root, it can still create files and
directories there, but it can't change ownership of them.
This feature comes from
https://bugzilla.gnome.org/show_bug.cgi?id=738954. As it turns out, this
feature was never completed. It only works on content objects and not
metadata objects, refs, deltas, summaries, etc. Rather than try to fix
all of those, remove the feature until someone has interest in
completing it.
Closes: #1754
Approved by: cgwalters
Actually testing the patch to add `--force-copy-zerosized` to
rpm-ostree tripped over the fact that it uses `--union-identical`,
and we just hit an assertion failure with that combination.
Fix this by copying over the logic we have for the hardlink case.
Closes: #1753
Approved by: jlebon
In rpm-ostree we've hit a few cases where hardlinking zero-sized
files causes us problems. The most prominent is lock files in
`/usr/etc`, such as `/usr/etc/selinux/semanage.LOCK`. If there
are two zero-sized lock files to grab, but they're hardlinked,
then locking will fail.
Another case here is if one is using ostree inside a container
and don't have access to FUSE (i.e. `rofiles-fuse`), then the
ostree hardlinking can cause files that aren't ordinarily hardlinked
to become so, and mutation of one mutates all. An example where
this is concerning is Python `__init__.py` files.
Now, these lock files should clearly not be in the tree to begin
with, but - we're not gaining a huge amount by hardlinking these
files either, so let's add an option to disable it.
Closes: #1752
Approved by: jlebon
Write to the journal when starting to finalize a staged deployment.
Combined with the "Transaction completed" message we already emit, this
makes it easy later on to determine whether the operation was successful
by inspecting the journal. This will be used by `rpm-ostree status`.
Closes: #1750
Approved by: cgwalters
There are use cases for libostree as a local content store
for content derived or delivered via other mechanisms (e.g. OCI
images, RPMs, etc.). rpm-ostree today imports RPMs into OSTree
branches, and puts the RPM header value as commit metadata.
Some of these can be quite large because the header includes
permissions for each file. Similarly, some OCI metadata is large.
Since there's no security issues with this, support committing
such content.
We still by default limit the size of metadata fetches, although
for good measure we make this configurable too via a new
`max-metadata-size` value.
Closes: https://github.com/ostreedev/ostree/issues/1721Closes: #1744
Approved by: jlebon
Currently the locking code checks if the value -1 was set for the config
key "lock-timeout-secs" and if so, a thread trying to acquire a lock
will block indefinitely. Positive values specify how long to attempt to
acquire a lock in a non-blocking way (the attempt is made once every
second). But when the value is read from the config file,
g_ascii_strtoull() is used, which converts it to an unsigned integer.
This commit makes libostree use g_ascii_strtoll() instead, so that it's
possible to set that key to -1 as intended.
Closes: #1737
Approved by: jlebon
Copying the xattrs on metadata objects is wrong in general, we
don't "own" them. Notably this would fail in the situation of
doing a pull from e.g. a `bare-user` source to a destination
that was on a different mount point (so we couldn't hardlink),
and the source had e.g. a `security.selinux` attribute.
Closes: #1734Closes: #1736
Approved by: jlebon
Just include the whole URL that failed if libcurl failed with something
elementary like CURLE_COULDNT_CONNECT or CURLE_COULDNT_RESOLVE_HOST.
Closes: #1731Closes: #1732
Approved by: cgwalters
There's some subtlety to this, we don't handle all cases.
But the 99% cases are using `--sysroot deploy` to create an
initial deployment, and then doing upgrades from inside
a booted deployment.
It was only the latter case that didn't work with a separate `/var`.
Fixing all of them would probably require libostree to learn
how to e.g. look at `/etc/fstab` (or worse, systemd mount units?)
and handle the mounting. I don't think we want to do anything
like that right now, since there are no active drivers for the
use case.
Closes: https://github.com/ostreedev/ostree/issues/1729Closes: #1730
Approved by: akiernan
Earlier, the actual reserved space (in blocks) were calculated inside the
transaction codepath ostree_repo_prepare_transaction(). However, while
reworking on ostree_repo_get_min_free_space_bytes() API, it was realized that
this calculation can be done independently from the transaction's codepaths, hence
enabling the usage for ostree_repo_get_min_free_space_bytes() API irrespective
of whether there is an ongoing transaction or not.
https://github.com/ostreedev/ostree/issues/1720Closes: #1722
Approved by: pwithnall
This commit defines a metadata key that tells clients to update their
remote config to add a collection ID. This functionality is currently
implemented in Flatpak for the key "xa.collection-id", but there are two
good reasons for moving the key to OSTree:
1) Servers such as Flathub shouldn't set xa.collection-id in their
metadata now or in the medium term future, because many users are still
using old versions of Flatpak and OSTree[1] which would hit various
bugs[2][3][4] on the P2P code paths that are enabled by collection IDs.
Defining a new key means that only clients running recent
(as-yet-unreleased) versions of Flatpak and OSTree will pay attention to
it and deploy the collection ID, leaving the users on old versions
unaffected.
2) OSTree is as "invested" in collection IDs as Flatpak, so there's no
reason the key should be defined in Flatpak rather than here. According
to Philip Withnall, the reason the key was put in Flatpak originally was
that at the time there was uncertainty about tying OSTree to collection
IDs.
[1] https://ahayzen.com/direct/flathub.html#downloadsbyflatpakstacked
[2] https://github.com/ostreedev/ostree/commit/e4e6d85ea
[3] https://github.com/flatpak/flatpak/commit/5813639f
[4] https://github.com/flatpak/flatpak/commit/5b21a5b7Closes: #1726
Approved by: pwithnall
There is no API method to remove a file or subdirectory from a MutableTree
besides directly manipulating the GHashTable returned by _get_files or
_get_subdirs. This isn't possible from an introspection binding that transforms
the returned GHashTable, and may also leave the tree checksum in an invalid
state. Introduce a new method so that removing files or subdirectories is
safe, and possible from bindings.
Closes: #1724
Approved by: jlebon
This fixes typos and grammar in the docs for OstreeRepo, and copies the
information about OSTREE_REPO_MODE_BARE_USER_ONLY from ostree-core.h
Closes: #1725
Approved by: jlebon
In the OstreeRepoFinderAvahi implementation,
ostree_avahi_service_build_repo_finder_result() is where the DNS-SD
records are processed and turned into OstreeRepoFinderResult objects.
Each result object is supposed to have a hash table mapping refs to
checksums, so this is accomplished by first adding a placeholder (a ref
mapping to a NULL checksum) for each ref matched by the bloom filter,
and later filling in the checksums using the remote's summary file,
which happens in get_checksums(). The problem is that there's no
guarantee all the checksums will be resolved (non-NULL), so the
ostree_repo_finder_result_new() call then hits an assertion failure in
is_valid_collection_ref_map() leading to a crash (in the case that one
or more refs had NULL checksums).
There are at least two situations where the ref checksum might not be
found in the peer remote's summary file:
1) The bloom filter match was a false positive. This is going to happen
sometimes by design.
2) The peer remote's summary is out of sync with its DNS-SD records.
This shouldn't normally happen but it's still good to be robust to the
possibility; in Endless OS nothing guarantees the atomicity of updating
the summary and DNS-SD records.
This commit changes libostree to be robust to the possibility of refs
missing from the peer remote's summary, by removing any that still have
a NULL checksum associated with them after the summary has been fetched
and processed.
The other OstreeRepoFinder implementations don't have this issue because
they use summary files directly and therefore always have access to the
checksum.
Closes: #1717
Approved by: pwithnall
when converted to bytes
In a subsequent commit, we add a public API to read the value of
min-free-space-* value in bytes. The value for free space check
is enforced in terms of block size instead of bytes. Therefore,
for consistency we check while preparing the transaction that the
value doesn't overflow when converted to bytes.
https://phabricator.endlessm.com/T23694Closes: #1715
Approved by: cgwalters
Debian and Debian-derived systems have their GRUB configuration file in
/boot/grub/grub.cfg, rather than /boot/grub2/grub.cfg. Detecting this
file is necessary to correctly generate GRUB boot configuration on
Debian systems.
Closes: #1714
Approved by: cgwalters
If a ref already exists, we are likely only a few commits behind the
current head of the ref, so it is probably better for bandwidth
consumption to pull the individual objects rather than the from-scratch
delta.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #1709
Approved by: cgwalters
Since 9dc6ddce08 it has not been true that
'new_config' was simply ref'd: it's serialized, and then re-parsed into
a new GKeyFile.
Closes: #1707
Approved by: jlebon
Add an invalid-cache test error flag to ensure that the code that checks
for and recovers from a corrupted summary cache is hit. This helps make
sure that the recovery path is actually used without resorting to
G_MESSAGES_DEBUG.
Closes: #1698
Approved by: cgwalters
If for some reason the cached summary doesn't match the cached signature
then fetch the remote summary and verify again. Since commit c4c2b5eb
this is unlikely to happen since the summary will only be cached if it
matches the signature. However, if the summary cache has been corrupted
for any other reason then it's best to be safe and fetch the remote
summary again.
This is essentially the corollary to c4c2b5eb. Where that commit helps
you from getting into the corrupted summary cache in the first place,
this helps you get out of it. Without this the client can get wedged
until a prune or the remote server republishes the summary.
Closes: #1698
Approved by: cgwalters
copy_option() unnecessarily passed ownership of the value
to g_variant_dict_insert_value, but that already refs, so it was leaked.
Closes: #1702
Approved by: cgwalters
We need to have the g_auto(GLnxDirFdIterator) inside the loop, or
we don't correctly clean up when iterating several times.
Closes: #1700
Approved by: cgwalters
Now that we have `auto-update-summary`, there is no point in having
`commit-update-summary`. The latter also only had an effect through
the `commit` CLI command, whereas the former is embedded directly in
libostree.
There is one corner case that slips through: `commit` would update the
summary file even if orphan commits were created, which we no longer do
here. I can't imagine anyone relying on this, so it seems safe to drop.
Closes: #1689Closes: #1693
Approved by: mwleeds
Mildly bikeshed, though I find the name `auto-update-summary` to be
easier to grok than `change-update-summary`. I think it's because it can
be read as "verb-verb-noun" rather than "noun-verb-noun".
Closes: #1693
Approved by: mwleeds
Normally, a configured remote will only serve refs with one associated
collection ID, but temporary remotes such as USB drives or LAN peers can
serve refs from multiple collection IDs which may use different GPG
keyrings. So the OstreeRepoFinderMount and OstreeRepoFinderAvahi classes
create dynamic OstreeRemote objects for each (uri, keyring) pair. So if
for example the USB mounted at /mnt/usb serves content from the
configured remotes "eos-apps" and "eos-sdk", the OstreeRepoFinderResult
array returned by ostree_repo_find_remotes_async() will have one result
with a remote called something like
file_mnt_usb_eos-apps.trustedkeys.gpg and the list of refs on the USB
that came from eos-apps, and another result with a remote
file_mnt_usb_eos-sdk.trustedkeys.gpg and the list of refs from eos-sdk.
Unfortunately while OstreeRepoFinderMount and OstreeRepoFinderAvahi
correctly only include refs in a result if the ref uses the associated
keyring, the find_remotes_cb() function used to clean up the set of
results looks at the remote summary file and includes every ref that's
in the intersection with the requested refs, regardless of whether it
uses a different remote's keyring. This leads to an error when you try
to pull from a USB containing refs from different collection IDs: the
pull using the wrong collection ID will error out with "Refspec not
found" and the result with the correct keyring will then be ignored "as
it has no relevant refs or they have already been pulled." So the pull
ultimately fails.
This commit fixes the issue by filtering refs coming from a dynamic
remote, so that only ones with the collection ID associated with the
keyring remote are examined. This only needs to be done for dynamic
remotes because you should be able to pull any ref from a configured
remote using its keyring. It's also only done when looking at the
collection map in the summary file, because LAN/USB remotes won't have a
"main" collection ID set (OSTREE_SUMMARY_COLLECTION_ID).
Closes: #1695
Approved by: pwithnall
This commits adds and implements a boolean repo config option called
"change-update-summary" which updates the summary file every time a ref
changes (additions, updates, and deletions).
The main impetus for this feature is that the `ostree create-usb` and
`flatpak create-usb` commands depend on the repo summary being up to
date. On the command line you can work around this by asking the user to
run `ostree summary --update` but in the case of GNOME Software calling
out to `flatpak create-usb` this wouldn't work because it's running as a
user and the repo is owned by root. That strategy also means flatpak
can't update the repo metadata refs for fear of invalidating the
summary.
Another use case for this relates to LAN updates. Specifically, the
component of eos-updater that generates DNS-SD records advertising ostree
refs depends on the repo summary being up to date.
Since ostree_repo_regenerate_summary() now takes an exclusive lock, this
should be safe to enable. However it's not enabled by default because of
the performance cost, and because it's more useful on clients than
servers (which likely have another mechanism for updating the summary).
Fixes https://github.com/ostreedev/ostree/issues/1664Closes: #1681
Approved by: jlebon
This ensures that commits aren't deleted and refs aren't added, removed,
or updated while the summary is being generated. This is in preparation
for adding a repo config option that will automatically regenerate the
summary on every ref change.
Closes: #1681
Approved by: jlebon
Using `MAX(0, $x)` here is useless since we're comparing against an
unsigned integer. Just unpack this and only subtract if it's safe to do
so.
Also, explicitly check for `fd >= 0` rather than just `!= -1` to be sure
it's a valid fd. And finally, explicitly check the return value of
`g_input_stream_read_all` as is done everywhere else in the tree and
make it clear that we're purposely ignoring the return value of `_flush`
here, but not in other places.
Discovered by Coverity.
Closes: #1692
Approved by: cgwalters
I initially was going to add a `G_DEFINE_AUTOPTR_CLEANUP_FUNC` for
`FetchStaticDeltaData`, but it honestly didn't seem worth mucking around
ownership everywhere and potentially getting it wrong.
Discovered by Coverity.
Closes: #1692
Approved by: cgwalters
In `write_metadata_object()`, make sure when creating tombstone commits
that we're actually passed an expected checksum to use.
In `write_dir_entry_to_mtree_internal()`, sanity check that `dfd_iter`
is indeed not `NULL` before trying to dereference it.
Discovered by Coverity.
Closes: #1692
Approved by: cgwalters
Since min_free_space_size_mb is considered before min_free_space_percent
in min_free_space_calculate_reserved_blocks(), it has to be considered
first when generating the error message in order for it to be accurate.
Closes: #1691
Approved by: jlebon
Previously, we would error out if both of the options were mentioned
in the config file (even if one of them is disabled with 0). There
were few suggestions that this behavior was not quite right.
Therefore, instead of throwing error and exiting, it's preferred to
warn the user. Hence, the solution that worked out is:
* Allow both options to exist simulateneously
* Check each config's value and decide:
* If both are present and are non-zero, warn the user. Also, prefer
to use min-free-space-size over the another.
* If both are absent, then use -percent=3% as fallback
* Every other case is valid hence, no warning
https://phabricator.endlessm.com/T13698
(cherry picked from commit be68991cf80f0aa1da7d36ab6e1d2c4d6c7cd3fb)
Signed-off-by: Robert McQueen <rob@endlessm.com>
Closes: #1685
Approved by: cgwalters
This is the inverse of https://github.com/ostreedev/ostree/pull/1558
aka commits cadece6c4f398ca61d21e497bd6e3fbb549f9cf6 and
3358698c86d80821d81443c906621c92672f99fb
Needed to fix `rpm-ostree kargs` test suite with default staging; skipping
a test here for now as eventually what we'll do is turn on the rpm-ostree
suite fully here.
Closes: #1677
Approved by: jlebon
Running `ostree commit --tree=ref=a --tree=dir=b` involves reading the
whole of a into an `OstreeMutableTree` before composing `b` on top. This
is inefficient if a is a complete rootfs and b is just touching one file.
We process O(size of a + size of b) directories rather than
O(number of touched dirs).
This commit makes `ostree commit` more efficient at composing multiple
directories together. With `ostree_mutable_tree_fill_empty_from_dirtree`
we create a lazy `OstreeMutableTree` which only reads the underlying
information from disk when needed. We don't need to read all the
subdirectories just to get the checksum of a tree already checked into the
repo.
This provides great speedups when composing a rootfs out of multiple other
rootfs as we do in our build system. We compose multiple containers
together with:
ostree commit --tree=ref=base-rootfs --tree=ref=container1 --tree=ref=container2
and it is much faster now.
As a test I ran
time ostree --repo=... commit --orphan --tree=ref=big-rootfs --tree=dir=modified_etc
Where modified_etc contained a modified sudoers file under /etc. I used
`strace` to count syscalls and I seperatly took timing measurements. To
test with a cold cache I ran
sync && echo 3 | sudo tee /proc/sys/vm/drop_caches
Results:
| | Before | After |
| -------------------- | ------ | ----- |
| Time (cold cache) | 8.1s | 0.12s |
| Time (warm cache) | 3.7s | 0.08s |
| `openat` calls | 53589 | 246 |
| `fgetxattr` calls | 78916 | 0 |
I'm not sure if this will have some negative interaction with the
`_ostree_repo_commit_modifier_apply` which is short-circuited here. I
think it was disabled for `--tree=ref=` anyway, but I'm not certain. All
the tests pass anyway.
I originally implemented this in terms of the `OstreeRepoFile` APIs, but
it was *way* less efficient, opening and reading many files unnecessarily.
Closes: #1643
Approved by: cgwalters
It is important that we use user-friendly error strings. The reason
being error strings are seen by users such as in GNOME Software's
error banner.
Closes: #1671
Approved by: jlebon
For `rpm-ostree ex livefs` we have a use case of pushing a rollback
deployment. There's no reason this should require deleting the staged
deployment (and doing so actually breaks livefs which tries to access
it as a data source).
I was initially very conservative here, but I think it ends up
being fairly easy to retain the staged deployment. We need to handle
two cases:
First, when the staged is *intentionally* deleted; here, we just need
to unlink the `/run` file, and then everything will be sync'd up after
reloading.
Second, (as in the livefs case) where we're retaining it,
e.g. adding a deployment to the end. What I realized here is that
we can have the code keep `new_deployments` as view without staged,
and then when we do the final reload we'll end up re-reading it from
disk anyways.
Closes: #1672
Approved by: jlebon
This implements a TODO item from
`ostree_mutable_tree_get_contents_checksum`. We now no-longer invalidate
the dirtree contents checksum at `get_contents_checksum` time - we
invalidate it when the mtree is modified. This is implemented by keeping
a pointer to the parent directory in each `OstreeMutableTree`. This gives
us stronger invariants on `contents_checksum`.
For even stronger guarantees about invariants we could make
`ostree_repo_write_mtree` or similar a member of `OstreeMutableTree` and
remove `ostree_mutable_tree_set_metadata_checksum`.
I think I've fixed a bug here too. We now invalidate parent's contents
checksum when our metadata checksum changes, whereas we didn't before.
Closes: #1655
Approved by: cgwalters
If ostree_repo_prepare_transaction() fails, we should reset the
repository’s state so that the failed call was essentially idempotent.
Do that by calling ostree_repo_abort_transaction() on the failure path.
Typically, the way for preparing a transaction to fail is for its
GCancellable to be triggered, rather than because any of the operations
involved in preparing a transaction are particularly failure prone.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #1647
Approved by: cgwalters
min-free-space-* act as a gating condition whether to we want hold onto caches in
repo/tmp. If it is found that the free-disk space is going below this threshold,
we flag it as an error and cleanup current boot's staging directory.
Closes: #1602
Approved by: jlebon
Here's a subtle bug in abort_transaction():
One of the policies of cleaning up is to skip the current boot's staging
directory. The responsible function for this is cleanup_tmpdir() which tries
to lock each of the tmpdir before deleting it. When it comes to the current
boot's staging dir, it tries to lock the directory(again!) but fails as there
is already a lockfile present. Just because the current boot's staging dir was
meant to be skipped, the bug never surfaced up and wasn't catastrohpic.
if (!_ostree_repo_try_lock_tmpdir (dfd, path, &lockfile, &did_lock, error))
return FALSE;
if (!did_lock)
return TRUE; /* Note early return */
...
if (g_str_has_prefix (path, self->stagedir_prefix))
return TRUE; /* Note early return */
The actual check for skipping staging dir for current boot was never reached
because the function returned at did_lock failure.
Therefore, execute cleanup_tmpdir() after releasing the lockfile in
abort_transaction() so that cleanup_tmpdir gets a chance to lock current boot's
staging directory and succeed.
Closes: #1602
Approved by: jlebon
Currently the BLS snippets are named ostree-$ID-$VARIANT_ID-$index.conf,
but the BLS config files are actually sorted by using the version field
which is the inverse of the index.
In most places, _ostree_sysroot_read_boot_loader_configs() is used to
get the BLS files and this function already returns them sorted by the
version field. The only place where the index trailing number is used is
in the ostree-grub-generator script that lists the BLS files to populate
the grub config file.
But for some bootloaders the BLS filename is the criteria for sorting by
taking the filename as a string version. So on these bootloaders the BLS
entries will be listed in the reverse order.
To avoid that, change the BLS snippets filename to have the version field
instead of the index and also to have the version before deployment name.
Make the filenames to be of the form ostree-$version-$ID-$VARIANT_ID.conf
so the version is before the deployment name.
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Closes: #1654
Approved by: cgwalters
If a function has a guint "out argument", passing a pointer to a gsize
is not, in general, valid. On an ILP64 platform there is no problem
since guint and gsize are identical, but on an LP64 platform it will
overwrite only the first word of the gsize, leaving the second word
unaffected. On little-endian machines, if the second word is
zero-initialized (as it is here), the result is numerically equal to
the guint, but on big-endian machines the result is around 4 billion
times what it should be, resulting in
ostree_repo_finder_config_resolve_async() reading past the end of
the array and causing undefined behaviour.
In practice this caused assertion failures (and consequently test
failures) on Debian's s390x (z/Architecture), ppc64 (64-bit PowerPC)
and sparc64 (64-bit SPARC) ports.
Closes: #1640
Signed-off-by: Simon McVittie <smcv@debian.org>
Closes: #1641
Approved by: cgwalters
A prelude to my understanding. Unfortunately `OstreeMutableTree` provides
little encapsulation, as each member has setters† so it's difficult to come
up with a list of invariants.
† `files` and `subdirs` only have getters, but the getters return mutable
references to the internals, so we still can't reason about invariants.
Closes: #1645
Approved by: jlebon
We special-case AVAHI_ERR_NO_DAEMON to not cause warnings, but if
we pass AVAHI_CLIENT_NO_FAIL to avahi_client_new, we never actually
see AVAHI_ERR_NO_DAEMON. Instead, we will get AVAHI_ERR_BAD_STATE
when we try to use the client.
Closes: #1618
Signed-off-by: Simon McVittie <smcv@debian.org>
Closes: #1639
Approved by: cgwalters
During the pull, there is an explicit check for free space on disk
vs. the size of uncompressed delta; But while writing the new content
objects that are generated, they have to honor min-free-space-* checks
too. We enforce this check in _bare_content_commit as that is where
we can know the final size of the new content object.
Closes: #1614
Approved by: jlebon
We were referencing the txn bits outside of the lock in the error
path. Generally shouldn't matter, but e.g. Rust wouldn't let us do this, and
race detector tooling will warn about it.
Closes: #1632
Approved by: jlebon
I generally like having variables include their units where applicable;
timer variables having `_secs` or `_ms`, etc.
Closes: #1632
Approved by: jlebon
Systemd units using ConditionNeedsUpdate run if the mtime of .updated in
the specified directory is newer than /usr. Since /usr has an mtime of
0, there's no way to have an older .updated file. Systemd units
typically specify ConditionNeedsUpdate=/etc or ConditionNeedsUpdate=/var
to support stateless systems like ostree.
Remove the file from the new deployment's /etc and the OS's /var
regardless of where they came from to ensure that these systemd units
run when booting new deployments. This will provide a method to run
services only on upgrade.
Closes: #1628https://bugzilla.gnome.org/show_bug.cgi?id=752950Closes: #1631
Approved by: cgwalters
Currently when I run `ostree prune` it hits a seg fault when the
hash_func is used (in this case g_str_hash) from the call stack
_ostree_repo_prune_tmp() -> g_hash_table_contains() ->
g_hash_table_lookup_node(). So the key, in this case dent->d_name, must
be corrupt in some way.
glnx_dirfd_iterator_next_dent() uses readdir() to get the dirent struct.
And according to the man page for readdir(3), "POSIX.1 explicitly notes
that this field should not be used as an lvalue" (in reference to
d_name). So this commit avoids modifying d_name in place and copies it
instead. This seems to avoid the seg fault.
Closes: #1627
Approved by: jlebon
If there’s a problem while aborting a transaction, store the error but
don’t report it until the end of the function — do a best effort at
clearing the rest of the transaction state first (since most of it
cannot fail).
If cleanup_tmpdir() fails (which, arguably, should not be a
showstopper), this allows a caller to recover and start a new
transaction in future.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #1626
Approved by: jlebon
Similar to min-free-space-percent but it supports specific sizes
(in MB, GB or TB). Also, making min-free-space-percent and -size
mutually exclusive.
min-free-space-percent does not give a fine tuning of the free disk
space that a user might decide to keep. It can translate to very large
size (e.g. 1% = ~10GB on 1TB HDD) or very small (e.g. 1% = ~330MB on 32GB
system like Endless devices). Hence, it makes sense to introduce a config
option to honor specific size as per the user.
Closes: #1616
Approved by: jlebon
We need to include libglnx.h in places where ostree-autocleanups.h is
included, so that we get backports of G_DEFINE_AUTOPTR_CLEANUP_FUNC and
friends.
Closes: #1615
Approved by: jlebon
This reverts commit f1d9196076.
Since libglnx.h does not get installed, it can't be included in
ostree-autocleanups.h, which is included by ostree.h.
Closes: #1615
Approved by: jlebon
Currently the API that allows P2P operations (e.g. pulling an ostree ref
from a LAN or USB source) is hidden behind the configure flag
--enable-experimental-api. This commit makes the API public and makes
that flag essentially a no-op (leaving it in place in case we want to
use it again in the future). The P2P API has been tested over the last
several months and proven to work.
This means that since we're no longer using the "experimental" feature
flag, P2P builds of Flatpak will fail when using versions of OSTree from
this commit onwards, until Flatpak is patched in the near future. If you
want to build Flatpak < 0.11.8 with P2P enabled and link against OSTree
2018.6, you'll have to patch Flatpak. However, since Flatpak won't yet
have a hard dependency on OSTree 2018.6, it needs a new way to determine
if the P2P API in OSTree is available, so this commit adds a "p2p"
feature flag. This way the feature set is more semantically correct than
if we had continued to use the "experimental" feature flag.
In addition to making the P2P API public, this commit makes the P2P unit
tests run by default, removes the f27-experimental CI instance that's no
longer needed, changes a few man pages to reflect the changes, and
updates the bash completion script to accept the new commands and
options.
Closes: #1596
Approved by: cgwalters
This commit includes libglnx.h in ostree-autocleanups.h, so we get the
g_autoptr backports wherever they're needed. Also, remove the "#include
libglnx.h" lines elsewhere that are no longer needed.
Closes: #1596
Approved by: cgwalters
Use the recently introduced architecture for retrying network requests
on transient failure to do the same for delta superblock requests, now
that they’re queued.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #1600
Approved by: jlebon
Just like all the other requests made for delta parts and objects by the
pull code, use a queue for delta superblocks. Currently this doesn’t do
any prioritisation or retries after transient failures, but it could do
in future.
This means that delta superblocks are now subject to the parallel
request limit in the fetcher, which was a problem highlighted here:
https://github.com/ostreedev/ostree/pull/1453#discussion_r168321706.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #1600
Approved by: jlebon
Alexander Larsson
Colin Walters
Matthew Leeds
rfairley
Jonathan Lebon
Giuseppe Scrivano
Umang Jain
Dan Nicholson
Robert McQueen
Felix Krull
Laurent Bonnans
Philip Withnall
Will Thompson
William Manley
Javier Martinez Canillas
Simon McVittie